diff --git a/source/libs/tdb/src/sqlite/btmutex.c b/source/libs/tdb/src/sqlite/btmutex.c new file mode 100644 index 0000000000..275a93ff21 --- /dev/null +++ b/source/libs/tdb/src/sqlite/btmutex.c @@ -0,0 +1,308 @@ +/* +** 2007 August 27 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains code used to implement mutexes on Btree objects. +** This code really belongs in btree.c. But btree.c is getting too +** big and we want to break it down some. This packaged seemed like +** a good breakout. +*/ +#include "btreeInt.h" +#ifndef SQLITE_OMIT_SHARED_CACHE +#if SQLITE_THREADSAFE + +/* +** Obtain the BtShared mutex associated with B-Tree handle p. Also, +** set BtShared.db to the database handle associated with p and the +** p->locked boolean to true. +*/ +static void lockBtreeMutex(Btree *p){ + assert( p->locked==0 ); + assert( sqlite3_mutex_notheld(p->pBt->mutex) ); + assert( sqlite3_mutex_held(p->db->mutex) ); + + sqlite3_mutex_enter(p->pBt->mutex); + p->pBt->db = p->db; + p->locked = 1; +} + +/* +** Release the BtShared mutex associated with B-Tree handle p and +** clear the p->locked boolean. +*/ +static void SQLITE_NOINLINE unlockBtreeMutex(Btree *p){ + BtShared *pBt = p->pBt; + assert( p->locked==1 ); + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( sqlite3_mutex_held(p->db->mutex) ); + assert( p->db==pBt->db ); + + sqlite3_mutex_leave(pBt->mutex); + p->locked = 0; +} + +/* Forward reference */ +static void SQLITE_NOINLINE btreeLockCarefully(Btree *p); + +/* +** Enter a mutex on the given BTree object. +** +** If the object is not sharable, then no mutex is ever required +** and this routine is a no-op. The underlying mutex is non-recursive. +** But we keep a reference count in Btree.wantToLock so the behavior +** of this interface is recursive. +** +** To avoid deadlocks, multiple Btrees are locked in the same order +** by all database connections. The p->pNext is a list of other +** Btrees belonging to the same database connection as the p Btree +** which need to be locked after p. If we cannot get a lock on +** p, then first unlock all of the others on p->pNext, then wait +** for the lock to become available on p, then relock all of the +** subsequent Btrees that desire a lock. +*/ +void sqlite3BtreeEnter(Btree *p){ + /* Some basic sanity checking on the Btree. The list of Btrees + ** connected by pNext and pPrev should be in sorted order by + ** Btree.pBt value. All elements of the list should belong to + ** the same connection. Only shared Btrees are on the list. */ + assert( p->pNext==0 || p->pNext->pBt>p->pBt ); + assert( p->pPrev==0 || p->pPrev->pBtpBt ); + assert( p->pNext==0 || p->pNext->db==p->db ); + assert( p->pPrev==0 || p->pPrev->db==p->db ); + assert( p->sharable || (p->pNext==0 && p->pPrev==0) ); + + /* Check for locking consistency */ + assert( !p->locked || p->wantToLock>0 ); + assert( p->sharable || p->wantToLock==0 ); + + /* We should already hold a lock on the database connection */ + assert( sqlite3_mutex_held(p->db->mutex) ); + + /* Unless the database is sharable and unlocked, then BtShared.db + ** should already be set correctly. */ + assert( (p->locked==0 && p->sharable) || p->pBt->db==p->db ); + + if( !p->sharable ) return; + p->wantToLock++; + if( p->locked ) return; + btreeLockCarefully(p); +} + +/* This is a helper function for sqlite3BtreeLock(). By moving +** complex, but seldom used logic, out of sqlite3BtreeLock() and +** into this routine, we avoid unnecessary stack pointer changes +** and thus help the sqlite3BtreeLock() routine to run much faster +** in the common case. +*/ +static void SQLITE_NOINLINE btreeLockCarefully(Btree *p){ + Btree *pLater; + + /* In most cases, we should be able to acquire the lock we + ** want without having to go through the ascending lock + ** procedure that follows. Just be sure not to block. + */ + if( sqlite3_mutex_try(p->pBt->mutex)==SQLITE_OK ){ + p->pBt->db = p->db; + p->locked = 1; + return; + } + + /* To avoid deadlock, first release all locks with a larger + ** BtShared address. Then acquire our lock. Then reacquire + ** the other BtShared locks that we used to hold in ascending + ** order. + */ + for(pLater=p->pNext; pLater; pLater=pLater->pNext){ + assert( pLater->sharable ); + assert( pLater->pNext==0 || pLater->pNext->pBt>pLater->pBt ); + assert( !pLater->locked || pLater->wantToLock>0 ); + if( pLater->locked ){ + unlockBtreeMutex(pLater); + } + } + lockBtreeMutex(p); + for(pLater=p->pNext; pLater; pLater=pLater->pNext){ + if( pLater->wantToLock ){ + lockBtreeMutex(pLater); + } + } +} + + +/* +** Exit the recursive mutex on a Btree. +*/ +void sqlite3BtreeLeave(Btree *p){ + assert( sqlite3_mutex_held(p->db->mutex) ); + if( p->sharable ){ + assert( p->wantToLock>0 ); + p->wantToLock--; + if( p->wantToLock==0 ){ + unlockBtreeMutex(p); + } + } +} + +#ifndef NDEBUG +/* +** Return true if the BtShared mutex is held on the btree, or if the +** B-Tree is not marked as sharable. +** +** This routine is used only from within assert() statements. +*/ +int sqlite3BtreeHoldsMutex(Btree *p){ + assert( p->sharable==0 || p->locked==0 || p->wantToLock>0 ); + assert( p->sharable==0 || p->locked==0 || p->db==p->pBt->db ); + assert( p->sharable==0 || p->locked==0 || sqlite3_mutex_held(p->pBt->mutex) ); + assert( p->sharable==0 || p->locked==0 || sqlite3_mutex_held(p->db->mutex) ); + + return (p->sharable==0 || p->locked); +} +#endif + + +/* +** Enter the mutex on every Btree associated with a database +** connection. This is needed (for example) prior to parsing +** a statement since we will be comparing table and column names +** against all schemas and we do not want those schemas being +** reset out from under us. +** +** There is a corresponding leave-all procedures. +** +** Enter the mutexes in accending order by BtShared pointer address +** to avoid the possibility of deadlock when two threads with +** two or more btrees in common both try to lock all their btrees +** at the same instant. +*/ +static void SQLITE_NOINLINE btreeEnterAll(sqlite3 *db){ + int i; + int skipOk = 1; + Btree *p; + assert( sqlite3_mutex_held(db->mutex) ); + for(i=0; inDb; i++){ + p = db->aDb[i].pBt; + if( p && p->sharable ){ + sqlite3BtreeEnter(p); + skipOk = 0; + } + } + db->noSharedCache = skipOk; +} +void sqlite3BtreeEnterAll(sqlite3 *db){ + if( db->noSharedCache==0 ) btreeEnterAll(db); +} +static void SQLITE_NOINLINE btreeLeaveAll(sqlite3 *db){ + int i; + Btree *p; + assert( sqlite3_mutex_held(db->mutex) ); + for(i=0; inDb; i++){ + p = db->aDb[i].pBt; + if( p ) sqlite3BtreeLeave(p); + } +} +void sqlite3BtreeLeaveAll(sqlite3 *db){ + if( db->noSharedCache==0 ) btreeLeaveAll(db); +} + +#ifndef NDEBUG +/* +** Return true if the current thread holds the database connection +** mutex and all required BtShared mutexes. +** +** This routine is used inside assert() statements only. +*/ +int sqlite3BtreeHoldsAllMutexes(sqlite3 *db){ + int i; + if( !sqlite3_mutex_held(db->mutex) ){ + return 0; + } + for(i=0; inDb; i++){ + Btree *p; + p = db->aDb[i].pBt; + if( p && p->sharable && + (p->wantToLock==0 || !sqlite3_mutex_held(p->pBt->mutex)) ){ + return 0; + } + } + return 1; +} +#endif /* NDEBUG */ + +#ifndef NDEBUG +/* +** Return true if the correct mutexes are held for accessing the +** db->aDb[iDb].pSchema structure. The mutexes required for schema +** access are: +** +** (1) The mutex on db +** (2) if iDb!=1, then the mutex on db->aDb[iDb].pBt. +** +** If pSchema is not NULL, then iDb is computed from pSchema and +** db using sqlite3SchemaToIndex(). +*/ +int sqlite3SchemaMutexHeld(sqlite3 *db, int iDb, Schema *pSchema){ + Btree *p; + assert( db!=0 ); + if( pSchema ) iDb = sqlite3SchemaToIndex(db, pSchema); + assert( iDb>=0 && iDbnDb ); + if( !sqlite3_mutex_held(db->mutex) ) return 0; + if( iDb==1 ) return 1; + p = db->aDb[iDb].pBt; + assert( p!=0 ); + return p->sharable==0 || p->locked==1; +} +#endif /* NDEBUG */ + +#else /* SQLITE_THREADSAFE>0 above. SQLITE_THREADSAFE==0 below */ +/* +** The following are special cases for mutex enter routines for use +** in single threaded applications that use shared cache. Except for +** these two routines, all mutex operations are no-ops in that case and +** are null #defines in btree.h. +** +** If shared cache is disabled, then all btree mutex routines, including +** the ones below, are no-ops and are null #defines in btree.h. +*/ + +void sqlite3BtreeEnter(Btree *p){ + p->pBt->db = p->db; +} +void sqlite3BtreeEnterAll(sqlite3 *db){ + int i; + for(i=0; inDb; i++){ + Btree *p = db->aDb[i].pBt; + if( p ){ + p->pBt->db = p->db; + } + } +} +#endif /* if SQLITE_THREADSAFE */ + +#ifndef SQLITE_OMIT_INCRBLOB +/* +** Enter a mutex on a Btree given a cursor owned by that Btree. +** +** These entry points are used by incremental I/O only. Enter() is required +** any time OMIT_SHARED_CACHE is not defined, regardless of whether or not +** the build is threadsafe. Leave() is only required by threadsafe builds. +*/ +void sqlite3BtreeEnterCursor(BtCursor *pCur){ + sqlite3BtreeEnter(pCur->pBtree); +} +# if SQLITE_THREADSAFE +void sqlite3BtreeLeaveCursor(BtCursor *pCur){ + sqlite3BtreeLeave(pCur->pBtree); +} +# endif +#endif /* ifndef SQLITE_OMIT_INCRBLOB */ + +#endif /* ifndef SQLITE_OMIT_SHARED_CACHE */ diff --git a/source/libs/tdb/src/sqlite/btree.c b/source/libs/tdb/src/sqlite/btree.c new file mode 100644 index 0000000000..32696f53e6 --- /dev/null +++ b/source/libs/tdb/src/sqlite/btree.c @@ -0,0 +1,10869 @@ +/* +** 2004 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements an external (disk-based) database using BTrees. +** See the header comment on "btreeInt.h" for additional information. +** Including a description of file format and an overview of operation. +*/ +#include "btreeInt.h" + +/* +** The header string that appears at the beginning of every +** SQLite database. +*/ +static const char zMagicHeader[] = SQLITE_FILE_HEADER; + +/* +** Set this global variable to 1 to enable tracing using the TRACE +** macro. +*/ +#if 0 +int sqlite3BtreeTrace=1; /* True to enable tracing */ +# define TRACE(X) if(sqlite3BtreeTrace){printf X;fflush(stdout);} +#else +# define TRACE(X) +#endif + +/* +** Extract a 2-byte big-endian integer from an array of unsigned bytes. +** But if the value is zero, make it 65536. +** +** This routine is used to extract the "offset to cell content area" value +** from the header of a btree page. If the page size is 65536 and the page +** is empty, the offset should be 65536, but the 2-byte value stores zero. +** This routine makes the necessary adjustment to 65536. +*/ +#define get2byteNotZero(X) (((((int)get2byte(X))-1)&0xffff)+1) + +/* +** Values passed as the 5th argument to allocateBtreePage() +*/ +#define BTALLOC_ANY 0 /* Allocate any page */ +#define BTALLOC_EXACT 1 /* Allocate exact page if possible */ +#define BTALLOC_LE 2 /* Allocate any page <= the parameter */ + +/* +** Macro IfNotOmitAV(x) returns (x) if SQLITE_OMIT_AUTOVACUUM is not +** defined, or 0 if it is. For example: +** +** bIncrVacuum = IfNotOmitAV(pBtShared->incrVacuum); +*/ +#ifndef SQLITE_OMIT_AUTOVACUUM +#define IfNotOmitAV(expr) (expr) +#else +#define IfNotOmitAV(expr) 0 +#endif + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** A list of BtShared objects that are eligible for participation +** in shared cache. This variable has file scope during normal builds, +** but the test harness needs to access it so we make it global for +** test builds. +** +** Access to this variable is protected by SQLITE_MUTEX_STATIC_MAIN. +*/ +#ifdef SQLITE_TEST +BtShared *SQLITE_WSD sqlite3SharedCacheList = 0; +#else +static BtShared *SQLITE_WSD sqlite3SharedCacheList = 0; +#endif +#endif /* SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Enable or disable the shared pager and schema features. +** +** This routine has no effect on existing database connections. +** The shared cache setting effects only future calls to +** sqlite3_open(), sqlite3_open16(), or sqlite3_open_v2(). +*/ +int sqlite3_enable_shared_cache(int enable){ + sqlite3GlobalConfig.sharedCacheEnabled = enable; + return SQLITE_OK; +} +#endif + + + +#ifdef SQLITE_OMIT_SHARED_CACHE + /* + ** The functions querySharedCacheTableLock(), setSharedCacheTableLock(), + ** and clearAllSharedCacheTableLocks() + ** manipulate entries in the BtShared.pLock linked list used to store + ** shared-cache table level locks. If the library is compiled with the + ** shared-cache feature disabled, then there is only ever one user + ** of each BtShared structure and so this locking is not necessary. + ** So define the lock related functions as no-ops. + */ + #define querySharedCacheTableLock(a,b,c) SQLITE_OK + #define setSharedCacheTableLock(a,b,c) SQLITE_OK + #define clearAllSharedCacheTableLocks(a) + #define downgradeAllSharedCacheTableLocks(a) + #define hasSharedCacheTableLock(a,b,c,d) 1 + #define hasReadConflicts(a, b) 0 +#endif + +#ifdef SQLITE_DEBUG +/* +** Return and reset the seek counter for a Btree object. +*/ +sqlite3_uint64 sqlite3BtreeSeekCount(Btree *pBt){ + u64 n = pBt->nSeek; + pBt->nSeek = 0; + return n; +} +#endif + +/* +** Implementation of the SQLITE_CORRUPT_PAGE() macro. Takes a single +** (MemPage*) as an argument. The (MemPage*) must not be NULL. +** +** If SQLITE_DEBUG is not defined, then this macro is equivalent to +** SQLITE_CORRUPT_BKPT. Or, if SQLITE_DEBUG is set, then the log message +** normally produced as a side-effect of SQLITE_CORRUPT_BKPT is augmented +** with the page number and filename associated with the (MemPage*). +*/ +#ifdef SQLITE_DEBUG +int corruptPageError(int lineno, MemPage *p){ + char *zMsg; + sqlite3BeginBenignMalloc(); + zMsg = sqlite3_mprintf("database corruption page %d of %s", + (int)p->pgno, sqlite3PagerFilename(p->pBt->pPager, 0) + ); + sqlite3EndBenignMalloc(); + if( zMsg ){ + sqlite3ReportError(SQLITE_CORRUPT, lineno, zMsg); + } + sqlite3_free(zMsg); + return SQLITE_CORRUPT_BKPT; +} +# define SQLITE_CORRUPT_PAGE(pMemPage) corruptPageError(__LINE__, pMemPage) +#else +# define SQLITE_CORRUPT_PAGE(pMemPage) SQLITE_CORRUPT_PGNO(pMemPage->pgno) +#endif + +#ifndef SQLITE_OMIT_SHARED_CACHE + +#ifdef SQLITE_DEBUG +/* +**** This function is only used as part of an assert() statement. *** +** +** Check to see if pBtree holds the required locks to read or write to the +** table with root page iRoot. Return 1 if it does and 0 if not. +** +** For example, when writing to a table with root-page iRoot via +** Btree connection pBtree: +** +** assert( hasSharedCacheTableLock(pBtree, iRoot, 0, WRITE_LOCK) ); +** +** When writing to an index that resides in a sharable database, the +** caller should have first obtained a lock specifying the root page of +** the corresponding table. This makes things a bit more complicated, +** as this module treats each table as a separate structure. To determine +** the table corresponding to the index being written, this +** function has to search through the database schema. +** +** Instead of a lock on the table/index rooted at page iRoot, the caller may +** hold a write-lock on the schema table (root page 1). This is also +** acceptable. +*/ +static int hasSharedCacheTableLock( + Btree *pBtree, /* Handle that must hold lock */ + Pgno iRoot, /* Root page of b-tree */ + int isIndex, /* True if iRoot is the root of an index b-tree */ + int eLockType /* Required lock type (READ_LOCK or WRITE_LOCK) */ +){ + Schema *pSchema = (Schema *)pBtree->pBt->pSchema; + Pgno iTab = 0; + BtLock *pLock; + + /* If this database is not shareable, or if the client is reading + ** and has the read-uncommitted flag set, then no lock is required. + ** Return true immediately. + */ + if( (pBtree->sharable==0) + || (eLockType==READ_LOCK && (pBtree->db->flags & SQLITE_ReadUncommit)) + ){ + return 1; + } + + /* If the client is reading or writing an index and the schema is + ** not loaded, then it is too difficult to actually check to see if + ** the correct locks are held. So do not bother - just return true. + ** This case does not come up very often anyhow. + */ + if( isIndex && (!pSchema || (pSchema->schemaFlags&DB_SchemaLoaded)==0) ){ + return 1; + } + + /* Figure out the root-page that the lock should be held on. For table + ** b-trees, this is just the root page of the b-tree being read or + ** written. For index b-trees, it is the root page of the associated + ** table. */ + if( isIndex ){ + HashElem *p; + int bSeen = 0; + for(p=sqliteHashFirst(&pSchema->idxHash); p; p=sqliteHashNext(p)){ + Index *pIdx = (Index *)sqliteHashData(p); + if( pIdx->tnum==(int)iRoot ){ + if( bSeen ){ + /* Two or more indexes share the same root page. There must + ** be imposter tables. So just return true. The assert is not + ** useful in that case. */ + return 1; + } + iTab = pIdx->pTable->tnum; + bSeen = 1; + } + } + }else{ + iTab = iRoot; + } + + /* Search for the required lock. Either a write-lock on root-page iTab, a + ** write-lock on the schema table, or (if the client is reading) a + ** read-lock on iTab will suffice. Return 1 if any of these are found. */ + for(pLock=pBtree->pBt->pLock; pLock; pLock=pLock->pNext){ + if( pLock->pBtree==pBtree + && (pLock->iTable==iTab || (pLock->eLock==WRITE_LOCK && pLock->iTable==1)) + && pLock->eLock>=eLockType + ){ + return 1; + } + } + + /* Failed to find the required lock. */ + return 0; +} +#endif /* SQLITE_DEBUG */ + +#ifdef SQLITE_DEBUG +/* +**** This function may be used as part of assert() statements only. **** +** +** Return true if it would be illegal for pBtree to write into the +** table or index rooted at iRoot because other shared connections are +** simultaneously reading that same table or index. +** +** It is illegal for pBtree to write if some other Btree object that +** shares the same BtShared object is currently reading or writing +** the iRoot table. Except, if the other Btree object has the +** read-uncommitted flag set, then it is OK for the other object to +** have a read cursor. +** +** For example, before writing to any part of the table or index +** rooted at page iRoot, one should call: +** +** assert( !hasReadConflicts(pBtree, iRoot) ); +*/ +static int hasReadConflicts(Btree *pBtree, Pgno iRoot){ + BtCursor *p; + for(p=pBtree->pBt->pCursor; p; p=p->pNext){ + if( p->pgnoRoot==iRoot + && p->pBtree!=pBtree + && 0==(p->pBtree->db->flags & SQLITE_ReadUncommit) + ){ + return 1; + } + } + return 0; +} +#endif /* #ifdef SQLITE_DEBUG */ + +/* +** Query to see if Btree handle p may obtain a lock of type eLock +** (READ_LOCK or WRITE_LOCK) on the table with root-page iTab. Return +** SQLITE_OK if the lock may be obtained (by calling +** setSharedCacheTableLock()), or SQLITE_LOCKED if not. +*/ +static int querySharedCacheTableLock(Btree *p, Pgno iTab, u8 eLock){ + BtShared *pBt = p->pBt; + BtLock *pIter; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( eLock==READ_LOCK || eLock==WRITE_LOCK ); + assert( p->db!=0 ); + assert( !(p->db->flags&SQLITE_ReadUncommit)||eLock==WRITE_LOCK||iTab==1 ); + + /* If requesting a write-lock, then the Btree must have an open write + ** transaction on this file. And, obviously, for this to be so there + ** must be an open write transaction on the file itself. + */ + assert( eLock==READ_LOCK || (p==pBt->pWriter && p->inTrans==TRANS_WRITE) ); + assert( eLock==READ_LOCK || pBt->inTransaction==TRANS_WRITE ); + + /* This routine is a no-op if the shared-cache is not enabled */ + if( !p->sharable ){ + return SQLITE_OK; + } + + /* If some other connection is holding an exclusive lock, the + ** requested lock may not be obtained. + */ + if( pBt->pWriter!=p && (pBt->btsFlags & BTS_EXCLUSIVE)!=0 ){ + sqlite3ConnectionBlocked(p->db, pBt->pWriter->db); + return SQLITE_LOCKED_SHAREDCACHE; + } + + for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){ + /* The condition (pIter->eLock!=eLock) in the following if(...) + ** statement is a simplification of: + ** + ** (eLock==WRITE_LOCK || pIter->eLock==WRITE_LOCK) + ** + ** since we know that if eLock==WRITE_LOCK, then no other connection + ** may hold a WRITE_LOCK on any table in this file (since there can + ** only be a single writer). + */ + assert( pIter->eLock==READ_LOCK || pIter->eLock==WRITE_LOCK ); + assert( eLock==READ_LOCK || pIter->pBtree==p || pIter->eLock==READ_LOCK); + if( pIter->pBtree!=p && pIter->iTable==iTab && pIter->eLock!=eLock ){ + sqlite3ConnectionBlocked(p->db, pIter->pBtree->db); + if( eLock==WRITE_LOCK ){ + assert( p==pBt->pWriter ); + pBt->btsFlags |= BTS_PENDING; + } + return SQLITE_LOCKED_SHAREDCACHE; + } + } + return SQLITE_OK; +} +#endif /* !SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Add a lock on the table with root-page iTable to the shared-btree used +** by Btree handle p. Parameter eLock must be either READ_LOCK or +** WRITE_LOCK. +** +** This function assumes the following: +** +** (a) The specified Btree object p is connected to a sharable +** database (one with the BtShared.sharable flag set), and +** +** (b) No other Btree objects hold a lock that conflicts +** with the requested lock (i.e. querySharedCacheTableLock() has +** already been called and returned SQLITE_OK). +** +** SQLITE_OK is returned if the lock is added successfully. SQLITE_NOMEM +** is returned if a malloc attempt fails. +*/ +static int setSharedCacheTableLock(Btree *p, Pgno iTable, u8 eLock){ + BtShared *pBt = p->pBt; + BtLock *pLock = 0; + BtLock *pIter; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( eLock==READ_LOCK || eLock==WRITE_LOCK ); + assert( p->db!=0 ); + + /* A connection with the read-uncommitted flag set will never try to + ** obtain a read-lock using this function. The only read-lock obtained + ** by a connection in read-uncommitted mode is on the sqlite_schema + ** table, and that lock is obtained in BtreeBeginTrans(). */ + assert( 0==(p->db->flags&SQLITE_ReadUncommit) || eLock==WRITE_LOCK ); + + /* This function should only be called on a sharable b-tree after it + ** has been determined that no other b-tree holds a conflicting lock. */ + assert( p->sharable ); + assert( SQLITE_OK==querySharedCacheTableLock(p, iTable, eLock) ); + + /* First search the list for an existing lock on this table. */ + for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){ + if( pIter->iTable==iTable && pIter->pBtree==p ){ + pLock = pIter; + break; + } + } + + /* If the above search did not find a BtLock struct associating Btree p + ** with table iTable, allocate one and link it into the list. + */ + if( !pLock ){ + pLock = (BtLock *)sqlite3MallocZero(sizeof(BtLock)); + if( !pLock ){ + return SQLITE_NOMEM_BKPT; + } + pLock->iTable = iTable; + pLock->pBtree = p; + pLock->pNext = pBt->pLock; + pBt->pLock = pLock; + } + + /* Set the BtLock.eLock variable to the maximum of the current lock + ** and the requested lock. This means if a write-lock was already held + ** and a read-lock requested, we don't incorrectly downgrade the lock. + */ + assert( WRITE_LOCK>READ_LOCK ); + if( eLock>pLock->eLock ){ + pLock->eLock = eLock; + } + + return SQLITE_OK; +} +#endif /* !SQLITE_OMIT_SHARED_CACHE */ + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Release all the table locks (locks obtained via calls to +** the setSharedCacheTableLock() procedure) held by Btree object p. +** +** This function assumes that Btree p has an open read or write +** transaction. If it does not, then the BTS_PENDING flag +** may be incorrectly cleared. +*/ +static void clearAllSharedCacheTableLocks(Btree *p){ + BtShared *pBt = p->pBt; + BtLock **ppIter = &pBt->pLock; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( p->sharable || 0==*ppIter ); + assert( p->inTrans>0 ); + + while( *ppIter ){ + BtLock *pLock = *ppIter; + assert( (pBt->btsFlags & BTS_EXCLUSIVE)==0 || pBt->pWriter==pLock->pBtree ); + assert( pLock->pBtree->inTrans>=pLock->eLock ); + if( pLock->pBtree==p ){ + *ppIter = pLock->pNext; + assert( pLock->iTable!=1 || pLock==&p->lock ); + if( pLock->iTable!=1 ){ + sqlite3_free(pLock); + } + }else{ + ppIter = &pLock->pNext; + } + } + + assert( (pBt->btsFlags & BTS_PENDING)==0 || pBt->pWriter ); + if( pBt->pWriter==p ){ + pBt->pWriter = 0; + pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING); + }else if( pBt->nTransaction==2 ){ + /* This function is called when Btree p is concluding its + ** transaction. If there currently exists a writer, and p is not + ** that writer, then the number of locks held by connections other + ** than the writer must be about to drop to zero. In this case + ** set the BTS_PENDING flag to 0. + ** + ** If there is not currently a writer, then BTS_PENDING must + ** be zero already. So this next line is harmless in that case. + */ + pBt->btsFlags &= ~BTS_PENDING; + } +} + +/* +** This function changes all write-locks held by Btree p into read-locks. +*/ +static void downgradeAllSharedCacheTableLocks(Btree *p){ + BtShared *pBt = p->pBt; + if( pBt->pWriter==p ){ + BtLock *pLock; + pBt->pWriter = 0; + pBt->btsFlags &= ~(BTS_EXCLUSIVE|BTS_PENDING); + for(pLock=pBt->pLock; pLock; pLock=pLock->pNext){ + assert( pLock->eLock==READ_LOCK || pLock->pBtree==p ); + pLock->eLock = READ_LOCK; + } + } +} + +#endif /* SQLITE_OMIT_SHARED_CACHE */ + +static void releasePage(MemPage *pPage); /* Forward reference */ +static void releasePageOne(MemPage *pPage); /* Forward reference */ +static void releasePageNotNull(MemPage *pPage); /* Forward reference */ + +/* +***** This routine is used inside of assert() only **** +** +** Verify that the cursor holds the mutex on its BtShared +*/ +#ifdef SQLITE_DEBUG +static int cursorHoldsMutex(BtCursor *p){ + return sqlite3_mutex_held(p->pBt->mutex); +} + +/* Verify that the cursor and the BtShared agree about what is the current +** database connetion. This is important in shared-cache mode. If the database +** connection pointers get out-of-sync, it is possible for routines like +** btreeInitPage() to reference an stale connection pointer that references a +** a connection that has already closed. This routine is used inside assert() +** statements only and for the purpose of double-checking that the btree code +** does keep the database connection pointers up-to-date. +*/ +static int cursorOwnsBtShared(BtCursor *p){ + assert( cursorHoldsMutex(p) ); + return (p->pBtree->db==p->pBt->db); +} +#endif + +/* +** Invalidate the overflow cache of the cursor passed as the first argument. +** on the shared btree structure pBt. +*/ +#define invalidateOverflowCache(pCur) (pCur->curFlags &= ~BTCF_ValidOvfl) + +/* +** Invalidate the overflow page-list cache for all cursors opened +** on the shared btree structure pBt. +*/ +static void invalidateAllOverflowCache(BtShared *pBt){ + BtCursor *p; + assert( sqlite3_mutex_held(pBt->mutex) ); + for(p=pBt->pCursor; p; p=p->pNext){ + invalidateOverflowCache(p); + } +} + +#ifndef SQLITE_OMIT_INCRBLOB +/* +** This function is called before modifying the contents of a table +** to invalidate any incrblob cursors that are open on the +** row or one of the rows being modified. +** +** If argument isClearTable is true, then the entire contents of the +** table is about to be deleted. In this case invalidate all incrblob +** cursors open on any row within the table with root-page pgnoRoot. +** +** Otherwise, if argument isClearTable is false, then the row with +** rowid iRow is being replaced or deleted. In this case invalidate +** only those incrblob cursors open on that specific row. +*/ +static void invalidateIncrblobCursors( + Btree *pBtree, /* The database file to check */ + Pgno pgnoRoot, /* The table that might be changing */ + i64 iRow, /* The rowid that might be changing */ + int isClearTable /* True if all rows are being deleted */ +){ + BtCursor *p; + assert( pBtree->hasIncrblobCur ); + assert( sqlite3BtreeHoldsMutex(pBtree) ); + pBtree->hasIncrblobCur = 0; + for(p=pBtree->pBt->pCursor; p; p=p->pNext){ + if( (p->curFlags & BTCF_Incrblob)!=0 ){ + pBtree->hasIncrblobCur = 1; + if( p->pgnoRoot==pgnoRoot && (isClearTable || p->info.nKey==iRow) ){ + p->eState = CURSOR_INVALID; + } + } + } +} + +#else + /* Stub function when INCRBLOB is omitted */ + #define invalidateIncrblobCursors(w,x,y,z) +#endif /* SQLITE_OMIT_INCRBLOB */ + +/* +** Set bit pgno of the BtShared.pHasContent bitvec. This is called +** when a page that previously contained data becomes a free-list leaf +** page. +** +** The BtShared.pHasContent bitvec exists to work around an obscure +** bug caused by the interaction of two useful IO optimizations surrounding +** free-list leaf pages: +** +** 1) When all data is deleted from a page and the page becomes +** a free-list leaf page, the page is not written to the database +** (as free-list leaf pages contain no meaningful data). Sometimes +** such a page is not even journalled (as it will not be modified, +** why bother journalling it?). +** +** 2) When a free-list leaf page is reused, its content is not read +** from the database or written to the journal file (why should it +** be, if it is not at all meaningful?). +** +** By themselves, these optimizations work fine and provide a handy +** performance boost to bulk delete or insert operations. However, if +** a page is moved to the free-list and then reused within the same +** transaction, a problem comes up. If the page is not journalled when +** it is moved to the free-list and it is also not journalled when it +** is extracted from the free-list and reused, then the original data +** may be lost. In the event of a rollback, it may not be possible +** to restore the database to its original configuration. +** +** The solution is the BtShared.pHasContent bitvec. Whenever a page is +** moved to become a free-list leaf page, the corresponding bit is +** set in the bitvec. Whenever a leaf page is extracted from the free-list, +** optimization 2 above is omitted if the corresponding bit is already +** set in BtShared.pHasContent. The contents of the bitvec are cleared +** at the end of every transaction. +*/ +static int btreeSetHasContent(BtShared *pBt, Pgno pgno){ + int rc = SQLITE_OK; + if( !pBt->pHasContent ){ + assert( pgno<=pBt->nPage ); + pBt->pHasContent = sqlite3BitvecCreate(pBt->nPage); + if( !pBt->pHasContent ){ + rc = SQLITE_NOMEM_BKPT; + } + } + if( rc==SQLITE_OK && pgno<=sqlite3BitvecSize(pBt->pHasContent) ){ + rc = sqlite3BitvecSet(pBt->pHasContent, pgno); + } + return rc; +} + +/* +** Query the BtShared.pHasContent vector. +** +** This function is called when a free-list leaf page is removed from the +** free-list for reuse. It returns false if it is safe to retrieve the +** page from the pager layer with the 'no-content' flag set. True otherwise. +*/ +static int btreeGetHasContent(BtShared *pBt, Pgno pgno){ + Bitvec *p = pBt->pHasContent; + return p && (pgno>sqlite3BitvecSize(p) || sqlite3BitvecTestNotNull(p, pgno)); +} + +/* +** Clear (destroy) the BtShared.pHasContent bitvec. This should be +** invoked at the conclusion of each write-transaction. +*/ +static void btreeClearHasContent(BtShared *pBt){ + sqlite3BitvecDestroy(pBt->pHasContent); + pBt->pHasContent = 0; +} + +/* +** Release all of the apPage[] pages for a cursor. +*/ +static void btreeReleaseAllCursorPages(BtCursor *pCur){ + int i; + if( pCur->iPage>=0 ){ + for(i=0; iiPage; i++){ + releasePageNotNull(pCur->apPage[i]); + } + releasePageNotNull(pCur->pPage); + pCur->iPage = -1; + } +} + +/* +** The cursor passed as the only argument must point to a valid entry +** when this function is called (i.e. have eState==CURSOR_VALID). This +** function saves the current cursor key in variables pCur->nKey and +** pCur->pKey. SQLITE_OK is returned if successful or an SQLite error +** code otherwise. +** +** If the cursor is open on an intkey table, then the integer key +** (the rowid) is stored in pCur->nKey and pCur->pKey is left set to +** NULL. If the cursor is open on a non-intkey table, then pCur->pKey is +** set to point to a malloced buffer pCur->nKey bytes in size containing +** the key. +*/ +static int saveCursorKey(BtCursor *pCur){ + int rc = SQLITE_OK; + assert( CURSOR_VALID==pCur->eState ); + assert( 0==pCur->pKey ); + assert( cursorHoldsMutex(pCur) ); + + if( pCur->curIntKey ){ + /* Only the rowid is required for a table btree */ + pCur->nKey = sqlite3BtreeIntegerKey(pCur); + }else{ + /* For an index btree, save the complete key content. It is possible + ** that the current key is corrupt. In that case, it is possible that + ** the sqlite3VdbeRecordUnpack() function may overread the buffer by + ** up to the size of 1 varint plus 1 8-byte value when the cursor + ** position is restored. Hence the 17 bytes of padding allocated + ** below. */ + void *pKey; + pCur->nKey = sqlite3BtreePayloadSize(pCur); + pKey = sqlite3Malloc( pCur->nKey + 9 + 8 ); + if( pKey ){ + rc = sqlite3BtreePayload(pCur, 0, (int)pCur->nKey, pKey); + if( rc==SQLITE_OK ){ + memset(((u8*)pKey)+pCur->nKey, 0, 9+8); + pCur->pKey = pKey; + }else{ + sqlite3_free(pKey); + } + }else{ + rc = SQLITE_NOMEM_BKPT; + } + } + assert( !pCur->curIntKey || !pCur->pKey ); + return rc; +} + +/* +** Save the current cursor position in the variables BtCursor.nKey +** and BtCursor.pKey. The cursor's state is set to CURSOR_REQUIRESEEK. +** +** The caller must ensure that the cursor is valid (has eState==CURSOR_VALID) +** prior to calling this routine. +*/ +static int saveCursorPosition(BtCursor *pCur){ + int rc; + + assert( CURSOR_VALID==pCur->eState || CURSOR_SKIPNEXT==pCur->eState ); + assert( 0==pCur->pKey ); + assert( cursorHoldsMutex(pCur) ); + + if( pCur->curFlags & BTCF_Pinned ){ + return SQLITE_CONSTRAINT_PINNED; + } + if( pCur->eState==CURSOR_SKIPNEXT ){ + pCur->eState = CURSOR_VALID; + }else{ + pCur->skipNext = 0; + } + + rc = saveCursorKey(pCur); + if( rc==SQLITE_OK ){ + btreeReleaseAllCursorPages(pCur); + pCur->eState = CURSOR_REQUIRESEEK; + } + + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl|BTCF_AtLast); + return rc; +} + +/* Forward reference */ +static int SQLITE_NOINLINE saveCursorsOnList(BtCursor*,Pgno,BtCursor*); + +/* +** Save the positions of all cursors (except pExcept) that are open on +** the table with root-page iRoot. "Saving the cursor position" means that +** the location in the btree is remembered in such a way that it can be +** moved back to the same spot after the btree has been modified. This +** routine is called just before cursor pExcept is used to modify the +** table, for example in BtreeDelete() or BtreeInsert(). +** +** If there are two or more cursors on the same btree, then all such +** cursors should have their BTCF_Multiple flag set. The btreeCursor() +** routine enforces that rule. This routine only needs to be called in +** the uncommon case when pExpect has the BTCF_Multiple flag set. +** +** If pExpect!=NULL and if no other cursors are found on the same root-page, +** then the BTCF_Multiple flag on pExpect is cleared, to avoid another +** pointless call to this routine. +** +** Implementation note: This routine merely checks to see if any cursors +** need to be saved. It calls out to saveCursorsOnList() in the (unusual) +** event that cursors are in need to being saved. +*/ +static int saveAllCursors(BtShared *pBt, Pgno iRoot, BtCursor *pExcept){ + BtCursor *p; + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pExcept==0 || pExcept->pBt==pBt ); + for(p=pBt->pCursor; p; p=p->pNext){ + if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) ) break; + } + if( p ) return saveCursorsOnList(p, iRoot, pExcept); + if( pExcept ) pExcept->curFlags &= ~BTCF_Multiple; + return SQLITE_OK; +} + +/* This helper routine to saveAllCursors does the actual work of saving +** the cursors if and when a cursor is found that actually requires saving. +** The common case is that no cursors need to be saved, so this routine is +** broken out from its caller to avoid unnecessary stack pointer movement. +*/ +static int SQLITE_NOINLINE saveCursorsOnList( + BtCursor *p, /* The first cursor that needs saving */ + Pgno iRoot, /* Only save cursor with this iRoot. Save all if zero */ + BtCursor *pExcept /* Do not save this cursor */ +){ + do{ + if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) ){ + if( p->eState==CURSOR_VALID || p->eState==CURSOR_SKIPNEXT ){ + int rc = saveCursorPosition(p); + if( SQLITE_OK!=rc ){ + return rc; + } + }else{ + testcase( p->iPage>=0 ); + btreeReleaseAllCursorPages(p); + } + } + p = p->pNext; + }while( p ); + return SQLITE_OK; +} + +/* +** Clear the current cursor position. +*/ +void sqlite3BtreeClearCursor(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + sqlite3_free(pCur->pKey); + pCur->pKey = 0; + pCur->eState = CURSOR_INVALID; +} + +/* +** In this version of BtreeMoveto, pKey is a packed index record +** such as is generated by the OP_MakeRecord opcode. Unpack the +** record and then call BtreeMovetoUnpacked() to do the work. +*/ +static int btreeMoveto( + BtCursor *pCur, /* Cursor open on the btree to be searched */ + const void *pKey, /* Packed key if the btree is an index */ + i64 nKey, /* Integer key for tables. Size of pKey for indices */ + int bias, /* Bias search to the high end */ + int *pRes /* Write search results here */ +){ + int rc; /* Status code */ + UnpackedRecord *pIdxKey; /* Unpacked index key */ + + if( pKey ){ + KeyInfo *pKeyInfo = pCur->pKeyInfo; + assert( nKey==(i64)(int)nKey ); + pIdxKey = sqlite3VdbeAllocUnpackedRecord(pKeyInfo); + if( pIdxKey==0 ) return SQLITE_NOMEM_BKPT; + sqlite3VdbeRecordUnpack(pKeyInfo, (int)nKey, pKey, pIdxKey); + if( pIdxKey->nField==0 || pIdxKey->nField>pKeyInfo->nAllField ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + rc = sqlite3BtreeIndexMoveto(pCur, pIdxKey, pRes); + } + sqlite3DbFree(pCur->pKeyInfo->db, pIdxKey); + }else{ + pIdxKey = 0; + rc = sqlite3BtreeTableMoveto(pCur, nKey, bias, pRes); + } + return rc; +} + +/* +** Restore the cursor to the position it was in (or as close to as possible) +** when saveCursorPosition() was called. Note that this call deletes the +** saved position info stored by saveCursorPosition(), so there can be +** at most one effective restoreCursorPosition() call after each +** saveCursorPosition(). +*/ +static int btreeRestoreCursorPosition(BtCursor *pCur){ + int rc; + int skipNext = 0; + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState>=CURSOR_REQUIRESEEK ); + if( pCur->eState==CURSOR_FAULT ){ + return pCur->skipNext; + } + pCur->eState = CURSOR_INVALID; + if( sqlite3FaultSim(410) ){ + rc = SQLITE_IOERR; + }else{ + rc = btreeMoveto(pCur, pCur->pKey, pCur->nKey, 0, &skipNext); + } + if( rc==SQLITE_OK ){ + sqlite3_free(pCur->pKey); + pCur->pKey = 0; + assert( pCur->eState==CURSOR_VALID || pCur->eState==CURSOR_INVALID ); + if( skipNext ) pCur->skipNext = skipNext; + if( pCur->skipNext && pCur->eState==CURSOR_VALID ){ + pCur->eState = CURSOR_SKIPNEXT; + } + } + return rc; +} + +#define restoreCursorPosition(p) \ + (p->eState>=CURSOR_REQUIRESEEK ? \ + btreeRestoreCursorPosition(p) : \ + SQLITE_OK) + +/* +** Determine whether or not a cursor has moved from the position where +** it was last placed, or has been invalidated for any other reason. +** Cursors can move when the row they are pointing at is deleted out +** from under them, for example. Cursor might also move if a btree +** is rebalanced. +** +** Calling this routine with a NULL cursor pointer returns false. +** +** Use the separate sqlite3BtreeCursorRestore() routine to restore a cursor +** back to where it ought to be if this routine returns true. +*/ +int sqlite3BtreeCursorHasMoved(BtCursor *pCur){ + assert( EIGHT_BYTE_ALIGNMENT(pCur) + || pCur==sqlite3BtreeFakeValidCursor() ); + assert( offsetof(BtCursor, eState)==0 ); + assert( sizeof(pCur->eState)==1 ); + return CURSOR_VALID != *(u8*)pCur; +} + +/* +** Return a pointer to a fake BtCursor object that will always answer +** false to the sqlite3BtreeCursorHasMoved() routine above. The fake +** cursor returned must not be used with any other Btree interface. +*/ +BtCursor *sqlite3BtreeFakeValidCursor(void){ + static u8 fakeCursor = CURSOR_VALID; + assert( offsetof(BtCursor, eState)==0 ); + return (BtCursor*)&fakeCursor; +} + +/* +** This routine restores a cursor back to its original position after it +** has been moved by some outside activity (such as a btree rebalance or +** a row having been deleted out from under the cursor). +** +** On success, the *pDifferentRow parameter is false if the cursor is left +** pointing at exactly the same row. *pDifferntRow is the row the cursor +** was pointing to has been deleted, forcing the cursor to point to some +** nearby row. +** +** This routine should only be called for a cursor that just returned +** TRUE from sqlite3BtreeCursorHasMoved(). +*/ +int sqlite3BtreeCursorRestore(BtCursor *pCur, int *pDifferentRow){ + int rc; + + assert( pCur!=0 ); + assert( pCur->eState!=CURSOR_VALID ); + rc = restoreCursorPosition(pCur); + if( rc ){ + *pDifferentRow = 1; + return rc; + } + if( pCur->eState!=CURSOR_VALID ){ + *pDifferentRow = 1; + }else{ + *pDifferentRow = 0; + } + return SQLITE_OK; +} + +#ifdef SQLITE_ENABLE_CURSOR_HINTS +/* +** Provide hints to the cursor. The particular hint given (and the type +** and number of the varargs parameters) is determined by the eHintType +** parameter. See the definitions of the BTREE_HINT_* macros for details. +*/ +void sqlite3BtreeCursorHint(BtCursor *pCur, int eHintType, ...){ + /* Used only by system that substitute their own storage engine */ +} +#endif + +/* +** Provide flag hints to the cursor. +*/ +void sqlite3BtreeCursorHintFlags(BtCursor *pCur, unsigned x){ + assert( x==BTREE_SEEK_EQ || x==BTREE_BULKLOAD || x==0 ); + pCur->hints = x; +} + + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** Given a page number of a regular database page, return the page +** number for the pointer-map page that contains the entry for the +** input page number. +** +** Return 0 (not a valid page) for pgno==1 since there is +** no pointer map associated with page 1. The integrity_check logic +** requires that ptrmapPageno(*,1)!=1. +*/ +static Pgno ptrmapPageno(BtShared *pBt, Pgno pgno){ + int nPagesPerMapPage; + Pgno iPtrMap, ret; + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pgno<2 ) return 0; + nPagesPerMapPage = (pBt->usableSize/5)+1; + iPtrMap = (pgno-2)/nPagesPerMapPage; + ret = (iPtrMap*nPagesPerMapPage) + 2; + if( ret==PENDING_BYTE_PAGE(pBt) ){ + ret++; + } + return ret; +} + +/* +** Write an entry into the pointer map. +** +** This routine updates the pointer map entry for page number 'key' +** so that it maps to type 'eType' and parent page number 'pgno'. +** +** If *pRC is initially non-zero (non-SQLITE_OK) then this routine is +** a no-op. If an error occurs, the appropriate error code is written +** into *pRC. +*/ +static void ptrmapPut(BtShared *pBt, Pgno key, u8 eType, Pgno parent, int *pRC){ + DbPage *pDbPage; /* The pointer map page */ + u8 *pPtrmap; /* The pointer map data */ + Pgno iPtrmap; /* The pointer map page number */ + int offset; /* Offset in pointer map page */ + int rc; /* Return code from subfunctions */ + + if( *pRC ) return; + + assert( sqlite3_mutex_held(pBt->mutex) ); + /* The super-journal page number must never be used as a pointer map page */ + assert( 0==PTRMAP_ISPAGE(pBt, PENDING_BYTE_PAGE(pBt)) ); + + assert( pBt->autoVacuum ); + if( key==0 ){ + *pRC = SQLITE_CORRUPT_BKPT; + return; + } + iPtrmap = PTRMAP_PAGENO(pBt, key); + rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage, 0); + if( rc!=SQLITE_OK ){ + *pRC = rc; + return; + } + if( ((char*)sqlite3PagerGetExtra(pDbPage))[0]!=0 ){ + /* The first byte of the extra data is the MemPage.isInit byte. + ** If that byte is set, it means this page is also being used + ** as a btree page. */ + *pRC = SQLITE_CORRUPT_BKPT; + goto ptrmap_exit; + } + offset = PTRMAP_PTROFFSET(iPtrmap, key); + if( offset<0 ){ + *pRC = SQLITE_CORRUPT_BKPT; + goto ptrmap_exit; + } + assert( offset <= (int)pBt->usableSize-5 ); + pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage); + + if( eType!=pPtrmap[offset] || get4byte(&pPtrmap[offset+1])!=parent ){ + TRACE(("PTRMAP_UPDATE: %d->(%d,%d)\n", key, eType, parent)); + *pRC= rc = sqlite3PagerWrite(pDbPage); + if( rc==SQLITE_OK ){ + pPtrmap[offset] = eType; + put4byte(&pPtrmap[offset+1], parent); + } + } + +ptrmap_exit: + sqlite3PagerUnref(pDbPage); +} + +/* +** Read an entry from the pointer map. +** +** This routine retrieves the pointer map entry for page 'key', writing +** the type and parent page number to *pEType and *pPgno respectively. +** An error code is returned if something goes wrong, otherwise SQLITE_OK. +*/ +static int ptrmapGet(BtShared *pBt, Pgno key, u8 *pEType, Pgno *pPgno){ + DbPage *pDbPage; /* The pointer map page */ + int iPtrmap; /* Pointer map page index */ + u8 *pPtrmap; /* Pointer map page data */ + int offset; /* Offset of entry in pointer map */ + int rc; + + assert( sqlite3_mutex_held(pBt->mutex) ); + + iPtrmap = PTRMAP_PAGENO(pBt, key); + rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage, 0); + if( rc!=0 ){ + return rc; + } + pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage); + + offset = PTRMAP_PTROFFSET(iPtrmap, key); + if( offset<0 ){ + sqlite3PagerUnref(pDbPage); + return SQLITE_CORRUPT_BKPT; + } + assert( offset <= (int)pBt->usableSize-5 ); + assert( pEType!=0 ); + *pEType = pPtrmap[offset]; + if( pPgno ) *pPgno = get4byte(&pPtrmap[offset+1]); + + sqlite3PagerUnref(pDbPage); + if( *pEType<1 || *pEType>5 ) return SQLITE_CORRUPT_PGNO(iPtrmap); + return SQLITE_OK; +} + +#else /* if defined SQLITE_OMIT_AUTOVACUUM */ + #define ptrmapPut(w,x,y,z,rc) + #define ptrmapGet(w,x,y,z) SQLITE_OK + #define ptrmapPutOvflPtr(x, y, z, rc) +#endif + +/* +** Given a btree page and a cell index (0 means the first cell on +** the page, 1 means the second cell, and so forth) return a pointer +** to the cell content. +** +** findCellPastPtr() does the same except it skips past the initial +** 4-byte child pointer found on interior pages, if there is one. +** +** This routine works only for pages that do not contain overflow cells. +*/ +#define findCell(P,I) \ + ((P)->aData + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)]))) +#define findCellPastPtr(P,I) \ + ((P)->aDataOfst + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)]))) + + +/* +** This is common tail processing for btreeParseCellPtr() and +** btreeParseCellPtrIndex() for the case when the cell does not fit entirely +** on a single B-tree page. Make necessary adjustments to the CellInfo +** structure. +*/ +static SQLITE_NOINLINE void btreeParseCellAdjustSizeForOverflow( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + /* If the payload will not fit completely on the local page, we have + ** to decide how much to store locally and how much to spill onto + ** overflow pages. The strategy is to minimize the amount of unused + ** space on overflow pages while keeping the amount of local storage + ** in between minLocal and maxLocal. + ** + ** Warning: changing the way overflow payload is distributed in any + ** way will result in an incompatible file format. + */ + int minLocal; /* Minimum amount of payload held locally */ + int maxLocal; /* Maximum amount of payload held locally */ + int surplus; /* Overflow payload available for local storage */ + + minLocal = pPage->minLocal; + maxLocal = pPage->maxLocal; + surplus = minLocal + (pInfo->nPayload - minLocal)%(pPage->pBt->usableSize-4); + testcase( surplus==maxLocal ); + testcase( surplus==maxLocal+1 ); + if( surplus <= maxLocal ){ + pInfo->nLocal = (u16)surplus; + }else{ + pInfo->nLocal = (u16)minLocal; + } + pInfo->nSize = (u16)(&pInfo->pPayload[pInfo->nLocal] - pCell) + 4; +} + +/* +** Given a record with nPayload bytes of payload stored within btree +** page pPage, return the number of bytes of payload stored locally. +*/ +static int btreePayloadToLocal(MemPage *pPage, i64 nPayload){ + int maxLocal; /* Maximum amount of payload held locally */ + maxLocal = pPage->maxLocal; + if( nPayload<=maxLocal ){ + return nPayload; + }else{ + int minLocal; /* Minimum amount of payload held locally */ + int surplus; /* Overflow payload available for local storage */ + minLocal = pPage->minLocal; + surplus = minLocal + (nPayload - minLocal)%(pPage->pBt->usableSize-4); + return ( surplus <= maxLocal ) ? surplus : minLocal; + } +} + +/* +** The following routines are implementations of the MemPage.xParseCell() +** method. +** +** Parse a cell content block and fill in the CellInfo structure. +** +** btreeParseCellPtr() => table btree leaf nodes +** btreeParseCellNoPayload() => table btree internal nodes +** btreeParseCellPtrIndex() => index btree nodes +** +** There is also a wrapper function btreeParseCell() that works for +** all MemPage types and that references the cell by index rather than +** by pointer. +*/ +static void btreeParseCellPtrNoPayload( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->leaf==0 ); + assert( pPage->childPtrSize==4 ); +#ifndef SQLITE_DEBUG + UNUSED_PARAMETER(pPage); +#endif + pInfo->nSize = 4 + getVarint(&pCell[4], (u64*)&pInfo->nKey); + pInfo->nPayload = 0; + pInfo->nLocal = 0; + pInfo->pPayload = 0; + return; +} +static void btreeParseCellPtr( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + u8 *pIter; /* For scanning through pCell */ + u32 nPayload; /* Number of bytes of cell payload */ + u64 iKey; /* Extracted Key value */ + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->leaf==0 || pPage->leaf==1 ); + assert( pPage->intKeyLeaf ); + assert( pPage->childPtrSize==0 ); + pIter = pCell; + + /* The next block of code is equivalent to: + ** + ** pIter += getVarint32(pIter, nPayload); + ** + ** The code is inlined to avoid a function call. + */ + nPayload = *pIter; + if( nPayload>=0x80 ){ + u8 *pEnd = &pIter[8]; + nPayload &= 0x7f; + do{ + nPayload = (nPayload<<7) | (*++pIter & 0x7f); + }while( (*pIter)>=0x80 && pIternKey); + ** + ** The code is inlined to avoid a function call. + */ + iKey = *pIter; + if( iKey>=0x80 ){ + u8 *pEnd = &pIter[7]; + iKey &= 0x7f; + while(1){ + iKey = (iKey<<7) | (*++pIter & 0x7f); + if( (*pIter)<0x80 ) break; + if( pIter>=pEnd ){ + iKey = (iKey<<8) | *++pIter; + break; + } + } + } + pIter++; + + pInfo->nKey = *(i64*)&iKey; + pInfo->nPayload = nPayload; + pInfo->pPayload = pIter; + testcase( nPayload==pPage->maxLocal ); + testcase( nPayload==pPage->maxLocal+1 ); + if( nPayload<=pPage->maxLocal ){ + /* This is the (easy) common case where the entire payload fits + ** on the local page. No overflow is required. + */ + pInfo->nSize = nPayload + (u16)(pIter - pCell); + if( pInfo->nSize<4 ) pInfo->nSize = 4; + pInfo->nLocal = (u16)nPayload; + }else{ + btreeParseCellAdjustSizeForOverflow(pPage, pCell, pInfo); + } +} +static void btreeParseCellPtrIndex( + MemPage *pPage, /* Page containing the cell */ + u8 *pCell, /* Pointer to the cell text. */ + CellInfo *pInfo /* Fill in this structure */ +){ + u8 *pIter; /* For scanning through pCell */ + u32 nPayload; /* Number of bytes of cell payload */ + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->leaf==0 || pPage->leaf==1 ); + assert( pPage->intKeyLeaf==0 ); + pIter = pCell + pPage->childPtrSize; + nPayload = *pIter; + if( nPayload>=0x80 ){ + u8 *pEnd = &pIter[8]; + nPayload &= 0x7f; + do{ + nPayload = (nPayload<<7) | (*++pIter & 0x7f); + }while( *(pIter)>=0x80 && pIternKey = nPayload; + pInfo->nPayload = nPayload; + pInfo->pPayload = pIter; + testcase( nPayload==pPage->maxLocal ); + testcase( nPayload==pPage->maxLocal+1 ); + if( nPayload<=pPage->maxLocal ){ + /* This is the (easy) common case where the entire payload fits + ** on the local page. No overflow is required. + */ + pInfo->nSize = nPayload + (u16)(pIter - pCell); + if( pInfo->nSize<4 ) pInfo->nSize = 4; + pInfo->nLocal = (u16)nPayload; + }else{ + btreeParseCellAdjustSizeForOverflow(pPage, pCell, pInfo); + } +} +static void btreeParseCell( + MemPage *pPage, /* Page containing the cell */ + int iCell, /* The cell index. First cell is 0 */ + CellInfo *pInfo /* Fill in this structure */ +){ + pPage->xParseCell(pPage, findCell(pPage, iCell), pInfo); +} + +/* +** The following routines are implementations of the MemPage.xCellSize +** method. +** +** Compute the total number of bytes that a Cell needs in the cell +** data area of the btree-page. The return number includes the cell +** data header and the local payload, but not any overflow page or +** the space used by the cell pointer. +** +** cellSizePtrNoPayload() => table internal nodes +** cellSizePtr() => all index nodes & table leaf nodes +*/ +static u16 cellSizePtr(MemPage *pPage, u8 *pCell){ + u8 *pIter = pCell + pPage->childPtrSize; /* For looping over bytes of pCell */ + u8 *pEnd; /* End mark for a varint */ + u32 nSize; /* Size value to return */ + +#ifdef SQLITE_DEBUG + /* The value returned by this function should always be the same as + ** the (CellInfo.nSize) value found by doing a full parse of the + ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of + ** this function verifies that this invariant is not violated. */ + CellInfo debuginfo; + pPage->xParseCell(pPage, pCell, &debuginfo); +#endif + + nSize = *pIter; + if( nSize>=0x80 ){ + pEnd = &pIter[8]; + nSize &= 0x7f; + do{ + nSize = (nSize<<7) | (*++pIter & 0x7f); + }while( *(pIter)>=0x80 && pIterintKey ){ + /* pIter now points at the 64-bit integer key value, a variable length + ** integer. The following block moves pIter to point at the first byte + ** past the end of the key value. */ + pEnd = &pIter[9]; + while( (*pIter++)&0x80 && pItermaxLocal ); + testcase( nSize==pPage->maxLocal+1 ); + if( nSize<=pPage->maxLocal ){ + nSize += (u32)(pIter - pCell); + if( nSize<4 ) nSize = 4; + }else{ + int minLocal = pPage->minLocal; + nSize = minLocal + (nSize - minLocal) % (pPage->pBt->usableSize - 4); + testcase( nSize==pPage->maxLocal ); + testcase( nSize==pPage->maxLocal+1 ); + if( nSize>pPage->maxLocal ){ + nSize = minLocal; + } + nSize += 4 + (u16)(pIter - pCell); + } + assert( nSize==debuginfo.nSize || CORRUPT_DB ); + return (u16)nSize; +} +static u16 cellSizePtrNoPayload(MemPage *pPage, u8 *pCell){ + u8 *pIter = pCell + 4; /* For looping over bytes of pCell */ + u8 *pEnd; /* End mark for a varint */ + +#ifdef SQLITE_DEBUG + /* The value returned by this function should always be the same as + ** the (CellInfo.nSize) value found by doing a full parse of the + ** cell. If SQLITE_DEBUG is defined, an assert() at the bottom of + ** this function verifies that this invariant is not violated. */ + CellInfo debuginfo; + pPage->xParseCell(pPage, pCell, &debuginfo); +#else + UNUSED_PARAMETER(pPage); +#endif + + assert( pPage->childPtrSize==4 ); + pEnd = pIter + 9; + while( (*pIter++)&0x80 && pIterxCellSize(pPage, findCell(pPage, iCell)); +} +#endif + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** The cell pCell is currently part of page pSrc but will ultimately be part +** of pPage. (pSrc and pPager are often the same.) If pCell contains a +** pointer to an overflow page, insert an entry into the pointer-map for +** the overflow page that will be valid after pCell has been moved to pPage. +*/ +static void ptrmapPutOvflPtr(MemPage *pPage, MemPage *pSrc, u8 *pCell,int *pRC){ + CellInfo info; + if( *pRC ) return; + assert( pCell!=0 ); + pPage->xParseCell(pPage, pCell, &info); + if( info.nLocalaDataEnd, pCell, pCell+info.nLocal) ){ + testcase( pSrc!=pPage ); + *pRC = SQLITE_CORRUPT_BKPT; + return; + } + ovfl = get4byte(&pCell[info.nSize-4]); + ptrmapPut(pPage->pBt, ovfl, PTRMAP_OVERFLOW1, pPage->pgno, pRC); + } +} +#endif + + +/* +** Defragment the page given. This routine reorganizes cells within the +** page so that there are no free-blocks on the free-block list. +** +** Parameter nMaxFrag is the maximum amount of fragmented space that may be +** present in the page after this routine returns. +** +** EVIDENCE-OF: R-44582-60138 SQLite may from time to time reorganize a +** b-tree page so that there are no freeblocks or fragment bytes, all +** unused bytes are contained in the unallocated space region, and all +** cells are packed tightly at the end of the page. +*/ +static int defragmentPage(MemPage *pPage, int nMaxFrag){ + int i; /* Loop counter */ + int pc; /* Address of the i-th cell */ + int hdr; /* Offset to the page header */ + int size; /* Size of a cell */ + int usableSize; /* Number of usable bytes on a page */ + int cellOffset; /* Offset to the cell pointer array */ + int cbrk; /* Offset to the cell content area */ + int nCell; /* Number of cells on the page */ + unsigned char *data; /* The page data */ + unsigned char *temp; /* Temp area for cell content */ + unsigned char *src; /* Source of content */ + int iCellFirst; /* First allowable cell index */ + int iCellLast; /* Last possible cell index */ + int iCellStart; /* First cell offset in input */ + + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( pPage->pBt!=0 ); + assert( pPage->pBt->usableSize <= SQLITE_MAX_PAGE_SIZE ); + assert( pPage->nOverflow==0 ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + temp = 0; + src = data = pPage->aData; + hdr = pPage->hdrOffset; + cellOffset = pPage->cellOffset; + nCell = pPage->nCell; + assert( nCell==get2byte(&data[hdr+3]) || CORRUPT_DB ); + iCellFirst = cellOffset + 2*nCell; + usableSize = pPage->pBt->usableSize; + + /* This block handles pages with two or fewer free blocks and nMaxFrag + ** or fewer fragmented bytes. In this case it is faster to move the + ** two (or one) blocks of cells using memmove() and add the required + ** offsets to each pointer in the cell-pointer array than it is to + ** reconstruct the entire page. */ + if( (int)data[hdr+7]<=nMaxFrag ){ + int iFree = get2byte(&data[hdr+1]); + if( iFree>usableSize-4 ) return SQLITE_CORRUPT_PAGE(pPage); + if( iFree ){ + int iFree2 = get2byte(&data[iFree]); + if( iFree2>usableSize-4 ) return SQLITE_CORRUPT_PAGE(pPage); + if( 0==iFree2 || (data[iFree2]==0 && data[iFree2+1]==0) ){ + u8 *pEnd = &data[cellOffset + nCell*2]; + u8 *pAddr; + int sz2 = 0; + int sz = get2byte(&data[iFree+2]); + int top = get2byte(&data[hdr+5]); + if( top>=iFree ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + if( iFree2 ){ + if( iFree+sz>iFree2 ) return SQLITE_CORRUPT_PAGE(pPage); + sz2 = get2byte(&data[iFree2+2]); + if( iFree2+sz2 > usableSize ) return SQLITE_CORRUPT_PAGE(pPage); + memmove(&data[iFree+sz+sz2], &data[iFree+sz], iFree2-(iFree+sz)); + sz += sz2; + }else if( NEVER(iFree+sz>usableSize) ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + + cbrk = top+sz; + assert( cbrk+(iFree-top) <= usableSize ); + memmove(&data[cbrk], &data[top], iFree-top); + for(pAddr=&data[cellOffset]; pAddriCellLast ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( pc>=iCellStart && pc<=iCellLast ); + size = pPage->xCellSize(pPage, &src[pc]); + cbrk -= size; + if( cbrkusableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( cbrk+size<=usableSize && cbrk>=iCellStart ); + testcase( cbrk+size==usableSize ); + testcase( pc+size==usableSize ); + put2byte(pAddr, cbrk); + if( temp==0 ){ + if( cbrk==pc ) continue; + temp = sqlite3PagerTempSpace(pPage->pBt->pPager); + memcpy(&temp[iCellStart], &data[iCellStart], usableSize - iCellStart); + src = temp; + } + memcpy(&data[cbrk], &src[pc], size); + } + data[hdr+7] = 0; + + defragment_out: + assert( pPage->nFree>=0 ); + if( data[hdr+7]+cbrk-iCellFirst!=pPage->nFree ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( cbrk>=iCellFirst ); + put2byte(&data[hdr+5], cbrk); + data[hdr+1] = 0; + data[hdr+2] = 0; + memset(&data[iCellFirst], 0, cbrk-iCellFirst); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + return SQLITE_OK; +} + +/* +** Search the free-list on page pPg for space to store a cell nByte bytes in +** size. If one can be found, return a pointer to the space and remove it +** from the free-list. +** +** If no suitable space can be found on the free-list, return NULL. +** +** This function may detect corruption within pPg. If corruption is +** detected then *pRc is set to SQLITE_CORRUPT and NULL is returned. +** +** Slots on the free list that are between 1 and 3 bytes larger than nByte +** will be ignored if adding the extra space to the fragmentation count +** causes the fragmentation count to exceed 60. +*/ +static u8 *pageFindSlot(MemPage *pPg, int nByte, int *pRc){ + const int hdr = pPg->hdrOffset; /* Offset to page header */ + u8 * const aData = pPg->aData; /* Page data */ + int iAddr = hdr + 1; /* Address of ptr to pc */ + int pc = get2byte(&aData[iAddr]); /* Address of a free slot */ + int x; /* Excess size of the slot */ + int maxPC = pPg->pBt->usableSize - nByte; /* Max address for a usable slot */ + int size; /* Size of the free slot */ + + assert( pc>0 ); + while( pc<=maxPC ){ + /* EVIDENCE-OF: R-22710-53328 The third and fourth bytes of each + ** freeblock form a big-endian integer which is the size of the freeblock + ** in bytes, including the 4-byte header. */ + size = get2byte(&aData[pc+2]); + if( (x = size - nByte)>=0 ){ + testcase( x==4 ); + testcase( x==3 ); + if( x<4 ){ + /* EVIDENCE-OF: R-11498-58022 In a well-formed b-tree page, the total + ** number of bytes in fragments may not exceed 60. */ + if( aData[hdr+7]>57 ) return 0; + + /* Remove the slot from the free-list. Update the number of + ** fragmented bytes within the page. */ + memcpy(&aData[iAddr], &aData[pc], 2); + aData[hdr+7] += (u8)x; + }else if( x+pc > maxPC ){ + /* This slot extends off the end of the usable part of the page */ + *pRc = SQLITE_CORRUPT_PAGE(pPg); + return 0; + }else{ + /* The slot remains on the free-list. Reduce its size to account + ** for the portion used by the new allocation. */ + put2byte(&aData[pc+2], x); + } + return &aData[pc + x]; + } + iAddr = pc; + pc = get2byte(&aData[pc]); + if( pc<=iAddr+size ){ + if( pc ){ + /* The next slot in the chain is not past the end of the current slot */ + *pRc = SQLITE_CORRUPT_PAGE(pPg); + } + return 0; + } + } + if( pc>maxPC+nByte-4 ){ + /* The free slot chain extends off the end of the page */ + *pRc = SQLITE_CORRUPT_PAGE(pPg); + } + return 0; +} + +/* +** Allocate nByte bytes of space from within the B-Tree page passed +** as the first argument. Write into *pIdx the index into pPage->aData[] +** of the first byte of allocated space. Return either SQLITE_OK or +** an error code (usually SQLITE_CORRUPT). +** +** The caller guarantees that there is sufficient space to make the +** allocation. This routine might need to defragment in order to bring +** all the space together, however. This routine will avoid using +** the first two bytes past the cell pointer area since presumably this +** allocation is being made in order to insert a new cell, so we will +** also end up needing a new cell pointer. +*/ +static int allocateSpace(MemPage *pPage, int nByte, int *pIdx){ + const int hdr = pPage->hdrOffset; /* Local cache of pPage->hdrOffset */ + u8 * const data = pPage->aData; /* Local cache of pPage->aData */ + int top; /* First byte of cell content area */ + int rc = SQLITE_OK; /* Integer return code */ + int gap; /* First byte of gap between cell pointers and cell content */ + + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( pPage->pBt ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( nByte>=0 ); /* Minimum cell size is 4 */ + assert( pPage->nFree>=nByte ); + assert( pPage->nOverflow==0 ); + assert( nByte < (int)(pPage->pBt->usableSize-8) ); + + assert( pPage->cellOffset == hdr + 12 - 4*pPage->leaf ); + gap = pPage->cellOffset + 2*pPage->nCell; + assert( gap<=65536 ); + /* EVIDENCE-OF: R-29356-02391 If the database uses a 65536-byte page size + ** and the reserved space is zero (the usual value for reserved space) + ** then the cell content offset of an empty page wants to be 65536. + ** However, that integer is too large to be stored in a 2-byte unsigned + ** integer, so a value of 0 is used in its place. */ + top = get2byte(&data[hdr+5]); + assert( top<=(int)pPage->pBt->usableSize ); /* by btreeComputeFreeSpace() */ + if( gap>top ){ + if( top==0 && pPage->pBt->usableSize==65536 ){ + top = 65536; + }else{ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + + /* If there is enough space between gap and top for one more cell pointer, + ** and if the freelist is not empty, then search the + ** freelist looking for a slot big enough to satisfy the request. + */ + testcase( gap+2==top ); + testcase( gap+1==top ); + testcase( gap==top ); + if( (data[hdr+2] || data[hdr+1]) && gap+2<=top ){ + u8 *pSpace = pageFindSlot(pPage, nByte, &rc); + if( pSpace ){ + int g2; + assert( pSpace+nByte<=data+pPage->pBt->usableSize ); + *pIdx = g2 = (int)(pSpace-data); + if( g2<=gap ){ + return SQLITE_CORRUPT_PAGE(pPage); + }else{ + return SQLITE_OK; + } + }else if( rc ){ + return rc; + } + } + + /* The request could not be fulfilled using a freelist slot. Check + ** to see if defragmentation is necessary. + */ + testcase( gap+2+nByte==top ); + if( gap+2+nByte>top ){ + assert( pPage->nCell>0 || CORRUPT_DB ); + assert( pPage->nFree>=0 ); + rc = defragmentPage(pPage, MIN(4, pPage->nFree - (2+nByte))); + if( rc ) return rc; + top = get2byteNotZero(&data[hdr+5]); + assert( gap+2+nByte<=top ); + } + + + /* Allocate memory from the gap in between the cell pointer array + ** and the cell content area. The btreeComputeFreeSpace() call has already + ** validated the freelist. Given that the freelist is valid, there + ** is no way that the allocation can extend off the end of the page. + ** The assert() below verifies the previous sentence. + */ + top -= nByte; + put2byte(&data[hdr+5], top); + assert( top+nByte <= (int)pPage->pBt->usableSize ); + *pIdx = top; + return SQLITE_OK; +} + +/* +** Return a section of the pPage->aData to the freelist. +** The first byte of the new free block is pPage->aData[iStart] +** and the size of the block is iSize bytes. +** +** Adjacent freeblocks are coalesced. +** +** Even though the freeblock list was checked by btreeComputeFreeSpace(), +** that routine will not detect overlap between cells or freeblocks. Nor +** does it detect cells or freeblocks that encrouch into the reserved bytes +** at the end of the page. So do additional corruption checks inside this +** routine and return SQLITE_CORRUPT if any problems are found. +*/ +static int freeSpace(MemPage *pPage, u16 iStart, u16 iSize){ + u16 iPtr; /* Address of ptr to next freeblock */ + u16 iFreeBlk; /* Address of the next freeblock */ + u8 hdr; /* Page header size. 0 or 100 */ + u8 nFrag = 0; /* Reduction in fragmentation */ + u16 iOrigSize = iSize; /* Original value of iSize */ + u16 x; /* Offset to cell content area */ + u32 iEnd = iStart + iSize; /* First byte past the iStart buffer */ + unsigned char *data = pPage->aData; /* Page content */ + + assert( pPage->pBt!=0 ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( CORRUPT_DB || iStart>=pPage->hdrOffset+6+pPage->childPtrSize ); + assert( CORRUPT_DB || iEnd <= pPage->pBt->usableSize ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( iSize>=4 ); /* Minimum cell size is 4 */ + assert( iStart<=pPage->pBt->usableSize-4 ); + + /* The list of freeblocks must be in ascending order. Find the + ** spot on the list where iStart should be inserted. + */ + hdr = pPage->hdrOffset; + iPtr = hdr + 1; + if( data[iPtr+1]==0 && data[iPtr]==0 ){ + iFreeBlk = 0; /* Shortcut for the case when the freelist is empty */ + }else{ + while( (iFreeBlk = get2byte(&data[iPtr]))pPage->pBt->usableSize-4 ){ /* TH3: corrupt081.100 */ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( iFreeBlk>iPtr || iFreeBlk==0 ); + + /* At this point: + ** iFreeBlk: First freeblock after iStart, or zero if none + ** iPtr: The address of a pointer to iFreeBlk + ** + ** Check to see if iFreeBlk should be coalesced onto the end of iStart. + */ + if( iFreeBlk && iEnd+3>=iFreeBlk ){ + nFrag = iFreeBlk - iEnd; + if( iEnd>iFreeBlk ) return SQLITE_CORRUPT_PAGE(pPage); + iEnd = iFreeBlk + get2byte(&data[iFreeBlk+2]); + if( iEnd > pPage->pBt->usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + iSize = iEnd - iStart; + iFreeBlk = get2byte(&data[iFreeBlk]); + } + + /* If iPtr is another freeblock (that is, if iPtr is not the freelist + ** pointer in the page header) then check to see if iStart should be + ** coalesced onto the end of iPtr. + */ + if( iPtr>hdr+1 ){ + int iPtrEnd = iPtr + get2byte(&data[iPtr+2]); + if( iPtrEnd+3>=iStart ){ + if( iPtrEnd>iStart ) return SQLITE_CORRUPT_PAGE(pPage); + nFrag += iStart - iPtrEnd; + iSize = iEnd - iPtr; + iStart = iPtr; + } + } + if( nFrag>data[hdr+7] ) return SQLITE_CORRUPT_PAGE(pPage); + data[hdr+7] -= nFrag; + } + x = get2byte(&data[hdr+5]); + if( iStart<=x ){ + /* The new freeblock is at the beginning of the cell content area, + ** so just extend the cell content area rather than create another + ** freelist entry */ + if( iStartpBt->btsFlags & BTS_FAST_SECURE ){ + /* Overwrite deleted information with zeros when the secure_delete + ** option is enabled */ + memset(&data[iStart], 0, iSize); + } + put2byte(&data[iStart], iFreeBlk); + put2byte(&data[iStart+2], iSize); + pPage->nFree += iOrigSize; + return SQLITE_OK; +} + +/* +** Decode the flags byte (the first byte of the header) for a page +** and initialize fields of the MemPage structure accordingly. +** +** Only the following combinations are supported. Anything different +** indicates a corrupt database files: +** +** PTF_ZERODATA +** PTF_ZERODATA | PTF_LEAF +** PTF_LEAFDATA | PTF_INTKEY +** PTF_LEAFDATA | PTF_INTKEY | PTF_LEAF +*/ +static int decodeFlags(MemPage *pPage, int flagByte){ + BtShared *pBt; /* A copy of pPage->pBt */ + + assert( pPage->hdrOffset==(pPage->pgno==1 ? 100 : 0) ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + pPage->leaf = (u8)(flagByte>>3); assert( PTF_LEAF == 1<<3 ); + flagByte &= ~PTF_LEAF; + pPage->childPtrSize = 4-4*pPage->leaf; + pPage->xCellSize = cellSizePtr; + pBt = pPage->pBt; + if( flagByte==(PTF_LEAFDATA | PTF_INTKEY) ){ + /* EVIDENCE-OF: R-07291-35328 A value of 5 (0x05) means the page is an + ** interior table b-tree page. */ + assert( (PTF_LEAFDATA|PTF_INTKEY)==5 ); + /* EVIDENCE-OF: R-26900-09176 A value of 13 (0x0d) means the page is a + ** leaf table b-tree page. */ + assert( (PTF_LEAFDATA|PTF_INTKEY|PTF_LEAF)==13 ); + pPage->intKey = 1; + if( pPage->leaf ){ + pPage->intKeyLeaf = 1; + pPage->xParseCell = btreeParseCellPtr; + }else{ + pPage->intKeyLeaf = 0; + pPage->xCellSize = cellSizePtrNoPayload; + pPage->xParseCell = btreeParseCellPtrNoPayload; + } + pPage->maxLocal = pBt->maxLeaf; + pPage->minLocal = pBt->minLeaf; + }else if( flagByte==PTF_ZERODATA ){ + /* EVIDENCE-OF: R-43316-37308 A value of 2 (0x02) means the page is an + ** interior index b-tree page. */ + assert( (PTF_ZERODATA)==2 ); + /* EVIDENCE-OF: R-59615-42828 A value of 10 (0x0a) means the page is a + ** leaf index b-tree page. */ + assert( (PTF_ZERODATA|PTF_LEAF)==10 ); + pPage->intKey = 0; + pPage->intKeyLeaf = 0; + pPage->xParseCell = btreeParseCellPtrIndex; + pPage->maxLocal = pBt->maxLocal; + pPage->minLocal = pBt->minLocal; + }else{ + /* EVIDENCE-OF: R-47608-56469 Any other value for the b-tree page type is + ** an error. */ + return SQLITE_CORRUPT_PAGE(pPage); + } + pPage->max1bytePayload = pBt->max1bytePayload; + return SQLITE_OK; +} + +/* +** Compute the amount of freespace on the page. In other words, fill +** in the pPage->nFree field. +*/ +static int btreeComputeFreeSpace(MemPage *pPage){ + int pc; /* Address of a freeblock within pPage->aData[] */ + u8 hdr; /* Offset to beginning of page header */ + u8 *data; /* Equal to pPage->aData */ + int usableSize; /* Amount of usable space on each page */ + int nFree; /* Number of unused bytes on the page */ + int top; /* First byte of the cell content area */ + int iCellFirst; /* First allowable cell or freeblock offset */ + int iCellLast; /* Last possible cell or freeblock offset */ + + assert( pPage->pBt!=0 ); + assert( pPage->pBt->db!=0 ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) ); + assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) ); + assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) ); + assert( pPage->isInit==1 ); + assert( pPage->nFree<0 ); + + usableSize = pPage->pBt->usableSize; + hdr = pPage->hdrOffset; + data = pPage->aData; + /* EVIDENCE-OF: R-58015-48175 The two-byte integer at offset 5 designates + ** the start of the cell content area. A zero value for this integer is + ** interpreted as 65536. */ + top = get2byteNotZero(&data[hdr+5]); + iCellFirst = hdr + 8 + pPage->childPtrSize + 2*pPage->nCell; + iCellLast = usableSize - 4; + + /* Compute the total free space on the page + ** EVIDENCE-OF: R-23588-34450 The two-byte integer at offset 1 gives the + ** start of the first freeblock on the page, or is zero if there are no + ** freeblocks. */ + pc = get2byte(&data[hdr+1]); + nFree = data[hdr+7] + top; /* Init nFree to non-freeblock free space */ + if( pc>0 ){ + u32 next, size; + if( pciCellLast ){ + /* Freeblock off the end of the page */ + return SQLITE_CORRUPT_PAGE(pPage); + } + next = get2byte(&data[pc]); + size = get2byte(&data[pc+2]); + nFree = nFree + size; + if( next<=pc+size+3 ) break; + pc = next; + } + if( next>0 ){ + /* Freeblock not in ascending order */ + return SQLITE_CORRUPT_PAGE(pPage); + } + if( pc+size>(unsigned int)usableSize ){ + /* Last freeblock extends past page end */ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + + /* At this point, nFree contains the sum of the offset to the start + ** of the cell-content area plus the number of free bytes within + ** the cell-content area. If this is greater than the usable-size + ** of the page, then the page must be corrupted. This check also + ** serves to verify that the offset to the start of the cell-content + ** area, according to the page header, lies within the page. + */ + if( nFree>usableSize || nFreenFree = (u16)(nFree - iCellFirst); + return SQLITE_OK; +} + +/* +** Do additional sanity check after btreeInitPage() if +** PRAGMA cell_size_check=ON +*/ +static SQLITE_NOINLINE int btreeCellSizeCheck(MemPage *pPage){ + int iCellFirst; /* First allowable cell or freeblock offset */ + int iCellLast; /* Last possible cell or freeblock offset */ + int i; /* Index into the cell pointer array */ + int sz; /* Size of a cell */ + int pc; /* Address of a freeblock within pPage->aData[] */ + u8 *data; /* Equal to pPage->aData */ + int usableSize; /* Maximum usable space on the page */ + int cellOffset; /* Start of cell content area */ + + iCellFirst = pPage->cellOffset + 2*pPage->nCell; + usableSize = pPage->pBt->usableSize; + iCellLast = usableSize - 4; + data = pPage->aData; + cellOffset = pPage->cellOffset; + if( !pPage->leaf ) iCellLast--; + for(i=0; inCell; i++){ + pc = get2byteAligned(&data[cellOffset+i*2]); + testcase( pc==iCellFirst ); + testcase( pc==iCellLast ); + if( pciCellLast ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + sz = pPage->xCellSize(pPage, &data[pc]); + testcase( pc+sz==usableSize ); + if( pc+sz>usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + return SQLITE_OK; +} + +/* +** Initialize the auxiliary information for a disk block. +** +** Return SQLITE_OK on success. If we see that the page does +** not contain a well-formed database page, then return +** SQLITE_CORRUPT. Note that a return of SQLITE_OK does not +** guarantee that the page is well-formed. It only shows that +** we failed to detect any corruption. +*/ +static int btreeInitPage(MemPage *pPage){ + u8 *data; /* Equal to pPage->aData */ + BtShared *pBt; /* The main btree structure */ + + assert( pPage->pBt!=0 ); + assert( pPage->pBt->db!=0 ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) ); + assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) ); + assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) ); + assert( pPage->isInit==0 ); + + pBt = pPage->pBt; + data = pPage->aData + pPage->hdrOffset; + /* EVIDENCE-OF: R-28594-02890 The one-byte flag at offset 0 indicating + ** the b-tree page type. */ + if( decodeFlags(pPage, data[0]) ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( pBt->pageSize>=512 && pBt->pageSize<=65536 ); + pPage->maskPage = (u16)(pBt->pageSize - 1); + pPage->nOverflow = 0; + pPage->cellOffset = pPage->hdrOffset + 8 + pPage->childPtrSize; + pPage->aCellIdx = data + pPage->childPtrSize + 8; + pPage->aDataEnd = pPage->aData + pBt->usableSize; + pPage->aDataOfst = pPage->aData + pPage->childPtrSize; + /* EVIDENCE-OF: R-37002-32774 The two-byte integer at offset 3 gives the + ** number of cells on the page. */ + pPage->nCell = get2byte(&data[3]); + if( pPage->nCell>MX_CELL(pBt) ){ + /* To many cells for a single page. The page must be corrupt */ + return SQLITE_CORRUPT_PAGE(pPage); + } + testcase( pPage->nCell==MX_CELL(pBt) ); + /* EVIDENCE-OF: R-24089-57979 If a page contains no cells (which is only + ** possible for a root page of a table that contains no rows) then the + ** offset to the cell content area will equal the page size minus the + ** bytes of reserved space. */ + assert( pPage->nCell>0 + || get2byteNotZero(&data[5])==(int)pBt->usableSize + || CORRUPT_DB ); + pPage->nFree = -1; /* Indicate that this value is yet uncomputed */ + pPage->isInit = 1; + if( pBt->db->flags & SQLITE_CellSizeCk ){ + return btreeCellSizeCheck(pPage); + } + return SQLITE_OK; +} + +/* +** Set up a raw page so that it looks like a database page holding +** no entries. +*/ +static void zeroPage(MemPage *pPage, int flags){ + unsigned char *data = pPage->aData; + BtShared *pBt = pPage->pBt; + u8 hdr = pPage->hdrOffset; + u16 first; + + assert( sqlite3PagerPagenumber(pPage->pDbPage)==pPage->pgno ); + assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage ); + assert( sqlite3PagerGetData(pPage->pDbPage) == data ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pBt->btsFlags & BTS_FAST_SECURE ){ + memset(&data[hdr], 0, pBt->usableSize - hdr); + } + data[hdr] = (char)flags; + first = hdr + ((flags&PTF_LEAF)==0 ? 12 : 8); + memset(&data[hdr+1], 0, 4); + data[hdr+7] = 0; + put2byte(&data[hdr+5], pBt->usableSize); + pPage->nFree = (u16)(pBt->usableSize - first); + decodeFlags(pPage, flags); + pPage->cellOffset = first; + pPage->aDataEnd = &data[pBt->usableSize]; + pPage->aCellIdx = &data[first]; + pPage->aDataOfst = &data[pPage->childPtrSize]; + pPage->nOverflow = 0; + assert( pBt->pageSize>=512 && pBt->pageSize<=65536 ); + pPage->maskPage = (u16)(pBt->pageSize - 1); + pPage->nCell = 0; + pPage->isInit = 1; +} + + +/* +** Convert a DbPage obtained from the pager into a MemPage used by +** the btree layer. +*/ +static MemPage *btreePageFromDbPage(DbPage *pDbPage, Pgno pgno, BtShared *pBt){ + MemPage *pPage = (MemPage*)sqlite3PagerGetExtra(pDbPage); + if( pgno!=pPage->pgno ){ + pPage->aData = sqlite3PagerGetData(pDbPage); + pPage->pDbPage = pDbPage; + pPage->pBt = pBt; + pPage->pgno = pgno; + pPage->hdrOffset = pgno==1 ? 100 : 0; + } + assert( pPage->aData==sqlite3PagerGetData(pDbPage) ); + return pPage; +} + +/* +** Get a page from the pager. Initialize the MemPage.pBt and +** MemPage.aData elements if needed. See also: btreeGetUnusedPage(). +** +** If the PAGER_GET_NOCONTENT flag is set, it means that we do not care +** about the content of the page at this time. So do not go to the disk +** to fetch the content. Just fill in the content with zeros for now. +** If in the future we call sqlite3PagerWrite() on this page, that +** means we have started to be concerned about content and the disk +** read should occur at that point. +*/ +static int btreeGetPage( + BtShared *pBt, /* The btree */ + Pgno pgno, /* Number of the page to fetch */ + MemPage **ppPage, /* Return the page in this parameter */ + int flags /* PAGER_GET_NOCONTENT or PAGER_GET_READONLY */ +){ + int rc; + DbPage *pDbPage; + + assert( flags==0 || flags==PAGER_GET_NOCONTENT || flags==PAGER_GET_READONLY ); + assert( sqlite3_mutex_held(pBt->mutex) ); + rc = sqlite3PagerGet(pBt->pPager, pgno, (DbPage**)&pDbPage, flags); + if( rc ) return rc; + *ppPage = btreePageFromDbPage(pDbPage, pgno, pBt); + return SQLITE_OK; +} + +/* +** Retrieve a page from the pager cache. If the requested page is not +** already in the pager cache return NULL. Initialize the MemPage.pBt and +** MemPage.aData elements if needed. +*/ +static MemPage *btreePageLookup(BtShared *pBt, Pgno pgno){ + DbPage *pDbPage; + assert( sqlite3_mutex_held(pBt->mutex) ); + pDbPage = sqlite3PagerLookup(pBt->pPager, pgno); + if( pDbPage ){ + return btreePageFromDbPage(pDbPage, pgno, pBt); + } + return 0; +} + +/* +** Return the size of the database file in pages. If there is any kind of +** error, return ((unsigned int)-1). +*/ +static Pgno btreePagecount(BtShared *pBt){ + return pBt->nPage; +} +Pgno sqlite3BtreeLastPage(Btree *p){ + assert( sqlite3BtreeHoldsMutex(p) ); + return btreePagecount(p->pBt); +} + +/* +** Get a page from the pager and initialize it. +** +** If pCur!=0 then the page is being fetched as part of a moveToChild() +** call. Do additional sanity checking on the page in this case. +** And if the fetch fails, this routine must decrement pCur->iPage. +** +** The page is fetched as read-write unless pCur is not NULL and is +** a read-only cursor. +** +** If an error occurs, then *ppPage is undefined. It +** may remain unchanged, or it may be set to an invalid value. +*/ +static int getAndInitPage( + BtShared *pBt, /* The database file */ + Pgno pgno, /* Number of the page to get */ + MemPage **ppPage, /* Write the page pointer here */ + BtCursor *pCur, /* Cursor to receive the page, or NULL */ + int bReadOnly /* True for a read-only page */ +){ + int rc; + DbPage *pDbPage; + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pCur==0 || ppPage==&pCur->pPage ); + assert( pCur==0 || bReadOnly==pCur->curPagerFlags ); + assert( pCur==0 || pCur->iPage>0 ); + + if( pgno>btreePagecount(pBt) ){ + rc = SQLITE_CORRUPT_BKPT; + goto getAndInitPage_error1; + } + rc = sqlite3PagerGet(pBt->pPager, pgno, (DbPage**)&pDbPage, bReadOnly); + if( rc ){ + goto getAndInitPage_error1; + } + *ppPage = (MemPage*)sqlite3PagerGetExtra(pDbPage); + if( (*ppPage)->isInit==0 ){ + btreePageFromDbPage(pDbPage, pgno, pBt); + rc = btreeInitPage(*ppPage); + if( rc!=SQLITE_OK ){ + goto getAndInitPage_error2; + } + } + assert( (*ppPage)->pgno==pgno ); + assert( (*ppPage)->aData==sqlite3PagerGetData(pDbPage) ); + + /* If obtaining a child page for a cursor, we must verify that the page is + ** compatible with the root page. */ + if( pCur && ((*ppPage)->nCell<1 || (*ppPage)->intKey!=pCur->curIntKey) ){ + rc = SQLITE_CORRUPT_PGNO(pgno); + goto getAndInitPage_error2; + } + return SQLITE_OK; + +getAndInitPage_error2: + releasePage(*ppPage); +getAndInitPage_error1: + if( pCur ){ + pCur->iPage--; + pCur->pPage = pCur->apPage[pCur->iPage]; + } + testcase( pgno==0 ); + assert( pgno!=0 || rc==SQLITE_CORRUPT ); + return rc; +} + +/* +** Release a MemPage. This should be called once for each prior +** call to btreeGetPage. +** +** Page1 is a special case and must be released using releasePageOne(). +*/ +static void releasePageNotNull(MemPage *pPage){ + assert( pPage->aData ); + assert( pPage->pBt ); + assert( pPage->pDbPage!=0 ); + assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage ); + assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + sqlite3PagerUnrefNotNull(pPage->pDbPage); +} +static void releasePage(MemPage *pPage){ + if( pPage ) releasePageNotNull(pPage); +} +static void releasePageOne(MemPage *pPage){ + assert( pPage!=0 ); + assert( pPage->aData ); + assert( pPage->pBt ); + assert( pPage->pDbPage!=0 ); + assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage ); + assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + sqlite3PagerUnrefPageOne(pPage->pDbPage); +} + +/* +** Get an unused page. +** +** This works just like btreeGetPage() with the addition: +** +** * If the page is already in use for some other purpose, immediately +** release it and return an SQLITE_CURRUPT error. +** * Make sure the isInit flag is clear +*/ +static int btreeGetUnusedPage( + BtShared *pBt, /* The btree */ + Pgno pgno, /* Number of the page to fetch */ + MemPage **ppPage, /* Return the page in this parameter */ + int flags /* PAGER_GET_NOCONTENT or PAGER_GET_READONLY */ +){ + int rc = btreeGetPage(pBt, pgno, ppPage, flags); + if( rc==SQLITE_OK ){ + if( sqlite3PagerPageRefcount((*ppPage)->pDbPage)>1 ){ + releasePage(*ppPage); + *ppPage = 0; + return SQLITE_CORRUPT_BKPT; + } + (*ppPage)->isInit = 0; + }else{ + *ppPage = 0; + } + return rc; +} + + +/* +** During a rollback, when the pager reloads information into the cache +** so that the cache is restored to its original state at the start of +** the transaction, for each page restored this routine is called. +** +** This routine needs to reset the extra data section at the end of the +** page to agree with the restored data. +*/ +static void pageReinit(DbPage *pData){ + MemPage *pPage; + pPage = (MemPage *)sqlite3PagerGetExtra(pData); + assert( sqlite3PagerPageRefcount(pData)>0 ); + if( pPage->isInit ){ + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + pPage->isInit = 0; + if( sqlite3PagerPageRefcount(pData)>1 ){ + /* pPage might not be a btree page; it might be an overflow page + ** or ptrmap page or a free page. In those cases, the following + ** call to btreeInitPage() will likely return SQLITE_CORRUPT. + ** But no harm is done by this. And it is very important that + ** btreeInitPage() be called on every btree page so we make + ** the call for every page that comes in for re-initing. */ + btreeInitPage(pPage); + } + } +} + +/* +** Invoke the busy handler for a btree. +*/ +static int btreeInvokeBusyHandler(void *pArg){ + BtShared *pBt = (BtShared*)pArg; + assert( pBt->db ); + assert( sqlite3_mutex_held(pBt->db->mutex) ); + return sqlite3InvokeBusyHandler(&pBt->db->busyHandler); +} + +/* +** Open a database file. +** +** zFilename is the name of the database file. If zFilename is NULL +** then an ephemeral database is created. The ephemeral database might +** be exclusively in memory, or it might use a disk-based memory cache. +** Either way, the ephemeral database will be automatically deleted +** when sqlite3BtreeClose() is called. +** +** If zFilename is ":memory:" then an in-memory database is created +** that is automatically destroyed when it is closed. +** +** The "flags" parameter is a bitmask that might contain bits like +** BTREE_OMIT_JOURNAL and/or BTREE_MEMORY. +** +** If the database is already opened in the same database connection +** and we are in shared cache mode, then the open will fail with an +** SQLITE_CONSTRAINT error. We cannot allow two or more BtShared +** objects in the same database connection since doing so will lead +** to problems with locking. +*/ +int sqlite3BtreeOpen( + sqlite3_vfs *pVfs, /* VFS to use for this b-tree */ + const char *zFilename, /* Name of the file containing the BTree database */ + sqlite3 *db, /* Associated database handle */ + Btree **ppBtree, /* Pointer to new Btree object written here */ + int flags, /* Options */ + int vfsFlags /* Flags passed through to sqlite3_vfs.xOpen() */ +){ + BtShared *pBt = 0; /* Shared part of btree structure */ + Btree *p; /* Handle to return */ + sqlite3_mutex *mutexOpen = 0; /* Prevents a race condition. Ticket #3537 */ + int rc = SQLITE_OK; /* Result code from this function */ + u8 nReserve; /* Byte of unused space on each page */ + unsigned char zDbHeader[100]; /* Database header content */ + + /* True if opening an ephemeral, temporary database */ + const int isTempDb = zFilename==0 || zFilename[0]==0; + + /* Set the variable isMemdb to true for an in-memory database, or + ** false for a file-based database. + */ +#ifdef SQLITE_OMIT_MEMORYDB + const int isMemdb = 0; +#else + const int isMemdb = (zFilename && strcmp(zFilename, ":memory:")==0) + || (isTempDb && sqlite3TempInMemory(db)) + || (vfsFlags & SQLITE_OPEN_MEMORY)!=0; +#endif + + assert( db!=0 ); + assert( pVfs!=0 ); + assert( sqlite3_mutex_held(db->mutex) ); + assert( (flags&0xff)==flags ); /* flags fit in 8 bits */ + + /* Only a BTREE_SINGLE database can be BTREE_UNORDERED */ + assert( (flags & BTREE_UNORDERED)==0 || (flags & BTREE_SINGLE)!=0 ); + + /* A BTREE_SINGLE database is always a temporary and/or ephemeral */ + assert( (flags & BTREE_SINGLE)==0 || isTempDb ); + + if( isMemdb ){ + flags |= BTREE_MEMORY; + } + if( (vfsFlags & SQLITE_OPEN_MAIN_DB)!=0 && (isMemdb || isTempDb) ){ + vfsFlags = (vfsFlags & ~SQLITE_OPEN_MAIN_DB) | SQLITE_OPEN_TEMP_DB; + } + p = sqlite3MallocZero(sizeof(Btree)); + if( !p ){ + return SQLITE_NOMEM_BKPT; + } + p->inTrans = TRANS_NONE; + p->db = db; +#ifndef SQLITE_OMIT_SHARED_CACHE + p->lock.pBtree = p; + p->lock.iTable = 1; +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO) + /* + ** If this Btree is a candidate for shared cache, try to find an + ** existing BtShared object that we can share with + */ + if( isTempDb==0 && (isMemdb==0 || (vfsFlags&SQLITE_OPEN_URI)!=0) ){ + if( vfsFlags & SQLITE_OPEN_SHAREDCACHE ){ + int nFilename = sqlite3Strlen30(zFilename)+1; + int nFullPathname = pVfs->mxPathname+1; + char *zFullPathname = sqlite3Malloc(MAX(nFullPathname,nFilename)); + MUTEX_LOGIC( sqlite3_mutex *mutexShared; ) + + p->sharable = 1; + if( !zFullPathname ){ + sqlite3_free(p); + return SQLITE_NOMEM_BKPT; + } + if( isMemdb ){ + memcpy(zFullPathname, zFilename, nFilename); + }else{ + rc = sqlite3OsFullPathname(pVfs, zFilename, + nFullPathname, zFullPathname); + if( rc ){ + if( rc==SQLITE_OK_SYMLINK ){ + rc = SQLITE_OK; + }else{ + sqlite3_free(zFullPathname); + sqlite3_free(p); + return rc; + } + } + } +#if SQLITE_THREADSAFE + mutexOpen = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_OPEN); + sqlite3_mutex_enter(mutexOpen); + mutexShared = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MAIN); + sqlite3_mutex_enter(mutexShared); +#endif + for(pBt=GLOBAL(BtShared*,sqlite3SharedCacheList); pBt; pBt=pBt->pNext){ + assert( pBt->nRef>0 ); + if( 0==strcmp(zFullPathname, sqlite3PagerFilename(pBt->pPager, 0)) + && sqlite3PagerVfs(pBt->pPager)==pVfs ){ + int iDb; + for(iDb=db->nDb-1; iDb>=0; iDb--){ + Btree *pExisting = db->aDb[iDb].pBt; + if( pExisting && pExisting->pBt==pBt ){ + sqlite3_mutex_leave(mutexShared); + sqlite3_mutex_leave(mutexOpen); + sqlite3_free(zFullPathname); + sqlite3_free(p); + return SQLITE_CONSTRAINT; + } + } + p->pBt = pBt; + pBt->nRef++; + break; + } + } + sqlite3_mutex_leave(mutexShared); + sqlite3_free(zFullPathname); + } +#ifdef SQLITE_DEBUG + else{ + /* In debug mode, we mark all persistent databases as sharable + ** even when they are not. This exercises the locking code and + ** gives more opportunity for asserts(sqlite3_mutex_held()) + ** statements to find locking problems. + */ + p->sharable = 1; + } +#endif + } +#endif + if( pBt==0 ){ + /* + ** The following asserts make sure that structures used by the btree are + ** the right size. This is to guard against size changes that result + ** when compiling on a different architecture. + */ + assert( sizeof(i64)==8 ); + assert( sizeof(u64)==8 ); + assert( sizeof(u32)==4 ); + assert( sizeof(u16)==2 ); + assert( sizeof(Pgno)==4 ); + + pBt = sqlite3MallocZero( sizeof(*pBt) ); + if( pBt==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto btree_open_out; + } + rc = sqlite3PagerOpen(pVfs, &pBt->pPager, zFilename, + sizeof(MemPage), flags, vfsFlags, pageReinit); + if( rc==SQLITE_OK ){ + sqlite3PagerSetMmapLimit(pBt->pPager, db ? db->szMmap : 0); + rc = sqlite3PagerReadFileheader(pBt->pPager,sizeof(zDbHeader),zDbHeader); + } + if( rc!=SQLITE_OK ){ + goto btree_open_out; + } + pBt->openFlags = (u8)flags; + pBt->db = db; + sqlite3PagerSetBusyHandler(pBt->pPager, btreeInvokeBusyHandler, pBt); + p->pBt = pBt; + + pBt->pCursor = 0; + pBt->pPage1 = 0; + if( sqlite3PagerIsreadonly(pBt->pPager) ) pBt->btsFlags |= BTS_READ_ONLY; +#if defined(SQLITE_SECURE_DELETE) + pBt->btsFlags |= BTS_SECURE_DELETE; +#elif defined(SQLITE_FAST_SECURE_DELETE) + pBt->btsFlags |= BTS_OVERWRITE; +#endif + /* EVIDENCE-OF: R-51873-39618 The page size for a database file is + ** determined by the 2-byte integer located at an offset of 16 bytes from + ** the beginning of the database file. */ + pBt->pageSize = (zDbHeader[16]<<8) | (zDbHeader[17]<<16); + if( pBt->pageSize<512 || pBt->pageSize>SQLITE_MAX_PAGE_SIZE + || ((pBt->pageSize-1)&pBt->pageSize)!=0 ){ + pBt->pageSize = 0; +#ifndef SQLITE_OMIT_AUTOVACUUM + /* If the magic name ":memory:" will create an in-memory database, then + ** leave the autoVacuum mode at 0 (do not auto-vacuum), even if + ** SQLITE_DEFAULT_AUTOVACUUM is true. On the other hand, if + ** SQLITE_OMIT_MEMORYDB has been defined, then ":memory:" is just a + ** regular file-name. In this case the auto-vacuum applies as per normal. + */ + if( zFilename && !isMemdb ){ + pBt->autoVacuum = (SQLITE_DEFAULT_AUTOVACUUM ? 1 : 0); + pBt->incrVacuum = (SQLITE_DEFAULT_AUTOVACUUM==2 ? 1 : 0); + } +#endif + nReserve = 0; + }else{ + /* EVIDENCE-OF: R-37497-42412 The size of the reserved region is + ** determined by the one-byte unsigned integer found at an offset of 20 + ** into the database file header. */ + nReserve = zDbHeader[20]; + pBt->btsFlags |= BTS_PAGESIZE_FIXED; +#ifndef SQLITE_OMIT_AUTOVACUUM + pBt->autoVacuum = (get4byte(&zDbHeader[36 + 4*4])?1:0); + pBt->incrVacuum = (get4byte(&zDbHeader[36 + 7*4])?1:0); +#endif + } + rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, nReserve); + if( rc ) goto btree_open_out; + pBt->usableSize = pBt->pageSize - nReserve; + assert( (pBt->pageSize & 7)==0 ); /* 8-byte alignment of pageSize */ + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO) + /* Add the new BtShared object to the linked list sharable BtShareds. + */ + pBt->nRef = 1; + if( p->sharable ){ + MUTEX_LOGIC( sqlite3_mutex *mutexShared; ) + MUTEX_LOGIC( mutexShared = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MAIN);) + if( SQLITE_THREADSAFE && sqlite3GlobalConfig.bCoreMutex ){ + pBt->mutex = sqlite3MutexAlloc(SQLITE_MUTEX_FAST); + if( pBt->mutex==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto btree_open_out; + } + } + sqlite3_mutex_enter(mutexShared); + pBt->pNext = GLOBAL(BtShared*,sqlite3SharedCacheList); + GLOBAL(BtShared*,sqlite3SharedCacheList) = pBt; + sqlite3_mutex_leave(mutexShared); + } +#endif + } + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO) + /* If the new Btree uses a sharable pBtShared, then link the new + ** Btree into the list of all sharable Btrees for the same connection. + ** The list is kept in ascending order by pBt address. + */ + if( p->sharable ){ + int i; + Btree *pSib; + for(i=0; inDb; i++){ + if( (pSib = db->aDb[i].pBt)!=0 && pSib->sharable ){ + while( pSib->pPrev ){ pSib = pSib->pPrev; } + if( (uptr)p->pBt<(uptr)pSib->pBt ){ + p->pNext = pSib; + p->pPrev = 0; + pSib->pPrev = p; + }else{ + while( pSib->pNext && (uptr)pSib->pNext->pBt<(uptr)p->pBt ){ + pSib = pSib->pNext; + } + p->pNext = pSib->pNext; + p->pPrev = pSib; + if( p->pNext ){ + p->pNext->pPrev = p; + } + pSib->pNext = p; + } + break; + } + } + } +#endif + *ppBtree = p; + +btree_open_out: + if( rc!=SQLITE_OK ){ + if( pBt && pBt->pPager ){ + sqlite3PagerClose(pBt->pPager, 0); + } + sqlite3_free(pBt); + sqlite3_free(p); + *ppBtree = 0; + }else{ + sqlite3_file *pFile; + + /* If the B-Tree was successfully opened, set the pager-cache size to the + ** default value. Except, when opening on an existing shared pager-cache, + ** do not change the pager-cache size. + */ + if( sqlite3BtreeSchema(p, 0, 0)==0 ){ + sqlite3BtreeSetCacheSize(p, SQLITE_DEFAULT_CACHE_SIZE); + } + + pFile = sqlite3PagerFile(pBt->pPager); + if( pFile->pMethods ){ + sqlite3OsFileControlHint(pFile, SQLITE_FCNTL_PDB, (void*)&pBt->db); + } + } + if( mutexOpen ){ + assert( sqlite3_mutex_held(mutexOpen) ); + sqlite3_mutex_leave(mutexOpen); + } + assert( rc!=SQLITE_OK || sqlite3BtreeConnectionCount(*ppBtree)>0 ); + return rc; +} + +/* +** Decrement the BtShared.nRef counter. When it reaches zero, +** remove the BtShared structure from the sharing list. Return +** true if the BtShared.nRef counter reaches zero and return +** false if it is still positive. +*/ +static int removeFromSharingList(BtShared *pBt){ +#ifndef SQLITE_OMIT_SHARED_CACHE + MUTEX_LOGIC( sqlite3_mutex *pMainMtx; ) + BtShared *pList; + int removed = 0; + + assert( sqlite3_mutex_notheld(pBt->mutex) ); + MUTEX_LOGIC( pMainMtx = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_MAIN); ) + sqlite3_mutex_enter(pMainMtx); + pBt->nRef--; + if( pBt->nRef<=0 ){ + if( GLOBAL(BtShared*,sqlite3SharedCacheList)==pBt ){ + GLOBAL(BtShared*,sqlite3SharedCacheList) = pBt->pNext; + }else{ + pList = GLOBAL(BtShared*,sqlite3SharedCacheList); + while( ALWAYS(pList) && pList->pNext!=pBt ){ + pList=pList->pNext; + } + if( ALWAYS(pList) ){ + pList->pNext = pBt->pNext; + } + } + if( SQLITE_THREADSAFE ){ + sqlite3_mutex_free(pBt->mutex); + } + removed = 1; + } + sqlite3_mutex_leave(pMainMtx); + return removed; +#else + return 1; +#endif +} + +/* +** Make sure pBt->pTmpSpace points to an allocation of +** MX_CELL_SIZE(pBt) bytes with a 4-byte prefix for a left-child +** pointer. +*/ +static void allocateTempSpace(BtShared *pBt){ + if( !pBt->pTmpSpace ){ + pBt->pTmpSpace = sqlite3PageMalloc( pBt->pageSize ); + + /* One of the uses of pBt->pTmpSpace is to format cells before + ** inserting them into a leaf page (function fillInCell()). If + ** a cell is less than 4 bytes in size, it is rounded up to 4 bytes + ** by the various routines that manipulate binary cells. Which + ** can mean that fillInCell() only initializes the first 2 or 3 + ** bytes of pTmpSpace, but that the first 4 bytes are copied from + ** it into a database page. This is not actually a problem, but it + ** does cause a valgrind error when the 1 or 2 bytes of unitialized + ** data is passed to system call write(). So to avoid this error, + ** zero the first 4 bytes of temp space here. + ** + ** Also: Provide four bytes of initialized space before the + ** beginning of pTmpSpace as an area available to prepend the + ** left-child pointer to the beginning of a cell. + */ + if( pBt->pTmpSpace ){ + memset(pBt->pTmpSpace, 0, 8); + pBt->pTmpSpace += 4; + } + } +} + +/* +** Free the pBt->pTmpSpace allocation +*/ +static void freeTempSpace(BtShared *pBt){ + if( pBt->pTmpSpace ){ + pBt->pTmpSpace -= 4; + sqlite3PageFree(pBt->pTmpSpace); + pBt->pTmpSpace = 0; + } +} + +/* +** Close an open database and invalidate all cursors. +*/ +int sqlite3BtreeClose(Btree *p){ + BtShared *pBt = p->pBt; + + /* Close all cursors opened via this handle. */ + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + + /* Verify that no other cursors have this Btree open */ +#ifdef SQLITE_DEBUG + { + BtCursor *pCur = pBt->pCursor; + while( pCur ){ + BtCursor *pTmp = pCur; + pCur = pCur->pNext; + assert( pTmp->pBtree!=p ); + + } + } +#endif + + /* Rollback any active transaction and free the handle structure. + ** The call to sqlite3BtreeRollback() drops any table-locks held by + ** this handle. + */ + sqlite3BtreeRollback(p, SQLITE_OK, 0); + sqlite3BtreeLeave(p); + + /* If there are still other outstanding references to the shared-btree + ** structure, return now. The remainder of this procedure cleans + ** up the shared-btree. + */ + assert( p->wantToLock==0 && p->locked==0 ); + if( !p->sharable || removeFromSharingList(pBt) ){ + /* The pBt is no longer on the sharing list, so we can access + ** it without having to hold the mutex. + ** + ** Clean out and delete the BtShared object. + */ + assert( !pBt->pCursor ); + sqlite3PagerClose(pBt->pPager, p->db); + if( pBt->xFreeSchema && pBt->pSchema ){ + pBt->xFreeSchema(pBt->pSchema); + } + sqlite3DbFree(0, pBt->pSchema); + freeTempSpace(pBt); + sqlite3_free(pBt); + } + +#ifndef SQLITE_OMIT_SHARED_CACHE + assert( p->wantToLock==0 ); + assert( p->locked==0 ); + if( p->pPrev ) p->pPrev->pNext = p->pNext; + if( p->pNext ) p->pNext->pPrev = p->pPrev; +#endif + + sqlite3_free(p); + return SQLITE_OK; +} + +/* +** Change the "soft" limit on the number of pages in the cache. +** Unused and unmodified pages will be recycled when the number of +** pages in the cache exceeds this soft limit. But the size of the +** cache is allowed to grow larger than this limit if it contains +** dirty pages or pages still in active use. +*/ +int sqlite3BtreeSetCacheSize(Btree *p, int mxPage){ + BtShared *pBt = p->pBt; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + sqlite3PagerSetCachesize(pBt->pPager, mxPage); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} + +/* +** Change the "spill" limit on the number of pages in the cache. +** If the number of pages exceeds this limit during a write transaction, +** the pager might attempt to "spill" pages to the journal early in +** order to free up memory. +** +** The value returned is the current spill size. If zero is passed +** as an argument, no changes are made to the spill size setting, so +** using mxPage of 0 is a way to query the current spill size. +*/ +int sqlite3BtreeSetSpillSize(Btree *p, int mxPage){ + BtShared *pBt = p->pBt; + int res; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + res = sqlite3PagerSetSpillsize(pBt->pPager, mxPage); + sqlite3BtreeLeave(p); + return res; +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* +** Change the limit on the amount of the database file that may be +** memory mapped. +*/ +int sqlite3BtreeSetMmapLimit(Btree *p, sqlite3_int64 szMmap){ + BtShared *pBt = p->pBt; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + sqlite3PagerSetMmapLimit(pBt->pPager, szMmap); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + +/* +** Change the way data is synced to disk in order to increase or decrease +** how well the database resists damage due to OS crashes and power +** failures. Level 1 is the same as asynchronous (no syncs() occur and +** there is a high probability of damage) Level 2 is the default. There +** is a very low but non-zero probability of damage. Level 3 reduces the +** probability of damage to near zero but with a write performance reduction. +*/ +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +int sqlite3BtreeSetPagerFlags( + Btree *p, /* The btree to set the safety level on */ + unsigned pgFlags /* Various PAGER_* flags */ +){ + BtShared *pBt = p->pBt; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + sqlite3PagerSetFlags(pBt->pPager, pgFlags); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} +#endif + +/* +** Change the default pages size and the number of reserved bytes per page. +** Or, if the page size has already been fixed, return SQLITE_READONLY +** without changing anything. +** +** The page size must be a power of 2 between 512 and 65536. If the page +** size supplied does not meet this constraint then the page size is not +** changed. +** +** Page sizes are constrained to be a power of two so that the region +** of the database file used for locking (beginning at PENDING_BYTE, +** the first byte past the 1GB boundary, 0x40000000) needs to occur +** at the beginning of a page. +** +** If parameter nReserve is less than zero, then the number of reserved +** bytes per page is left unchanged. +** +** If the iFix!=0 then the BTS_PAGESIZE_FIXED flag is set so that the page size +** and autovacuum mode can no longer be changed. +*/ +int sqlite3BtreeSetPageSize(Btree *p, int pageSize, int nReserve, int iFix){ + int rc = SQLITE_OK; + int x; + BtShared *pBt = p->pBt; + assert( nReserve>=0 && nReserve<=255 ); + sqlite3BtreeEnter(p); + pBt->nReserveWanted = nReserve; + x = pBt->pageSize - pBt->usableSize; + if( nReservebtsFlags & BTS_PAGESIZE_FIXED ){ + sqlite3BtreeLeave(p); + return SQLITE_READONLY; + } + assert( nReserve>=0 && nReserve<=255 ); + if( pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE && + ((pageSize-1)&pageSize)==0 ){ + assert( (pageSize & 7)==0 ); + assert( !pBt->pCursor ); + if( nReserve>32 && pageSize==512 ) pageSize = 1024; + pBt->pageSize = (u32)pageSize; + freeTempSpace(pBt); + } + rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, nReserve); + pBt->usableSize = pBt->pageSize - (u16)nReserve; + if( iFix ) pBt->btsFlags |= BTS_PAGESIZE_FIXED; + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Return the currently defined page size +*/ +int sqlite3BtreeGetPageSize(Btree *p){ + return p->pBt->pageSize; +} + +/* +** This function is similar to sqlite3BtreeGetReserve(), except that it +** may only be called if it is guaranteed that the b-tree mutex is already +** held. +** +** This is useful in one special case in the backup API code where it is +** known that the shared b-tree mutex is held, but the mutex on the +** database handle that owns *p is not. In this case if sqlite3BtreeEnter() +** were to be called, it might collide with some other operation on the +** database handle that owns *p, causing undefined behavior. +*/ +int sqlite3BtreeGetReserveNoMutex(Btree *p){ + int n; + assert( sqlite3_mutex_held(p->pBt->mutex) ); + n = p->pBt->pageSize - p->pBt->usableSize; + return n; +} + +/* +** Return the number of bytes of space at the end of every page that +** are intentually left unused. This is the "reserved" space that is +** sometimes used by extensions. +** +** The value returned is the larger of the current reserve size and +** the latest reserve size requested by SQLITE_FILECTRL_RESERVE_BYTES. +** The amount of reserve can only grow - never shrink. +*/ +int sqlite3BtreeGetRequestedReserve(Btree *p){ + int n1, n2; + sqlite3BtreeEnter(p); + n1 = (int)p->pBt->nReserveWanted; + n2 = sqlite3BtreeGetReserveNoMutex(p); + sqlite3BtreeLeave(p); + return n1>n2 ? n1 : n2; +} + + +/* +** Set the maximum page count for a database if mxPage is positive. +** No changes are made if mxPage is 0 or negative. +** Regardless of the value of mxPage, return the maximum page count. +*/ +Pgno sqlite3BtreeMaxPageCount(Btree *p, Pgno mxPage){ + Pgno n; + sqlite3BtreeEnter(p); + n = sqlite3PagerMaxPageCount(p->pBt->pPager, mxPage); + sqlite3BtreeLeave(p); + return n; +} + +/* +** Change the values for the BTS_SECURE_DELETE and BTS_OVERWRITE flags: +** +** newFlag==0 Both BTS_SECURE_DELETE and BTS_OVERWRITE are cleared +** newFlag==1 BTS_SECURE_DELETE set and BTS_OVERWRITE is cleared +** newFlag==2 BTS_SECURE_DELETE cleared and BTS_OVERWRITE is set +** newFlag==(-1) No changes +** +** This routine acts as a query if newFlag is less than zero +** +** With BTS_OVERWRITE set, deleted content is overwritten by zeros, but +** freelist leaf pages are not written back to the database. Thus in-page +** deleted content is cleared, but freelist deleted content is not. +** +** With BTS_SECURE_DELETE, operation is like BTS_OVERWRITE with the addition +** that freelist leaf pages are written back into the database, increasing +** the amount of disk I/O. +*/ +int sqlite3BtreeSecureDelete(Btree *p, int newFlag){ + int b; + if( p==0 ) return 0; + sqlite3BtreeEnter(p); + assert( BTS_OVERWRITE==BTS_SECURE_DELETE*2 ); + assert( BTS_FAST_SECURE==(BTS_OVERWRITE|BTS_SECURE_DELETE) ); + if( newFlag>=0 ){ + p->pBt->btsFlags &= ~BTS_FAST_SECURE; + p->pBt->btsFlags |= BTS_SECURE_DELETE*newFlag; + } + b = (p->pBt->btsFlags & BTS_FAST_SECURE)/BTS_SECURE_DELETE; + sqlite3BtreeLeave(p); + return b; +} + +/* +** Change the 'auto-vacuum' property of the database. If the 'autoVacuum' +** parameter is non-zero, then auto-vacuum mode is enabled. If zero, it +** is disabled. The default value for the auto-vacuum property is +** determined by the SQLITE_DEFAULT_AUTOVACUUM macro. +*/ +int sqlite3BtreeSetAutoVacuum(Btree *p, int autoVacuum){ +#ifdef SQLITE_OMIT_AUTOVACUUM + return SQLITE_READONLY; +#else + BtShared *pBt = p->pBt; + int rc = SQLITE_OK; + u8 av = (u8)autoVacuum; + + sqlite3BtreeEnter(p); + if( (pBt->btsFlags & BTS_PAGESIZE_FIXED)!=0 && (av ?1:0)!=pBt->autoVacuum ){ + rc = SQLITE_READONLY; + }else{ + pBt->autoVacuum = av ?1:0; + pBt->incrVacuum = av==2 ?1:0; + } + sqlite3BtreeLeave(p); + return rc; +#endif +} + +/* +** Return the value of the 'auto-vacuum' property. If auto-vacuum is +** enabled 1 is returned. Otherwise 0. +*/ +int sqlite3BtreeGetAutoVacuum(Btree *p){ +#ifdef SQLITE_OMIT_AUTOVACUUM + return BTREE_AUTOVACUUM_NONE; +#else + int rc; + sqlite3BtreeEnter(p); + rc = ( + (!p->pBt->autoVacuum)?BTREE_AUTOVACUUM_NONE: + (!p->pBt->incrVacuum)?BTREE_AUTOVACUUM_FULL: + BTREE_AUTOVACUUM_INCR + ); + sqlite3BtreeLeave(p); + return rc; +#endif +} + +/* +** If the user has not set the safety-level for this database connection +** using "PRAGMA synchronous", and if the safety-level is not already +** set to the value passed to this function as the second parameter, +** set it so. +*/ +#if SQLITE_DEFAULT_SYNCHRONOUS!=SQLITE_DEFAULT_WAL_SYNCHRONOUS \ + && !defined(SQLITE_OMIT_WAL) +static void setDefaultSyncFlag(BtShared *pBt, u8 safety_level){ + sqlite3 *db; + Db *pDb; + if( (db=pBt->db)!=0 && (pDb=db->aDb)!=0 ){ + while( pDb->pBt==0 || pDb->pBt->pBt!=pBt ){ pDb++; } + if( pDb->bSyncSet==0 + && pDb->safety_level!=safety_level + && pDb!=&db->aDb[1] + ){ + pDb->safety_level = safety_level; + sqlite3PagerSetFlags(pBt->pPager, + pDb->safety_level | (db->flags & PAGER_FLAGS_MASK)); + } + } +} +#else +# define setDefaultSyncFlag(pBt,safety_level) +#endif + +/* Forward declaration */ +static int newDatabase(BtShared*); + + +/* +** Get a reference to pPage1 of the database file. This will +** also acquire a readlock on that file. +** +** SQLITE_OK is returned on success. If the file is not a +** well-formed database file, then SQLITE_CORRUPT is returned. +** SQLITE_BUSY is returned if the database is locked. SQLITE_NOMEM +** is returned if we run out of memory. +*/ +static int lockBtree(BtShared *pBt){ + int rc; /* Result code from subfunctions */ + MemPage *pPage1; /* Page 1 of the database file */ + u32 nPage; /* Number of pages in the database */ + u32 nPageFile = 0; /* Number of pages in the database file */ + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pBt->pPage1==0 ); + rc = sqlite3PagerSharedLock(pBt->pPager); + if( rc!=SQLITE_OK ) return rc; + rc = btreeGetPage(pBt, 1, &pPage1, 0); + if( rc!=SQLITE_OK ) return rc; + + /* Do some checking to help insure the file we opened really is + ** a valid database file. + */ + nPage = get4byte(28+(u8*)pPage1->aData); + sqlite3PagerPagecount(pBt->pPager, (int*)&nPageFile); + if( nPage==0 || memcmp(24+(u8*)pPage1->aData, 92+(u8*)pPage1->aData,4)!=0 ){ + nPage = nPageFile; + } + if( (pBt->db->flags & SQLITE_ResetDatabase)!=0 ){ + nPage = 0; + } + if( nPage>0 ){ + u32 pageSize; + u32 usableSize; + u8 *page1 = pPage1->aData; + rc = SQLITE_NOTADB; + /* EVIDENCE-OF: R-43737-39999 Every valid SQLite database file begins + ** with the following 16 bytes (in hex): 53 51 4c 69 74 65 20 66 6f 72 6d + ** 61 74 20 33 00. */ + if( memcmp(page1, zMagicHeader, 16)!=0 ){ + goto page1_init_failed; + } + +#ifdef SQLITE_OMIT_WAL + if( page1[18]>1 ){ + pBt->btsFlags |= BTS_READ_ONLY; + } + if( page1[19]>1 ){ + goto page1_init_failed; + } +#else + if( page1[18]>2 ){ + pBt->btsFlags |= BTS_READ_ONLY; + } + if( page1[19]>2 ){ + goto page1_init_failed; + } + + /* If the read version is set to 2, this database should be accessed + ** in WAL mode. If the log is not already open, open it now. Then + ** return SQLITE_OK and return without populating BtShared.pPage1. + ** The caller detects this and calls this function again. This is + ** required as the version of page 1 currently in the page1 buffer + ** may not be the latest version - there may be a newer one in the log + ** file. + */ + if( page1[19]==2 && (pBt->btsFlags & BTS_NO_WAL)==0 ){ + int isOpen = 0; + rc = sqlite3PagerOpenWal(pBt->pPager, &isOpen); + if( rc!=SQLITE_OK ){ + goto page1_init_failed; + }else{ + setDefaultSyncFlag(pBt, SQLITE_DEFAULT_WAL_SYNCHRONOUS+1); + if( isOpen==0 ){ + releasePageOne(pPage1); + return SQLITE_OK; + } + } + rc = SQLITE_NOTADB; + }else{ + setDefaultSyncFlag(pBt, SQLITE_DEFAULT_SYNCHRONOUS+1); + } +#endif + + /* EVIDENCE-OF: R-15465-20813 The maximum and minimum embedded payload + ** fractions and the leaf payload fraction values must be 64, 32, and 32. + ** + ** The original design allowed these amounts to vary, but as of + ** version 3.6.0, we require them to be fixed. + */ + if( memcmp(&page1[21], "\100\040\040",3)!=0 ){ + goto page1_init_failed; + } + /* EVIDENCE-OF: R-51873-39618 The page size for a database file is + ** determined by the 2-byte integer located at an offset of 16 bytes from + ** the beginning of the database file. */ + pageSize = (page1[16]<<8) | (page1[17]<<16); + /* EVIDENCE-OF: R-25008-21688 The size of a page is a power of two + ** between 512 and 65536 inclusive. */ + if( ((pageSize-1)&pageSize)!=0 + || pageSize>SQLITE_MAX_PAGE_SIZE + || pageSize<=256 + ){ + goto page1_init_failed; + } + pBt->btsFlags |= BTS_PAGESIZE_FIXED; + assert( (pageSize & 7)==0 ); + /* EVIDENCE-OF: R-59310-51205 The "reserved space" size in the 1-byte + ** integer at offset 20 is the number of bytes of space at the end of + ** each page to reserve for extensions. + ** + ** EVIDENCE-OF: R-37497-42412 The size of the reserved region is + ** determined by the one-byte unsigned integer found at an offset of 20 + ** into the database file header. */ + usableSize = pageSize - page1[20]; + if( (u32)pageSize!=pBt->pageSize ){ + /* After reading the first page of the database assuming a page size + ** of BtShared.pageSize, we have discovered that the page-size is + ** actually pageSize. Unlock the database, leave pBt->pPage1 at + ** zero and return SQLITE_OK. The caller will call this function + ** again with the correct page-size. + */ + releasePageOne(pPage1); + pBt->usableSize = usableSize; + pBt->pageSize = pageSize; + freeTempSpace(pBt); + rc = sqlite3PagerSetPagesize(pBt->pPager, &pBt->pageSize, + pageSize-usableSize); + return rc; + } + if( sqlite3WritableSchema(pBt->db)==0 && nPage>nPageFile ){ + rc = SQLITE_CORRUPT_BKPT; + goto page1_init_failed; + } + /* EVIDENCE-OF: R-28312-64704 However, the usable size is not allowed to + ** be less than 480. In other words, if the page size is 512, then the + ** reserved space size cannot exceed 32. */ + if( usableSize<480 ){ + goto page1_init_failed; + } + pBt->pageSize = pageSize; + pBt->usableSize = usableSize; +#ifndef SQLITE_OMIT_AUTOVACUUM + pBt->autoVacuum = (get4byte(&page1[36 + 4*4])?1:0); + pBt->incrVacuum = (get4byte(&page1[36 + 7*4])?1:0); +#endif + } + + /* maxLocal is the maximum amount of payload to store locally for + ** a cell. Make sure it is small enough so that at least minFanout + ** cells can will fit on one page. We assume a 10-byte page header. + ** Besides the payload, the cell must store: + ** 2-byte pointer to the cell + ** 4-byte child pointer + ** 9-byte nKey value + ** 4-byte nData value + ** 4-byte overflow page pointer + ** So a cell consists of a 2-byte pointer, a header which is as much as + ** 17 bytes long, 0 to N bytes of payload, and an optional 4 byte overflow + ** page pointer. + */ + pBt->maxLocal = (u16)((pBt->usableSize-12)*64/255 - 23); + pBt->minLocal = (u16)((pBt->usableSize-12)*32/255 - 23); + pBt->maxLeaf = (u16)(pBt->usableSize - 35); + pBt->minLeaf = (u16)((pBt->usableSize-12)*32/255 - 23); + if( pBt->maxLocal>127 ){ + pBt->max1bytePayload = 127; + }else{ + pBt->max1bytePayload = (u8)pBt->maxLocal; + } + assert( pBt->maxLeaf + 23 <= MX_CELL_SIZE(pBt) ); + pBt->pPage1 = pPage1; + pBt->nPage = nPage; + return SQLITE_OK; + +page1_init_failed: + releasePageOne(pPage1); + pBt->pPage1 = 0; + return rc; +} + +#ifndef NDEBUG +/* +** Return the number of cursors open on pBt. This is for use +** in assert() expressions, so it is only compiled if NDEBUG is not +** defined. +** +** Only write cursors are counted if wrOnly is true. If wrOnly is +** false then all cursors are counted. +** +** For the purposes of this routine, a cursor is any cursor that +** is capable of reading or writing to the database. Cursors that +** have been tripped into the CURSOR_FAULT state are not counted. +*/ +static int countValidCursors(BtShared *pBt, int wrOnly){ + BtCursor *pCur; + int r = 0; + for(pCur=pBt->pCursor; pCur; pCur=pCur->pNext){ + if( (wrOnly==0 || (pCur->curFlags & BTCF_WriteFlag)!=0) + && pCur->eState!=CURSOR_FAULT ) r++; + } + return r; +} +#endif + +/* +** If there are no outstanding cursors and we are not in the middle +** of a transaction but there is a read lock on the database, then +** this routine unrefs the first page of the database file which +** has the effect of releasing the read lock. +** +** If there is a transaction in progress, this routine is a no-op. +*/ +static void unlockBtreeIfUnused(BtShared *pBt){ + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( countValidCursors(pBt,0)==0 || pBt->inTransaction>TRANS_NONE ); + if( pBt->inTransaction==TRANS_NONE && pBt->pPage1!=0 ){ + MemPage *pPage1 = pBt->pPage1; + assert( pPage1->aData ); + assert( sqlite3PagerRefcount(pBt->pPager)==1 ); + pBt->pPage1 = 0; + releasePageOne(pPage1); + } +} + +/* +** If pBt points to an empty file then convert that empty file +** into a new empty database by initializing the first page of +** the database. +*/ +static int newDatabase(BtShared *pBt){ + MemPage *pP1; + unsigned char *data; + int rc; + + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pBt->nPage>0 ){ + return SQLITE_OK; + } + pP1 = pBt->pPage1; + assert( pP1!=0 ); + data = pP1->aData; + rc = sqlite3PagerWrite(pP1->pDbPage); + if( rc ) return rc; + memcpy(data, zMagicHeader, sizeof(zMagicHeader)); + assert( sizeof(zMagicHeader)==16 ); + data[16] = (u8)((pBt->pageSize>>8)&0xff); + data[17] = (u8)((pBt->pageSize>>16)&0xff); + data[18] = 1; + data[19] = 1; + assert( pBt->usableSize<=pBt->pageSize && pBt->usableSize+255>=pBt->pageSize); + data[20] = (u8)(pBt->pageSize - pBt->usableSize); + data[21] = 64; + data[22] = 32; + data[23] = 32; + memset(&data[24], 0, 100-24); + zeroPage(pP1, PTF_INTKEY|PTF_LEAF|PTF_LEAFDATA ); + pBt->btsFlags |= BTS_PAGESIZE_FIXED; +#ifndef SQLITE_OMIT_AUTOVACUUM + assert( pBt->autoVacuum==1 || pBt->autoVacuum==0 ); + assert( pBt->incrVacuum==1 || pBt->incrVacuum==0 ); + put4byte(&data[36 + 4*4], pBt->autoVacuum); + put4byte(&data[36 + 7*4], pBt->incrVacuum); +#endif + pBt->nPage = 1; + data[31] = 1; + return SQLITE_OK; +} + +/* +** Initialize the first page of the database file (creating a database +** consisting of a single page and no schema objects). Return SQLITE_OK +** if successful, or an SQLite error code otherwise. +*/ +int sqlite3BtreeNewDb(Btree *p){ + int rc; + sqlite3BtreeEnter(p); + p->pBt->nPage = 0; + rc = newDatabase(p->pBt); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Attempt to start a new transaction. A write-transaction +** is started if the second argument is nonzero, otherwise a read- +** transaction. If the second argument is 2 or more and exclusive +** transaction is started, meaning that no other process is allowed +** to access the database. A preexisting transaction may not be +** upgraded to exclusive by calling this routine a second time - the +** exclusivity flag only works for a new transaction. +** +** A write-transaction must be started before attempting any +** changes to the database. None of the following routines +** will work unless a transaction is started first: +** +** sqlite3BtreeCreateTable() +** sqlite3BtreeCreateIndex() +** sqlite3BtreeClearTable() +** sqlite3BtreeDropTable() +** sqlite3BtreeInsert() +** sqlite3BtreeDelete() +** sqlite3BtreeUpdateMeta() +** +** If an initial attempt to acquire the lock fails because of lock contention +** and the database was previously unlocked, then invoke the busy handler +** if there is one. But if there was previously a read-lock, do not +** invoke the busy handler - just return SQLITE_BUSY. SQLITE_BUSY is +** returned when there is already a read-lock in order to avoid a deadlock. +** +** Suppose there are two processes A and B. A has a read lock and B has +** a reserved lock. B tries to promote to exclusive but is blocked because +** of A's read lock. A tries to promote to reserved but is blocked by B. +** One or the other of the two processes must give way or there can be +** no progress. By returning SQLITE_BUSY and not invoking the busy callback +** when A already has a read lock, we encourage A to give up and let B +** proceed. +*/ +int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ + BtShared *pBt = p->pBt; + Pager *pPager = pBt->pPager; + int rc = SQLITE_OK; + + sqlite3BtreeEnter(p); + btreeIntegrity(p); + + /* If the btree is already in a write-transaction, or it + ** is already in a read-transaction and a read-transaction + ** is requested, this is a no-op. + */ + if( p->inTrans==TRANS_WRITE || (p->inTrans==TRANS_READ && !wrflag) ){ + goto trans_begun; + } + assert( pBt->inTransaction==TRANS_WRITE || IfNotOmitAV(pBt->bDoTruncate)==0 ); + + if( (p->db->flags & SQLITE_ResetDatabase) + && sqlite3PagerIsreadonly(pPager)==0 + ){ + pBt->btsFlags &= ~BTS_READ_ONLY; + } + + /* Write transactions are not possible on a read-only database */ + if( (pBt->btsFlags & BTS_READ_ONLY)!=0 && wrflag ){ + rc = SQLITE_READONLY; + goto trans_begun; + } + +#ifndef SQLITE_OMIT_SHARED_CACHE + { + sqlite3 *pBlock = 0; + /* If another database handle has already opened a write transaction + ** on this shared-btree structure and a second write transaction is + ** requested, return SQLITE_LOCKED. + */ + if( (wrflag && pBt->inTransaction==TRANS_WRITE) + || (pBt->btsFlags & BTS_PENDING)!=0 + ){ + pBlock = pBt->pWriter->db; + }else if( wrflag>1 ){ + BtLock *pIter; + for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){ + if( pIter->pBtree!=p ){ + pBlock = pIter->pBtree->db; + break; + } + } + } + if( pBlock ){ + sqlite3ConnectionBlocked(p->db, pBlock); + rc = SQLITE_LOCKED_SHAREDCACHE; + goto trans_begun; + } + } +#endif + + /* Any read-only or read-write transaction implies a read-lock on + ** page 1. So if some other shared-cache client already has a write-lock + ** on page 1, the transaction cannot be opened. */ + rc = querySharedCacheTableLock(p, SCHEMA_ROOT, READ_LOCK); + if( SQLITE_OK!=rc ) goto trans_begun; + + pBt->btsFlags &= ~BTS_INITIALLY_EMPTY; + if( pBt->nPage==0 ) pBt->btsFlags |= BTS_INITIALLY_EMPTY; + do { + sqlite3PagerWalDb(pPager, p->db); + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + /* If transitioning from no transaction directly to a write transaction, + ** block for the WRITER lock first if possible. */ + if( pBt->pPage1==0 && wrflag ){ + assert( pBt->inTransaction==TRANS_NONE ); + rc = sqlite3PagerWalWriteLock(pPager, 1); + if( rc!=SQLITE_BUSY && rc!=SQLITE_OK ) break; + } +#endif + + /* Call lockBtree() until either pBt->pPage1 is populated or + ** lockBtree() returns something other than SQLITE_OK. lockBtree() + ** may return SQLITE_OK but leave pBt->pPage1 set to 0 if after + ** reading page 1 it discovers that the page-size of the database + ** file is not pBt->pageSize. In this case lockBtree() will update + ** pBt->pageSize to the page-size of the file on disk. + */ + while( pBt->pPage1==0 && SQLITE_OK==(rc = lockBtree(pBt)) ); + + if( rc==SQLITE_OK && wrflag ){ + if( (pBt->btsFlags & BTS_READ_ONLY)!=0 ){ + rc = SQLITE_READONLY; + }else{ + rc = sqlite3PagerBegin(pPager, wrflag>1, sqlite3TempInMemory(p->db)); + if( rc==SQLITE_OK ){ + rc = newDatabase(pBt); + }else if( rc==SQLITE_BUSY_SNAPSHOT && pBt->inTransaction==TRANS_NONE ){ + /* if there was no transaction opened when this function was + ** called and SQLITE_BUSY_SNAPSHOT is returned, change the error + ** code to SQLITE_BUSY. */ + rc = SQLITE_BUSY; + } + } + } + + if( rc!=SQLITE_OK ){ + (void)sqlite3PagerWalWriteLock(pPager, 0); + unlockBtreeIfUnused(pBt); + } + }while( (rc&0xFF)==SQLITE_BUSY && pBt->inTransaction==TRANS_NONE && + btreeInvokeBusyHandler(pBt) ); + sqlite3PagerWalDb(pPager, 0); +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + if( rc==SQLITE_BUSY_TIMEOUT ) rc = SQLITE_BUSY; +#endif + + if( rc==SQLITE_OK ){ + if( p->inTrans==TRANS_NONE ){ + pBt->nTransaction++; +#ifndef SQLITE_OMIT_SHARED_CACHE + if( p->sharable ){ + assert( p->lock.pBtree==p && p->lock.iTable==1 ); + p->lock.eLock = READ_LOCK; + p->lock.pNext = pBt->pLock; + pBt->pLock = &p->lock; + } +#endif + } + p->inTrans = (wrflag?TRANS_WRITE:TRANS_READ); + if( p->inTrans>pBt->inTransaction ){ + pBt->inTransaction = p->inTrans; + } + if( wrflag ){ + MemPage *pPage1 = pBt->pPage1; +#ifndef SQLITE_OMIT_SHARED_CACHE + assert( !pBt->pWriter ); + pBt->pWriter = p; + pBt->btsFlags &= ~BTS_EXCLUSIVE; + if( wrflag>1 ) pBt->btsFlags |= BTS_EXCLUSIVE; +#endif + + /* If the db-size header field is incorrect (as it may be if an old + ** client has been writing the database file), update it now. Doing + ** this sooner rather than later means the database size can safely + ** re-read the database size from page 1 if a savepoint or transaction + ** rollback occurs within the transaction. + */ + if( pBt->nPage!=get4byte(&pPage1->aData[28]) ){ + rc = sqlite3PagerWrite(pPage1->pDbPage); + if( rc==SQLITE_OK ){ + put4byte(&pPage1->aData[28], pBt->nPage); + } + } + } + } + +trans_begun: + if( rc==SQLITE_OK ){ + if( pSchemaVersion ){ + *pSchemaVersion = get4byte(&pBt->pPage1->aData[40]); + } + if( wrflag ){ + /* This call makes sure that the pager has the correct number of + ** open savepoints. If the second parameter is greater than 0 and + ** the sub-journal is not already open, then it will be opened here. + */ + rc = sqlite3PagerOpenSavepoint(pPager, p->db->nSavepoint); + } + } + + btreeIntegrity(p); + sqlite3BtreeLeave(p); + return rc; +} + +#ifndef SQLITE_OMIT_AUTOVACUUM + +/* +** Set the pointer-map entries for all children of page pPage. Also, if +** pPage contains cells that point to overflow pages, set the pointer +** map entries for the overflow pages as well. +*/ +static int setChildPtrmaps(MemPage *pPage){ + int i; /* Counter variable */ + int nCell; /* Number of cells in page pPage */ + int rc; /* Return code */ + BtShared *pBt = pPage->pBt; + Pgno pgno = pPage->pgno; + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + rc = pPage->isInit ? SQLITE_OK : btreeInitPage(pPage); + if( rc!=SQLITE_OK ) return rc; + nCell = pPage->nCell; + + for(i=0; ileaf ){ + Pgno childPgno = get4byte(pCell); + ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno, &rc); + } + } + + if( !pPage->leaf ){ + Pgno childPgno = get4byte(&pPage->aData[pPage->hdrOffset+8]); + ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno, &rc); + } + + return rc; +} + +/* +** Somewhere on pPage is a pointer to page iFrom. Modify this pointer so +** that it points to iTo. Parameter eType describes the type of pointer to +** be modified, as follows: +** +** PTRMAP_BTREE: pPage is a btree-page. The pointer points at a child +** page of pPage. +** +** PTRMAP_OVERFLOW1: pPage is a btree-page. The pointer points at an overflow +** page pointed to by one of the cells on pPage. +** +** PTRMAP_OVERFLOW2: pPage is an overflow-page. The pointer points at the next +** overflow page in the list. +*/ +static int modifyPagePointer(MemPage *pPage, Pgno iFrom, Pgno iTo, u8 eType){ + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + if( eType==PTRMAP_OVERFLOW2 ){ + /* The pointer is always the first 4 bytes of the page in this case. */ + if( get4byte(pPage->aData)!=iFrom ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + put4byte(pPage->aData, iTo); + }else{ + int i; + int nCell; + int rc; + + rc = pPage->isInit ? SQLITE_OK : btreeInitPage(pPage); + if( rc ) return rc; + nCell = pPage->nCell; + + for(i=0; ixParseCell(pPage, pCell, &info); + if( info.nLocal pPage->aData+pPage->pBt->usableSize ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + if( iFrom==get4byte(pCell+info.nSize-4) ){ + put4byte(pCell+info.nSize-4, iTo); + break; + } + } + }else{ + if( get4byte(pCell)==iFrom ){ + put4byte(pCell, iTo); + break; + } + } + } + + if( i==nCell ){ + if( eType!=PTRMAP_BTREE || + get4byte(&pPage->aData[pPage->hdrOffset+8])!=iFrom ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + put4byte(&pPage->aData[pPage->hdrOffset+8], iTo); + } + } + return SQLITE_OK; +} + + +/* +** Move the open database page pDbPage to location iFreePage in the +** database. The pDbPage reference remains valid. +** +** The isCommit flag indicates that there is no need to remember that +** the journal needs to be sync()ed before database page pDbPage->pgno +** can be written to. The caller has already promised not to write to that +** page. +*/ +static int relocatePage( + BtShared *pBt, /* Btree */ + MemPage *pDbPage, /* Open page to move */ + u8 eType, /* Pointer map 'type' entry for pDbPage */ + Pgno iPtrPage, /* Pointer map 'page-no' entry for pDbPage */ + Pgno iFreePage, /* The location to move pDbPage to */ + int isCommit /* isCommit flag passed to sqlite3PagerMovepage */ +){ + MemPage *pPtrPage; /* The page that contains a pointer to pDbPage */ + Pgno iDbPage = pDbPage->pgno; + Pager *pPager = pBt->pPager; + int rc; + + assert( eType==PTRMAP_OVERFLOW2 || eType==PTRMAP_OVERFLOW1 || + eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ); + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( pDbPage->pBt==pBt ); + if( iDbPage<3 ) return SQLITE_CORRUPT_BKPT; + + /* Move page iDbPage from its current location to page number iFreePage */ + TRACE(("AUTOVACUUM: Moving %d to free page %d (ptr page %d type %d)\n", + iDbPage, iFreePage, iPtrPage, eType)); + rc = sqlite3PagerMovepage(pPager, pDbPage->pDbPage, iFreePage, isCommit); + if( rc!=SQLITE_OK ){ + return rc; + } + pDbPage->pgno = iFreePage; + + /* If pDbPage was a btree-page, then it may have child pages and/or cells + ** that point to overflow pages. The pointer map entries for all these + ** pages need to be changed. + ** + ** If pDbPage is an overflow page, then the first 4 bytes may store a + ** pointer to a subsequent overflow page. If this is the case, then + ** the pointer map needs to be updated for the subsequent overflow page. + */ + if( eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ){ + rc = setChildPtrmaps(pDbPage); + if( rc!=SQLITE_OK ){ + return rc; + } + }else{ + Pgno nextOvfl = get4byte(pDbPage->aData); + if( nextOvfl!=0 ){ + ptrmapPut(pBt, nextOvfl, PTRMAP_OVERFLOW2, iFreePage, &rc); + if( rc!=SQLITE_OK ){ + return rc; + } + } + } + + /* Fix the database pointer on page iPtrPage that pointed at iDbPage so + ** that it points at iFreePage. Also fix the pointer map entry for + ** iPtrPage. + */ + if( eType!=PTRMAP_ROOTPAGE ){ + rc = btreeGetPage(pBt, iPtrPage, &pPtrPage, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = sqlite3PagerWrite(pPtrPage->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(pPtrPage); + return rc; + } + rc = modifyPagePointer(pPtrPage, iDbPage, iFreePage, eType); + releasePage(pPtrPage); + if( rc==SQLITE_OK ){ + ptrmapPut(pBt, iFreePage, eType, iPtrPage, &rc); + } + } + return rc; +} + +/* Forward declaration required by incrVacuumStep(). */ +static int allocateBtreePage(BtShared *, MemPage **, Pgno *, Pgno, u8); + +/* +** Perform a single step of an incremental-vacuum. If successful, return +** SQLITE_OK. If there is no work to do (and therefore no point in +** calling this function again), return SQLITE_DONE. Or, if an error +** occurs, return some other error code. +** +** More specifically, this function attempts to re-organize the database so +** that the last page of the file currently in use is no longer in use. +** +** Parameter nFin is the number of pages that this database would contain +** were this function called until it returns SQLITE_DONE. +** +** If the bCommit parameter is non-zero, this function assumes that the +** caller will keep calling incrVacuumStep() until it returns SQLITE_DONE +** or an error. bCommit is passed true for an auto-vacuum-on-commit +** operation, or false for an incremental vacuum. +*/ +static int incrVacuumStep(BtShared *pBt, Pgno nFin, Pgno iLastPg, int bCommit){ + Pgno nFreeList; /* Number of pages still on the free-list */ + int rc; + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( iLastPg>nFin ); + + if( !PTRMAP_ISPAGE(pBt, iLastPg) && iLastPg!=PENDING_BYTE_PAGE(pBt) ){ + u8 eType; + Pgno iPtrPage; + + nFreeList = get4byte(&pBt->pPage1->aData[36]); + if( nFreeList==0 ){ + return SQLITE_DONE; + } + + rc = ptrmapGet(pBt, iLastPg, &eType, &iPtrPage); + if( rc!=SQLITE_OK ){ + return rc; + } + if( eType==PTRMAP_ROOTPAGE ){ + return SQLITE_CORRUPT_BKPT; + } + + if( eType==PTRMAP_FREEPAGE ){ + if( bCommit==0 ){ + /* Remove the page from the files free-list. This is not required + ** if bCommit is non-zero. In that case, the free-list will be + ** truncated to zero after this function returns, so it doesn't + ** matter if it still contains some garbage entries. + */ + Pgno iFreePg; + MemPage *pFreePg; + rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iLastPg, BTALLOC_EXACT); + if( rc!=SQLITE_OK ){ + return rc; + } + assert( iFreePg==iLastPg ); + releasePage(pFreePg); + } + } else { + Pgno iFreePg; /* Index of free page to move pLastPg to */ + MemPage *pLastPg; + u8 eMode = BTALLOC_ANY; /* Mode parameter for allocateBtreePage() */ + Pgno iNear = 0; /* nearby parameter for allocateBtreePage() */ + + rc = btreeGetPage(pBt, iLastPg, &pLastPg, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* If bCommit is zero, this loop runs exactly once and page pLastPg + ** is swapped with the first free page pulled off the free list. + ** + ** On the other hand, if bCommit is greater than zero, then keep + ** looping until a free-page located within the first nFin pages + ** of the file is found. + */ + if( bCommit==0 ){ + eMode = BTALLOC_LE; + iNear = nFin; + } + do { + MemPage *pFreePg; + rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iNear, eMode); + if( rc!=SQLITE_OK ){ + releasePage(pLastPg); + return rc; + } + releasePage(pFreePg); + }while( bCommit && iFreePg>nFin ); + assert( iFreePgbDoTruncate = 1; + pBt->nPage = iLastPg; + } + return SQLITE_OK; +} + +/* +** The database opened by the first argument is an auto-vacuum database +** nOrig pages in size containing nFree free pages. Return the expected +** size of the database in pages following an auto-vacuum operation. +*/ +static Pgno finalDbSize(BtShared *pBt, Pgno nOrig, Pgno nFree){ + int nEntry; /* Number of entries on one ptrmap page */ + Pgno nPtrmap; /* Number of PtrMap pages to be freed */ + Pgno nFin; /* Return value */ + + nEntry = pBt->usableSize/5; + nPtrmap = (nFree-nOrig+PTRMAP_PAGENO(pBt, nOrig)+nEntry)/nEntry; + nFin = nOrig - nFree - nPtrmap; + if( nOrig>PENDING_BYTE_PAGE(pBt) && nFinpBt; + + sqlite3BtreeEnter(p); + assert( pBt->inTransaction==TRANS_WRITE && p->inTrans==TRANS_WRITE ); + if( !pBt->autoVacuum ){ + rc = SQLITE_DONE; + }else{ + Pgno nOrig = btreePagecount(pBt); + Pgno nFree = get4byte(&pBt->pPage1->aData[36]); + Pgno nFin = finalDbSize(pBt, nOrig, nFree); + + if( nOrig=nOrig ){ + rc = SQLITE_CORRUPT_BKPT; + }else if( nFree>0 ){ + rc = saveAllCursors(pBt, 0, 0); + if( rc==SQLITE_OK ){ + invalidateAllOverflowCache(pBt); + rc = incrVacuumStep(pBt, nFin, nOrig, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + put4byte(&pBt->pPage1->aData[28], pBt->nPage); + } + }else{ + rc = SQLITE_DONE; + } + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** This routine is called prior to sqlite3PagerCommit when a transaction +** is committed for an auto-vacuum database. +*/ +static int autoVacuumCommit(Btree *p){ + int rc = SQLITE_OK; + Pager *pPager; + BtShared *pBt; + sqlite3 *db; + VVA_ONLY( int nRef ); + + assert( p!=0 ); + pBt = p->pBt; + pPager = pBt->pPager; + VVA_ONLY( nRef = sqlite3PagerRefcount(pPager); ) + + assert( sqlite3_mutex_held(pBt->mutex) ); + invalidateAllOverflowCache(pBt); + assert(pBt->autoVacuum); + if( !pBt->incrVacuum ){ + Pgno nFin; /* Number of pages in database after autovacuuming */ + Pgno nFree; /* Number of pages on the freelist initially */ + Pgno nVac; /* Number of pages to vacuum */ + Pgno iFree; /* The next page to be freed */ + Pgno nOrig; /* Database size before freeing */ + + nOrig = btreePagecount(pBt); + if( PTRMAP_ISPAGE(pBt, nOrig) || nOrig==PENDING_BYTE_PAGE(pBt) ){ + /* It is not possible to create a database for which the final page + ** is either a pointer-map page or the pending-byte page. If one + ** is encountered, this indicates corruption. + */ + return SQLITE_CORRUPT_BKPT; + } + + nFree = get4byte(&pBt->pPage1->aData[36]); + db = p->db; + if( db->xAutovacPages ){ + int iDb; + for(iDb=0; ALWAYS(iDbnDb); iDb++){ + if( db->aDb[iDb].pBt==p ) break; + } + nVac = db->xAutovacPages( + db->pAutovacPagesArg, + db->aDb[iDb].zDbSName, + nOrig, + nFree, + pBt->pageSize + ); + if( nVac>nFree ){ + nVac = nFree; + } + if( nVac==0 ){ + return SQLITE_OK; + } + }else{ + nVac = nFree; + } + nFin = finalDbSize(pBt, nOrig, nVac); + if( nFin>nOrig ) return SQLITE_CORRUPT_BKPT; + if( nFinnFin && rc==SQLITE_OK; iFree--){ + rc = incrVacuumStep(pBt, nFin, iFree, nVac==nFree); + } + if( (rc==SQLITE_DONE || rc==SQLITE_OK) && nFree>0 ){ + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( nVac==nFree ){ + put4byte(&pBt->pPage1->aData[32], 0); + put4byte(&pBt->pPage1->aData[36], 0); + } + put4byte(&pBt->pPage1->aData[28], nFin); + pBt->bDoTruncate = 1; + pBt->nPage = nFin; + } + if( rc!=SQLITE_OK ){ + sqlite3PagerRollback(pPager); + } + } + + assert( nRef>=sqlite3PagerRefcount(pPager) ); + return rc; +} + +#else /* ifndef SQLITE_OMIT_AUTOVACUUM */ +# define setChildPtrmaps(x) SQLITE_OK +#endif + +/* +** This routine does the first phase of a two-phase commit. This routine +** causes a rollback journal to be created (if it does not already exist) +** and populated with enough information so that if a power loss occurs +** the database can be restored to its original state by playing back +** the journal. Then the contents of the journal are flushed out to +** the disk. After the journal is safely on oxide, the changes to the +** database are written into the database file and flushed to oxide. +** At the end of this call, the rollback journal still exists on the +** disk and we are still holding all locks, so the transaction has not +** committed. See sqlite3BtreeCommitPhaseTwo() for the second phase of the +** commit process. +** +** This call is a no-op if no write-transaction is currently active on pBt. +** +** Otherwise, sync the database file for the btree pBt. zSuperJrnl points to +** the name of a super-journal file that should be written into the +** individual journal file, or is NULL, indicating no super-journal file +** (single database transaction). +** +** When this is called, the super-journal should already have been +** created, populated with this journal pointer and synced to disk. +** +** Once this is routine has returned, the only thing required to commit +** the write-transaction for this database file is to delete the journal. +*/ +int sqlite3BtreeCommitPhaseOne(Btree *p, const char *zSuperJrnl){ + int rc = SQLITE_OK; + if( p->inTrans==TRANS_WRITE ){ + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + rc = autoVacuumCommit(p); + if( rc!=SQLITE_OK ){ + sqlite3BtreeLeave(p); + return rc; + } + } + if( pBt->bDoTruncate ){ + sqlite3PagerTruncateImage(pBt->pPager, pBt->nPage); + } +#endif + rc = sqlite3PagerCommitPhaseOne(pBt->pPager, zSuperJrnl, 0); + sqlite3BtreeLeave(p); + } + return rc; +} + +/* +** This function is called from both BtreeCommitPhaseTwo() and BtreeRollback() +** at the conclusion of a transaction. +*/ +static void btreeEndTransaction(Btree *p){ + BtShared *pBt = p->pBt; + sqlite3 *db = p->db; + assert( sqlite3BtreeHoldsMutex(p) ); + +#ifndef SQLITE_OMIT_AUTOVACUUM + pBt->bDoTruncate = 0; +#endif + if( p->inTrans>TRANS_NONE && db->nVdbeRead>1 ){ + /* If there are other active statements that belong to this database + ** handle, downgrade to a read-only transaction. The other statements + ** may still be reading from the database. */ + downgradeAllSharedCacheTableLocks(p); + p->inTrans = TRANS_READ; + }else{ + /* If the handle had any kind of transaction open, decrement the + ** transaction count of the shared btree. If the transaction count + ** reaches 0, set the shared state to TRANS_NONE. The unlockBtreeIfUnused() + ** call below will unlock the pager. */ + if( p->inTrans!=TRANS_NONE ){ + clearAllSharedCacheTableLocks(p); + pBt->nTransaction--; + if( 0==pBt->nTransaction ){ + pBt->inTransaction = TRANS_NONE; + } + } + + /* Set the current transaction state to TRANS_NONE and unlock the + ** pager if this call closed the only read or write transaction. */ + p->inTrans = TRANS_NONE; + unlockBtreeIfUnused(pBt); + } + + btreeIntegrity(p); +} + +/* +** Commit the transaction currently in progress. +** +** This routine implements the second phase of a 2-phase commit. The +** sqlite3BtreeCommitPhaseOne() routine does the first phase and should +** be invoked prior to calling this routine. The sqlite3BtreeCommitPhaseOne() +** routine did all the work of writing information out to disk and flushing the +** contents so that they are written onto the disk platter. All this +** routine has to do is delete or truncate or zero the header in the +** the rollback journal (which causes the transaction to commit) and +** drop locks. +** +** Normally, if an error occurs while the pager layer is attempting to +** finalize the underlying journal file, this function returns an error and +** the upper layer will attempt a rollback. However, if the second argument +** is non-zero then this b-tree transaction is part of a multi-file +** transaction. In this case, the transaction has already been committed +** (by deleting a super-journal file) and the caller will ignore this +** functions return code. So, even if an error occurs in the pager layer, +** reset the b-tree objects internal state to indicate that the write +** transaction has been closed. This is quite safe, as the pager will have +** transitioned to the error state. +** +** This will release the write lock on the database file. If there +** are no active cursors, it also releases the read lock. +*/ +int sqlite3BtreeCommitPhaseTwo(Btree *p, int bCleanup){ + + if( p->inTrans==TRANS_NONE ) return SQLITE_OK; + sqlite3BtreeEnter(p); + btreeIntegrity(p); + + /* If the handle has a write-transaction open, commit the shared-btrees + ** transaction and set the shared state to TRANS_READ. + */ + if( p->inTrans==TRANS_WRITE ){ + int rc; + BtShared *pBt = p->pBt; + assert( pBt->inTransaction==TRANS_WRITE ); + assert( pBt->nTransaction>0 ); + rc = sqlite3PagerCommitPhaseTwo(pBt->pPager); + if( rc!=SQLITE_OK && bCleanup==0 ){ + sqlite3BtreeLeave(p); + return rc; + } + p->iBDataVersion--; /* Compensate for pPager->iDataVersion++; */ + pBt->inTransaction = TRANS_READ; + btreeClearHasContent(pBt); + } + + btreeEndTransaction(p); + sqlite3BtreeLeave(p); + return SQLITE_OK; +} + +/* +** Do both phases of a commit. +*/ +int sqlite3BtreeCommit(Btree *p){ + int rc; + sqlite3BtreeEnter(p); + rc = sqlite3BtreeCommitPhaseOne(p, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3BtreeCommitPhaseTwo(p, 0); + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** This routine sets the state to CURSOR_FAULT and the error +** code to errCode for every cursor on any BtShared that pBtree +** references. Or if the writeOnly flag is set to 1, then only +** trip write cursors and leave read cursors unchanged. +** +** Every cursor is a candidate to be tripped, including cursors +** that belong to other database connections that happen to be +** sharing the cache with pBtree. +** +** This routine gets called when a rollback occurs. If the writeOnly +** flag is true, then only write-cursors need be tripped - read-only +** cursors save their current positions so that they may continue +** following the rollback. Or, if writeOnly is false, all cursors are +** tripped. In general, writeOnly is false if the transaction being +** rolled back modified the database schema. In this case b-tree root +** pages may be moved or deleted from the database altogether, making +** it unsafe for read cursors to continue. +** +** If the writeOnly flag is true and an error is encountered while +** saving the current position of a read-only cursor, all cursors, +** including all read-cursors are tripped. +** +** SQLITE_OK is returned if successful, or if an error occurs while +** saving a cursor position, an SQLite error code. +*/ +int sqlite3BtreeTripAllCursors(Btree *pBtree, int errCode, int writeOnly){ + BtCursor *p; + int rc = SQLITE_OK; + + assert( (writeOnly==0 || writeOnly==1) && BTCF_WriteFlag==1 ); + if( pBtree ){ + sqlite3BtreeEnter(pBtree); + for(p=pBtree->pBt->pCursor; p; p=p->pNext){ + if( writeOnly && (p->curFlags & BTCF_WriteFlag)==0 ){ + if( p->eState==CURSOR_VALID || p->eState==CURSOR_SKIPNEXT ){ + rc = saveCursorPosition(p); + if( rc!=SQLITE_OK ){ + (void)sqlite3BtreeTripAllCursors(pBtree, rc, 0); + break; + } + } + }else{ + sqlite3BtreeClearCursor(p); + p->eState = CURSOR_FAULT; + p->skipNext = errCode; + } + btreeReleaseAllCursorPages(p); + } + sqlite3BtreeLeave(pBtree); + } + return rc; +} + +/* +** Set the pBt->nPage field correctly, according to the current +** state of the database. Assume pBt->pPage1 is valid. +*/ +static void btreeSetNPage(BtShared *pBt, MemPage *pPage1){ + int nPage = get4byte(&pPage1->aData[28]); + testcase( nPage==0 ); + if( nPage==0 ) sqlite3PagerPagecount(pBt->pPager, &nPage); + testcase( pBt->nPage!=nPage ); + pBt->nPage = nPage; +} + +/* +** Rollback the transaction in progress. +** +** If tripCode is not SQLITE_OK then cursors will be invalidated (tripped). +** Only write cursors are tripped if writeOnly is true but all cursors are +** tripped if writeOnly is false. Any attempt to use +** a tripped cursor will result in an error. +** +** This will release the write lock on the database file. If there +** are no active cursors, it also releases the read lock. +*/ +int sqlite3BtreeRollback(Btree *p, int tripCode, int writeOnly){ + int rc; + BtShared *pBt = p->pBt; + MemPage *pPage1; + + assert( writeOnly==1 || writeOnly==0 ); + assert( tripCode==SQLITE_ABORT_ROLLBACK || tripCode==SQLITE_OK ); + sqlite3BtreeEnter(p); + if( tripCode==SQLITE_OK ){ + rc = tripCode = saveAllCursors(pBt, 0, 0); + if( rc ) writeOnly = 0; + }else{ + rc = SQLITE_OK; + } + if( tripCode ){ + int rc2 = sqlite3BtreeTripAllCursors(p, tripCode, writeOnly); + assert( rc==SQLITE_OK || (writeOnly==0 && rc2==SQLITE_OK) ); + if( rc2!=SQLITE_OK ) rc = rc2; + } + btreeIntegrity(p); + + if( p->inTrans==TRANS_WRITE ){ + int rc2; + + assert( TRANS_WRITE==pBt->inTransaction ); + rc2 = sqlite3PagerRollback(pBt->pPager); + if( rc2!=SQLITE_OK ){ + rc = rc2; + } + + /* The rollback may have destroyed the pPage1->aData value. So + ** call btreeGetPage() on page 1 again to make + ** sure pPage1->aData is set correctly. */ + if( btreeGetPage(pBt, 1, &pPage1, 0)==SQLITE_OK ){ + btreeSetNPage(pBt, pPage1); + releasePageOne(pPage1); + } + assert( countValidCursors(pBt, 1)==0 ); + pBt->inTransaction = TRANS_READ; + btreeClearHasContent(pBt); + } + + btreeEndTransaction(p); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Start a statement subtransaction. The subtransaction can be rolled +** back independently of the main transaction. You must start a transaction +** before starting a subtransaction. The subtransaction is ended automatically +** if the main transaction commits or rolls back. +** +** Statement subtransactions are used around individual SQL statements +** that are contained within a BEGIN...COMMIT block. If a constraint +** error occurs within the statement, the effect of that one statement +** can be rolled back without having to rollback the entire transaction. +** +** A statement sub-transaction is implemented as an anonymous savepoint. The +** value passed as the second parameter is the total number of savepoints, +** including the new anonymous savepoint, open on the B-Tree. i.e. if there +** are no active savepoints and no other statement-transactions open, +** iStatement is 1. This anonymous savepoint can be released or rolled back +** using the sqlite3BtreeSavepoint() function. +*/ +int sqlite3BtreeBeginStmt(Btree *p, int iStatement){ + int rc; + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + assert( p->inTrans==TRANS_WRITE ); + assert( (pBt->btsFlags & BTS_READ_ONLY)==0 ); + assert( iStatement>0 ); + assert( iStatement>p->db->nSavepoint ); + assert( pBt->inTransaction==TRANS_WRITE ); + /* At the pager level, a statement transaction is a savepoint with + ** an index greater than all savepoints created explicitly using + ** SQL statements. It is illegal to open, release or rollback any + ** such savepoints while the statement transaction savepoint is active. + */ + rc = sqlite3PagerOpenSavepoint(pBt->pPager, iStatement); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** The second argument to this function, op, is always SAVEPOINT_ROLLBACK +** or SAVEPOINT_RELEASE. This function either releases or rolls back the +** savepoint identified by parameter iSavepoint, depending on the value +** of op. +** +** Normally, iSavepoint is greater than or equal to zero. However, if op is +** SAVEPOINT_ROLLBACK, then iSavepoint may also be -1. In this case the +** contents of the entire transaction are rolled back. This is different +** from a normal transaction rollback, as no locks are released and the +** transaction remains open. +*/ +int sqlite3BtreeSavepoint(Btree *p, int op, int iSavepoint){ + int rc = SQLITE_OK; + if( p && p->inTrans==TRANS_WRITE ){ + BtShared *pBt = p->pBt; + assert( op==SAVEPOINT_RELEASE || op==SAVEPOINT_ROLLBACK ); + assert( iSavepoint>=0 || (iSavepoint==-1 && op==SAVEPOINT_ROLLBACK) ); + sqlite3BtreeEnter(p); + if( op==SAVEPOINT_ROLLBACK ){ + rc = saveAllCursors(pBt, 0, 0); + } + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSavepoint(pBt->pPager, op, iSavepoint); + } + if( rc==SQLITE_OK ){ + if( iSavepoint<0 && (pBt->btsFlags & BTS_INITIALLY_EMPTY)!=0 ){ + pBt->nPage = 0; + } + rc = newDatabase(pBt); + btreeSetNPage(pBt, pBt->pPage1); + + /* pBt->nPage might be zero if the database was corrupt when + ** the transaction was started. Otherwise, it must be at least 1. */ + assert( CORRUPT_DB || pBt->nPage>0 ); + } + sqlite3BtreeLeave(p); + } + return rc; +} + +/* +** Create a new cursor for the BTree whose root is on the page +** iTable. If a read-only cursor is requested, it is assumed that +** the caller already has at least a read-only transaction open +** on the database already. If a write-cursor is requested, then +** the caller is assumed to have an open write transaction. +** +** If the BTREE_WRCSR bit of wrFlag is clear, then the cursor can only +** be used for reading. If the BTREE_WRCSR bit is set, then the cursor +** can be used for reading or for writing if other conditions for writing +** are also met. These are the conditions that must be met in order +** for writing to be allowed: +** +** 1: The cursor must have been opened with wrFlag containing BTREE_WRCSR +** +** 2: Other database connections that share the same pager cache +** but which are not in the READ_UNCOMMITTED state may not have +** cursors open with wrFlag==0 on the same table. Otherwise +** the changes made by this write cursor would be visible to +** the read cursors in the other database connection. +** +** 3: The database must be writable (not on read-only media) +** +** 4: There must be an active transaction. +** +** The BTREE_FORDELETE bit of wrFlag may optionally be set if BTREE_WRCSR +** is set. If FORDELETE is set, that is a hint to the implementation that +** this cursor will only be used to seek to and delete entries of an index +** as part of a larger DELETE statement. The FORDELETE hint is not used by +** this implementation. But in a hypothetical alternative storage engine +** in which index entries are automatically deleted when corresponding table +** rows are deleted, the FORDELETE flag is a hint that all SEEK and DELETE +** operations on this cursor can be no-ops and all READ operations can +** return a null row (2-bytes: 0x01 0x00). +** +** No checking is done to make sure that page iTable really is the +** root page of a b-tree. If it is not, then the cursor acquired +** will not work correctly. +** +** It is assumed that the sqlite3BtreeCursorZero() has been called +** on pCur to initialize the memory space prior to invoking this routine. +*/ +static int btreeCursor( + Btree *p, /* The btree */ + Pgno iTable, /* Root page of table to open */ + int wrFlag, /* 1 to write. 0 read-only */ + struct KeyInfo *pKeyInfo, /* First arg to comparison function */ + BtCursor *pCur /* Space for new cursor */ +){ + BtShared *pBt = p->pBt; /* Shared b-tree handle */ + BtCursor *pX; /* Looping over other all cursors */ + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( wrFlag==0 + || wrFlag==BTREE_WRCSR + || wrFlag==(BTREE_WRCSR|BTREE_FORDELETE) + ); + + /* The following assert statements verify that if this is a sharable + ** b-tree database, the connection is holding the required table locks, + ** and that no other connection has any open cursor that conflicts with + ** this lock. The iTable<1 term disables the check for corrupt schemas. */ + assert( hasSharedCacheTableLock(p, iTable, pKeyInfo!=0, (wrFlag?2:1)) + || iTable<1 ); + assert( wrFlag==0 || !hasReadConflicts(p, iTable) ); + + /* Assert that the caller has opened the required transaction. */ + assert( p->inTrans>TRANS_NONE ); + assert( wrFlag==0 || p->inTrans==TRANS_WRITE ); + assert( pBt->pPage1 && pBt->pPage1->aData ); + assert( wrFlag==0 || (pBt->btsFlags & BTS_READ_ONLY)==0 ); + + if( wrFlag ){ + allocateTempSpace(pBt); + if( pBt->pTmpSpace==0 ) return SQLITE_NOMEM_BKPT; + } + if( iTable<=1 ){ + if( iTable<1 ){ + return SQLITE_CORRUPT_BKPT; + }else if( btreePagecount(pBt)==0 ){ + assert( wrFlag==0 ); + iTable = 0; + } + } + + /* Now that no other errors can occur, finish filling in the BtCursor + ** variables and link the cursor into the BtShared list. */ + pCur->pgnoRoot = iTable; + pCur->iPage = -1; + pCur->pKeyInfo = pKeyInfo; + pCur->pBtree = p; + pCur->pBt = pBt; + pCur->curFlags = wrFlag ? BTCF_WriteFlag : 0; + pCur->curPagerFlags = wrFlag ? 0 : PAGER_GET_READONLY; + /* If there are two or more cursors on the same btree, then all such + ** cursors *must* have the BTCF_Multiple flag set. */ + for(pX=pBt->pCursor; pX; pX=pX->pNext){ + if( pX->pgnoRoot==iTable ){ + pX->curFlags |= BTCF_Multiple; + pCur->curFlags |= BTCF_Multiple; + } + } + pCur->pNext = pBt->pCursor; + pBt->pCursor = pCur; + pCur->eState = CURSOR_INVALID; + return SQLITE_OK; +} +static int btreeCursorWithLock( + Btree *p, /* The btree */ + Pgno iTable, /* Root page of table to open */ + int wrFlag, /* 1 to write. 0 read-only */ + struct KeyInfo *pKeyInfo, /* First arg to comparison function */ + BtCursor *pCur /* Space for new cursor */ +){ + int rc; + sqlite3BtreeEnter(p); + rc = btreeCursor(p, iTable, wrFlag, pKeyInfo, pCur); + sqlite3BtreeLeave(p); + return rc; +} +int sqlite3BtreeCursor( + Btree *p, /* The btree */ + Pgno iTable, /* Root page of table to open */ + int wrFlag, /* 1 to write. 0 read-only */ + struct KeyInfo *pKeyInfo, /* First arg to xCompare() */ + BtCursor *pCur /* Write new cursor here */ +){ + if( p->sharable ){ + return btreeCursorWithLock(p, iTable, wrFlag, pKeyInfo, pCur); + }else{ + return btreeCursor(p, iTable, wrFlag, pKeyInfo, pCur); + } +} + +/* +** Return the size of a BtCursor object in bytes. +** +** This interfaces is needed so that users of cursors can preallocate +** sufficient storage to hold a cursor. The BtCursor object is opaque +** to users so they cannot do the sizeof() themselves - they must call +** this routine. +*/ +int sqlite3BtreeCursorSize(void){ + return ROUND8(sizeof(BtCursor)); +} + +/* +** Initialize memory that will be converted into a BtCursor object. +** +** The simple approach here would be to memset() the entire object +** to zero. But it turns out that the apPage[] and aiIdx[] arrays +** do not need to be zeroed and they are large, so we can save a lot +** of run-time by skipping the initialization of those elements. +*/ +void sqlite3BtreeCursorZero(BtCursor *p){ + memset(p, 0, offsetof(BtCursor, BTCURSOR_FIRST_UNINIT)); +} + +/* +** Close a cursor. The read lock on the database file is released +** when the last cursor is closed. +*/ +int sqlite3BtreeCloseCursor(BtCursor *pCur){ + Btree *pBtree = pCur->pBtree; + if( pBtree ){ + BtShared *pBt = pCur->pBt; + sqlite3BtreeEnter(pBtree); + assert( pBt->pCursor!=0 ); + if( pBt->pCursor==pCur ){ + pBt->pCursor = pCur->pNext; + }else{ + BtCursor *pPrev = pBt->pCursor; + do{ + if( pPrev->pNext==pCur ){ + pPrev->pNext = pCur->pNext; + break; + } + pPrev = pPrev->pNext; + }while( ALWAYS(pPrev) ); + } + btreeReleaseAllCursorPages(pCur); + unlockBtreeIfUnused(pBt); + sqlite3_free(pCur->aOverflow); + sqlite3_free(pCur->pKey); + if( (pBt->openFlags & BTREE_SINGLE) && pBt->pCursor==0 ){ + /* Since the BtShared is not sharable, there is no need to + ** worry about the missing sqlite3BtreeLeave() call here. */ + assert( pBtree->sharable==0 ); + sqlite3BtreeClose(pBtree); + }else{ + sqlite3BtreeLeave(pBtree); + } + pCur->pBtree = 0; + } + return SQLITE_OK; +} + +/* +** Make sure the BtCursor* given in the argument has a valid +** BtCursor.info structure. If it is not already valid, call +** btreeParseCell() to fill it in. +** +** BtCursor.info is a cache of the information in the current cell. +** Using this cache reduces the number of calls to btreeParseCell(). +*/ +#ifndef NDEBUG + static int cellInfoEqual(CellInfo *a, CellInfo *b){ + if( a->nKey!=b->nKey ) return 0; + if( a->pPayload!=b->pPayload ) return 0; + if( a->nPayload!=b->nPayload ) return 0; + if( a->nLocal!=b->nLocal ) return 0; + if( a->nSize!=b->nSize ) return 0; + return 1; + } + static void assertCellInfo(BtCursor *pCur){ + CellInfo info; + memset(&info, 0, sizeof(info)); + btreeParseCell(pCur->pPage, pCur->ix, &info); + assert( CORRUPT_DB || cellInfoEqual(&info, &pCur->info) ); + } +#else + #define assertCellInfo(x) +#endif +static SQLITE_NOINLINE void getCellInfo(BtCursor *pCur){ + if( pCur->info.nSize==0 ){ + pCur->curFlags |= BTCF_ValidNKey; + btreeParseCell(pCur->pPage,pCur->ix,&pCur->info); + }else{ + assertCellInfo(pCur); + } +} + +#ifndef NDEBUG /* The next routine used only within assert() statements */ +/* +** Return true if the given BtCursor is valid. A valid cursor is one +** that is currently pointing to a row in a (non-empty) table. +** This is a verification routine is used only within assert() statements. +*/ +int sqlite3BtreeCursorIsValid(BtCursor *pCur){ + return pCur && pCur->eState==CURSOR_VALID; +} +#endif /* NDEBUG */ +int sqlite3BtreeCursorIsValidNN(BtCursor *pCur){ + assert( pCur!=0 ); + return pCur->eState==CURSOR_VALID; +} + +/* +** Return the value of the integer key or "rowid" for a table btree. +** This routine is only valid for a cursor that is pointing into a +** ordinary table btree. If the cursor points to an index btree or +** is invalid, the result of this routine is undefined. +*/ +i64 sqlite3BtreeIntegerKey(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->curIntKey ); + getCellInfo(pCur); + return pCur->info.nKey; +} + +/* +** Pin or unpin a cursor. +*/ +void sqlite3BtreeCursorPin(BtCursor *pCur){ + assert( (pCur->curFlags & BTCF_Pinned)==0 ); + pCur->curFlags |= BTCF_Pinned; +} +void sqlite3BtreeCursorUnpin(BtCursor *pCur){ + assert( (pCur->curFlags & BTCF_Pinned)!=0 ); + pCur->curFlags &= ~BTCF_Pinned; +} + +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC +/* +** Return the offset into the database file for the start of the +** payload to which the cursor is pointing. +*/ +i64 sqlite3BtreeOffset(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + getCellInfo(pCur); + return (i64)pCur->pBt->pageSize*((i64)pCur->pPage->pgno - 1) + + (i64)(pCur->info.pPayload - pCur->pPage->aData); +} +#endif /* SQLITE_ENABLE_OFFSET_SQL_FUNC */ + +/* +** Return the number of bytes of payload for the entry that pCur is +** currently pointing to. For table btrees, this will be the amount +** of data. For index btrees, this will be the size of the key. +** +** The caller must guarantee that the cursor is pointing to a non-NULL +** valid entry. In other words, the calling procedure must guarantee +** that the cursor has Cursor.eState==CURSOR_VALID. +*/ +u32 sqlite3BtreePayloadSize(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + getCellInfo(pCur); + return pCur->info.nPayload; +} + +/* +** Return an upper bound on the size of any record for the table +** that the cursor is pointing into. +** +** This is an optimization. Everything will still work if this +** routine always returns 2147483647 (which is the largest record +** that SQLite can handle) or more. But returning a smaller value might +** prevent large memory allocations when trying to interpret a +** corrupt datrabase. +** +** The current implementation merely returns the size of the underlying +** database file. +*/ +sqlite3_int64 sqlite3BtreeMaxRecordSize(BtCursor *pCur){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + return pCur->pBt->pageSize * (sqlite3_int64)pCur->pBt->nPage; +} + +/* +** Given the page number of an overflow page in the database (parameter +** ovfl), this function finds the page number of the next page in the +** linked list of overflow pages. If possible, it uses the auto-vacuum +** pointer-map data instead of reading the content of page ovfl to do so. +** +** If an error occurs an SQLite error code is returned. Otherwise: +** +** The page number of the next overflow page in the linked list is +** written to *pPgnoNext. If page ovfl is the last page in its linked +** list, *pPgnoNext is set to zero. +** +** If ppPage is not NULL, and a reference to the MemPage object corresponding +** to page number pOvfl was obtained, then *ppPage is set to point to that +** reference. It is the responsibility of the caller to call releasePage() +** on *ppPage to free the reference. In no reference was obtained (because +** the pointer-map was used to obtain the value for *pPgnoNext), then +** *ppPage is set to zero. +*/ +static int getOverflowPage( + BtShared *pBt, /* The database file */ + Pgno ovfl, /* Current overflow page number */ + MemPage **ppPage, /* OUT: MemPage handle (may be NULL) */ + Pgno *pPgnoNext /* OUT: Next overflow page number */ +){ + Pgno next = 0; + MemPage *pPage = 0; + int rc = SQLITE_OK; + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert(pPgnoNext); + +#ifndef SQLITE_OMIT_AUTOVACUUM + /* Try to find the next page in the overflow list using the + ** autovacuum pointer-map pages. Guess that the next page in + ** the overflow list is page number (ovfl+1). If that guess turns + ** out to be wrong, fall back to loading the data of page + ** number ovfl to determine the next page number. + */ + if( pBt->autoVacuum ){ + Pgno pgno; + Pgno iGuess = ovfl+1; + u8 eType; + + while( PTRMAP_ISPAGE(pBt, iGuess) || iGuess==PENDING_BYTE_PAGE(pBt) ){ + iGuess++; + } + + if( iGuess<=btreePagecount(pBt) ){ + rc = ptrmapGet(pBt, iGuess, &eType, &pgno); + if( rc==SQLITE_OK && eType==PTRMAP_OVERFLOW2 && pgno==ovfl ){ + next = iGuess; + rc = SQLITE_DONE; + } + } + } +#endif + + assert( next==0 || rc==SQLITE_DONE ); + if( rc==SQLITE_OK ){ + rc = btreeGetPage(pBt, ovfl, &pPage, (ppPage==0) ? PAGER_GET_READONLY : 0); + assert( rc==SQLITE_OK || pPage==0 ); + if( rc==SQLITE_OK ){ + next = get4byte(pPage->aData); + } + } + + *pPgnoNext = next; + if( ppPage ){ + *ppPage = pPage; + }else{ + releasePage(pPage); + } + return (rc==SQLITE_DONE ? SQLITE_OK : rc); +} + +/* +** Copy data from a buffer to a page, or from a page to a buffer. +** +** pPayload is a pointer to data stored on database page pDbPage. +** If argument eOp is false, then nByte bytes of data are copied +** from pPayload to the buffer pointed at by pBuf. If eOp is true, +** then sqlite3PagerWrite() is called on pDbPage and nByte bytes +** of data are copied from the buffer pBuf to pPayload. +** +** SQLITE_OK is returned on success, otherwise an error code. +*/ +static int copyPayload( + void *pPayload, /* Pointer to page data */ + void *pBuf, /* Pointer to buffer */ + int nByte, /* Number of bytes to copy */ + int eOp, /* 0 -> copy from page, 1 -> copy to page */ + DbPage *pDbPage /* Page containing pPayload */ +){ + if( eOp ){ + /* Copy data from buffer to page (a write operation) */ + int rc = sqlite3PagerWrite(pDbPage); + if( rc!=SQLITE_OK ){ + return rc; + } + memcpy(pPayload, pBuf, nByte); + }else{ + /* Copy data from page to buffer (a read operation) */ + memcpy(pBuf, pPayload, nByte); + } + return SQLITE_OK; +} + +/* +** This function is used to read or overwrite payload information +** for the entry that the pCur cursor is pointing to. The eOp +** argument is interpreted as follows: +** +** 0: The operation is a read. Populate the overflow cache. +** 1: The operation is a write. Populate the overflow cache. +** +** A total of "amt" bytes are read or written beginning at "offset". +** Data is read to or from the buffer pBuf. +** +** The content being read or written might appear on the main page +** or be scattered out on multiple overflow pages. +** +** If the current cursor entry uses one or more overflow pages +** this function may allocate space for and lazily populate +** the overflow page-list cache array (BtCursor.aOverflow). +** Subsequent calls use this cache to make seeking to the supplied offset +** more efficient. +** +** Once an overflow page-list cache has been allocated, it must be +** invalidated if some other cursor writes to the same table, or if +** the cursor is moved to a different row. Additionally, in auto-vacuum +** mode, the following events may invalidate an overflow page-list cache. +** +** * An incremental vacuum, +** * A commit in auto_vacuum="full" mode, +** * Creating a table (may require moving an overflow page). +*/ +static int accessPayload( + BtCursor *pCur, /* Cursor pointing to entry to read from */ + u32 offset, /* Begin reading this far into payload */ + u32 amt, /* Read this many bytes */ + unsigned char *pBuf, /* Write the bytes into this buffer */ + int eOp /* zero to read. non-zero to write. */ +){ + unsigned char *aPayload; + int rc = SQLITE_OK; + int iIdx = 0; + MemPage *pPage = pCur->pPage; /* Btree page of current entry */ + BtShared *pBt = pCur->pBt; /* Btree this cursor belongs to */ +#ifdef SQLITE_DIRECT_OVERFLOW_READ + unsigned char * const pBufStart = pBuf; /* Start of original out buffer */ +#endif + + assert( pPage ); + assert( eOp==0 || eOp==1 ); + assert( pCur->eState==CURSOR_VALID ); + if( pCur->ix>=pPage->nCell ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + assert( cursorHoldsMutex(pCur) ); + + getCellInfo(pCur); + aPayload = pCur->info.pPayload; + assert( offset+amt <= pCur->info.nPayload ); + + assert( aPayload > pPage->aData ); + if( (uptr)(aPayload - pPage->aData) > (pBt->usableSize - pCur->info.nLocal) ){ + /* Trying to read or write past the end of the data is an error. The + ** conditional above is really: + ** &aPayload[pCur->info.nLocal] > &pPage->aData[pBt->usableSize] + ** but is recast into its current form to avoid integer overflow problems + */ + return SQLITE_CORRUPT_PAGE(pPage); + } + + /* Check if data must be read/written to/from the btree page itself. */ + if( offsetinfo.nLocal ){ + int a = amt; + if( a+offset>pCur->info.nLocal ){ + a = pCur->info.nLocal - offset; + } + rc = copyPayload(&aPayload[offset], pBuf, a, eOp, pPage->pDbPage); + offset = 0; + pBuf += a; + amt -= a; + }else{ + offset -= pCur->info.nLocal; + } + + + if( rc==SQLITE_OK && amt>0 ){ + const u32 ovflSize = pBt->usableSize - 4; /* Bytes content per ovfl page */ + Pgno nextPage; + + nextPage = get4byte(&aPayload[pCur->info.nLocal]); + + /* If the BtCursor.aOverflow[] has not been allocated, allocate it now. + ** + ** The aOverflow[] array is sized at one entry for each overflow page + ** in the overflow chain. The page number of the first overflow page is + ** stored in aOverflow[0], etc. A value of 0 in the aOverflow[] array + ** means "not yet known" (the cache is lazily populated). + */ + if( (pCur->curFlags & BTCF_ValidOvfl)==0 ){ + int nOvfl = (pCur->info.nPayload-pCur->info.nLocal+ovflSize-1)/ovflSize; + if( pCur->aOverflow==0 + || nOvfl*(int)sizeof(Pgno) > sqlite3MallocSize(pCur->aOverflow) + ){ + Pgno *aNew = (Pgno*)sqlite3Realloc( + pCur->aOverflow, nOvfl*2*sizeof(Pgno) + ); + if( aNew==0 ){ + return SQLITE_NOMEM_BKPT; + }else{ + pCur->aOverflow = aNew; + } + } + memset(pCur->aOverflow, 0, nOvfl*sizeof(Pgno)); + pCur->curFlags |= BTCF_ValidOvfl; + }else{ + /* If the overflow page-list cache has been allocated and the + ** entry for the first required overflow page is valid, skip + ** directly to it. + */ + if( pCur->aOverflow[offset/ovflSize] ){ + iIdx = (offset/ovflSize); + nextPage = pCur->aOverflow[iIdx]; + offset = (offset%ovflSize); + } + } + + assert( rc==SQLITE_OK && amt>0 ); + while( nextPage ){ + /* If required, populate the overflow page-list cache. */ + if( nextPage > pBt->nPage ) return SQLITE_CORRUPT_BKPT; + assert( pCur->aOverflow[iIdx]==0 + || pCur->aOverflow[iIdx]==nextPage + || CORRUPT_DB ); + pCur->aOverflow[iIdx] = nextPage; + + if( offset>=ovflSize ){ + /* The only reason to read this page is to obtain the page + ** number for the next page in the overflow chain. The page + ** data is not required. So first try to lookup the overflow + ** page-list cache, if any, then fall back to the getOverflowPage() + ** function. + */ + assert( pCur->curFlags & BTCF_ValidOvfl ); + assert( pCur->pBtree->db==pBt->db ); + if( pCur->aOverflow[iIdx+1] ){ + nextPage = pCur->aOverflow[iIdx+1]; + }else{ + rc = getOverflowPage(pBt, nextPage, 0, &nextPage); + } + offset -= ovflSize; + }else{ + /* Need to read this page properly. It contains some of the + ** range of data that is being read (eOp==0) or written (eOp!=0). + */ + int a = amt; + if( a + offset > ovflSize ){ + a = ovflSize - offset; + } + +#ifdef SQLITE_DIRECT_OVERFLOW_READ + /* If all the following are true: + ** + ** 1) this is a read operation, and + ** 2) data is required from the start of this overflow page, and + ** 3) there are no dirty pages in the page-cache + ** 4) the database is file-backed, and + ** 5) the page is not in the WAL file + ** 6) at least 4 bytes have already been read into the output buffer + ** + ** then data can be read directly from the database file into the + ** output buffer, bypassing the page-cache altogether. This speeds + ** up loading large records that span many overflow pages. + */ + if( eOp==0 /* (1) */ + && offset==0 /* (2) */ + && sqlite3PagerDirectReadOk(pBt->pPager, nextPage) /* (3,4,5) */ + && &pBuf[-4]>=pBufStart /* (6) */ + ){ + sqlite3_file *fd = sqlite3PagerFile(pBt->pPager); + u8 aSave[4]; + u8 *aWrite = &pBuf[-4]; + assert( aWrite>=pBufStart ); /* due to (6) */ + memcpy(aSave, aWrite, 4); + rc = sqlite3OsRead(fd, aWrite, a+4, (i64)pBt->pageSize*(nextPage-1)); + if( rc && nextPage>pBt->nPage ) rc = SQLITE_CORRUPT_BKPT; + nextPage = get4byte(aWrite); + memcpy(aWrite, aSave, 4); + }else +#endif + + { + DbPage *pDbPage; + rc = sqlite3PagerGet(pBt->pPager, nextPage, &pDbPage, + (eOp==0 ? PAGER_GET_READONLY : 0) + ); + if( rc==SQLITE_OK ){ + aPayload = sqlite3PagerGetData(pDbPage); + nextPage = get4byte(aPayload); + rc = copyPayload(&aPayload[offset+4], pBuf, a, eOp, pDbPage); + sqlite3PagerUnref(pDbPage); + offset = 0; + } + } + amt -= a; + if( amt==0 ) return rc; + pBuf += a; + } + if( rc ) break; + iIdx++; + } + } + + if( rc==SQLITE_OK && amt>0 ){ + /* Overflow chain ends prematurely */ + return SQLITE_CORRUPT_PAGE(pPage); + } + return rc; +} + +/* +** Read part of the payload for the row at which that cursor pCur is currently +** pointing. "amt" bytes will be transferred into pBuf[]. The transfer +** begins at "offset". +** +** pCur can be pointing to either a table or an index b-tree. +** If pointing to a table btree, then the content section is read. If +** pCur is pointing to an index b-tree then the key section is read. +** +** For sqlite3BtreePayload(), the caller must ensure that pCur is pointing +** to a valid row in the table. For sqlite3BtreePayloadChecked(), the +** cursor might be invalid or might need to be restored before being read. +** +** Return SQLITE_OK on success or an error code if anything goes +** wrong. An error is returned if "offset+amt" is larger than +** the available payload. +*/ +int sqlite3BtreePayload(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){ + assert( cursorHoldsMutex(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->iPage>=0 && pCur->pPage ); + return accessPayload(pCur, offset, amt, (unsigned char*)pBuf, 0); +} + +/* +** This variant of sqlite3BtreePayload() works even if the cursor has not +** in the CURSOR_VALID state. It is only used by the sqlite3_blob_read() +** interface. +*/ +#ifndef SQLITE_OMIT_INCRBLOB +static SQLITE_NOINLINE int accessPayloadChecked( + BtCursor *pCur, + u32 offset, + u32 amt, + void *pBuf +){ + int rc; + if ( pCur->eState==CURSOR_INVALID ){ + return SQLITE_ABORT; + } + assert( cursorOwnsBtShared(pCur) ); + rc = btreeRestoreCursorPosition(pCur); + return rc ? rc : accessPayload(pCur, offset, amt, pBuf, 0); +} +int sqlite3BtreePayloadChecked(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){ + if( pCur->eState==CURSOR_VALID ){ + assert( cursorOwnsBtShared(pCur) ); + return accessPayload(pCur, offset, amt, pBuf, 0); + }else{ + return accessPayloadChecked(pCur, offset, amt, pBuf); + } +} +#endif /* SQLITE_OMIT_INCRBLOB */ + +/* +** Return a pointer to payload information from the entry that the +** pCur cursor is pointing to. The pointer is to the beginning of +** the key if index btrees (pPage->intKey==0) and is the data for +** table btrees (pPage->intKey==1). The number of bytes of available +** key/data is written into *pAmt. If *pAmt==0, then the value +** returned will not be a valid pointer. +** +** This routine is an optimization. It is common for the entire key +** and data to fit on the local page and for there to be no overflow +** pages. When that is so, this routine can be used to access the +** key and data without making a copy. If the key and/or data spills +** onto overflow pages, then accessPayload() must be used to reassemble +** the key/data and copy it into a preallocated buffer. +** +** The pointer returned by this routine looks directly into the cached +** page of the database. The data might change or move the next time +** any btree routine is called. +*/ +static const void *fetchPayload( + BtCursor *pCur, /* Cursor pointing to entry to read from */ + u32 *pAmt /* Write the number of available bytes here */ +){ + int amt; + assert( pCur!=0 && pCur->iPage>=0 && pCur->pPage); + assert( pCur->eState==CURSOR_VALID ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->ixpPage->nCell || CORRUPT_DB ); + assert( pCur->info.nSize>0 ); + assert( pCur->info.pPayload>pCur->pPage->aData || CORRUPT_DB ); + assert( pCur->info.pPayloadpPage->aDataEnd ||CORRUPT_DB); + amt = pCur->info.nLocal; + if( amt>(int)(pCur->pPage->aDataEnd - pCur->info.pPayload) ){ + /* There is too little space on the page for the expected amount + ** of local content. Database must be corrupt. */ + assert( CORRUPT_DB ); + amt = MAX(0, (int)(pCur->pPage->aDataEnd - pCur->info.pPayload)); + } + *pAmt = (u32)amt; + return (void*)pCur->info.pPayload; +} + + +/* +** For the entry that cursor pCur is point to, return as +** many bytes of the key or data as are available on the local +** b-tree page. Write the number of available bytes into *pAmt. +** +** The pointer returned is ephemeral. The key/data may move +** or be destroyed on the next call to any Btree routine, +** including calls from other threads against the same cache. +** Hence, a mutex on the BtShared should be held prior to calling +** this routine. +** +** These routines is used to get quick access to key and data +** in the common case where no overflow pages are used. +*/ +const void *sqlite3BtreePayloadFetch(BtCursor *pCur, u32 *pAmt){ + return fetchPayload(pCur, pAmt); +} + + +/* +** Move the cursor down to a new child page. The newPgno argument is the +** page number of the child page to move to. +** +** This function returns SQLITE_CORRUPT if the page-header flags field of +** the new child page does not match the flags field of the parent (i.e. +** if an intkey page appears to be the parent of a non-intkey page, or +** vice-versa). +*/ +static int moveToChild(BtCursor *pCur, u32 newPgno){ + BtShared *pBt = pCur->pBt; + + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->iPageiPage>=0 ); + if( pCur->iPage>=(BTCURSOR_MAX_DEPTH-1) ){ + return SQLITE_CORRUPT_BKPT; + } + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl); + pCur->aiIdx[pCur->iPage] = pCur->ix; + pCur->apPage[pCur->iPage] = pCur->pPage; + pCur->ix = 0; + pCur->iPage++; + return getAndInitPage(pBt, newPgno, &pCur->pPage, pCur, pCur->curPagerFlags); +} + +#ifdef SQLITE_DEBUG +/* +** Page pParent is an internal (non-leaf) tree page. This function +** asserts that page number iChild is the left-child if the iIdx'th +** cell in page pParent. Or, if iIdx is equal to the total number of +** cells in pParent, that page number iChild is the right-child of +** the page. +*/ +static void assertParentIndex(MemPage *pParent, int iIdx, Pgno iChild){ + if( CORRUPT_DB ) return; /* The conditions tested below might not be true + ** in a corrupt database */ + assert( iIdx<=pParent->nCell ); + if( iIdx==pParent->nCell ){ + assert( get4byte(&pParent->aData[pParent->hdrOffset+8])==iChild ); + }else{ + assert( get4byte(findCell(pParent, iIdx))==iChild ); + } +} +#else +# define assertParentIndex(x,y,z) +#endif + +/* +** Move the cursor up to the parent page. +** +** pCur->idx is set to the cell index that contains the pointer +** to the page we are coming from. If we are coming from the +** right-most child page then pCur->idx is set to one more than +** the largest cell index. +*/ +static void moveToParent(BtCursor *pCur){ + MemPage *pLeaf; + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->iPage>0 ); + assert( pCur->pPage ); + assertParentIndex( + pCur->apPage[pCur->iPage-1], + pCur->aiIdx[pCur->iPage-1], + pCur->pPage->pgno + ); + testcase( pCur->aiIdx[pCur->iPage-1] > pCur->apPage[pCur->iPage-1]->nCell ); + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl); + pCur->ix = pCur->aiIdx[pCur->iPage-1]; + pLeaf = pCur->pPage; + pCur->pPage = pCur->apPage[--pCur->iPage]; + releasePageNotNull(pLeaf); +} + +/* +** Move the cursor to point to the root page of its b-tree structure. +** +** If the table has a virtual root page, then the cursor is moved to point +** to the virtual root page instead of the actual root page. A table has a +** virtual root page when the actual root page contains no cells and a +** single child page. This can only happen with the table rooted at page 1. +** +** If the b-tree structure is empty, the cursor state is set to +** CURSOR_INVALID and this routine returns SQLITE_EMPTY. Otherwise, +** the cursor is set to point to the first cell located on the root +** (or virtual root) page and the cursor state is set to CURSOR_VALID. +** +** If this function returns successfully, it may be assumed that the +** page-header flags indicate that the [virtual] root-page is the expected +** kind of b-tree page (i.e. if when opening the cursor the caller did not +** specify a KeyInfo structure the flags byte is set to 0x05 or 0x0D, +** indicating a table b-tree, or if the caller did specify a KeyInfo +** structure the flags byte is set to 0x02 or 0x0A, indicating an index +** b-tree). +*/ +static int moveToRoot(BtCursor *pCur){ + MemPage *pRoot; + int rc = SQLITE_OK; + + assert( cursorOwnsBtShared(pCur) ); + assert( CURSOR_INVALID < CURSOR_REQUIRESEEK ); + assert( CURSOR_VALID < CURSOR_REQUIRESEEK ); + assert( CURSOR_FAULT > CURSOR_REQUIRESEEK ); + assert( pCur->eState < CURSOR_REQUIRESEEK || pCur->iPage<0 ); + assert( pCur->pgnoRoot>0 || pCur->iPage<0 ); + + if( pCur->iPage>=0 ){ + if( pCur->iPage ){ + releasePageNotNull(pCur->pPage); + while( --pCur->iPage ){ + releasePageNotNull(pCur->apPage[pCur->iPage]); + } + pCur->pPage = pCur->apPage[0]; + goto skip_init; + } + }else if( pCur->pgnoRoot==0 ){ + pCur->eState = CURSOR_INVALID; + return SQLITE_EMPTY; + }else{ + assert( pCur->iPage==(-1) ); + if( pCur->eState>=CURSOR_REQUIRESEEK ){ + if( pCur->eState==CURSOR_FAULT ){ + assert( pCur->skipNext!=SQLITE_OK ); + return pCur->skipNext; + } + sqlite3BtreeClearCursor(pCur); + } + rc = getAndInitPage(pCur->pBtree->pBt, pCur->pgnoRoot, &pCur->pPage, + 0, pCur->curPagerFlags); + if( rc!=SQLITE_OK ){ + pCur->eState = CURSOR_INVALID; + return rc; + } + pCur->iPage = 0; + pCur->curIntKey = pCur->pPage->intKey; + } + pRoot = pCur->pPage; + assert( pRoot->pgno==pCur->pgnoRoot ); + + /* If pCur->pKeyInfo is not NULL, then the caller that opened this cursor + ** expected to open it on an index b-tree. Otherwise, if pKeyInfo is + ** NULL, the caller expects a table b-tree. If this is not the case, + ** return an SQLITE_CORRUPT error. + ** + ** Earlier versions of SQLite assumed that this test could not fail + ** if the root page was already loaded when this function was called (i.e. + ** if pCur->iPage>=0). But this is not so if the database is corrupted + ** in such a way that page pRoot is linked into a second b-tree table + ** (or the freelist). */ + assert( pRoot->intKey==1 || pRoot->intKey==0 ); + if( pRoot->isInit==0 || (pCur->pKeyInfo==0)!=pRoot->intKey ){ + return SQLITE_CORRUPT_PAGE(pCur->pPage); + } + +skip_init: + pCur->ix = 0; + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidNKey|BTCF_ValidOvfl); + + pRoot = pCur->pPage; + if( pRoot->nCell>0 ){ + pCur->eState = CURSOR_VALID; + }else if( !pRoot->leaf ){ + Pgno subpage; + if( pRoot->pgno!=1 ) return SQLITE_CORRUPT_BKPT; + subpage = get4byte(&pRoot->aData[pRoot->hdrOffset+8]); + pCur->eState = CURSOR_VALID; + rc = moveToChild(pCur, subpage); + }else{ + pCur->eState = CURSOR_INVALID; + rc = SQLITE_EMPTY; + } + return rc; +} + +/* +** Move the cursor down to the left-most leaf entry beneath the +** entry to which it is currently pointing. +** +** The left-most leaf is the one with the smallest key - the first +** in ascending order. +*/ +static int moveToLeftmost(BtCursor *pCur){ + Pgno pgno; + int rc = SQLITE_OK; + MemPage *pPage; + + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + while( rc==SQLITE_OK && !(pPage = pCur->pPage)->leaf ){ + assert( pCur->ixnCell ); + pgno = get4byte(findCell(pPage, pCur->ix)); + rc = moveToChild(pCur, pgno); + } + return rc; +} + +/* +** Move the cursor down to the right-most leaf entry beneath the +** page to which it is currently pointing. Notice the difference +** between moveToLeftmost() and moveToRightmost(). moveToLeftmost() +** finds the left-most entry beneath the *entry* whereas moveToRightmost() +** finds the right-most entry beneath the *page*. +** +** The right-most entry is the one with the largest key - the last +** key in ascending order. +*/ +static int moveToRightmost(BtCursor *pCur){ + Pgno pgno; + int rc = SQLITE_OK; + MemPage *pPage = 0; + + assert( cursorOwnsBtShared(pCur) ); + assert( pCur->eState==CURSOR_VALID ); + while( !(pPage = pCur->pPage)->leaf ){ + pgno = get4byte(&pPage->aData[pPage->hdrOffset+8]); + pCur->ix = pPage->nCell; + rc = moveToChild(pCur, pgno); + if( rc ) return rc; + } + pCur->ix = pPage->nCell-1; + assert( pCur->info.nSize==0 ); + assert( (pCur->curFlags & BTCF_ValidNKey)==0 ); + return SQLITE_OK; +} + +/* Move the cursor to the first entry in the table. Return SQLITE_OK +** on success. Set *pRes to 0 if the cursor actually points to something +** or set *pRes to 1 if the table is empty. +*/ +int sqlite3BtreeFirst(BtCursor *pCur, int *pRes){ + int rc; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + rc = moveToRoot(pCur); + if( rc==SQLITE_OK ){ + assert( pCur->pPage->nCell>0 ); + *pRes = 0; + rc = moveToLeftmost(pCur); + }else if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = 1; + rc = SQLITE_OK; + } + return rc; +} + +/* Move the cursor to the last entry in the table. Return SQLITE_OK +** on success. Set *pRes to 0 if the cursor actually points to something +** or set *pRes to 1 if the table is empty. +*/ +int sqlite3BtreeLast(BtCursor *pCur, int *pRes){ + int rc; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + + /* If the cursor already points to the last entry, this is a no-op. */ + if( CURSOR_VALID==pCur->eState && (pCur->curFlags & BTCF_AtLast)!=0 ){ +#ifdef SQLITE_DEBUG + /* This block serves to assert() that the cursor really does point + ** to the last entry in the b-tree. */ + int ii; + for(ii=0; iiiPage; ii++){ + assert( pCur->aiIdx[ii]==pCur->apPage[ii]->nCell ); + } + assert( pCur->ix==pCur->pPage->nCell-1 || CORRUPT_DB ); + testcase( pCur->ix!=pCur->pPage->nCell-1 ); + /* ^-- dbsqlfuzz b92b72e4de80b5140c30ab71372ca719b8feb618 */ + assert( pCur->pPage->leaf ); +#endif + *pRes = 0; + return SQLITE_OK; + } + + rc = moveToRoot(pCur); + if( rc==SQLITE_OK ){ + assert( pCur->eState==CURSOR_VALID ); + *pRes = 0; + rc = moveToRightmost(pCur); + if( rc==SQLITE_OK ){ + pCur->curFlags |= BTCF_AtLast; + }else{ + pCur->curFlags &= ~BTCF_AtLast; + } + }else if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = 1; + rc = SQLITE_OK; + } + return rc; +} + +/* Move the cursor so that it points to an entry in a table (a.k.a INTKEY) +** table near the key intKey. Return a success code. +** +** If an exact match is not found, then the cursor is always +** left pointing at a leaf page which would hold the entry if it +** were present. The cursor might point to an entry that comes +** before or after the key. +** +** An integer is written into *pRes which is the result of +** comparing the key with the entry to which the cursor is +** pointing. The meaning of the integer written into +** *pRes is as follows: +** +** *pRes<0 The cursor is left pointing at an entry that +** is smaller than intKey or if the table is empty +** and the cursor is therefore left point to nothing. +** +** *pRes==0 The cursor is left pointing at an entry that +** exactly matches intKey. +** +** *pRes>0 The cursor is left pointing at an entry that +** is larger than intKey. +*/ +int sqlite3BtreeTableMoveto( + BtCursor *pCur, /* The cursor to be moved */ + i64 intKey, /* The table key */ + int biasRight, /* If true, bias the search to the high end */ + int *pRes /* Write search results here */ +){ + int rc; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + assert( pRes ); + assert( pCur->pKeyInfo==0 ); + assert( pCur->eState!=CURSOR_VALID || pCur->curIntKey!=0 ); + + /* If the cursor is already positioned at the point we are trying + ** to move to, then just return without doing any work */ + if( pCur->eState==CURSOR_VALID && (pCur->curFlags & BTCF_ValidNKey)!=0 ){ + if( pCur->info.nKey==intKey ){ + *pRes = 0; + return SQLITE_OK; + } + if( pCur->info.nKeycurFlags & BTCF_AtLast)!=0 ){ + *pRes = -1; + return SQLITE_OK; + } + /* If the requested key is one more than the previous key, then + ** try to get there using sqlite3BtreeNext() rather than a full + ** binary search. This is an optimization only. The correct answer + ** is still obtained without this case, only a little more slowely */ + if( pCur->info.nKey+1==intKey ){ + *pRes = 0; + rc = sqlite3BtreeNext(pCur, 0); + if( rc==SQLITE_OK ){ + getCellInfo(pCur); + if( pCur->info.nKey==intKey ){ + return SQLITE_OK; + } + }else if( rc!=SQLITE_DONE ){ + return rc; + } + } + } + } + +#ifdef SQLITE_DEBUG + pCur->pBtree->nSeek++; /* Performance measurement during testing */ +#endif + + rc = moveToRoot(pCur); + if( rc ){ + if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = -1; + return SQLITE_OK; + } + return rc; + } + assert( pCur->pPage ); + assert( pCur->pPage->isInit ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->pPage->nCell > 0 ); + assert( pCur->iPage==0 || pCur->apPage[0]->intKey==pCur->curIntKey ); + assert( pCur->curIntKey ); + + for(;;){ + int lwr, upr, idx, c; + Pgno chldPg; + MemPage *pPage = pCur->pPage; + u8 *pCell; /* Pointer to current cell in pPage */ + + /* pPage->nCell must be greater than zero. If this is the root-page + ** the cursor would have been INVALID above and this for(;;) loop + ** not run. If this is not the root-page, then the moveToChild() routine + ** would have already detected db corruption. Similarly, pPage must + ** be the right kind (index or table) of b-tree page. Otherwise + ** a moveToChild() or moveToRoot() call would have detected corruption. */ + assert( pPage->nCell>0 ); + assert( pPage->intKey ); + lwr = 0; + upr = pPage->nCell-1; + assert( biasRight==0 || biasRight==1 ); + idx = upr>>(1-biasRight); /* idx = biasRight ? upr : (lwr+upr)/2; */ + pCur->ix = (u16)idx; + for(;;){ + i64 nCellKey; + pCell = findCellPastPtr(pPage, idx); + if( pPage->intKeyLeaf ){ + while( 0x80 <= *(pCell++) ){ + if( pCell>=pPage->aDataEnd ){ + return SQLITE_CORRUPT_PAGE(pPage); + } + } + } + getVarint(pCell, (u64*)&nCellKey); + if( nCellKeyupr ){ c = -1; break; } + }else if( nCellKey>intKey ){ + upr = idx-1; + if( lwr>upr ){ c = +1; break; } + }else{ + assert( nCellKey==intKey ); + pCur->ix = (u16)idx; + if( !pPage->leaf ){ + lwr = idx; + goto moveto_table_next_layer; + }else{ + pCur->curFlags |= BTCF_ValidNKey; + pCur->info.nKey = nCellKey; + pCur->info.nSize = 0; + *pRes = 0; + return SQLITE_OK; + } + } + assert( lwr+upr>=0 ); + idx = (lwr+upr)>>1; /* idx = (lwr+upr)/2; */ + } + assert( lwr==upr+1 || !pPage->leaf ); + assert( pPage->isInit ); + if( pPage->leaf ){ + assert( pCur->ixpPage->nCell ); + pCur->ix = (u16)idx; + *pRes = c; + rc = SQLITE_OK; + goto moveto_table_finish; + } +moveto_table_next_layer: + if( lwr>=pPage->nCell ){ + chldPg = get4byte(&pPage->aData[pPage->hdrOffset+8]); + }else{ + chldPg = get4byte(findCell(pPage, lwr)); + } + pCur->ix = (u16)lwr; + rc = moveToChild(pCur, chldPg); + if( rc ) break; + } +moveto_table_finish: + pCur->info.nSize = 0; + assert( (pCur->curFlags & BTCF_ValidOvfl)==0 ); + return rc; +} + +/* Move the cursor so that it points to an entry in an index table +** near the key pIdxKey. Return a success code. +** +** If an exact match is not found, then the cursor is always +** left pointing at a leaf page which would hold the entry if it +** were present. The cursor might point to an entry that comes +** before or after the key. +** +** An integer is written into *pRes which is the result of +** comparing the key with the entry to which the cursor is +** pointing. The meaning of the integer written into +** *pRes is as follows: +** +** *pRes<0 The cursor is left pointing at an entry that +** is smaller than pIdxKey or if the table is empty +** and the cursor is therefore left point to nothing. +** +** *pRes==0 The cursor is left pointing at an entry that +** exactly matches pIdxKey. +** +** *pRes>0 The cursor is left pointing at an entry that +** is larger than pIdxKey. +** +** The pIdxKey->eqSeen field is set to 1 if there +** exists an entry in the table that exactly matches pIdxKey. +*/ +int sqlite3BtreeIndexMoveto( + BtCursor *pCur, /* The cursor to be moved */ + UnpackedRecord *pIdxKey, /* Unpacked index key */ + int *pRes /* Write search results here */ +){ + int rc; + RecordCompare xRecordCompare; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + assert( pRes ); + assert( pCur->pKeyInfo!=0 ); + +#ifdef SQLITE_DEBUG + pCur->pBtree->nSeek++; /* Performance measurement during testing */ +#endif + + xRecordCompare = sqlite3VdbeFindCompare(pIdxKey); + pIdxKey->errCode = 0; + assert( pIdxKey->default_rc==1 + || pIdxKey->default_rc==0 + || pIdxKey->default_rc==-1 + ); + + rc = moveToRoot(pCur); + if( rc ){ + if( rc==SQLITE_EMPTY ){ + assert( pCur->pgnoRoot==0 || pCur->pPage->nCell==0 ); + *pRes = -1; + return SQLITE_OK; + } + return rc; + } + assert( pCur->pPage ); + assert( pCur->pPage->isInit ); + assert( pCur->eState==CURSOR_VALID ); + assert( pCur->pPage->nCell > 0 ); + assert( pCur->iPage==0 || pCur->apPage[0]->intKey==pCur->curIntKey ); + assert( pCur->curIntKey || pIdxKey ); + for(;;){ + int lwr, upr, idx, c; + Pgno chldPg; + MemPage *pPage = pCur->pPage; + u8 *pCell; /* Pointer to current cell in pPage */ + + /* pPage->nCell must be greater than zero. If this is the root-page + ** the cursor would have been INVALID above and this for(;;) loop + ** not run. If this is not the root-page, then the moveToChild() routine + ** would have already detected db corruption. Similarly, pPage must + ** be the right kind (index or table) of b-tree page. Otherwise + ** a moveToChild() or moveToRoot() call would have detected corruption. */ + assert( pPage->nCell>0 ); + assert( pPage->intKey==(pIdxKey==0) ); + lwr = 0; + upr = pPage->nCell-1; + idx = upr>>1; /* idx = (lwr+upr)/2; */ + pCur->ix = (u16)idx; + for(;;){ + int nCell; /* Size of the pCell cell in bytes */ + pCell = findCellPastPtr(pPage, idx); + + /* The maximum supported page-size is 65536 bytes. This means that + ** the maximum number of record bytes stored on an index B-Tree + ** page is less than 16384 bytes and may be stored as a 2-byte + ** varint. This information is used to attempt to avoid parsing + ** the entire cell by checking for the cases where the record is + ** stored entirely within the b-tree page by inspecting the first + ** 2 bytes of the cell. + */ + nCell = pCell[0]; + if( nCell<=pPage->max1bytePayload ){ + /* This branch runs if the record-size field of the cell is a + ** single byte varint and the record fits entirely on the main + ** b-tree page. */ + testcase( pCell+nCell+1==pPage->aDataEnd ); + c = xRecordCompare(nCell, (void*)&pCell[1], pIdxKey); + }else if( !(pCell[1] & 0x80) + && (nCell = ((nCell&0x7f)<<7) + pCell[1])<=pPage->maxLocal + ){ + /* The record-size field is a 2 byte varint and the record + ** fits entirely on the main b-tree page. */ + testcase( pCell+nCell+2==pPage->aDataEnd ); + c = xRecordCompare(nCell, (void*)&pCell[2], pIdxKey); + }else{ + /* The record flows over onto one or more overflow pages. In + ** this case the whole cell needs to be parsed, a buffer allocated + ** and accessPayload() used to retrieve the record into the + ** buffer before VdbeRecordCompare() can be called. + ** + ** If the record is corrupt, the xRecordCompare routine may read + ** up to two varints past the end of the buffer. An extra 18 + ** bytes of padding is allocated at the end of the buffer in + ** case this happens. */ + void *pCellKey; + u8 * const pCellBody = pCell - pPage->childPtrSize; + const int nOverrun = 18; /* Size of the overrun padding */ + pPage->xParseCell(pPage, pCellBody, &pCur->info); + nCell = (int)pCur->info.nKey; + testcase( nCell<0 ); /* True if key size is 2^32 or more */ + testcase( nCell==0 ); /* Invalid key size: 0x80 0x80 0x00 */ + testcase( nCell==1 ); /* Invalid key size: 0x80 0x80 0x01 */ + testcase( nCell==2 ); /* Minimum legal index key size */ + if( nCell<2 || nCell/pCur->pBt->usableSize>pCur->pBt->nPage ){ + rc = SQLITE_CORRUPT_PAGE(pPage); + goto moveto_index_finish; + } + pCellKey = sqlite3Malloc( nCell+nOverrun ); + if( pCellKey==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto moveto_index_finish; + } + pCur->ix = (u16)idx; + rc = accessPayload(pCur, 0, nCell, (unsigned char*)pCellKey, 0); + memset(((u8*)pCellKey)+nCell,0,nOverrun); /* Fix uninit warnings */ + pCur->curFlags &= ~BTCF_ValidOvfl; + if( rc ){ + sqlite3_free(pCellKey); + goto moveto_index_finish; + } + c = sqlite3VdbeRecordCompare(nCell, pCellKey, pIdxKey); + sqlite3_free(pCellKey); + } + assert( + (pIdxKey->errCode!=SQLITE_CORRUPT || c==0) + && (pIdxKey->errCode!=SQLITE_NOMEM || pCur->pBtree->db->mallocFailed) + ); + if( c<0 ){ + lwr = idx+1; + }else if( c>0 ){ + upr = idx-1; + }else{ + assert( c==0 ); + *pRes = 0; + rc = SQLITE_OK; + pCur->ix = (u16)idx; + if( pIdxKey->errCode ) rc = SQLITE_CORRUPT_BKPT; + goto moveto_index_finish; + } + if( lwr>upr ) break; + assert( lwr+upr>=0 ); + idx = (lwr+upr)>>1; /* idx = (lwr+upr)/2 */ + } + assert( lwr==upr+1 || (pPage->intKey && !pPage->leaf) ); + assert( pPage->isInit ); + if( pPage->leaf ){ + assert( pCur->ixpPage->nCell ); + pCur->ix = (u16)idx; + *pRes = c; + rc = SQLITE_OK; + goto moveto_index_finish; + } + if( lwr>=pPage->nCell ){ + chldPg = get4byte(&pPage->aData[pPage->hdrOffset+8]); + }else{ + chldPg = get4byte(findCell(pPage, lwr)); + } + pCur->ix = (u16)lwr; + rc = moveToChild(pCur, chldPg); + if( rc ) break; + } +moveto_index_finish: + pCur->info.nSize = 0; + assert( (pCur->curFlags & BTCF_ValidOvfl)==0 ); + return rc; +} + + +/* +** Return TRUE if the cursor is not pointing at an entry of the table. +** +** TRUE will be returned after a call to sqlite3BtreeNext() moves +** past the last entry in the table or sqlite3BtreePrev() moves past +** the first entry. TRUE is also returned if the table is empty. +*/ +int sqlite3BtreeEof(BtCursor *pCur){ + /* TODO: What if the cursor is in CURSOR_REQUIRESEEK but all table entries + ** have been deleted? This API will need to change to return an error code + ** as well as the boolean result value. + */ + return (CURSOR_VALID!=pCur->eState); +} + +/* +** Return an estimate for the number of rows in the table that pCur is +** pointing to. Return a negative number if no estimate is currently +** available. +*/ +i64 sqlite3BtreeRowCountEst(BtCursor *pCur){ + i64 n; + u8 i; + + assert( cursorOwnsBtShared(pCur) ); + assert( sqlite3_mutex_held(pCur->pBtree->db->mutex) ); + + /* Currently this interface is only called by the OP_IfSmaller + ** opcode, and it that case the cursor will always be valid and + ** will always point to a leaf node. */ + if( NEVER(pCur->eState!=CURSOR_VALID) ) return -1; + if( NEVER(pCur->pPage->leaf==0) ) return -1; + + n = pCur->pPage->nCell; + for(i=0; iiPage; i++){ + n *= pCur->apPage[i]->nCell; + } + return n; +} + +/* +** Advance the cursor to the next entry in the database. +** Return value: +** +** SQLITE_OK success +** SQLITE_DONE cursor is already pointing at the last element +** otherwise some kind of error occurred +** +** The main entry point is sqlite3BtreeNext(). That routine is optimized +** for the common case of merely incrementing the cell counter BtCursor.aiIdx +** to the next cell on the current page. The (slower) btreeNext() helper +** routine is called when it is necessary to move to a different page or +** to restore the cursor. +** +** If bit 0x01 of the F argument in sqlite3BtreeNext(C,F) is 1, then the +** cursor corresponds to an SQL index and this routine could have been +** skipped if the SQL index had been a unique index. The F argument +** is a hint to the implement. SQLite btree implementation does not use +** this hint, but COMDB2 does. +*/ +static SQLITE_NOINLINE int btreeNext(BtCursor *pCur){ + int rc; + int idx; + MemPage *pPage; + + assert( cursorOwnsBtShared(pCur) ); + if( pCur->eState!=CURSOR_VALID ){ + assert( (pCur->curFlags & BTCF_ValidOvfl)==0 ); + rc = restoreCursorPosition(pCur); + if( rc!=SQLITE_OK ){ + return rc; + } + if( CURSOR_INVALID==pCur->eState ){ + return SQLITE_DONE; + } + if( pCur->eState==CURSOR_SKIPNEXT ){ + pCur->eState = CURSOR_VALID; + if( pCur->skipNext>0 ) return SQLITE_OK; + } + } + + pPage = pCur->pPage; + idx = ++pCur->ix; + if( !pPage->isInit || sqlite3FaultSim(412) ){ + /* The only known way for this to happen is for there to be a + ** recursive SQL function that does a DELETE operation as part of a + ** SELECT which deletes content out from under an active cursor + ** in a corrupt database file where the table being DELETE-ed from + ** has pages in common with the table being queried. See TH3 + ** module cov1/btree78.test testcase 220 (2018-06-08) for an + ** example. */ + return SQLITE_CORRUPT_BKPT; + } + + if( idx>=pPage->nCell ){ + if( !pPage->leaf ){ + rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8])); + if( rc ) return rc; + return moveToLeftmost(pCur); + } + do{ + if( pCur->iPage==0 ){ + pCur->eState = CURSOR_INVALID; + return SQLITE_DONE; + } + moveToParent(pCur); + pPage = pCur->pPage; + }while( pCur->ix>=pPage->nCell ); + if( pPage->intKey ){ + return sqlite3BtreeNext(pCur, 0); + }else{ + return SQLITE_OK; + } + } + if( pPage->leaf ){ + return SQLITE_OK; + }else{ + return moveToLeftmost(pCur); + } +} +int sqlite3BtreeNext(BtCursor *pCur, int flags){ + MemPage *pPage; + UNUSED_PARAMETER( flags ); /* Used in COMDB2 but not native SQLite */ + assert( cursorOwnsBtShared(pCur) ); + assert( flags==0 || flags==1 ); + pCur->info.nSize = 0; + pCur->curFlags &= ~(BTCF_ValidNKey|BTCF_ValidOvfl); + if( pCur->eState!=CURSOR_VALID ) return btreeNext(pCur); + pPage = pCur->pPage; + if( (++pCur->ix)>=pPage->nCell ){ + pCur->ix--; + return btreeNext(pCur); + } + if( pPage->leaf ){ + return SQLITE_OK; + }else{ + return moveToLeftmost(pCur); + } +} + +/* +** Step the cursor to the back to the previous entry in the database. +** Return values: +** +** SQLITE_OK success +** SQLITE_DONE the cursor is already on the first element of the table +** otherwise some kind of error occurred +** +** The main entry point is sqlite3BtreePrevious(). That routine is optimized +** for the common case of merely decrementing the cell counter BtCursor.aiIdx +** to the previous cell on the current page. The (slower) btreePrevious() +** helper routine is called when it is necessary to move to a different page +** or to restore the cursor. +** +** If bit 0x01 of the F argument to sqlite3BtreePrevious(C,F) is 1, then +** the cursor corresponds to an SQL index and this routine could have been +** skipped if the SQL index had been a unique index. The F argument is a +** hint to the implement. The native SQLite btree implementation does not +** use this hint, but COMDB2 does. +*/ +static SQLITE_NOINLINE int btreePrevious(BtCursor *pCur){ + int rc; + MemPage *pPage; + + assert( cursorOwnsBtShared(pCur) ); + assert( (pCur->curFlags & (BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey))==0 ); + assert( pCur->info.nSize==0 ); + if( pCur->eState!=CURSOR_VALID ){ + rc = restoreCursorPosition(pCur); + if( rc!=SQLITE_OK ){ + return rc; + } + if( CURSOR_INVALID==pCur->eState ){ + return SQLITE_DONE; + } + if( CURSOR_SKIPNEXT==pCur->eState ){ + pCur->eState = CURSOR_VALID; + if( pCur->skipNext<0 ) return SQLITE_OK; + } + } + + pPage = pCur->pPage; + assert( pPage->isInit ); + if( !pPage->leaf ){ + int idx = pCur->ix; + rc = moveToChild(pCur, get4byte(findCell(pPage, idx))); + if( rc ) return rc; + rc = moveToRightmost(pCur); + }else{ + while( pCur->ix==0 ){ + if( pCur->iPage==0 ){ + pCur->eState = CURSOR_INVALID; + return SQLITE_DONE; + } + moveToParent(pCur); + } + assert( pCur->info.nSize==0 ); + assert( (pCur->curFlags & (BTCF_ValidOvfl))==0 ); + + pCur->ix--; + pPage = pCur->pPage; + if( pPage->intKey && !pPage->leaf ){ + rc = sqlite3BtreePrevious(pCur, 0); + }else{ + rc = SQLITE_OK; + } + } + return rc; +} +int sqlite3BtreePrevious(BtCursor *pCur, int flags){ + assert( cursorOwnsBtShared(pCur) ); + assert( flags==0 || flags==1 ); + UNUSED_PARAMETER( flags ); /* Used in COMDB2 but not native SQLite */ + pCur->curFlags &= ~(BTCF_AtLast|BTCF_ValidOvfl|BTCF_ValidNKey); + pCur->info.nSize = 0; + if( pCur->eState!=CURSOR_VALID + || pCur->ix==0 + || pCur->pPage->leaf==0 + ){ + return btreePrevious(pCur); + } + pCur->ix--; + return SQLITE_OK; +} + +/* +** Allocate a new page from the database file. +** +** The new page is marked as dirty. (In other words, sqlite3PagerWrite() +** has already been called on the new page.) The new page has also +** been referenced and the calling routine is responsible for calling +** sqlite3PagerUnref() on the new page when it is done. +** +** SQLITE_OK is returned on success. Any other return value indicates +** an error. *ppPage is set to NULL in the event of an error. +** +** If the "nearby" parameter is not 0, then an effort is made to +** locate a page close to the page number "nearby". This can be used in an +** attempt to keep related pages close to each other in the database file, +** which in turn can make database access faster. +** +** If the eMode parameter is BTALLOC_EXACT and the nearby page exists +** anywhere on the free-list, then it is guaranteed to be returned. If +** eMode is BTALLOC_LT then the page returned will be less than or equal +** to nearby if any such page exists. If eMode is BTALLOC_ANY then there +** are no restrictions on which page is returned. +*/ +static int allocateBtreePage( + BtShared *pBt, /* The btree */ + MemPage **ppPage, /* Store pointer to the allocated page here */ + Pgno *pPgno, /* Store the page number here */ + Pgno nearby, /* Search for a page near this one */ + u8 eMode /* BTALLOC_EXACT, BTALLOC_LT, or BTALLOC_ANY */ +){ + MemPage *pPage1; + int rc; + u32 n; /* Number of pages on the freelist */ + u32 k; /* Number of leaves on the trunk of the freelist */ + MemPage *pTrunk = 0; + MemPage *pPrevTrunk = 0; + Pgno mxPage; /* Total size of the database file */ + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( eMode==BTALLOC_ANY || (nearby>0 && IfNotOmitAV(pBt->autoVacuum)) ); + pPage1 = pBt->pPage1; + mxPage = btreePagecount(pBt); + /* EVIDENCE-OF: R-05119-02637 The 4-byte big-endian integer at offset 36 + ** stores stores the total number of pages on the freelist. */ + n = get4byte(&pPage1->aData[36]); + testcase( n==mxPage-1 ); + if( n>=mxPage ){ + return SQLITE_CORRUPT_BKPT; + } + if( n>0 ){ + /* There are pages on the freelist. Reuse one of those pages. */ + Pgno iTrunk; + u8 searchList = 0; /* If the free-list must be searched for 'nearby' */ + u32 nSearch = 0; /* Count of the number of search attempts */ + + /* If eMode==BTALLOC_EXACT and a query of the pointer-map + ** shows that the page 'nearby' is somewhere on the free-list, then + ** the entire-list will be searched for that page. + */ +#ifndef SQLITE_OMIT_AUTOVACUUM + if( eMode==BTALLOC_EXACT ){ + if( nearby<=mxPage ){ + u8 eType; + assert( nearby>0 ); + assert( pBt->autoVacuum ); + rc = ptrmapGet(pBt, nearby, &eType, 0); + if( rc ) return rc; + if( eType==PTRMAP_FREEPAGE ){ + searchList = 1; + } + } + }else if( eMode==BTALLOC_LE ){ + searchList = 1; + } +#endif + + /* Decrement the free-list count by 1. Set iTrunk to the index of the + ** first free-list trunk page. iPrevTrunk is initially 1. + */ + rc = sqlite3PagerWrite(pPage1->pDbPage); + if( rc ) return rc; + put4byte(&pPage1->aData[36], n-1); + + /* The code within this loop is run only once if the 'searchList' variable + ** is not true. Otherwise, it runs once for each trunk-page on the + ** free-list until the page 'nearby' is located (eMode==BTALLOC_EXACT) + ** or until a page less than 'nearby' is located (eMode==BTALLOC_LT) + */ + do { + pPrevTrunk = pTrunk; + if( pPrevTrunk ){ + /* EVIDENCE-OF: R-01506-11053 The first integer on a freelist trunk page + ** is the page number of the next freelist trunk page in the list or + ** zero if this is the last freelist trunk page. */ + iTrunk = get4byte(&pPrevTrunk->aData[0]); + }else{ + /* EVIDENCE-OF: R-59841-13798 The 4-byte big-endian integer at offset 32 + ** stores the page number of the first page of the freelist, or zero if + ** the freelist is empty. */ + iTrunk = get4byte(&pPage1->aData[32]); + } + testcase( iTrunk==mxPage ); + if( iTrunk>mxPage || nSearch++ > n ){ + rc = SQLITE_CORRUPT_PGNO(pPrevTrunk ? pPrevTrunk->pgno : 1); + }else{ + rc = btreeGetUnusedPage(pBt, iTrunk, &pTrunk, 0); + } + if( rc ){ + pTrunk = 0; + goto end_allocate_page; + } + assert( pTrunk!=0 ); + assert( pTrunk->aData!=0 ); + /* EVIDENCE-OF: R-13523-04394 The second integer on a freelist trunk page + ** is the number of leaf page pointers to follow. */ + k = get4byte(&pTrunk->aData[4]); + if( k==0 && !searchList ){ + /* The trunk has no leaves and the list is not being searched. + ** So extract the trunk page itself and use it as the newly + ** allocated page */ + assert( pPrevTrunk==0 ); + rc = sqlite3PagerWrite(pTrunk->pDbPage); + if( rc ){ + goto end_allocate_page; + } + *pPgno = iTrunk; + memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4); + *ppPage = pTrunk; + pTrunk = 0; + TRACE(("ALLOCATE: %d trunk - %d free pages left\n", *pPgno, n-1)); + }else if( k>(u32)(pBt->usableSize/4 - 2) ){ + /* Value of k is out of range. Database corruption */ + rc = SQLITE_CORRUPT_PGNO(iTrunk); + goto end_allocate_page; +#ifndef SQLITE_OMIT_AUTOVACUUM + }else if( searchList + && (nearby==iTrunk || (iTrunkpDbPage); + if( rc ){ + goto end_allocate_page; + } + if( k==0 ){ + if( !pPrevTrunk ){ + memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4); + }else{ + rc = sqlite3PagerWrite(pPrevTrunk->pDbPage); + if( rc!=SQLITE_OK ){ + goto end_allocate_page; + } + memcpy(&pPrevTrunk->aData[0], &pTrunk->aData[0], 4); + } + }else{ + /* The trunk page is required by the caller but it contains + ** pointers to free-list leaves. The first leaf becomes a trunk + ** page in this case. + */ + MemPage *pNewTrunk; + Pgno iNewTrunk = get4byte(&pTrunk->aData[8]); + if( iNewTrunk>mxPage ){ + rc = SQLITE_CORRUPT_PGNO(iTrunk); + goto end_allocate_page; + } + testcase( iNewTrunk==mxPage ); + rc = btreeGetUnusedPage(pBt, iNewTrunk, &pNewTrunk, 0); + if( rc!=SQLITE_OK ){ + goto end_allocate_page; + } + rc = sqlite3PagerWrite(pNewTrunk->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(pNewTrunk); + goto end_allocate_page; + } + memcpy(&pNewTrunk->aData[0], &pTrunk->aData[0], 4); + put4byte(&pNewTrunk->aData[4], k-1); + memcpy(&pNewTrunk->aData[8], &pTrunk->aData[12], (k-1)*4); + releasePage(pNewTrunk); + if( !pPrevTrunk ){ + assert( sqlite3PagerIswriteable(pPage1->pDbPage) ); + put4byte(&pPage1->aData[32], iNewTrunk); + }else{ + rc = sqlite3PagerWrite(pPrevTrunk->pDbPage); + if( rc ){ + goto end_allocate_page; + } + put4byte(&pPrevTrunk->aData[0], iNewTrunk); + } + } + pTrunk = 0; + TRACE(("ALLOCATE: %d trunk - %d free pages left\n", *pPgno, n-1)); +#endif + }else if( k>0 ){ + /* Extract a leaf from the trunk */ + u32 closest; + Pgno iPage; + unsigned char *aData = pTrunk->aData; + if( nearby>0 ){ + u32 i; + closest = 0; + if( eMode==BTALLOC_LE ){ + for(i=0; imxPage || iPage<2 ){ + rc = SQLITE_CORRUPT_PGNO(iTrunk); + goto end_allocate_page; + } + testcase( iPage==mxPage ); + if( !searchList + || (iPage==nearby || (iPagepgno, n-1)); + rc = sqlite3PagerWrite(pTrunk->pDbPage); + if( rc ) goto end_allocate_page; + if( closestpDbPage); + if( rc!=SQLITE_OK ){ + releasePage(*ppPage); + *ppPage = 0; + } + } + searchList = 0; + } + } + releasePage(pPrevTrunk); + pPrevTrunk = 0; + }while( searchList ); + }else{ + /* There are no pages on the freelist, so append a new page to the + ** database image. + ** + ** Normally, new pages allocated by this block can be requested from the + ** pager layer with the 'no-content' flag set. This prevents the pager + ** from trying to read the pages content from disk. However, if the + ** current transaction has already run one or more incremental-vacuum + ** steps, then the page we are about to allocate may contain content + ** that is required in the event of a rollback. In this case, do + ** not set the no-content flag. This causes the pager to load and journal + ** the current page content before overwriting it. + ** + ** Note that the pager will not actually attempt to load or journal + ** content for any page that really does lie past the end of the database + ** file on disk. So the effects of disabling the no-content optimization + ** here are confined to those pages that lie between the end of the + ** database image and the end of the database file. + */ + int bNoContent = (0==IfNotOmitAV(pBt->bDoTruncate))? PAGER_GET_NOCONTENT:0; + + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( rc ) return rc; + pBt->nPage++; + if( pBt->nPage==PENDING_BYTE_PAGE(pBt) ) pBt->nPage++; + +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum && PTRMAP_ISPAGE(pBt, pBt->nPage) ){ + /* If *pPgno refers to a pointer-map page, allocate two new pages + ** at the end of the file instead of one. The first allocated page + ** becomes a new pointer-map page, the second is used by the caller. + */ + MemPage *pPg = 0; + TRACE(("ALLOCATE: %d from end of file (pointer-map page)\n", pBt->nPage)); + assert( pBt->nPage!=PENDING_BYTE_PAGE(pBt) ); + rc = btreeGetUnusedPage(pBt, pBt->nPage, &pPg, bNoContent); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pPg->pDbPage); + releasePage(pPg); + } + if( rc ) return rc; + pBt->nPage++; + if( pBt->nPage==PENDING_BYTE_PAGE(pBt) ){ pBt->nPage++; } + } +#endif + put4byte(28 + (u8*)pBt->pPage1->aData, pBt->nPage); + *pPgno = pBt->nPage; + + assert( *pPgno!=PENDING_BYTE_PAGE(pBt) ); + rc = btreeGetUnusedPage(pBt, *pPgno, ppPage, bNoContent); + if( rc ) return rc; + rc = sqlite3PagerWrite((*ppPage)->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(*ppPage); + *ppPage = 0; + } + TRACE(("ALLOCATE: %d from end of file\n", *pPgno)); + } + + assert( CORRUPT_DB || *pPgno!=PENDING_BYTE_PAGE(pBt) ); + +end_allocate_page: + releasePage(pTrunk); + releasePage(pPrevTrunk); + assert( rc!=SQLITE_OK || sqlite3PagerPageRefcount((*ppPage)->pDbPage)<=1 ); + assert( rc!=SQLITE_OK || (*ppPage)->isInit==0 ); + return rc; +} + +/* +** This function is used to add page iPage to the database file free-list. +** It is assumed that the page is not already a part of the free-list. +** +** The value passed as the second argument to this function is optional. +** If the caller happens to have a pointer to the MemPage object +** corresponding to page iPage handy, it may pass it as the second value. +** Otherwise, it may pass NULL. +** +** If a pointer to a MemPage object is passed as the second argument, +** its reference count is not altered by this function. +*/ +static int freePage2(BtShared *pBt, MemPage *pMemPage, Pgno iPage){ + MemPage *pTrunk = 0; /* Free-list trunk page */ + Pgno iTrunk = 0; /* Page number of free-list trunk page */ + MemPage *pPage1 = pBt->pPage1; /* Local reference to page 1 */ + MemPage *pPage; /* Page being freed. May be NULL. */ + int rc; /* Return Code */ + u32 nFree; /* Initial number of pages on free-list */ + + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( CORRUPT_DB || iPage>1 ); + assert( !pMemPage || pMemPage->pgno==iPage ); + + if( NEVER(iPage<2) || iPage>pBt->nPage ){ + return SQLITE_CORRUPT_BKPT; + } + if( pMemPage ){ + pPage = pMemPage; + sqlite3PagerRef(pPage->pDbPage); + }else{ + pPage = btreePageLookup(pBt, iPage); + } + + /* Increment the free page count on pPage1 */ + rc = sqlite3PagerWrite(pPage1->pDbPage); + if( rc ) goto freepage_out; + nFree = get4byte(&pPage1->aData[36]); + put4byte(&pPage1->aData[36], nFree+1); + + if( pBt->btsFlags & BTS_SECURE_DELETE ){ + /* If the secure_delete option is enabled, then + ** always fully overwrite deleted information with zeros. + */ + if( (!pPage && ((rc = btreeGetPage(pBt, iPage, &pPage, 0))!=0) ) + || ((rc = sqlite3PagerWrite(pPage->pDbPage))!=0) + ){ + goto freepage_out; + } + memset(pPage->aData, 0, pPage->pBt->pageSize); + } + + /* If the database supports auto-vacuum, write an entry in the pointer-map + ** to indicate that the page is free. + */ + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, iPage, PTRMAP_FREEPAGE, 0, &rc); + if( rc ) goto freepage_out; + } + + /* Now manipulate the actual database free-list structure. There are two + ** possibilities. If the free-list is currently empty, or if the first + ** trunk page in the free-list is full, then this page will become a + ** new free-list trunk page. Otherwise, it will become a leaf of the + ** first trunk page in the current free-list. This block tests if it + ** is possible to add the page as a new free-list leaf. + */ + if( nFree!=0 ){ + u32 nLeaf; /* Initial number of leaf cells on trunk page */ + + iTrunk = get4byte(&pPage1->aData[32]); + if( iTrunk>btreePagecount(pBt) ){ + rc = SQLITE_CORRUPT_BKPT; + goto freepage_out; + } + rc = btreeGetPage(pBt, iTrunk, &pTrunk, 0); + if( rc!=SQLITE_OK ){ + goto freepage_out; + } + + nLeaf = get4byte(&pTrunk->aData[4]); + assert( pBt->usableSize>32 ); + if( nLeaf > (u32)pBt->usableSize/4 - 2 ){ + rc = SQLITE_CORRUPT_BKPT; + goto freepage_out; + } + if( nLeaf < (u32)pBt->usableSize/4 - 8 ){ + /* In this case there is room on the trunk page to insert the page + ** being freed as a new leaf. + ** + ** Note that the trunk page is not really full until it contains + ** usableSize/4 - 2 entries, not usableSize/4 - 8 entries as we have + ** coded. But due to a coding error in versions of SQLite prior to + ** 3.6.0, databases with freelist trunk pages holding more than + ** usableSize/4 - 8 entries will be reported as corrupt. In order + ** to maintain backwards compatibility with older versions of SQLite, + ** we will continue to restrict the number of entries to usableSize/4 - 8 + ** for now. At some point in the future (once everyone has upgraded + ** to 3.6.0 or later) we should consider fixing the conditional above + ** to read "usableSize/4-2" instead of "usableSize/4-8". + ** + ** EVIDENCE-OF: R-19920-11576 However, newer versions of SQLite still + ** avoid using the last six entries in the freelist trunk page array in + ** order that database files created by newer versions of SQLite can be + ** read by older versions of SQLite. + */ + rc = sqlite3PagerWrite(pTrunk->pDbPage); + if( rc==SQLITE_OK ){ + put4byte(&pTrunk->aData[4], nLeaf+1); + put4byte(&pTrunk->aData[8+nLeaf*4], iPage); + if( pPage && (pBt->btsFlags & BTS_SECURE_DELETE)==0 ){ + sqlite3PagerDontWrite(pPage->pDbPage); + } + rc = btreeSetHasContent(pBt, iPage); + } + TRACE(("FREE-PAGE: %d leaf on trunk page %d\n",pPage->pgno,pTrunk->pgno)); + goto freepage_out; + } + } + + /* If control flows to this point, then it was not possible to add the + ** the page being freed as a leaf page of the first trunk in the free-list. + ** Possibly because the free-list is empty, or possibly because the + ** first trunk in the free-list is full. Either way, the page being freed + ** will become the new first trunk page in the free-list. + */ + if( pPage==0 && SQLITE_OK!=(rc = btreeGetPage(pBt, iPage, &pPage, 0)) ){ + goto freepage_out; + } + rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc!=SQLITE_OK ){ + goto freepage_out; + } + put4byte(pPage->aData, iTrunk); + put4byte(&pPage->aData[4], 0); + put4byte(&pPage1->aData[32], iPage); + TRACE(("FREE-PAGE: %d new trunk page replacing %d\n", pPage->pgno, iTrunk)); + +freepage_out: + if( pPage ){ + pPage->isInit = 0; + } + releasePage(pPage); + releasePage(pTrunk); + return rc; +} +static void freePage(MemPage *pPage, int *pRC){ + if( (*pRC)==SQLITE_OK ){ + *pRC = freePage2(pPage->pBt, pPage, pPage->pgno); + } +} + +/* +** Free the overflow pages associated with the given Cell. +*/ +static SQLITE_NOINLINE int clearCellOverflow( + MemPage *pPage, /* The page that contains the Cell */ + unsigned char *pCell, /* First byte of the Cell */ + CellInfo *pInfo /* Size information about the cell */ +){ + BtShared *pBt; + Pgno ovflPgno; + int rc; + int nOvfl; + u32 ovflPageSize; + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pInfo->nLocal!=pInfo->nPayload ); + testcase( pCell + pInfo->nSize == pPage->aDataEnd ); + testcase( pCell + (pInfo->nSize-1) == pPage->aDataEnd ); + if( pCell + pInfo->nSize > pPage->aDataEnd ){ + /* Cell extends past end of page */ + return SQLITE_CORRUPT_PAGE(pPage); + } + ovflPgno = get4byte(pCell + pInfo->nSize - 4); + pBt = pPage->pBt; + assert( pBt->usableSize > 4 ); + ovflPageSize = pBt->usableSize - 4; + nOvfl = (pInfo->nPayload - pInfo->nLocal + ovflPageSize - 1)/ovflPageSize; + assert( nOvfl>0 || + (CORRUPT_DB && (pInfo->nPayload + ovflPageSize)btreePagecount(pBt) ){ + /* 0 is not a legal page number and page 1 cannot be an + ** overflow page. Therefore if ovflPgno<2 or past the end of the + ** file the database must be corrupt. */ + return SQLITE_CORRUPT_BKPT; + } + if( nOvfl ){ + rc = getOverflowPage(pBt, ovflPgno, &pOvfl, &iNext); + if( rc ) return rc; + } + + if( ( pOvfl || ((pOvfl = btreePageLookup(pBt, ovflPgno))!=0) ) + && sqlite3PagerPageRefcount(pOvfl->pDbPage)!=1 + ){ + /* There is no reason any cursor should have an outstanding reference + ** to an overflow page belonging to a cell that is being deleted/updated. + ** So if there exists more than one reference to this page, then it + ** must not really be an overflow page and the database must be corrupt. + ** It is helpful to detect this before calling freePage2(), as + ** freePage2() may zero the page contents if secure-delete mode is + ** enabled. If this 'overflow' page happens to be a page that the + ** caller is iterating through or using in some other way, this + ** can be problematic. + */ + rc = SQLITE_CORRUPT_BKPT; + }else{ + rc = freePage2(pBt, pOvfl, ovflPgno); + } + + if( pOvfl ){ + sqlite3PagerUnref(pOvfl->pDbPage); + } + if( rc ) return rc; + ovflPgno = iNext; + } + return SQLITE_OK; +} + +/* Call xParseCell to compute the size of a cell. If the cell contains +** overflow, then invoke cellClearOverflow to clear out that overflow. +** STore the result code (SQLITE_OK or some error code) in rc. +** +** Implemented as macro to force inlining for performance. +*/ +#define BTREE_CLEAR_CELL(rc, pPage, pCell, sInfo) \ + pPage->xParseCell(pPage, pCell, &sInfo); \ + if( sInfo.nLocal!=sInfo.nPayload ){ \ + rc = clearCellOverflow(pPage, pCell, &sInfo); \ + }else{ \ + rc = SQLITE_OK; \ + } + + +/* +** Create the byte sequence used to represent a cell on page pPage +** and write that byte sequence into pCell[]. Overflow pages are +** allocated and filled in as necessary. The calling procedure +** is responsible for making sure sufficient space has been allocated +** for pCell[]. +** +** Note that pCell does not necessary need to point to the pPage->aData +** area. pCell might point to some temporary storage. The cell will +** be constructed in this temporary area then copied into pPage->aData +** later. +*/ +static int fillInCell( + MemPage *pPage, /* The page that contains the cell */ + unsigned char *pCell, /* Complete text of the cell */ + const BtreePayload *pX, /* Payload with which to construct the cell */ + int *pnSize /* Write cell size here */ +){ + int nPayload; + const u8 *pSrc; + int nSrc, n, rc, mn; + int spaceLeft; + MemPage *pToRelease; + unsigned char *pPrior; + unsigned char *pPayload; + BtShared *pBt; + Pgno pgnoOvfl; + int nHeader; + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + + /* pPage is not necessarily writeable since pCell might be auxiliary + ** buffer space that is separate from the pPage buffer area */ + assert( pCellaData || pCell>=&pPage->aData[pPage->pBt->pageSize] + || sqlite3PagerIswriteable(pPage->pDbPage) ); + + /* Fill in the header. */ + nHeader = pPage->childPtrSize; + if( pPage->intKey ){ + nPayload = pX->nData + pX->nZero; + pSrc = pX->pData; + nSrc = pX->nData; + assert( pPage->intKeyLeaf ); /* fillInCell() only called for leaves */ + nHeader += putVarint32(&pCell[nHeader], nPayload); + nHeader += putVarint(&pCell[nHeader], *(u64*)&pX->nKey); + }else{ + assert( pX->nKey<=0x7fffffff && pX->pKey!=0 ); + nSrc = nPayload = (int)pX->nKey; + pSrc = pX->pKey; + nHeader += putVarint32(&pCell[nHeader], nPayload); + } + + /* Fill in the payload */ + pPayload = &pCell[nHeader]; + if( nPayload<=pPage->maxLocal ){ + /* This is the common case where everything fits on the btree page + ** and no overflow pages are required. */ + n = nHeader + nPayload; + testcase( n==3 ); + testcase( n==4 ); + if( n<4 ) n = 4; + *pnSize = n; + assert( nSrc<=nPayload ); + testcase( nSrcminLocal; + n = mn + (nPayload - mn) % (pPage->pBt->usableSize - 4); + testcase( n==pPage->maxLocal ); + testcase( n==pPage->maxLocal+1 ); + if( n > pPage->maxLocal ) n = mn; + spaceLeft = n; + *pnSize = n + nHeader + 4; + pPrior = &pCell[nHeader+n]; + pToRelease = 0; + pgnoOvfl = 0; + pBt = pPage->pBt; + + /* At this point variables should be set as follows: + ** + ** nPayload Total payload size in bytes + ** pPayload Begin writing payload here + ** spaceLeft Space available at pPayload. If nPayload>spaceLeft, + ** that means content must spill into overflow pages. + ** *pnSize Size of the local cell (not counting overflow pages) + ** pPrior Where to write the pgno of the first overflow page + ** + ** Use a call to btreeParseCellPtr() to verify that the values above + ** were computed correctly. + */ +#ifdef SQLITE_DEBUG + { + CellInfo info; + pPage->xParseCell(pPage, pCell, &info); + assert( nHeader==(int)(info.pPayload - pCell) ); + assert( info.nKey==pX->nKey ); + assert( *pnSize == info.nSize ); + assert( spaceLeft == info.nLocal ); + } +#endif + + /* Write the payload into the local Cell and any extra into overflow pages */ + while( 1 ){ + n = nPayload; + if( n>spaceLeft ) n = spaceLeft; + + /* If pToRelease is not zero than pPayload points into the data area + ** of pToRelease. Make sure pToRelease is still writeable. */ + assert( pToRelease==0 || sqlite3PagerIswriteable(pToRelease->pDbPage) ); + + /* If pPayload is part of the data area of pPage, then make sure pPage + ** is still writeable */ + assert( pPayloadaData || pPayload>=&pPage->aData[pBt->pageSize] + || sqlite3PagerIswriteable(pPage->pDbPage) ); + + if( nSrc>=n ){ + memcpy(pPayload, pSrc, n); + }else if( nSrc>0 ){ + n = nSrc; + memcpy(pPayload, pSrc, n); + }else{ + memset(pPayload, 0, n); + } + nPayload -= n; + if( nPayload<=0 ) break; + pPayload += n; + pSrc += n; + nSrc -= n; + spaceLeft -= n; + if( spaceLeft==0 ){ + MemPage *pOvfl = 0; +#ifndef SQLITE_OMIT_AUTOVACUUM + Pgno pgnoPtrmap = pgnoOvfl; /* Overflow page pointer-map entry page */ + if( pBt->autoVacuum ){ + do{ + pgnoOvfl++; + } while( + PTRMAP_ISPAGE(pBt, pgnoOvfl) || pgnoOvfl==PENDING_BYTE_PAGE(pBt) + ); + } +#endif + rc = allocateBtreePage(pBt, &pOvfl, &pgnoOvfl, pgnoOvfl, 0); +#ifndef SQLITE_OMIT_AUTOVACUUM + /* If the database supports auto-vacuum, and the second or subsequent + ** overflow page is being allocated, add an entry to the pointer-map + ** for that page now. + ** + ** If this is the first overflow page, then write a partial entry + ** to the pointer-map. If we write nothing to this pointer-map slot, + ** then the optimistic overflow chain processing in clearCell() + ** may misinterpret the uninitialized values and delete the + ** wrong pages from the database. + */ + if( pBt->autoVacuum && rc==SQLITE_OK ){ + u8 eType = (pgnoPtrmap?PTRMAP_OVERFLOW2:PTRMAP_OVERFLOW1); + ptrmapPut(pBt, pgnoOvfl, eType, pgnoPtrmap, &rc); + if( rc ){ + releasePage(pOvfl); + } + } +#endif + if( rc ){ + releasePage(pToRelease); + return rc; + } + + /* If pToRelease is not zero than pPrior points into the data area + ** of pToRelease. Make sure pToRelease is still writeable. */ + assert( pToRelease==0 || sqlite3PagerIswriteable(pToRelease->pDbPage) ); + + /* If pPrior is part of the data area of pPage, then make sure pPage + ** is still writeable */ + assert( pPrioraData || pPrior>=&pPage->aData[pBt->pageSize] + || sqlite3PagerIswriteable(pPage->pDbPage) ); + + put4byte(pPrior, pgnoOvfl); + releasePage(pToRelease); + pToRelease = pOvfl; + pPrior = pOvfl->aData; + put4byte(pPrior, 0); + pPayload = &pOvfl->aData[4]; + spaceLeft = pBt->usableSize - 4; + } + } + releasePage(pToRelease); + return SQLITE_OK; +} + +/* +** Remove the i-th cell from pPage. This routine effects pPage only. +** The cell content is not freed or deallocated. It is assumed that +** the cell content has been copied someplace else. This routine just +** removes the reference to the cell from pPage. +** +** "sz" must be the number of bytes in the cell. +*/ +static void dropCell(MemPage *pPage, int idx, int sz, int *pRC){ + u32 pc; /* Offset to cell content of cell being deleted */ + u8 *data; /* pPage->aData */ + u8 *ptr; /* Used to move bytes around within data[] */ + int rc; /* The return code */ + int hdr; /* Beginning of the header. 0 most pages. 100 page 1 */ + + if( *pRC ) return; + assert( idx>=0 && idxnCell ); + assert( CORRUPT_DB || sz==cellSize(pPage, idx) ); + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( pPage->nFree>=0 ); + data = pPage->aData; + ptr = &pPage->aCellIdx[2*idx]; + pc = get2byte(ptr); + hdr = pPage->hdrOffset; + testcase( pc==get2byte(&data[hdr+5]) ); + testcase( pc+sz==pPage->pBt->usableSize ); + if( pc+sz > pPage->pBt->usableSize ){ + *pRC = SQLITE_CORRUPT_BKPT; + return; + } + rc = freeSpace(pPage, pc, sz); + if( rc ){ + *pRC = rc; + return; + } + pPage->nCell--; + if( pPage->nCell==0 ){ + memset(&data[hdr+1], 0, 4); + data[hdr+7] = 0; + put2byte(&data[hdr+5], pPage->pBt->usableSize); + pPage->nFree = pPage->pBt->usableSize - pPage->hdrOffset + - pPage->childPtrSize - 8; + }else{ + memmove(ptr, ptr+2, 2*(pPage->nCell - idx)); + put2byte(&data[hdr+3], pPage->nCell); + pPage->nFree += 2; + } +} + +/* +** Insert a new cell on pPage at cell index "i". pCell points to the +** content of the cell. +** +** If the cell content will fit on the page, then put it there. If it +** will not fit, then make a copy of the cell content into pTemp if +** pTemp is not null. Regardless of pTemp, allocate a new entry +** in pPage->apOvfl[] and make it point to the cell content (either +** in pTemp or the original pCell) and also record its index. +** Allocating a new entry in pPage->aCell[] implies that +** pPage->nOverflow is incremented. +** +** *pRC must be SQLITE_OK when this routine is called. +*/ +static void insertCell( + MemPage *pPage, /* Page into which we are copying */ + int i, /* New cell becomes the i-th cell of the page */ + u8 *pCell, /* Content of the new cell */ + int sz, /* Bytes of content in pCell */ + u8 *pTemp, /* Temp storage space for pCell, if needed */ + Pgno iChild, /* If non-zero, replace first 4 bytes with this value */ + int *pRC /* Read and write return code from here */ +){ + int idx = 0; /* Where to write new cell content in data[] */ + int j; /* Loop counter */ + u8 *data; /* The content of the whole page */ + u8 *pIns; /* The point in pPage->aCellIdx[] where no cell inserted */ + + assert( *pRC==SQLITE_OK ); + assert( i>=0 && i<=pPage->nCell+pPage->nOverflow ); + assert( MX_CELL(pPage->pBt)<=10921 ); + assert( pPage->nCell<=MX_CELL(pPage->pBt) || CORRUPT_DB ); + assert( pPage->nOverflow<=ArraySize(pPage->apOvfl) ); + assert( ArraySize(pPage->apOvfl)==ArraySize(pPage->aiOvfl) ); + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( sz==pPage->xCellSize(pPage, pCell) || CORRUPT_DB ); + assert( pPage->nFree>=0 ); + if( pPage->nOverflow || sz+2>pPage->nFree ){ + if( pTemp ){ + memcpy(pTemp, pCell, sz); + pCell = pTemp; + } + if( iChild ){ + put4byte(pCell, iChild); + } + j = pPage->nOverflow++; + /* Comparison against ArraySize-1 since we hold back one extra slot + ** as a contingency. In other words, never need more than 3 overflow + ** slots but 4 are allocated, just to be safe. */ + assert( j < ArraySize(pPage->apOvfl)-1 ); + pPage->apOvfl[j] = pCell; + pPage->aiOvfl[j] = (u16)i; + + /* When multiple overflows occur, they are always sequential and in + ** sorted order. This invariants arise because multiple overflows can + ** only occur when inserting divider cells into the parent page during + ** balancing, and the dividers are adjacent and sorted. + */ + assert( j==0 || pPage->aiOvfl[j-1]<(u16)i ); /* Overflows in sorted order */ + assert( j==0 || i==pPage->aiOvfl[j-1]+1 ); /* Overflows are sequential */ + }else{ + int rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc!=SQLITE_OK ){ + *pRC = rc; + return; + } + assert( sqlite3PagerIswriteable(pPage->pDbPage) ); + data = pPage->aData; + assert( &data[pPage->cellOffset]==pPage->aCellIdx ); + rc = allocateSpace(pPage, sz, &idx); + if( rc ){ *pRC = rc; return; } + /* The allocateSpace() routine guarantees the following properties + ** if it returns successfully */ + assert( idx >= 0 ); + assert( idx >= pPage->cellOffset+2*pPage->nCell+2 || CORRUPT_DB ); + assert( idx+sz <= (int)pPage->pBt->usableSize ); + pPage->nFree -= (u16)(2 + sz); + if( iChild ){ + /* In a corrupt database where an entry in the cell index section of + ** a btree page has a value of 3 or less, the pCell value might point + ** as many as 4 bytes in front of the start of the aData buffer for + ** the source page. Make sure this does not cause problems by not + ** reading the first 4 bytes */ + memcpy(&data[idx+4], pCell+4, sz-4); + put4byte(&data[idx], iChild); + }else{ + memcpy(&data[idx], pCell, sz); + } + pIns = pPage->aCellIdx + i*2; + memmove(pIns+2, pIns, 2*(pPage->nCell - i)); + put2byte(pIns, idx); + pPage->nCell++; + /* increment the cell count */ + if( (++data[pPage->hdrOffset+4])==0 ) data[pPage->hdrOffset+3]++; + assert( get2byte(&data[pPage->hdrOffset+3])==pPage->nCell || CORRUPT_DB ); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pPage->pBt->autoVacuum ){ + /* The cell may contain a pointer to an overflow page. If so, write + ** the entry for the overflow page into the pointer map. + */ + ptrmapPutOvflPtr(pPage, pPage, pCell, pRC); + } +#endif + } +} + +/* +** The following parameters determine how many adjacent pages get involved +** in a balancing operation. NN is the number of neighbors on either side +** of the page that participate in the balancing operation. NB is the +** total number of pages that participate, including the target page and +** NN neighbors on either side. +** +** The minimum value of NN is 1 (of course). Increasing NN above 1 +** (to 2 or 3) gives a modest improvement in SELECT and DELETE performance +** in exchange for a larger degradation in INSERT and UPDATE performance. +** The value of NN appears to give the best results overall. +** +** (Later:) The description above makes it seem as if these values are +** tunable - as if you could change them and recompile and it would all work. +** But that is unlikely. NB has been 3 since the inception of SQLite and +** we have never tested any other value. +*/ +#define NN 1 /* Number of neighbors on either side of pPage */ +#define NB 3 /* (NN*2+1): Total pages involved in the balance */ + +/* +** A CellArray object contains a cache of pointers and sizes for a +** consecutive sequence of cells that might be held on multiple pages. +** +** The cells in this array are the divider cell or cells from the pParent +** page plus up to three child pages. There are a total of nCell cells. +** +** pRef is a pointer to one of the pages that contributes cells. This is +** used to access information such as MemPage.intKey and MemPage.pBt->pageSize +** which should be common to all pages that contribute cells to this array. +** +** apCell[] and szCell[] hold, respectively, pointers to the start of each +** cell and the size of each cell. Some of the apCell[] pointers might refer +** to overflow cells. In other words, some apCel[] pointers might not point +** to content area of the pages. +** +** A szCell[] of zero means the size of that cell has not yet been computed. +** +** The cells come from as many as four different pages: +** +** ----------- +** | Parent | +** ----------- +** / | \ +** / | \ +** --------- --------- --------- +** |Child-1| |Child-2| |Child-3| +** --------- --------- --------- +** +** The order of cells is in the array is for an index btree is: +** +** 1. All cells from Child-1 in order +** 2. The first divider cell from Parent +** 3. All cells from Child-2 in order +** 4. The second divider cell from Parent +** 5. All cells from Child-3 in order +** +** For a table-btree (with rowids) the items 2 and 4 are empty because +** content exists only in leaves and there are no divider cells. +** +** For an index btree, the apEnd[] array holds pointer to the end of page +** for Child-1, the Parent, Child-2, the Parent (again), and Child-3, +** respectively. The ixNx[] array holds the number of cells contained in +** each of these 5 stages, and all stages to the left. Hence: +** +** ixNx[0] = Number of cells in Child-1. +** ixNx[1] = Number of cells in Child-1 plus 1 for first divider. +** ixNx[2] = Number of cells in Child-1 and Child-2 + 1 for 1st divider. +** ixNx[3] = Number of cells in Child-1 and Child-2 + both divider cells +** ixNx[4] = Total number of cells. +** +** For a table-btree, the concept is similar, except only apEnd[0]..apEnd[2] +** are used and they point to the leaf pages only, and the ixNx value are: +** +** ixNx[0] = Number of cells in Child-1. +** ixNx[1] = Number of cells in Child-1 and Child-2. +** ixNx[2] = Total number of cells. +** +** Sometimes when deleting, a child page can have zero cells. In those +** cases, ixNx[] entries with higher indexes, and the corresponding apEnd[] +** entries, shift down. The end result is that each ixNx[] entry should +** be larger than the previous +*/ +typedef struct CellArray CellArray; +struct CellArray { + int nCell; /* Number of cells in apCell[] */ + MemPage *pRef; /* Reference page */ + u8 **apCell; /* All cells begin balanced */ + u16 *szCell; /* Local size of all cells in apCell[] */ + u8 *apEnd[NB*2]; /* MemPage.aDataEnd values */ + int ixNx[NB*2]; /* Index of at which we move to the next apEnd[] */ +}; + +/* +** Make sure the cell sizes at idx, idx+1, ..., idx+N-1 have been +** computed. +*/ +static void populateCellCache(CellArray *p, int idx, int N){ + assert( idx>=0 && idx+N<=p->nCell ); + while( N>0 ){ + assert( p->apCell[idx]!=0 ); + if( p->szCell[idx]==0 ){ + p->szCell[idx] = p->pRef->xCellSize(p->pRef, p->apCell[idx]); + }else{ + assert( CORRUPT_DB || + p->szCell[idx]==p->pRef->xCellSize(p->pRef, p->apCell[idx]) ); + } + idx++; + N--; + } +} + +/* +** Return the size of the Nth element of the cell array +*/ +static SQLITE_NOINLINE u16 computeCellSize(CellArray *p, int N){ + assert( N>=0 && NnCell ); + assert( p->szCell[N]==0 ); + p->szCell[N] = p->pRef->xCellSize(p->pRef, p->apCell[N]); + return p->szCell[N]; +} +static u16 cachedCellSize(CellArray *p, int N){ + assert( N>=0 && NnCell ); + if( p->szCell[N] ) return p->szCell[N]; + return computeCellSize(p, N); +} + +/* +** Array apCell[] contains pointers to nCell b-tree page cells. The +** szCell[] array contains the size in bytes of each cell. This function +** replaces the current contents of page pPg with the contents of the cell +** array. +** +** Some of the cells in apCell[] may currently be stored in pPg. This +** function works around problems caused by this by making a copy of any +** such cells before overwriting the page data. +** +** The MemPage.nFree field is invalidated by this function. It is the +** responsibility of the caller to set it correctly. +*/ +static int rebuildPage( + CellArray *pCArray, /* Content to be added to page pPg */ + int iFirst, /* First cell in pCArray to use */ + int nCell, /* Final number of cells on page */ + MemPage *pPg /* The page to be reconstructed */ +){ + const int hdr = pPg->hdrOffset; /* Offset of header on pPg */ + u8 * const aData = pPg->aData; /* Pointer to data for pPg */ + const int usableSize = pPg->pBt->usableSize; + u8 * const pEnd = &aData[usableSize]; + int i = iFirst; /* Which cell to copy from pCArray*/ + u32 j; /* Start of cell content area */ + int iEnd = i+nCell; /* Loop terminator */ + u8 *pCellptr = pPg->aCellIdx; + u8 *pTmp = sqlite3PagerTempSpace(pPg->pBt->pPager); + u8 *pData; + int k; /* Current slot in pCArray->apEnd[] */ + u8 *pSrcEnd; /* Current pCArray->apEnd[k] value */ + + assert( i(u32)usableSize) ){ j = 0; } + memcpy(&pTmp[j], &aData[j], usableSize - j); + + for(k=0; pCArray->ixNx[k]<=i && ALWAYS(kapEnd[k]; + + pData = pEnd; + while( 1/*exit by break*/ ){ + u8 *pCell = pCArray->apCell[i]; + u16 sz = pCArray->szCell[i]; + assert( sz>0 ); + if( SQLITE_WITHIN(pCell,aData+j,pEnd) ){ + if( ((uptr)(pCell+sz))>(uptr)pEnd ) return SQLITE_CORRUPT_BKPT; + pCell = &pTmp[pCell - aData]; + }else if( (uptr)(pCell+sz)>(uptr)pSrcEnd + && (uptr)(pCell)<(uptr)pSrcEnd + ){ + return SQLITE_CORRUPT_BKPT; + } + + pData -= sz; + put2byte(pCellptr, (pData - aData)); + pCellptr += 2; + if( pData < pCellptr ) return SQLITE_CORRUPT_BKPT; + memmove(pData, pCell, sz); + assert( sz==pPg->xCellSize(pPg, pCell) || CORRUPT_DB ); + i++; + if( i>=iEnd ) break; + if( pCArray->ixNx[k]<=i ){ + k++; + pSrcEnd = pCArray->apEnd[k]; + } + } + + /* The pPg->nFree field is now set incorrectly. The caller will fix it. */ + pPg->nCell = nCell; + pPg->nOverflow = 0; + + put2byte(&aData[hdr+1], 0); + put2byte(&aData[hdr+3], pPg->nCell); + put2byte(&aData[hdr+5], pData - aData); + aData[hdr+7] = 0x00; + return SQLITE_OK; +} + +/* +** The pCArray objects contains pointers to b-tree cells and the cell sizes. +** This function attempts to add the cells stored in the array to page pPg. +** If it cannot (because the page needs to be defragmented before the cells +** will fit), non-zero is returned. Otherwise, if the cells are added +** successfully, zero is returned. +** +** Argument pCellptr points to the first entry in the cell-pointer array +** (part of page pPg) to populate. After cell apCell[0] is written to the +** page body, a 16-bit offset is written to pCellptr. And so on, for each +** cell in the array. It is the responsibility of the caller to ensure +** that it is safe to overwrite this part of the cell-pointer array. +** +** When this function is called, *ppData points to the start of the +** content area on page pPg. If the size of the content area is extended, +** *ppData is updated to point to the new start of the content area +** before returning. +** +** Finally, argument pBegin points to the byte immediately following the +** end of the space required by this page for the cell-pointer area (for +** all cells - not just those inserted by the current call). If the content +** area must be extended to before this point in order to accomodate all +** cells in apCell[], then the cells do not fit and non-zero is returned. +*/ +static int pageInsertArray( + MemPage *pPg, /* Page to add cells to */ + u8 *pBegin, /* End of cell-pointer array */ + u8 **ppData, /* IN/OUT: Page content-area pointer */ + u8 *pCellptr, /* Pointer to cell-pointer area */ + int iFirst, /* Index of first cell to add */ + int nCell, /* Number of cells to add to pPg */ + CellArray *pCArray /* Array of cells */ +){ + int i = iFirst; /* Loop counter - cell index to insert */ + u8 *aData = pPg->aData; /* Complete page */ + u8 *pData = *ppData; /* Content area. A subset of aData[] */ + int iEnd = iFirst + nCell; /* End of loop. One past last cell to ins */ + int k; /* Current slot in pCArray->apEnd[] */ + u8 *pEnd; /* Maximum extent of cell data */ + assert( CORRUPT_DB || pPg->hdrOffset==0 ); /* Never called on page 1 */ + if( iEnd<=iFirst ) return 0; + for(k=0; pCArray->ixNx[k]<=i && ALWAYS(kapEnd[k]; + while( 1 /*Exit by break*/ ){ + int sz, rc; + u8 *pSlot; + assert( pCArray->szCell[i]!=0 ); + sz = pCArray->szCell[i]; + if( (aData[1]==0 && aData[2]==0) || (pSlot = pageFindSlot(pPg,sz,&rc))==0 ){ + if( (pData - pBegin)apCell[i] will never overlap on a well-formed + ** database. But they might for a corrupt database. Hence use memmove() + ** since memcpy() sends SIGABORT with overlapping buffers on OpenBSD */ + assert( (pSlot+sz)<=pCArray->apCell[i] + || pSlot>=(pCArray->apCell[i]+sz) + || CORRUPT_DB ); + if( (uptr)(pCArray->apCell[i]+sz)>(uptr)pEnd + && (uptr)(pCArray->apCell[i])<(uptr)pEnd + ){ + assert( CORRUPT_DB ); + (void)SQLITE_CORRUPT_BKPT; + return 1; + } + memmove(pSlot, pCArray->apCell[i], sz); + put2byte(pCellptr, (pSlot - aData)); + pCellptr += 2; + i++; + if( i>=iEnd ) break; + if( pCArray->ixNx[k]<=i ){ + k++; + pEnd = pCArray->apEnd[k]; + } + } + *ppData = pData; + return 0; +} + +/* +** The pCArray object contains pointers to b-tree cells and their sizes. +** +** This function adds the space associated with each cell in the array +** that is currently stored within the body of pPg to the pPg free-list. +** The cell-pointers and other fields of the page are not updated. +** +** This function returns the total number of cells added to the free-list. +*/ +static int pageFreeArray( + MemPage *pPg, /* Page to edit */ + int iFirst, /* First cell to delete */ + int nCell, /* Cells to delete */ + CellArray *pCArray /* Array of cells */ +){ + u8 * const aData = pPg->aData; + u8 * const pEnd = &aData[pPg->pBt->usableSize]; + u8 * const pStart = &aData[pPg->hdrOffset + 8 + pPg->childPtrSize]; + int nRet = 0; + int i; + int iEnd = iFirst + nCell; + u8 *pFree = 0; + int szFree = 0; + + for(i=iFirst; iapCell[i]; + if( SQLITE_WITHIN(pCell, pStart, pEnd) ){ + int sz; + /* No need to use cachedCellSize() here. The sizes of all cells that + ** are to be freed have already been computing while deciding which + ** cells need freeing */ + sz = pCArray->szCell[i]; assert( sz>0 ); + if( pFree!=(pCell + sz) ){ + if( pFree ){ + assert( pFree>aData && (pFree - aData)<65536 ); + freeSpace(pPg, (u16)(pFree - aData), szFree); + } + pFree = pCell; + szFree = sz; + if( pFree+sz>pEnd ){ + return 0; + } + }else{ + pFree = pCell; + szFree += sz; + } + nRet++; + } + } + if( pFree ){ + assert( pFree>aData && (pFree - aData)<65536 ); + freeSpace(pPg, (u16)(pFree - aData), szFree); + } + return nRet; +} + +/* +** pCArray contains pointers to and sizes of all cells in the page being +** balanced. The current page, pPg, has pPg->nCell cells starting with +** pCArray->apCell[iOld]. After balancing, this page should hold nNew cells +** starting at apCell[iNew]. +** +** This routine makes the necessary adjustments to pPg so that it contains +** the correct cells after being balanced. +** +** The pPg->nFree field is invalid when this function returns. It is the +** responsibility of the caller to set it correctly. +*/ +static int editPage( + MemPage *pPg, /* Edit this page */ + int iOld, /* Index of first cell currently on page */ + int iNew, /* Index of new first cell on page */ + int nNew, /* Final number of cells on page */ + CellArray *pCArray /* Array of cells and sizes */ +){ + u8 * const aData = pPg->aData; + const int hdr = pPg->hdrOffset; + u8 *pBegin = &pPg->aCellIdx[nNew * 2]; + int nCell = pPg->nCell; /* Cells stored on pPg */ + u8 *pData; + u8 *pCellptr; + int i; + int iOldEnd = iOld + pPg->nCell + pPg->nOverflow; + int iNewEnd = iNew + nNew; + +#ifdef SQLITE_DEBUG + u8 *pTmp = sqlite3PagerTempSpace(pPg->pBt->pPager); + memcpy(pTmp, aData, pPg->pBt->usableSize); +#endif + + /* Remove cells from the start and end of the page */ + assert( nCell>=0 ); + if( iOldnCell) ) return SQLITE_CORRUPT_BKPT; + memmove(pPg->aCellIdx, &pPg->aCellIdx[nShift*2], nCell*2); + nCell -= nShift; + } + if( iNewEnd < iOldEnd ){ + int nTail = pageFreeArray(pPg, iNewEnd, iOldEnd - iNewEnd, pCArray); + assert( nCell>=nTail ); + nCell -= nTail; + } + + pData = &aData[get2byteNotZero(&aData[hdr+5])]; + if( pDatapPg->aDataEnd) ) goto editpage_fail; + + /* Add cells to the start of the page */ + if( iNew=0 ); + pCellptr = pPg->aCellIdx; + memmove(&pCellptr[nAdd*2], pCellptr, nCell*2); + if( pageInsertArray( + pPg, pBegin, &pData, pCellptr, + iNew, nAdd, pCArray + ) ) goto editpage_fail; + nCell += nAdd; + } + + /* Add any overflow cells */ + for(i=0; inOverflow; i++){ + int iCell = (iOld + pPg->aiOvfl[i]) - iNew; + if( iCell>=0 && iCellaCellIdx[iCell * 2]; + if( nCell>iCell ){ + memmove(&pCellptr[2], pCellptr, (nCell - iCell) * 2); + } + nCell++; + cachedCellSize(pCArray, iCell+iNew); + if( pageInsertArray( + pPg, pBegin, &pData, pCellptr, + iCell+iNew, 1, pCArray + ) ) goto editpage_fail; + } + } + + /* Append cells to the end of the page */ + assert( nCell>=0 ); + pCellptr = &pPg->aCellIdx[nCell*2]; + if( pageInsertArray( + pPg, pBegin, &pData, pCellptr, + iNew+nCell, nNew-nCell, pCArray + ) ) goto editpage_fail; + + pPg->nCell = nNew; + pPg->nOverflow = 0; + + put2byte(&aData[hdr+3], pPg->nCell); + put2byte(&aData[hdr+5], pData - aData); + +#ifdef SQLITE_DEBUG + for(i=0; iapCell[i+iNew]; + int iOff = get2byteAligned(&pPg->aCellIdx[i*2]); + if( SQLITE_WITHIN(pCell, aData, &aData[pPg->pBt->usableSize]) ){ + pCell = &pTmp[pCell - aData]; + } + assert( 0==memcmp(pCell, &aData[iOff], + pCArray->pRef->xCellSize(pCArray->pRef, pCArray->apCell[i+iNew])) ); + } +#endif + + return SQLITE_OK; + editpage_fail: + /* Unable to edit this page. Rebuild it from scratch instead. */ + populateCellCache(pCArray, iNew, nNew); + return rebuildPage(pCArray, iNew, nNew, pPg); +} + + +#ifndef SQLITE_OMIT_QUICKBALANCE +/* +** This version of balance() handles the common special case where +** a new entry is being inserted on the extreme right-end of the +** tree, in other words, when the new entry will become the largest +** entry in the tree. +** +** Instead of trying to balance the 3 right-most leaf pages, just add +** a new page to the right-hand side and put the one new entry in +** that page. This leaves the right side of the tree somewhat +** unbalanced. But odds are that we will be inserting new entries +** at the end soon afterwards so the nearly empty page will quickly +** fill up. On average. +** +** pPage is the leaf page which is the right-most page in the tree. +** pParent is its parent. pPage must have a single overflow entry +** which is also the right-most entry on the page. +** +** The pSpace buffer is used to store a temporary copy of the divider +** cell that will be inserted into pParent. Such a cell consists of a 4 +** byte page number followed by a variable length integer. In other +** words, at most 13 bytes. Hence the pSpace buffer must be at +** least 13 bytes in size. +*/ +static int balance_quick(MemPage *pParent, MemPage *pPage, u8 *pSpace){ + BtShared *const pBt = pPage->pBt; /* B-Tree Database */ + MemPage *pNew; /* Newly allocated page */ + int rc; /* Return Code */ + Pgno pgnoNew; /* Page number of pNew */ + + assert( sqlite3_mutex_held(pPage->pBt->mutex) ); + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + assert( pPage->nOverflow==1 ); + + if( pPage->nCell==0 ) return SQLITE_CORRUPT_BKPT; /* dbfuzz001.test */ + assert( pPage->nFree>=0 ); + assert( pParent->nFree>=0 ); + + /* Allocate a new page. This page will become the right-sibling of + ** pPage. Make the parent page writable, so that the new divider cell + ** may be inserted. If both these operations are successful, proceed. + */ + rc = allocateBtreePage(pBt, &pNew, &pgnoNew, 0, 0); + + if( rc==SQLITE_OK ){ + + u8 *pOut = &pSpace[4]; + u8 *pCell = pPage->apOvfl[0]; + u16 szCell = pPage->xCellSize(pPage, pCell); + u8 *pStop; + CellArray b; + + assert( sqlite3PagerIswriteable(pNew->pDbPage) ); + assert( CORRUPT_DB || pPage->aData[0]==(PTF_INTKEY|PTF_LEAFDATA|PTF_LEAF) ); + zeroPage(pNew, PTF_INTKEY|PTF_LEAFDATA|PTF_LEAF); + b.nCell = 1; + b.pRef = pPage; + b.apCell = &pCell; + b.szCell = &szCell; + b.apEnd[0] = pPage->aDataEnd; + b.ixNx[0] = 2; + rc = rebuildPage(&b, 0, 1, pNew); + if( NEVER(rc) ){ + releasePage(pNew); + return rc; + } + pNew->nFree = pBt->usableSize - pNew->cellOffset - 2 - szCell; + + /* If this is an auto-vacuum database, update the pointer map + ** with entries for the new page, and any pointer from the + ** cell on the page to an overflow page. If either of these + ** operations fails, the return code is set, but the contents + ** of the parent page are still manipulated by thh code below. + ** That is Ok, at this point the parent page is guaranteed to + ** be marked as dirty. Returning an error code will cause a + ** rollback, undoing any changes made to the parent page. + */ + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, pgnoNew, PTRMAP_BTREE, pParent->pgno, &rc); + if( szCell>pNew->minLocal ){ + ptrmapPutOvflPtr(pNew, pNew, pCell, &rc); + } + } + + /* Create a divider cell to insert into pParent. The divider cell + ** consists of a 4-byte page number (the page number of pPage) and + ** a variable length key value (which must be the same value as the + ** largest key on pPage). + ** + ** To find the largest key value on pPage, first find the right-most + ** cell on pPage. The first two fields of this cell are the + ** record-length (a variable length integer at most 32-bits in size) + ** and the key value (a variable length integer, may have any value). + ** The first of the while(...) loops below skips over the record-length + ** field. The second while(...) loop copies the key value from the + ** cell on pPage into the pSpace buffer. + */ + pCell = findCell(pPage, pPage->nCell-1); + pStop = &pCell[9]; + while( (*(pCell++)&0x80) && pCellnCell, pSpace, (int)(pOut-pSpace), + 0, pPage->pgno, &rc); + } + + /* Set the right-child pointer of pParent to point to the new page. */ + put4byte(&pParent->aData[pParent->hdrOffset+8], pgnoNew); + + /* Release the reference to the new page. */ + releasePage(pNew); + } + + return rc; +} +#endif /* SQLITE_OMIT_QUICKBALANCE */ + +#if 0 +/* +** This function does not contribute anything to the operation of SQLite. +** it is sometimes activated temporarily while debugging code responsible +** for setting pointer-map entries. +*/ +static int ptrmapCheckPages(MemPage **apPage, int nPage){ + int i, j; + for(i=0; ipBt; + assert( pPage->isInit ); + + for(j=0; jnCell; j++){ + CellInfo info; + u8 *z; + + z = findCell(pPage, j); + pPage->xParseCell(pPage, z, &info); + if( info.nLocalpgno && e==PTRMAP_OVERFLOW1 ); + } + if( !pPage->leaf ){ + Pgno child = get4byte(z); + ptrmapGet(pBt, child, &e, &n); + assert( n==pPage->pgno && e==PTRMAP_BTREE ); + } + } + if( !pPage->leaf ){ + Pgno child = get4byte(&pPage->aData[pPage->hdrOffset+8]); + ptrmapGet(pBt, child, &e, &n); + assert( n==pPage->pgno && e==PTRMAP_BTREE ); + } + } + return 1; +} +#endif + +/* +** This function is used to copy the contents of the b-tree node stored +** on page pFrom to page pTo. If page pFrom was not a leaf page, then +** the pointer-map entries for each child page are updated so that the +** parent page stored in the pointer map is page pTo. If pFrom contained +** any cells with overflow page pointers, then the corresponding pointer +** map entries are also updated so that the parent page is page pTo. +** +** If pFrom is currently carrying any overflow cells (entries in the +** MemPage.apOvfl[] array), they are not copied to pTo. +** +** Before returning, page pTo is reinitialized using btreeInitPage(). +** +** The performance of this function is not critical. It is only used by +** the balance_shallower() and balance_deeper() procedures, neither of +** which are called often under normal circumstances. +*/ +static void copyNodeContent(MemPage *pFrom, MemPage *pTo, int *pRC){ + if( (*pRC)==SQLITE_OK ){ + BtShared * const pBt = pFrom->pBt; + u8 * const aFrom = pFrom->aData; + u8 * const aTo = pTo->aData; + int const iFromHdr = pFrom->hdrOffset; + int const iToHdr = ((pTo->pgno==1) ? 100 : 0); + int rc; + int iData; + + + assert( pFrom->isInit ); + assert( pFrom->nFree>=iToHdr ); + assert( get2byte(&aFrom[iFromHdr+5]) <= (int)pBt->usableSize ); + + /* Copy the b-tree node content from page pFrom to page pTo. */ + iData = get2byte(&aFrom[iFromHdr+5]); + memcpy(&aTo[iData], &aFrom[iData], pBt->usableSize-iData); + memcpy(&aTo[iToHdr], &aFrom[iFromHdr], pFrom->cellOffset + 2*pFrom->nCell); + + /* Reinitialize page pTo so that the contents of the MemPage structure + ** match the new data. The initialization of pTo can actually fail under + ** fairly obscure circumstances, even though it is a copy of initialized + ** page pFrom. + */ + pTo->isInit = 0; + rc = btreeInitPage(pTo); + if( rc==SQLITE_OK ) rc = btreeComputeFreeSpace(pTo); + if( rc!=SQLITE_OK ){ + *pRC = rc; + return; + } + + /* If this is an auto-vacuum database, update the pointer-map entries + ** for any b-tree or overflow pages that pTo now contains the pointers to. + */ + if( ISAUTOVACUUM ){ + *pRC = setChildPtrmaps(pTo); + } + } +} + +/* +** This routine redistributes cells on the iParentIdx'th child of pParent +** (hereafter "the page") and up to 2 siblings so that all pages have about the +** same amount of free space. Usually a single sibling on either side of the +** page are used in the balancing, though both siblings might come from one +** side if the page is the first or last child of its parent. If the page +** has fewer than 2 siblings (something which can only happen if the page +** is a root page or a child of a root page) then all available siblings +** participate in the balancing. +** +** The number of siblings of the page might be increased or decreased by +** one or two in an effort to keep pages nearly full but not over full. +** +** Note that when this routine is called, some of the cells on the page +** might not actually be stored in MemPage.aData[]. This can happen +** if the page is overfull. This routine ensures that all cells allocated +** to the page and its siblings fit into MemPage.aData[] before returning. +** +** In the course of balancing the page and its siblings, cells may be +** inserted into or removed from the parent page (pParent). Doing so +** may cause the parent page to become overfull or underfull. If this +** happens, it is the responsibility of the caller to invoke the correct +** balancing routine to fix this problem (see the balance() routine). +** +** If this routine fails for any reason, it might leave the database +** in a corrupted state. So if this routine fails, the database should +** be rolled back. +** +** The third argument to this function, aOvflSpace, is a pointer to a +** buffer big enough to hold one page. If while inserting cells into the parent +** page (pParent) the parent page becomes overfull, this buffer is +** used to store the parent's overflow cells. Because this function inserts +** a maximum of four divider cells into the parent page, and the maximum +** size of a cell stored within an internal node is always less than 1/4 +** of the page-size, the aOvflSpace[] buffer is guaranteed to be large +** enough for all overflow cells. +** +** If aOvflSpace is set to a null pointer, this function returns +** SQLITE_NOMEM. +*/ +static int balance_nonroot( + MemPage *pParent, /* Parent page of siblings being balanced */ + int iParentIdx, /* Index of "the page" in pParent */ + u8 *aOvflSpace, /* page-size bytes of space for parent ovfl */ + int isRoot, /* True if pParent is a root-page */ + int bBulk /* True if this call is part of a bulk load */ +){ + BtShared *pBt; /* The whole database */ + int nMaxCells = 0; /* Allocated size of apCell, szCell, aFrom. */ + int nNew = 0; /* Number of pages in apNew[] */ + int nOld; /* Number of pages in apOld[] */ + int i, j, k; /* Loop counters */ + int nxDiv; /* Next divider slot in pParent->aCell[] */ + int rc = SQLITE_OK; /* The return code */ + u16 leafCorrection; /* 4 if pPage is a leaf. 0 if not */ + int leafData; /* True if pPage is a leaf of a LEAFDATA tree */ + int usableSpace; /* Bytes in pPage beyond the header */ + int pageFlags; /* Value of pPage->aData[0] */ + int iSpace1 = 0; /* First unused byte of aSpace1[] */ + int iOvflSpace = 0; /* First unused byte of aOvflSpace[] */ + int szScratch; /* Size of scratch memory requested */ + MemPage *apOld[NB]; /* pPage and up to two siblings */ + MemPage *apNew[NB+2]; /* pPage and up to NB siblings after balancing */ + u8 *pRight; /* Location in parent of right-sibling pointer */ + u8 *apDiv[NB-1]; /* Divider cells in pParent */ + int cntNew[NB+2]; /* Index in b.paCell[] of cell after i-th page */ + int cntOld[NB+2]; /* Old index in b.apCell[] */ + int szNew[NB+2]; /* Combined size of cells placed on i-th page */ + u8 *aSpace1; /* Space for copies of dividers cells */ + Pgno pgno; /* Temp var to store a page number in */ + u8 abDone[NB+2]; /* True after i'th new page is populated */ + Pgno aPgno[NB+2]; /* Page numbers of new pages before shuffling */ + Pgno aPgOrder[NB+2]; /* Copy of aPgno[] used for sorting pages */ + u16 aPgFlags[NB+2]; /* flags field of new pages before shuffling */ + CellArray b; /* Parsed information on cells being balanced */ + + memset(abDone, 0, sizeof(abDone)); + memset(&b, 0, sizeof(b)); + pBt = pParent->pBt; + assert( sqlite3_mutex_held(pBt->mutex) ); + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + + /* At this point pParent may have at most one overflow cell. And if + ** this overflow cell is present, it must be the cell with + ** index iParentIdx. This scenario comes about when this function + ** is called (indirectly) from sqlite3BtreeDelete(). + */ + assert( pParent->nOverflow==0 || pParent->nOverflow==1 ); + assert( pParent->nOverflow==0 || pParent->aiOvfl[0]==iParentIdx ); + + if( !aOvflSpace ){ + return SQLITE_NOMEM_BKPT; + } + assert( pParent->nFree>=0 ); + + /* Find the sibling pages to balance. Also locate the cells in pParent + ** that divide the siblings. An attempt is made to find NN siblings on + ** either side of pPage. More siblings are taken from one side, however, + ** if there are fewer than NN siblings on the other side. If pParent + ** has NB or fewer children then all children of pParent are taken. + ** + ** This loop also drops the divider cells from the parent page. This + ** way, the remainder of the function does not have to deal with any + ** overflow cells in the parent page, since if any existed they will + ** have already been removed. + */ + i = pParent->nOverflow + pParent->nCell; + if( i<2 ){ + nxDiv = 0; + }else{ + assert( bBulk==0 || bBulk==1 ); + if( iParentIdx==0 ){ + nxDiv = 0; + }else if( iParentIdx==i ){ + nxDiv = i-2+bBulk; + }else{ + nxDiv = iParentIdx-1; + } + i = 2-bBulk; + } + nOld = i+1; + if( (i+nxDiv-pParent->nOverflow)==pParent->nCell ){ + pRight = &pParent->aData[pParent->hdrOffset+8]; + }else{ + pRight = findCell(pParent, i+nxDiv-pParent->nOverflow); + } + pgno = get4byte(pRight); + while( 1 ){ + if( rc==SQLITE_OK ){ + rc = getAndInitPage(pBt, pgno, &apOld[i], 0, 0); + } + if( rc ){ + memset(apOld, 0, (i+1)*sizeof(MemPage*)); + goto balance_cleanup; + } + if( apOld[i]->nFree<0 ){ + rc = btreeComputeFreeSpace(apOld[i]); + if( rc ){ + memset(apOld, 0, (i)*sizeof(MemPage*)); + goto balance_cleanup; + } + } + nMaxCells += apOld[i]->nCell + ArraySize(pParent->apOvfl); + if( (i--)==0 ) break; + + if( pParent->nOverflow && i+nxDiv==pParent->aiOvfl[0] ){ + apDiv[i] = pParent->apOvfl[0]; + pgno = get4byte(apDiv[i]); + szNew[i] = pParent->xCellSize(pParent, apDiv[i]); + pParent->nOverflow = 0; + }else{ + apDiv[i] = findCell(pParent, i+nxDiv-pParent->nOverflow); + pgno = get4byte(apDiv[i]); + szNew[i] = pParent->xCellSize(pParent, apDiv[i]); + + /* Drop the cell from the parent page. apDiv[i] still points to + ** the cell within the parent, even though it has been dropped. + ** This is safe because dropping a cell only overwrites the first + ** four bytes of it, and this function does not need the first + ** four bytes of the divider cell. So the pointer is safe to use + ** later on. + ** + ** But not if we are in secure-delete mode. In secure-delete mode, + ** the dropCell() routine will overwrite the entire cell with zeroes. + ** In this case, temporarily copy the cell into the aOvflSpace[] + ** buffer. It will be copied out again as soon as the aSpace[] buffer + ** is allocated. */ + if( pBt->btsFlags & BTS_FAST_SECURE ){ + int iOff; + + /* If the following if() condition is not true, the db is corrupted. + ** The call to dropCell() below will detect this. */ + iOff = SQLITE_PTR_TO_INT(apDiv[i]) - SQLITE_PTR_TO_INT(pParent->aData); + if( (iOff+szNew[i])<=(int)pBt->usableSize ){ + memcpy(&aOvflSpace[iOff], apDiv[i], szNew[i]); + apDiv[i] = &aOvflSpace[apDiv[i]-pParent->aData]; + } + } + dropCell(pParent, i+nxDiv-pParent->nOverflow, szNew[i], &rc); + } + } + + /* Make nMaxCells a multiple of 4 in order to preserve 8-byte + ** alignment */ + nMaxCells = (nMaxCells + 3)&~3; + + /* + ** Allocate space for memory structures + */ + szScratch = + nMaxCells*sizeof(u8*) /* b.apCell */ + + nMaxCells*sizeof(u16) /* b.szCell */ + + pBt->pageSize; /* aSpace1 */ + + assert( szScratch<=7*(int)pBt->pageSize ); + b.apCell = sqlite3StackAllocRaw(0, szScratch ); + if( b.apCell==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto balance_cleanup; + } + b.szCell = (u16*)&b.apCell[nMaxCells]; + aSpace1 = (u8*)&b.szCell[nMaxCells]; + assert( EIGHT_BYTE_ALIGNMENT(aSpace1) ); + + /* + ** Load pointers to all cells on sibling pages and the divider cells + ** into the local b.apCell[] array. Make copies of the divider cells + ** into space obtained from aSpace1[]. The divider cells have already + ** been removed from pParent. + ** + ** If the siblings are on leaf pages, then the child pointers of the + ** divider cells are stripped from the cells before they are copied + ** into aSpace1[]. In this way, all cells in b.apCell[] are without + ** child pointers. If siblings are not leaves, then all cell in + ** b.apCell[] include child pointers. Either way, all cells in b.apCell[] + ** are alike. + ** + ** leafCorrection: 4 if pPage is a leaf. 0 if pPage is not a leaf. + ** leafData: 1 if pPage holds key+data and pParent holds only keys. + */ + b.pRef = apOld[0]; + leafCorrection = b.pRef->leaf*4; + leafData = b.pRef->intKeyLeaf; + for(i=0; inCell; + u8 *aData = pOld->aData; + u16 maskPage = pOld->maskPage; + u8 *piCell = aData + pOld->cellOffset; + u8 *piEnd; + VVA_ONLY( int nCellAtStart = b.nCell; ) + + /* Verify that all sibling pages are of the same "type" (table-leaf, + ** table-interior, index-leaf, or index-interior). + */ + if( pOld->aData[0]!=apOld[0]->aData[0] ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + + /* Load b.apCell[] with pointers to all cells in pOld. If pOld + ** contains overflow cells, include them in the b.apCell[] array + ** in the correct spot. + ** + ** Note that when there are multiple overflow cells, it is always the + ** case that they are sequential and adjacent. This invariant arises + ** because multiple overflows can only occurs when inserting divider + ** cells into a parent on a prior balance, and divider cells are always + ** adjacent and are inserted in order. There is an assert() tagged + ** with "NOTE 1" in the overflow cell insertion loop to prove this + ** invariant. + ** + ** This must be done in advance. Once the balance starts, the cell + ** offset section of the btree page will be overwritten and we will no + ** long be able to find the cells if a pointer to each cell is not saved + ** first. + */ + memset(&b.szCell[b.nCell], 0, sizeof(b.szCell[0])*(limit+pOld->nOverflow)); + if( pOld->nOverflow>0 ){ + if( NEVER(limitaiOvfl[0]) ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + limit = pOld->aiOvfl[0]; + for(j=0; jnOverflow; k++){ + assert( k==0 || pOld->aiOvfl[k-1]+1==pOld->aiOvfl[k] );/* NOTE 1 */ + b.apCell[b.nCell] = pOld->apOvfl[k]; + b.nCell++; + } + } + piEnd = aData + pOld->cellOffset + 2*pOld->nCell; + while( piCellnCell+pOld->nOverflow) ); + + cntOld[i] = b.nCell; + if( imaxLocal+23 ); + assert( iSpace1 <= (int)pBt->pageSize ); + memcpy(pTemp, apDiv[i], sz); + b.apCell[b.nCell] = pTemp+leafCorrection; + assert( leafCorrection==0 || leafCorrection==4 ); + b.szCell[b.nCell] = b.szCell[b.nCell] - leafCorrection; + if( !pOld->leaf ){ + assert( leafCorrection==0 ); + assert( pOld->hdrOffset==0 || CORRUPT_DB ); + /* The right pointer of the child page pOld becomes the left + ** pointer of the divider cell */ + memcpy(b.apCell[b.nCell], &pOld->aData[8], 4); + }else{ + assert( leafCorrection==4 ); + while( b.szCell[b.nCell]<4 ){ + /* Do not allow any cells smaller than 4 bytes. If a smaller cell + ** does exist, pad it with 0x00 bytes. */ + assert( b.szCell[b.nCell]==3 || CORRUPT_DB ); + assert( b.apCell[b.nCell]==&aSpace1[iSpace1-3] || CORRUPT_DB ); + aSpace1[iSpace1++] = 0x00; + b.szCell[b.nCell]++; + } + } + b.nCell++; + } + } + + /* + ** Figure out the number of pages needed to hold all b.nCell cells. + ** Store this number in "k". Also compute szNew[] which is the total + ** size of all cells on the i-th page and cntNew[] which is the index + ** in b.apCell[] of the cell that divides page i from page i+1. + ** cntNew[k] should equal b.nCell. + ** + ** Values computed by this block: + ** + ** k: The total number of sibling pages + ** szNew[i]: Spaced used on the i-th sibling page. + ** cntNew[i]: Index in b.apCell[] and b.szCell[] for the first cell to + ** the right of the i-th sibling page. + ** usableSpace: Number of bytes of space available on each sibling. + ** + */ + usableSpace = pBt->usableSize - 12 + leafCorrection; + for(i=k=0; iaDataEnd; + b.ixNx[k] = cntOld[i]; + if( k && b.ixNx[k]==b.ixNx[k-1] ){ + k--; /* Omit b.ixNx[] entry for child pages with no cells */ + } + if( !leafData ){ + k++; + b.apEnd[k] = pParent->aDataEnd; + b.ixNx[k] = cntOld[i]+1; + } + assert( p->nFree>=0 ); + szNew[i] = usableSpace - p->nFree; + for(j=0; jnOverflow; j++){ + szNew[i] += 2 + p->xCellSize(p, p->apOvfl[j]); + } + cntNew[i] = cntOld[i]; + } + k = nOld; + for(i=0; iusableSpace ){ + if( i+1>=k ){ + k = i+2; + if( k>NB+2 ){ rc = SQLITE_CORRUPT_BKPT; goto balance_cleanup; } + szNew[k-1] = 0; + cntNew[k-1] = b.nCell; + } + sz = 2 + cachedCellSize(&b, cntNew[i]-1); + szNew[i] -= sz; + if( !leafData ){ + if( cntNew[i]usableSpace ) break; + szNew[i] += sz; + cntNew[i]++; + if( !leafData ){ + if( cntNew[i]=b.nCell ){ + k = i+1; + }else if( cntNew[i] <= (i>0 ? cntNew[i-1] : 0) ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + } + + /* + ** The packing computed by the previous block is biased toward the siblings + ** on the left side (siblings with smaller keys). The left siblings are + ** always nearly full, while the right-most sibling might be nearly empty. + ** The next block of code attempts to adjust the packing of siblings to + ** get a better balance. + ** + ** This adjustment is more than an optimization. The packing above might + ** be so out of balance as to be illegal. For example, the right-most + ** sibling might be completely empty. This adjustment is not optional. + */ + for(i=k-1; i>0; i--){ + int szRight = szNew[i]; /* Size of sibling on the right */ + int szLeft = szNew[i-1]; /* Size of sibling on the left */ + int r; /* Index of right-most cell in left sibling */ + int d; /* Index of first cell to the left of right sibling */ + + r = cntNew[i-1] - 1; + d = r + 1 - leafData; + (void)cachedCellSize(&b, d); + do{ + assert( d szLeft-(b.szCell[r]+(i==k-1?0:2)))){ + break; + } + szRight += b.szCell[d] + 2; + szLeft -= b.szCell[r] + 2; + cntNew[i-1] = r; + r--; + d--; + }while( r>=0 ); + szNew[i] = szRight; + szNew[i-1] = szLeft; + if( cntNew[i-1] <= (i>1 ? cntNew[i-2] : 0) ){ + rc = SQLITE_CORRUPT_BKPT; + goto balance_cleanup; + } + } + + /* Sanity check: For a non-corrupt database file one of the follwing + ** must be true: + ** (1) We found one or more cells (cntNew[0])>0), or + ** (2) pPage is a virtual root page. A virtual root page is when + ** the real root page is page 1 and we are the only child of + ** that page. + */ + assert( cntNew[0]>0 || (pParent->pgno==1 && pParent->nCell==0) || CORRUPT_DB); + TRACE(("BALANCE: old: %d(nc=%d) %d(nc=%d) %d(nc=%d)\n", + apOld[0]->pgno, apOld[0]->nCell, + nOld>=2 ? apOld[1]->pgno : 0, nOld>=2 ? apOld[1]->nCell : 0, + nOld>=3 ? apOld[2]->pgno : 0, nOld>=3 ? apOld[2]->nCell : 0 + )); + + /* + ** Allocate k new pages. Reuse old pages where possible. + */ + pageFlags = apOld[0]->aData[0]; + for(i=0; ipDbPage); + nNew++; + if( sqlite3PagerPageRefcount(pNew->pDbPage)!=1+(i==(iParentIdx-nxDiv)) + && rc==SQLITE_OK + ){ + rc = SQLITE_CORRUPT_BKPT; + } + if( rc ) goto balance_cleanup; + }else{ + assert( i>0 ); + rc = allocateBtreePage(pBt, &pNew, &pgno, (bBulk ? 1 : pgno), 0); + if( rc ) goto balance_cleanup; + zeroPage(pNew, pageFlags); + apNew[i] = pNew; + nNew++; + cntOld[i] = b.nCell; + + /* Set the pointer-map entry for the new sibling page. */ + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, pNew->pgno, PTRMAP_BTREE, pParent->pgno, &rc); + if( rc!=SQLITE_OK ){ + goto balance_cleanup; + } + } + } + } + + /* + ** Reassign page numbers so that the new pages are in ascending order. + ** This helps to keep entries in the disk file in order so that a scan + ** of the table is closer to a linear scan through the file. That in turn + ** helps the operating system to deliver pages from the disk more rapidly. + ** + ** An O(n^2) insertion sort algorithm is used, but since n is never more + ** than (NB+2) (a small constant), that should not be a problem. + ** + ** When NB==3, this one optimization makes the database about 25% faster + ** for large insertions and deletions. + */ + for(i=0; ipgno; + aPgFlags[i] = apNew[i]->pDbPage->flags; + for(j=0; ji ){ + sqlite3PagerRekey(apNew[iBest]->pDbPage, pBt->nPage+iBest+1, 0); + } + sqlite3PagerRekey(apNew[i]->pDbPage, pgno, aPgFlags[iBest]); + apNew[i]->pgno = pgno; + } + } + + TRACE(("BALANCE: new: %d(%d nc=%d) %d(%d nc=%d) %d(%d nc=%d) " + "%d(%d nc=%d) %d(%d nc=%d)\n", + apNew[0]->pgno, szNew[0], cntNew[0], + nNew>=2 ? apNew[1]->pgno : 0, nNew>=2 ? szNew[1] : 0, + nNew>=2 ? cntNew[1] - cntNew[0] - !leafData : 0, + nNew>=3 ? apNew[2]->pgno : 0, nNew>=3 ? szNew[2] : 0, + nNew>=3 ? cntNew[2] - cntNew[1] - !leafData : 0, + nNew>=4 ? apNew[3]->pgno : 0, nNew>=4 ? szNew[3] : 0, + nNew>=4 ? cntNew[3] - cntNew[2] - !leafData : 0, + nNew>=5 ? apNew[4]->pgno : 0, nNew>=5 ? szNew[4] : 0, + nNew>=5 ? cntNew[4] - cntNew[3] - !leafData : 0 + )); + + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + assert( nNew>=1 && nNew<=ArraySize(apNew) ); + assert( apNew[nNew-1]!=0 ); + put4byte(pRight, apNew[nNew-1]->pgno); + + /* If the sibling pages are not leaves, ensure that the right-child pointer + ** of the right-most new sibling page is set to the value that was + ** originally in the same field of the right-most old sibling page. */ + if( (pageFlags & PTF_LEAF)==0 && nOld!=nNew ){ + MemPage *pOld = (nNew>nOld ? apNew : apOld)[nOld-1]; + memcpy(&apNew[nNew-1]->aData[8], &pOld->aData[8], 4); + } + + /* Make any required updates to pointer map entries associated with + ** cells stored on sibling pages following the balance operation. Pointer + ** map entries associated with divider cells are set by the insertCell() + ** routine. The associated pointer map entries are: + ** + ** a) if the cell contains a reference to an overflow chain, the + ** entry associated with the first page in the overflow chain, and + ** + ** b) if the sibling pages are not leaves, the child page associated + ** with the cell. + ** + ** If the sibling pages are not leaves, then the pointer map entry + ** associated with the right-child of each sibling may also need to be + ** updated. This happens below, after the sibling pages have been + ** populated, not here. + */ + if( ISAUTOVACUUM ){ + MemPage *pOld; + MemPage *pNew = pOld = apNew[0]; + int cntOldNext = pNew->nCell + pNew->nOverflow; + int iNew = 0; + int iOld = 0; + + for(i=0; i=0 && iOldnCell + pOld->nOverflow + !leafData; + } + if( i==cntNew[iNew] ){ + pNew = apNew[++iNew]; + if( !leafData ) continue; + } + + /* Cell pCell is destined for new sibling page pNew. Originally, it + ** was either part of sibling page iOld (possibly an overflow cell), + ** or else the divider cell to the left of sibling page iOld. So, + ** if sibling page iOld had the same page number as pNew, and if + ** pCell really was a part of sibling page iOld (not a divider or + ** overflow cell), we can skip updating the pointer map entries. */ + if( iOld>=nNew + || pNew->pgno!=aPgno[iOld] + || !SQLITE_WITHIN(pCell,pOld->aData,pOld->aDataEnd) + ){ + if( !leafCorrection ){ + ptrmapPut(pBt, get4byte(pCell), PTRMAP_BTREE, pNew->pgno, &rc); + } + if( cachedCellSize(&b,i)>pNew->minLocal ){ + ptrmapPutOvflPtr(pNew, pOld, pCell, &rc); + } + if( rc ) goto balance_cleanup; + } + } + } + + /* Insert new divider cells into pParent. */ + for(i=0; ileaf ){ + memcpy(&pNew->aData[8], pCell, 4); + }else if( leafData ){ + /* If the tree is a leaf-data tree, and the siblings are leaves, + ** then there is no divider cell in b.apCell[]. Instead, the divider + ** cell consists of the integer key for the right-most cell of + ** the sibling-page assembled above only. + */ + CellInfo info; + j--; + pNew->xParseCell(pNew, b.apCell[j], &info); + pCell = pTemp; + sz = 4 + putVarint(&pCell[4], info.nKey); + pTemp = 0; + }else{ + pCell -= 4; + /* Obscure case for non-leaf-data trees: If the cell at pCell was + ** previously stored on a leaf node, and its reported size was 4 + ** bytes, then it may actually be smaller than this + ** (see btreeParseCellPtr(), 4 bytes is the minimum size of + ** any cell). But it is important to pass the correct size to + ** insertCell(), so reparse the cell now. + ** + ** This can only happen for b-trees used to evaluate "IN (SELECT ...)" + ** and WITHOUT ROWID tables with exactly one column which is the + ** primary key. + */ + if( b.szCell[j]==4 ){ + assert(leafCorrection==4); + sz = pParent->xCellSize(pParent, pCell); + } + } + iOvflSpace += sz; + assert( sz<=pBt->maxLocal+23 ); + assert( iOvflSpace <= (int)pBt->pageSize ); + for(k=0; b.ixNx[k]<=i && ALWAYS(kpgno, &rc); + if( rc!=SQLITE_OK ) goto balance_cleanup; + assert( sqlite3PagerIswriteable(pParent->pDbPage) ); + } + + /* Now update the actual sibling pages. The order in which they are updated + ** is important, as this code needs to avoid disrupting any page from which + ** cells may still to be read. In practice, this means: + ** + ** (1) If cells are moving left (from apNew[iPg] to apNew[iPg-1]) + ** then it is not safe to update page apNew[iPg] until after + ** the left-hand sibling apNew[iPg-1] has been updated. + ** + ** (2) If cells are moving right (from apNew[iPg] to apNew[iPg+1]) + ** then it is not safe to update page apNew[iPg] until after + ** the right-hand sibling apNew[iPg+1] has been updated. + ** + ** If neither of the above apply, the page is safe to update. + ** + ** The iPg value in the following loop starts at nNew-1 goes down + ** to 0, then back up to nNew-1 again, thus making two passes over + ** the pages. On the initial downward pass, only condition (1) above + ** needs to be tested because (2) will always be true from the previous + ** step. On the upward pass, both conditions are always true, so the + ** upwards pass simply processes pages that were missed on the downward + ** pass. + */ + for(i=1-nNew; i=0 && iPg=0 /* On the upwards pass, or... */ + || cntOld[iPg-1]>=cntNew[iPg-1] /* Condition (1) is true */ + ){ + int iNew; + int iOld; + int nNewCell; + + /* Verify condition (1): If cells are moving left, update iPg + ** only after iPg-1 has already been updated. */ + assert( iPg==0 || cntOld[iPg-1]>=cntNew[iPg-1] || abDone[iPg-1] ); + + /* Verify condition (2): If cells are moving right, update iPg + ** only after iPg+1 has already been updated. */ + assert( cntNew[iPg]>=cntOld[iPg] || abDone[iPg+1] ); + + if( iPg==0 ){ + iNew = iOld = 0; + nNewCell = cntNew[0]; + }else{ + iOld = iPgnFree = usableSpace-szNew[iPg]; + assert( apNew[iPg]->nOverflow==0 ); + assert( apNew[iPg]->nCell==nNewCell ); + } + } + + /* All pages have been processed exactly once */ + assert( memcmp(abDone, "\01\01\01\01\01", nNew)==0 ); + + assert( nOld>0 ); + assert( nNew>0 ); + + if( isRoot && pParent->nCell==0 && pParent->hdrOffset<=apNew[0]->nFree ){ + /* The root page of the b-tree now contains no cells. The only sibling + ** page is the right-child of the parent. Copy the contents of the + ** child page into the parent, decreasing the overall height of the + ** b-tree structure by one. This is described as the "balance-shallower" + ** sub-algorithm in some documentation. + ** + ** If this is an auto-vacuum database, the call to copyNodeContent() + ** sets all pointer-map entries corresponding to database image pages + ** for which the pointer is stored within the content being copied. + ** + ** It is critical that the child page be defragmented before being + ** copied into the parent, because if the parent is page 1 then it will + ** by smaller than the child due to the database header, and so all the + ** free space needs to be up front. + */ + assert( nNew==1 || CORRUPT_DB ); + rc = defragmentPage(apNew[0], -1); + testcase( rc!=SQLITE_OK ); + assert( apNew[0]->nFree == + (get2byteNotZero(&apNew[0]->aData[5]) - apNew[0]->cellOffset + - apNew[0]->nCell*2) + || rc!=SQLITE_OK + ); + copyNodeContent(apNew[0], pParent, &rc); + freePage(apNew[0], &rc); + }else if( ISAUTOVACUUM && !leafCorrection ){ + /* Fix the pointer map entries associated with the right-child of each + ** sibling page. All other pointer map entries have already been taken + ** care of. */ + for(i=0; iaData[8]); + ptrmapPut(pBt, key, PTRMAP_BTREE, apNew[i]->pgno, &rc); + } + } + + assert( pParent->isInit ); + TRACE(("BALANCE: finished: old=%d new=%d cells=%d\n", + nOld, nNew, b.nCell)); + + /* Free any old pages that were not reused as new pages. + */ + for(i=nNew; iisInit ){ + /* The ptrmapCheckPages() contains assert() statements that verify that + ** all pointer map pages are set correctly. This is helpful while + ** debugging. This is usually disabled because a corrupt database may + ** cause an assert() statement to fail. */ + ptrmapCheckPages(apNew, nNew); + ptrmapCheckPages(&pParent, 1); + } +#endif + + /* + ** Cleanup before returning. + */ +balance_cleanup: + sqlite3StackFree(0, b.apCell); + for(i=0; ipBt; /* The BTree */ + + assert( pRoot->nOverflow>0 ); + assert( sqlite3_mutex_held(pBt->mutex) ); + + /* Make pRoot, the root page of the b-tree, writable. Allocate a new + ** page that will become the new right-child of pPage. Copy the contents + ** of the node stored on pRoot into the new child page. + */ + rc = sqlite3PagerWrite(pRoot->pDbPage); + if( rc==SQLITE_OK ){ + rc = allocateBtreePage(pBt,&pChild,&pgnoChild,pRoot->pgno,0); + copyNodeContent(pRoot, pChild, &rc); + if( ISAUTOVACUUM ){ + ptrmapPut(pBt, pgnoChild, PTRMAP_BTREE, pRoot->pgno, &rc); + } + } + if( rc ){ + *ppChild = 0; + releasePage(pChild); + return rc; + } + assert( sqlite3PagerIswriteable(pChild->pDbPage) ); + assert( sqlite3PagerIswriteable(pRoot->pDbPage) ); + assert( pChild->nCell==pRoot->nCell || CORRUPT_DB ); + + TRACE(("BALANCE: copy root %d into %d\n", pRoot->pgno, pChild->pgno)); + + /* Copy the overflow cells from pRoot to pChild */ + memcpy(pChild->aiOvfl, pRoot->aiOvfl, + pRoot->nOverflow*sizeof(pRoot->aiOvfl[0])); + memcpy(pChild->apOvfl, pRoot->apOvfl, + pRoot->nOverflow*sizeof(pRoot->apOvfl[0])); + pChild->nOverflow = pRoot->nOverflow; + + /* Zero the contents of pRoot. Then install pChild as the right-child. */ + zeroPage(pRoot, pChild->aData[0] & ~PTF_LEAF); + put4byte(&pRoot->aData[pRoot->hdrOffset+8], pgnoChild); + + *ppChild = pChild; + return SQLITE_OK; +} + +/* +** Return SQLITE_CORRUPT if any cursor other than pCur is currently valid +** on the same B-tree as pCur. +** +** This can occur if a database is corrupt with two or more SQL tables +** pointing to the same b-tree. If an insert occurs on one SQL table +** and causes a BEFORE TRIGGER to do a secondary insert on the other SQL +** table linked to the same b-tree. If the secondary insert causes a +** rebalance, that can change content out from under the cursor on the +** first SQL table, violating invariants on the first insert. +*/ +static int anotherValidCursor(BtCursor *pCur){ + BtCursor *pOther; + for(pOther=pCur->pBt->pCursor; pOther; pOther=pOther->pNext){ + if( pOther!=pCur + && pOther->eState==CURSOR_VALID + && pOther->pPage==pCur->pPage + ){ + return SQLITE_CORRUPT_BKPT; + } + } + return SQLITE_OK; +} + +/* +** The page that pCur currently points to has just been modified in +** some way. This function figures out if this modification means the +** tree needs to be balanced, and if so calls the appropriate balancing +** routine. Balancing routines are: +** +** balance_quick() +** balance_deeper() +** balance_nonroot() +*/ +static int balance(BtCursor *pCur){ + int rc = SQLITE_OK; + const int nMin = pCur->pBt->usableSize * 2 / 3; + u8 aBalanceQuickSpace[13]; + u8 *pFree = 0; + + VVA_ONLY( int balance_quick_called = 0 ); + VVA_ONLY( int balance_deeper_called = 0 ); + + do { + int iPage; + MemPage *pPage = pCur->pPage; + + if( NEVER(pPage->nFree<0) && btreeComputeFreeSpace(pPage) ) break; + if( pPage->nOverflow==0 && pPage->nFree<=nMin ){ + break; + }else if( (iPage = pCur->iPage)==0 ){ + if( pPage->nOverflow && (rc = anotherValidCursor(pCur))==SQLITE_OK ){ + /* The root page of the b-tree is overfull. In this case call the + ** balance_deeper() function to create a new child for the root-page + ** and copy the current contents of the root-page to it. The + ** next iteration of the do-loop will balance the child page. + */ + assert( balance_deeper_called==0 ); + VVA_ONLY( balance_deeper_called++ ); + rc = balance_deeper(pPage, &pCur->apPage[1]); + if( rc==SQLITE_OK ){ + pCur->iPage = 1; + pCur->ix = 0; + pCur->aiIdx[0] = 0; + pCur->apPage[0] = pPage; + pCur->pPage = pCur->apPage[1]; + assert( pCur->pPage->nOverflow ); + } + }else{ + break; + } + }else{ + MemPage * const pParent = pCur->apPage[iPage-1]; + int const iIdx = pCur->aiIdx[iPage-1]; + + rc = sqlite3PagerWrite(pParent->pDbPage); + if( rc==SQLITE_OK && pParent->nFree<0 ){ + rc = btreeComputeFreeSpace(pParent); + } + if( rc==SQLITE_OK ){ +#ifndef SQLITE_OMIT_QUICKBALANCE + if( pPage->intKeyLeaf + && pPage->nOverflow==1 + && pPage->aiOvfl[0]==pPage->nCell + && pParent->pgno!=1 + && pParent->nCell==iIdx + ){ + /* Call balance_quick() to create a new sibling of pPage on which + ** to store the overflow cell. balance_quick() inserts a new cell + ** into pParent, which may cause pParent overflow. If this + ** happens, the next iteration of the do-loop will balance pParent + ** use either balance_nonroot() or balance_deeper(). Until this + ** happens, the overflow cell is stored in the aBalanceQuickSpace[] + ** buffer. + ** + ** The purpose of the following assert() is to check that only a + ** single call to balance_quick() is made for each call to this + ** function. If this were not verified, a subtle bug involving reuse + ** of the aBalanceQuickSpace[] might sneak in. + */ + assert( balance_quick_called==0 ); + VVA_ONLY( balance_quick_called++ ); + rc = balance_quick(pParent, pPage, aBalanceQuickSpace); + }else +#endif + { + /* In this case, call balance_nonroot() to redistribute cells + ** between pPage and up to 2 of its sibling pages. This involves + ** modifying the contents of pParent, which may cause pParent to + ** become overfull or underfull. The next iteration of the do-loop + ** will balance the parent page to correct this. + ** + ** If the parent page becomes overfull, the overflow cell or cells + ** are stored in the pSpace buffer allocated immediately below. + ** A subsequent iteration of the do-loop will deal with this by + ** calling balance_nonroot() (balance_deeper() may be called first, + ** but it doesn't deal with overflow cells - just moves them to a + ** different page). Once this subsequent call to balance_nonroot() + ** has completed, it is safe to release the pSpace buffer used by + ** the previous call, as the overflow cell data will have been + ** copied either into the body of a database page or into the new + ** pSpace buffer passed to the latter call to balance_nonroot(). + */ + u8 *pSpace = sqlite3PageMalloc(pCur->pBt->pageSize); + rc = balance_nonroot(pParent, iIdx, pSpace, iPage==1, + pCur->hints&BTREE_BULKLOAD); + if( pFree ){ + /* If pFree is not NULL, it points to the pSpace buffer used + ** by a previous call to balance_nonroot(). Its contents are + ** now stored either on real database pages or within the + ** new pSpace buffer, so it may be safely freed here. */ + sqlite3PageFree(pFree); + } + + /* The pSpace buffer will be freed after the next call to + ** balance_nonroot(), or just before this function returns, whichever + ** comes first. */ + pFree = pSpace; + } + } + + pPage->nOverflow = 0; + + /* The next iteration of the do-loop balances the parent page. */ + releasePage(pPage); + pCur->iPage--; + assert( pCur->iPage>=0 ); + pCur->pPage = pCur->apPage[pCur->iPage]; + } + }while( rc==SQLITE_OK ); + + if( pFree ){ + sqlite3PageFree(pFree); + } + return rc; +} + +/* Overwrite content from pX into pDest. Only do the write if the +** content is different from what is already there. +*/ +static int btreeOverwriteContent( + MemPage *pPage, /* MemPage on which writing will occur */ + u8 *pDest, /* Pointer to the place to start writing */ + const BtreePayload *pX, /* Source of data to write */ + int iOffset, /* Offset of first byte to write */ + int iAmt /* Number of bytes to be written */ +){ + int nData = pX->nData - iOffset; + if( nData<=0 ){ + /* Overwritting with zeros */ + int i; + for(i=0; ipDbPage); + if( rc ) return rc; + memset(pDest + i, 0, iAmt - i); + } + }else{ + if( nDatapData) + iOffset, iAmt)!=0 ){ + int rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc ) return rc; + /* In a corrupt database, it is possible for the source and destination + ** buffers to overlap. This is harmless since the database is already + ** corrupt but it does cause valgrind and ASAN warnings. So use + ** memmove(). */ + memmove(pDest, ((u8*)pX->pData) + iOffset, iAmt); + } + } + return SQLITE_OK; +} + +/* +** Overwrite the cell that cursor pCur is pointing to with fresh content +** contained in pX. +*/ +static int btreeOverwriteCell(BtCursor *pCur, const BtreePayload *pX){ + int iOffset; /* Next byte of pX->pData to write */ + int nTotal = pX->nData + pX->nZero; /* Total bytes of to write */ + int rc; /* Return code */ + MemPage *pPage = pCur->pPage; /* Page being written */ + BtShared *pBt; /* Btree */ + Pgno ovflPgno; /* Next overflow page to write */ + u32 ovflPageSize; /* Size to write on overflow page */ + + if( pCur->info.pPayload + pCur->info.nLocal > pPage->aDataEnd + || pCur->info.pPayload < pPage->aData + pPage->cellOffset + ){ + return SQLITE_CORRUPT_BKPT; + } + /* Overwrite the local portion first */ + rc = btreeOverwriteContent(pPage, pCur->info.pPayload, pX, + 0, pCur->info.nLocal); + if( rc ) return rc; + if( pCur->info.nLocal==nTotal ) return SQLITE_OK; + + /* Now overwrite the overflow pages */ + iOffset = pCur->info.nLocal; + assert( nTotal>=0 ); + assert( iOffset>=0 ); + ovflPgno = get4byte(pCur->info.pPayload + iOffset); + pBt = pPage->pBt; + ovflPageSize = pBt->usableSize - 4; + do{ + rc = btreeGetPage(pBt, ovflPgno, &pPage, 0); + if( rc ) return rc; + if( sqlite3PagerPageRefcount(pPage->pDbPage)!=1 || pPage->isInit ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + if( iOffset+ovflPageSize<(u32)nTotal ){ + ovflPgno = get4byte(pPage->aData); + }else{ + ovflPageSize = nTotal - iOffset; + } + rc = btreeOverwriteContent(pPage, pPage->aData+4, pX, + iOffset, ovflPageSize); + } + sqlite3PagerUnref(pPage->pDbPage); + if( rc ) return rc; + iOffset += ovflPageSize; + }while( iOffset0 then pCur points to a cell +** that is larger than (pKey,nKey). +** +** If seekResult==0, that means pCur is pointing at some unknown location. +** In that case, this routine must seek the cursor to the correct insertion +** point for (pKey,nKey) before doing the insertion. For index btrees, +** if pX->nMem is non-zero, then pX->aMem contains pointers to the unpacked +** key values and pX->aMem can be used instead of pX->pKey to avoid having +** to decode the key. +*/ +int sqlite3BtreeInsert( + BtCursor *pCur, /* Insert data into the table of this cursor */ + const BtreePayload *pX, /* Content of the row to be inserted */ + int flags, /* True if this is likely an append */ + int seekResult /* Result of prior MovetoUnpacked() call */ +){ + int rc; + int loc = seekResult; /* -1: before desired location +1: after */ + int szNew = 0; + int idx; + MemPage *pPage; + Btree *p = pCur->pBtree; + BtShared *pBt = p->pBt; + unsigned char *oldCell; + unsigned char *newCell = 0; + + assert( (flags & (BTREE_SAVEPOSITION|BTREE_APPEND|BTREE_PREFORMAT))==flags ); + assert( (flags & BTREE_PREFORMAT)==0 || seekResult || pCur->pKeyInfo==0 ); + + if( pCur->eState==CURSOR_FAULT ){ + assert( pCur->skipNext!=SQLITE_OK ); + return pCur->skipNext; + } + + assert( cursorOwnsBtShared(pCur) ); + assert( (pCur->curFlags & BTCF_WriteFlag)!=0 + && pBt->inTransaction==TRANS_WRITE + && (pBt->btsFlags & BTS_READ_ONLY)==0 ); + assert( hasSharedCacheTableLock(p, pCur->pgnoRoot, pCur->pKeyInfo!=0, 2) ); + + /* Assert that the caller has been consistent. If this cursor was opened + ** expecting an index b-tree, then the caller should be inserting blob + ** keys with no associated data. If the cursor was opened expecting an + ** intkey table, the caller should be inserting integer keys with a + ** blob of associated data. */ + assert( (flags & BTREE_PREFORMAT) || (pX->pKey==0)==(pCur->pKeyInfo==0) ); + + /* Save the positions of any other cursors open on this table. + ** + ** In some cases, the call to btreeMoveto() below is a no-op. For + ** example, when inserting data into a table with auto-generated integer + ** keys, the VDBE layer invokes sqlite3BtreeLast() to figure out the + ** integer key to use. It then calls this function to actually insert the + ** data into the intkey B-Tree. In this case btreeMoveto() recognizes + ** that the cursor is already where it needs to be and returns without + ** doing any work. To avoid thwarting these optimizations, it is important + ** not to clear the cursor here. + */ + if( pCur->curFlags & BTCF_Multiple ){ + rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur); + if( rc ) return rc; + if( loc && pCur->iPage<0 ){ + /* This can only happen if the schema is corrupt such that there is more + ** than one table or index with the same root page as used by the cursor. + ** Which can only happen if the SQLITE_NoSchemaError flag was set when + ** the schema was loaded. This cannot be asserted though, as a user might + ** set the flag, load the schema, and then unset the flag. */ + return SQLITE_CORRUPT_BKPT; + } + } + + if( pCur->pKeyInfo==0 ){ + assert( pX->pKey==0 ); + /* If this is an insert into a table b-tree, invalidate any incrblob + ** cursors open on the row being replaced */ + if( p->hasIncrblobCur ){ + invalidateIncrblobCursors(p, pCur->pgnoRoot, pX->nKey, 0); + } + + /* If BTREE_SAVEPOSITION is set, the cursor must already be pointing + ** to a row with the same key as the new entry being inserted. + */ +#ifdef SQLITE_DEBUG + if( flags & BTREE_SAVEPOSITION ){ + assert( pCur->curFlags & BTCF_ValidNKey ); + assert( pX->nKey==pCur->info.nKey ); + assert( loc==0 ); + } +#endif + + /* On the other hand, BTREE_SAVEPOSITION==0 does not imply + ** that the cursor is not pointing to a row to be overwritten. + ** So do a complete check. + */ + if( (pCur->curFlags&BTCF_ValidNKey)!=0 && pX->nKey==pCur->info.nKey ){ + /* The cursor is pointing to the entry that is to be + ** overwritten */ + assert( pX->nData>=0 && pX->nZero>=0 ); + if( pCur->info.nSize!=0 + && pCur->info.nPayload==(u32)pX->nData+pX->nZero + ){ + /* New entry is the same size as the old. Do an overwrite */ + return btreeOverwriteCell(pCur, pX); + } + assert( loc==0 ); + }else if( loc==0 ){ + /* The cursor is *not* pointing to the cell to be overwritten, nor + ** to an adjacent cell. Move the cursor so that it is pointing either + ** to the cell to be overwritten or an adjacent cell. + */ + rc = sqlite3BtreeTableMoveto(pCur, pX->nKey, + (flags & BTREE_APPEND)!=0, &loc); + if( rc ) return rc; + } + }else{ + /* This is an index or a WITHOUT ROWID table */ + + /* If BTREE_SAVEPOSITION is set, the cursor must already be pointing + ** to a row with the same key as the new entry being inserted. + */ + assert( (flags & BTREE_SAVEPOSITION)==0 || loc==0 ); + + /* If the cursor is not already pointing either to the cell to be + ** overwritten, or if a new cell is being inserted, if the cursor is + ** not pointing to an immediately adjacent cell, then move the cursor + ** so that it does. + */ + if( loc==0 && (flags & BTREE_SAVEPOSITION)==0 ){ + if( pX->nMem ){ + UnpackedRecord r; + r.pKeyInfo = pCur->pKeyInfo; + r.aMem = pX->aMem; + r.nField = pX->nMem; + r.default_rc = 0; + r.eqSeen = 0; + rc = sqlite3BtreeIndexMoveto(pCur, &r, &loc); + }else{ + rc = btreeMoveto(pCur, pX->pKey, pX->nKey, + (flags & BTREE_APPEND)!=0, &loc); + } + if( rc ) return rc; + } + + /* If the cursor is currently pointing to an entry to be overwritten + ** and the new content is the same as as the old, then use the + ** overwrite optimization. + */ + if( loc==0 ){ + getCellInfo(pCur); + if( pCur->info.nKey==pX->nKey ){ + BtreePayload x2; + x2.pData = pX->pKey; + x2.nData = pX->nKey; + x2.nZero = 0; + return btreeOverwriteCell(pCur, &x2); + } + } + } + assert( pCur->eState==CURSOR_VALID + || (pCur->eState==CURSOR_INVALID && loc) + || CORRUPT_DB ); + + pPage = pCur->pPage; + assert( pPage->intKey || pX->nKey>=0 || (flags & BTREE_PREFORMAT) ); + assert( pPage->leaf || !pPage->intKey ); + if( pPage->nFree<0 ){ + if( NEVER(pCur->eState>CURSOR_INVALID) ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + rc = btreeComputeFreeSpace(pPage); + } + if( rc ) return rc; + } + + TRACE(("INSERT: table=%d nkey=%lld ndata=%d page=%d %s\n", + pCur->pgnoRoot, pX->nKey, pX->nData, pPage->pgno, + loc==0 ? "overwrite" : "new entry")); + assert( pPage->isInit ); + newCell = pBt->pTmpSpace; + assert( newCell!=0 ); + if( flags & BTREE_PREFORMAT ){ + rc = SQLITE_OK; + szNew = pBt->nPreformatSize; + if( szNew<4 ) szNew = 4; + if( ISAUTOVACUUM && szNew>pPage->maxLocal ){ + CellInfo info; + pPage->xParseCell(pPage, newCell, &info); + if( info.nPayload!=info.nLocal ){ + Pgno ovfl = get4byte(&newCell[szNew-4]); + ptrmapPut(pBt, ovfl, PTRMAP_OVERFLOW1, pPage->pgno, &rc); + } + } + }else{ + rc = fillInCell(pPage, newCell, pX, &szNew); + } + if( rc ) goto end_insert; + assert( szNew==pPage->xCellSize(pPage, newCell) ); + assert( szNew <= MX_CELL_SIZE(pBt) ); + idx = pCur->ix; + if( loc==0 ){ + CellInfo info; + assert( idx>=0 ); + if( idx>=pPage->nCell ){ + return SQLITE_CORRUPT_BKPT; + } + rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc ){ + goto end_insert; + } + oldCell = findCell(pPage, idx); + if( !pPage->leaf ){ + memcpy(newCell, oldCell, 4); + } + BTREE_CLEAR_CELL(rc, pPage, oldCell, info); + testcase( pCur->curFlags & BTCF_ValidOvfl ); + invalidateOverflowCache(pCur); + if( info.nSize==szNew && info.nLocal==info.nPayload + && (!ISAUTOVACUUM || szNewminLocal) + ){ + /* Overwrite the old cell with the new if they are the same size. + ** We could also try to do this if the old cell is smaller, then add + ** the leftover space to the free list. But experiments show that + ** doing that is no faster then skipping this optimization and just + ** calling dropCell() and insertCell(). + ** + ** This optimization cannot be used on an autovacuum database if the + ** new entry uses overflow pages, as the insertCell() call below is + ** necessary to add the PTRMAP_OVERFLOW1 pointer-map entry. */ + assert( rc==SQLITE_OK ); /* clearCell never fails when nLocal==nPayload */ + if( oldCell < pPage->aData+pPage->hdrOffset+10 ){ + return SQLITE_CORRUPT_BKPT; + } + if( oldCell+szNew > pPage->aDataEnd ){ + return SQLITE_CORRUPT_BKPT; + } + memcpy(oldCell, newCell, szNew); + return SQLITE_OK; + } + dropCell(pPage, idx, info.nSize, &rc); + if( rc ) goto end_insert; + }else if( loc<0 && pPage->nCell>0 ){ + assert( pPage->leaf ); + idx = ++pCur->ix; + pCur->curFlags &= ~BTCF_ValidNKey; + }else{ + assert( pPage->leaf ); + } + insertCell(pPage, idx, newCell, szNew, 0, 0, &rc); + assert( pPage->nOverflow==0 || rc==SQLITE_OK ); + assert( rc!=SQLITE_OK || pPage->nCell>0 || pPage->nOverflow>0 ); + + /* If no error has occurred and pPage has an overflow cell, call balance() + ** to redistribute the cells within the tree. Since balance() may move + ** the cursor, zero the BtCursor.info.nSize and BTCF_ValidNKey + ** variables. + ** + ** Previous versions of SQLite called moveToRoot() to move the cursor + ** back to the root page as balance() used to invalidate the contents + ** of BtCursor.apPage[] and BtCursor.aiIdx[]. Instead of doing that, + ** set the cursor state to "invalid". This makes common insert operations + ** slightly faster. + ** + ** There is a subtle but important optimization here too. When inserting + ** multiple records into an intkey b-tree using a single cursor (as can + ** happen while processing an "INSERT INTO ... SELECT" statement), it + ** is advantageous to leave the cursor pointing to the last entry in + ** the b-tree if possible. If the cursor is left pointing to the last + ** entry in the table, and the next row inserted has an integer key + ** larger than the largest existing key, it is possible to insert the + ** row without seeking the cursor. This can be a big performance boost. + */ + pCur->info.nSize = 0; + if( pPage->nOverflow ){ + assert( rc==SQLITE_OK ); + pCur->curFlags &= ~(BTCF_ValidNKey); + rc = balance(pCur); + + /* Must make sure nOverflow is reset to zero even if the balance() + ** fails. Internal data structure corruption will result otherwise. + ** Also, set the cursor state to invalid. This stops saveCursorPosition() + ** from trying to save the current position of the cursor. */ + pCur->pPage->nOverflow = 0; + pCur->eState = CURSOR_INVALID; + if( (flags & BTREE_SAVEPOSITION) && rc==SQLITE_OK ){ + btreeReleaseAllCursorPages(pCur); + if( pCur->pKeyInfo ){ + assert( pCur->pKey==0 ); + pCur->pKey = sqlite3Malloc( pX->nKey ); + if( pCur->pKey==0 ){ + rc = SQLITE_NOMEM; + }else{ + memcpy(pCur->pKey, pX->pKey, pX->nKey); + } + } + pCur->eState = CURSOR_REQUIRESEEK; + pCur->nKey = pX->nKey; + } + } + assert( pCur->iPage<0 || pCur->pPage->nOverflow==0 ); + +end_insert: + return rc; +} + +/* +** This function is used as part of copying the current row from cursor +** pSrc into cursor pDest. If the cursors are open on intkey tables, then +** parameter iKey is used as the rowid value when the record is copied +** into pDest. Otherwise, the record is copied verbatim. +** +** This function does not actually write the new value to cursor pDest. +** Instead, it creates and populates any required overflow pages and +** writes the data for the new cell into the BtShared.pTmpSpace buffer +** for the destination database. The size of the cell, in bytes, is left +** in BtShared.nPreformatSize. The caller completes the insertion by +** calling sqlite3BtreeInsert() with the BTREE_PREFORMAT flag specified. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +int sqlite3BtreeTransferRow(BtCursor *pDest, BtCursor *pSrc, i64 iKey){ + int rc = SQLITE_OK; + BtShared *pBt = pDest->pBt; + u8 *aOut = pBt->pTmpSpace; /* Pointer to next output buffer */ + const u8 *aIn; /* Pointer to next input buffer */ + u32 nIn; /* Size of input buffer aIn[] */ + u32 nRem; /* Bytes of data still to copy */ + + getCellInfo(pSrc); + aOut += putVarint32(aOut, pSrc->info.nPayload); + if( pDest->pKeyInfo==0 ) aOut += putVarint(aOut, iKey); + nIn = pSrc->info.nLocal; + aIn = pSrc->info.pPayload; + if( aIn+nIn>pSrc->pPage->aDataEnd ){ + return SQLITE_CORRUPT_BKPT; + } + nRem = pSrc->info.nPayload; + if( nIn==nRem && nInpPage->maxLocal ){ + memcpy(aOut, aIn, nIn); + pBt->nPreformatSize = nIn + (aOut - pBt->pTmpSpace); + }else{ + Pager *pSrcPager = pSrc->pBt->pPager; + u8 *pPgnoOut = 0; + Pgno ovflIn = 0; + DbPage *pPageIn = 0; + MemPage *pPageOut = 0; + u32 nOut; /* Size of output buffer aOut[] */ + + nOut = btreePayloadToLocal(pDest->pPage, pSrc->info.nPayload); + pBt->nPreformatSize = nOut + (aOut - pBt->pTmpSpace); + if( nOutinfo.nPayload ){ + pPgnoOut = &aOut[nOut]; + pBt->nPreformatSize += 4; + } + + if( nRem>nIn ){ + if( aIn+nIn+4>pSrc->pPage->aDataEnd ){ + return SQLITE_CORRUPT_BKPT; + } + ovflIn = get4byte(&pSrc->info.pPayload[nIn]); + } + + do { + nRem -= nOut; + do{ + assert( nOut>0 ); + if( nIn>0 ){ + int nCopy = MIN(nOut, nIn); + memcpy(aOut, aIn, nCopy); + nOut -= nCopy; + nIn -= nCopy; + aOut += nCopy; + aIn += nCopy; + } + if( nOut>0 ){ + sqlite3PagerUnref(pPageIn); + pPageIn = 0; + rc = sqlite3PagerGet(pSrcPager, ovflIn, &pPageIn, PAGER_GET_READONLY); + if( rc==SQLITE_OK ){ + aIn = (const u8*)sqlite3PagerGetData(pPageIn); + ovflIn = get4byte(aIn); + aIn += 4; + nIn = pSrc->pBt->usableSize - 4; + } + } + }while( rc==SQLITE_OK && nOut>0 ); + + if( rc==SQLITE_OK && nRem>0 && ALWAYS(pPgnoOut) ){ + Pgno pgnoNew; + MemPage *pNew = 0; + rc = allocateBtreePage(pBt, &pNew, &pgnoNew, 0, 0); + put4byte(pPgnoOut, pgnoNew); + if( ISAUTOVACUUM && pPageOut ){ + ptrmapPut(pBt, pgnoNew, PTRMAP_OVERFLOW2, pPageOut->pgno, &rc); + } + releasePage(pPageOut); + pPageOut = pNew; + if( pPageOut ){ + pPgnoOut = pPageOut->aData; + put4byte(pPgnoOut, 0); + aOut = &pPgnoOut[4]; + nOut = MIN(pBt->usableSize - 4, nRem); + } + } + }while( nRem>0 && rc==SQLITE_OK ); + + releasePage(pPageOut); + sqlite3PagerUnref(pPageIn); + } + + return rc; +} + +/* +** Delete the entry that the cursor is pointing to. +** +** If the BTREE_SAVEPOSITION bit of the flags parameter is zero, then +** the cursor is left pointing at an arbitrary location after the delete. +** But if that bit is set, then the cursor is left in a state such that +** the next call to BtreeNext() or BtreePrev() moves it to the same row +** as it would have been on if the call to BtreeDelete() had been omitted. +** +** The BTREE_AUXDELETE bit of flags indicates that is one of several deletes +** associated with a single table entry and its indexes. Only one of those +** deletes is considered the "primary" delete. The primary delete occurs +** on a cursor that is not a BTREE_FORDELETE cursor. All but one delete +** operation on non-FORDELETE cursors is tagged with the AUXDELETE flag. +** The BTREE_AUXDELETE bit is a hint that is not used by this implementation, +** but which might be used by alternative storage engines. +*/ +int sqlite3BtreeDelete(BtCursor *pCur, u8 flags){ + Btree *p = pCur->pBtree; + BtShared *pBt = p->pBt; + int rc; /* Return code */ + MemPage *pPage; /* Page to delete cell from */ + unsigned char *pCell; /* Pointer to cell to delete */ + int iCellIdx; /* Index of cell to delete */ + int iCellDepth; /* Depth of node containing pCell */ + CellInfo info; /* Size of the cell being deleted */ + int bSkipnext = 0; /* Leaf cursor in SKIPNEXT state */ + u8 bPreserve = flags & BTREE_SAVEPOSITION; /* Keep cursor valid */ + + assert( cursorOwnsBtShared(pCur) ); + assert( pBt->inTransaction==TRANS_WRITE ); + assert( (pBt->btsFlags & BTS_READ_ONLY)==0 ); + assert( pCur->curFlags & BTCF_WriteFlag ); + assert( hasSharedCacheTableLock(p, pCur->pgnoRoot, pCur->pKeyInfo!=0, 2) ); + assert( !hasReadConflicts(p, pCur->pgnoRoot) ); + assert( (flags & ~(BTREE_SAVEPOSITION | BTREE_AUXDELETE))==0 ); + if( pCur->eState==CURSOR_REQUIRESEEK ){ + rc = btreeRestoreCursorPosition(pCur); + assert( rc!=SQLITE_OK || CORRUPT_DB || pCur->eState==CURSOR_VALID ); + if( rc || pCur->eState!=CURSOR_VALID ) return rc; + } + assert( CORRUPT_DB || pCur->eState==CURSOR_VALID ); + + iCellDepth = pCur->iPage; + iCellIdx = pCur->ix; + pPage = pCur->pPage; + pCell = findCell(pPage, iCellIdx); + if( pPage->nFree<0 && btreeComputeFreeSpace(pPage) ) return SQLITE_CORRUPT; + + /* If the bPreserve flag is set to true, then the cursor position must + ** be preserved following this delete operation. If the current delete + ** will cause a b-tree rebalance, then this is done by saving the cursor + ** key and leaving the cursor in CURSOR_REQUIRESEEK state before + ** returning. + ** + ** Or, if the current delete will not cause a rebalance, then the cursor + ** will be left in CURSOR_SKIPNEXT state pointing to the entry immediately + ** before or after the deleted entry. In this case set bSkipnext to true. */ + if( bPreserve ){ + if( !pPage->leaf + || (pPage->nFree+cellSizePtr(pPage,pCell)+2)>(int)(pBt->usableSize*2/3) + || pPage->nCell==1 /* See dbfuzz001.test for a test case */ + ){ + /* A b-tree rebalance will be required after deleting this entry. + ** Save the cursor key. */ + rc = saveCursorKey(pCur); + if( rc ) return rc; + }else{ + bSkipnext = 1; + } + } + + /* If the page containing the entry to delete is not a leaf page, move + ** the cursor to the largest entry in the tree that is smaller than + ** the entry being deleted. This cell will replace the cell being deleted + ** from the internal node. The 'previous' entry is used for this instead + ** of the 'next' entry, as the previous entry is always a part of the + ** sub-tree headed by the child page of the cell being deleted. This makes + ** balancing the tree following the delete operation easier. */ + if( !pPage->leaf ){ + rc = sqlite3BtreePrevious(pCur, 0); + assert( rc!=SQLITE_DONE ); + if( rc ) return rc; + } + + /* Save the positions of any other cursors open on this table before + ** making any modifications. */ + if( pCur->curFlags & BTCF_Multiple ){ + rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur); + if( rc ) return rc; + } + + /* If this is a delete operation to remove a row from a table b-tree, + ** invalidate any incrblob cursors open on the row being deleted. */ + if( pCur->pKeyInfo==0 && p->hasIncrblobCur ){ + invalidateIncrblobCursors(p, pCur->pgnoRoot, pCur->info.nKey, 0); + } + + /* Make the page containing the entry to be deleted writable. Then free any + ** overflow pages associated with the entry and finally remove the cell + ** itself from within the page. */ + rc = sqlite3PagerWrite(pPage->pDbPage); + if( rc ) return rc; + BTREE_CLEAR_CELL(rc, pPage, pCell, info); + dropCell(pPage, iCellIdx, info.nSize, &rc); + if( rc ) return rc; + + /* If the cell deleted was not located on a leaf page, then the cursor + ** is currently pointing to the largest entry in the sub-tree headed + ** by the child-page of the cell that was just deleted from an internal + ** node. The cell from the leaf node needs to be moved to the internal + ** node to replace the deleted cell. */ + if( !pPage->leaf ){ + MemPage *pLeaf = pCur->pPage; + int nCell; + Pgno n; + unsigned char *pTmp; + + if( pLeaf->nFree<0 ){ + rc = btreeComputeFreeSpace(pLeaf); + if( rc ) return rc; + } + if( iCellDepthiPage-1 ){ + n = pCur->apPage[iCellDepth+1]->pgno; + }else{ + n = pCur->pPage->pgno; + } + pCell = findCell(pLeaf, pLeaf->nCell-1); + if( pCell<&pLeaf->aData[4] ) return SQLITE_CORRUPT_BKPT; + nCell = pLeaf->xCellSize(pLeaf, pCell); + assert( MX_CELL_SIZE(pBt) >= nCell ); + pTmp = pBt->pTmpSpace; + assert( pTmp!=0 ); + rc = sqlite3PagerWrite(pLeaf->pDbPage); + if( rc==SQLITE_OK ){ + insertCell(pPage, iCellIdx, pCell-4, nCell+4, pTmp, n, &rc); + } + dropCell(pLeaf, pLeaf->nCell-1, nCell, &rc); + if( rc ) return rc; + } + + /* Balance the tree. If the entry deleted was located on a leaf page, + ** then the cursor still points to that page. In this case the first + ** call to balance() repairs the tree, and the if(...) condition is + ** never true. + ** + ** Otherwise, if the entry deleted was on an internal node page, then + ** pCur is pointing to the leaf page from which a cell was removed to + ** replace the cell deleted from the internal node. This is slightly + ** tricky as the leaf node may be underfull, and the internal node may + ** be either under or overfull. In this case run the balancing algorithm + ** on the leaf node first. If the balance proceeds far enough up the + ** tree that we can be sure that any problem in the internal node has + ** been corrected, so be it. Otherwise, after balancing the leaf node, + ** walk the cursor up the tree to the internal node and balance it as + ** well. */ + rc = balance(pCur); + if( rc==SQLITE_OK && pCur->iPage>iCellDepth ){ + releasePageNotNull(pCur->pPage); + pCur->iPage--; + while( pCur->iPage>iCellDepth ){ + releasePage(pCur->apPage[pCur->iPage--]); + } + pCur->pPage = pCur->apPage[pCur->iPage]; + rc = balance(pCur); + } + + if( rc==SQLITE_OK ){ + if( bSkipnext ){ + assert( bPreserve && (pCur->iPage==iCellDepth || CORRUPT_DB) ); + assert( pPage==pCur->pPage || CORRUPT_DB ); + assert( (pPage->nCell>0 || CORRUPT_DB) && iCellIdx<=pPage->nCell ); + pCur->eState = CURSOR_SKIPNEXT; + if( iCellIdx>=pPage->nCell ){ + pCur->skipNext = -1; + pCur->ix = pPage->nCell-1; + }else{ + pCur->skipNext = 1; + } + }else{ + rc = moveToRoot(pCur); + if( bPreserve ){ + btreeReleaseAllCursorPages(pCur); + pCur->eState = CURSOR_REQUIRESEEK; + } + if( rc==SQLITE_EMPTY ) rc = SQLITE_OK; + } + } + return rc; +} + +/* +** Create a new BTree table. Write into *piTable the page +** number for the root page of the new table. +** +** The type of type is determined by the flags parameter. Only the +** following values of flags are currently in use. Other values for +** flags might not work: +** +** BTREE_INTKEY|BTREE_LEAFDATA Used for SQL tables with rowid keys +** BTREE_ZERODATA Used for SQL indices +*/ +static int btreeCreateTable(Btree *p, Pgno *piTable, int createTabFlags){ + BtShared *pBt = p->pBt; + MemPage *pRoot; + Pgno pgnoRoot; + int rc; + int ptfFlags; /* Page-type flage for the root page of new table */ + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( pBt->inTransaction==TRANS_WRITE ); + assert( (pBt->btsFlags & BTS_READ_ONLY)==0 ); + +#ifdef SQLITE_OMIT_AUTOVACUUM + rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0); + if( rc ){ + return rc; + } +#else + if( pBt->autoVacuum ){ + Pgno pgnoMove; /* Move a page here to make room for the root-page */ + MemPage *pPageMove; /* The page to move to. */ + + /* Creating a new table may probably require moving an existing database + ** to make room for the new tables root page. In case this page turns + ** out to be an overflow page, delete all overflow page-map caches + ** held by open cursors. + */ + invalidateAllOverflowCache(pBt); + + /* Read the value of meta[3] from the database to determine where the + ** root page of the new table should go. meta[3] is the largest root-page + ** created so far, so the new root-page is (meta[3]+1). + */ + sqlite3BtreeGetMeta(p, BTREE_LARGEST_ROOT_PAGE, &pgnoRoot); + if( pgnoRoot>btreePagecount(pBt) ){ + return SQLITE_CORRUPT_BKPT; + } + pgnoRoot++; + + /* The new root-page may not be allocated on a pointer-map page, or the + ** PENDING_BYTE page. + */ + while( pgnoRoot==PTRMAP_PAGENO(pBt, pgnoRoot) || + pgnoRoot==PENDING_BYTE_PAGE(pBt) ){ + pgnoRoot++; + } + assert( pgnoRoot>=3 ); + + /* Allocate a page. The page that currently resides at pgnoRoot will + ** be moved to the allocated page (unless the allocated page happens + ** to reside at pgnoRoot). + */ + rc = allocateBtreePage(pBt, &pPageMove, &pgnoMove, pgnoRoot, BTALLOC_EXACT); + if( rc!=SQLITE_OK ){ + return rc; + } + + if( pgnoMove!=pgnoRoot ){ + /* pgnoRoot is the page that will be used for the root-page of + ** the new table (assuming an error did not occur). But we were + ** allocated pgnoMove. If required (i.e. if it was not allocated + ** by extending the file), the current page at position pgnoMove + ** is already journaled. + */ + u8 eType = 0; + Pgno iPtrPage = 0; + + /* Save the positions of any open cursors. This is required in + ** case they are holding a reference to an xFetch reference + ** corresponding to page pgnoRoot. */ + rc = saveAllCursors(pBt, 0, 0); + releasePage(pPageMove); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Move the page currently at pgnoRoot to pgnoMove. */ + rc = btreeGetPage(pBt, pgnoRoot, &pRoot, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = ptrmapGet(pBt, pgnoRoot, &eType, &iPtrPage); + if( eType==PTRMAP_ROOTPAGE || eType==PTRMAP_FREEPAGE ){ + rc = SQLITE_CORRUPT_BKPT; + } + if( rc!=SQLITE_OK ){ + releasePage(pRoot); + return rc; + } + assert( eType!=PTRMAP_ROOTPAGE ); + assert( eType!=PTRMAP_FREEPAGE ); + rc = relocatePage(pBt, pRoot, eType, iPtrPage, pgnoMove, 0); + releasePage(pRoot); + + /* Obtain the page at pgnoRoot */ + if( rc!=SQLITE_OK ){ + return rc; + } + rc = btreeGetPage(pBt, pgnoRoot, &pRoot, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = sqlite3PagerWrite(pRoot->pDbPage); + if( rc!=SQLITE_OK ){ + releasePage(pRoot); + return rc; + } + }else{ + pRoot = pPageMove; + } + + /* Update the pointer-map and meta-data with the new root-page number. */ + ptrmapPut(pBt, pgnoRoot, PTRMAP_ROOTPAGE, 0, &rc); + if( rc ){ + releasePage(pRoot); + return rc; + } + + /* When the new root page was allocated, page 1 was made writable in + ** order either to increase the database filesize, or to decrement the + ** freelist count. Hence, the sqlite3BtreeUpdateMeta() call cannot fail. + */ + assert( sqlite3PagerIswriteable(pBt->pPage1->pDbPage) ); + rc = sqlite3BtreeUpdateMeta(p, 4, pgnoRoot); + if( NEVER(rc) ){ + releasePage(pRoot); + return rc; + } + + }else{ + rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0); + if( rc ) return rc; + } +#endif + assert( sqlite3PagerIswriteable(pRoot->pDbPage) ); + if( createTabFlags & BTREE_INTKEY ){ + ptfFlags = PTF_INTKEY | PTF_LEAFDATA | PTF_LEAF; + }else{ + ptfFlags = PTF_ZERODATA | PTF_LEAF; + } + zeroPage(pRoot, ptfFlags); + sqlite3PagerUnref(pRoot->pDbPage); + assert( (pBt->openFlags & BTREE_SINGLE)==0 || pgnoRoot==2 ); + *piTable = pgnoRoot; + return SQLITE_OK; +} +int sqlite3BtreeCreateTable(Btree *p, Pgno *piTable, int flags){ + int rc; + sqlite3BtreeEnter(p); + rc = btreeCreateTable(p, piTable, flags); + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Erase the given database page and all its children. Return +** the page to the freelist. +*/ +static int clearDatabasePage( + BtShared *pBt, /* The BTree that contains the table */ + Pgno pgno, /* Page number to clear */ + int freePageFlag, /* Deallocate page if true */ + i64 *pnChange /* Add number of Cells freed to this counter */ +){ + MemPage *pPage; + int rc; + unsigned char *pCell; + int i; + int hdr; + CellInfo info; + + assert( sqlite3_mutex_held(pBt->mutex) ); + if( pgno>btreePagecount(pBt) ){ + return SQLITE_CORRUPT_BKPT; + } + rc = getAndInitPage(pBt, pgno, &pPage, 0, 0); + if( rc ) return rc; + if( (pBt->openFlags & BTREE_SINGLE)==0 + && sqlite3PagerPageRefcount(pPage->pDbPage)!=1 + ){ + rc = SQLITE_CORRUPT_BKPT; + goto cleardatabasepage_out; + } + hdr = pPage->hdrOffset; + for(i=0; inCell; i++){ + pCell = findCell(pPage, i); + if( !pPage->leaf ){ + rc = clearDatabasePage(pBt, get4byte(pCell), 1, pnChange); + if( rc ) goto cleardatabasepage_out; + } + BTREE_CLEAR_CELL(rc, pPage, pCell, info); + if( rc ) goto cleardatabasepage_out; + } + if( !pPage->leaf ){ + rc = clearDatabasePage(pBt, get4byte(&pPage->aData[hdr+8]), 1, pnChange); + if( rc ) goto cleardatabasepage_out; + if( pPage->intKey ) pnChange = 0; + } + if( pnChange ){ + testcase( !pPage->intKey ); + *pnChange += pPage->nCell; + } + if( freePageFlag ){ + freePage(pPage, &rc); + }else if( (rc = sqlite3PagerWrite(pPage->pDbPage))==0 ){ + zeroPage(pPage, pPage->aData[hdr] | PTF_LEAF); + } + +cleardatabasepage_out: + releasePage(pPage); + return rc; +} + +/* +** Delete all information from a single table in the database. iTable is +** the page number of the root of the table. After this routine returns, +** the root page is empty, but still exists. +** +** This routine will fail with SQLITE_LOCKED if there are any open +** read cursors on the table. Open write cursors are moved to the +** root of the table. +** +** If pnChange is not NULL, then the integer value pointed to by pnChange +** is incremented by the number of entries in the table. +*/ +int sqlite3BtreeClearTable(Btree *p, int iTable, i64 *pnChange){ + int rc; + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + assert( p->inTrans==TRANS_WRITE ); + + rc = saveAllCursors(pBt, (Pgno)iTable, 0); + + if( SQLITE_OK==rc ){ + /* Invalidate all incrblob cursors open on table iTable (assuming iTable + ** is the root of a table b-tree - if it is not, the following call is + ** a no-op). */ + if( p->hasIncrblobCur ){ + invalidateIncrblobCursors(p, (Pgno)iTable, 0, 1); + } + rc = clearDatabasePage(pBt, (Pgno)iTable, 0, pnChange); + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** Delete all information from the single table that pCur is open on. +** +** This routine only work for pCur on an ephemeral table. +*/ +int sqlite3BtreeClearTableOfCursor(BtCursor *pCur){ + return sqlite3BtreeClearTable(pCur->pBtree, pCur->pgnoRoot, 0); +} + +/* +** Erase all information in a table and add the root of the table to +** the freelist. Except, the root of the principle table (the one on +** page 1) is never added to the freelist. +** +** This routine will fail with SQLITE_LOCKED if there are any open +** cursors on the table. +** +** If AUTOVACUUM is enabled and the page at iTable is not the last +** root page in the database file, then the last root page +** in the database file is moved into the slot formerly occupied by +** iTable and that last slot formerly occupied by the last root page +** is added to the freelist instead of iTable. In this say, all +** root pages are kept at the beginning of the database file, which +** is necessary for AUTOVACUUM to work right. *piMoved is set to the +** page number that used to be the last root page in the file before +** the move. If no page gets moved, *piMoved is set to 0. +** The last root page is recorded in meta[3] and the value of +** meta[3] is updated by this procedure. +*/ +static int btreeDropTable(Btree *p, Pgno iTable, int *piMoved){ + int rc; + MemPage *pPage = 0; + BtShared *pBt = p->pBt; + + assert( sqlite3BtreeHoldsMutex(p) ); + assert( p->inTrans==TRANS_WRITE ); + assert( iTable>=2 ); + if( iTable>btreePagecount(pBt) ){ + return SQLITE_CORRUPT_BKPT; + } + + rc = sqlite3BtreeClearTable(p, iTable, 0); + if( rc ) return rc; + rc = btreeGetPage(pBt, (Pgno)iTable, &pPage, 0); + if( NEVER(rc) ){ + releasePage(pPage); + return rc; + } + + *piMoved = 0; + +#ifdef SQLITE_OMIT_AUTOVACUUM + freePage(pPage, &rc); + releasePage(pPage); +#else + if( pBt->autoVacuum ){ + Pgno maxRootPgno; + sqlite3BtreeGetMeta(p, BTREE_LARGEST_ROOT_PAGE, &maxRootPgno); + + if( iTable==maxRootPgno ){ + /* If the table being dropped is the table with the largest root-page + ** number in the database, put the root page on the free list. + */ + freePage(pPage, &rc); + releasePage(pPage); + if( rc!=SQLITE_OK ){ + return rc; + } + }else{ + /* The table being dropped does not have the largest root-page + ** number in the database. So move the page that does into the + ** gap left by the deleted root-page. + */ + MemPage *pMove; + releasePage(pPage); + rc = btreeGetPage(pBt, maxRootPgno, &pMove, 0); + if( rc!=SQLITE_OK ){ + return rc; + } + rc = relocatePage(pBt, pMove, PTRMAP_ROOTPAGE, 0, iTable, 0); + releasePage(pMove); + if( rc!=SQLITE_OK ){ + return rc; + } + pMove = 0; + rc = btreeGetPage(pBt, maxRootPgno, &pMove, 0); + freePage(pMove, &rc); + releasePage(pMove); + if( rc!=SQLITE_OK ){ + return rc; + } + *piMoved = maxRootPgno; + } + + /* Set the new 'max-root-page' value in the database header. This + ** is the old value less one, less one more if that happens to + ** be a root-page number, less one again if that is the + ** PENDING_BYTE_PAGE. + */ + maxRootPgno--; + while( maxRootPgno==PENDING_BYTE_PAGE(pBt) + || PTRMAP_ISPAGE(pBt, maxRootPgno) ){ + maxRootPgno--; + } + assert( maxRootPgno!=PENDING_BYTE_PAGE(pBt) ); + + rc = sqlite3BtreeUpdateMeta(p, 4, maxRootPgno); + }else{ + freePage(pPage, &rc); + releasePage(pPage); + } +#endif + return rc; +} +int sqlite3BtreeDropTable(Btree *p, int iTable, int *piMoved){ + int rc; + sqlite3BtreeEnter(p); + rc = btreeDropTable(p, iTable, piMoved); + sqlite3BtreeLeave(p); + return rc; +} + + +/* +** This function may only be called if the b-tree connection already +** has a read or write transaction open on the database. +** +** Read the meta-information out of a database file. Meta[0] +** is the number of free pages currently in the database. Meta[1] +** through meta[15] are available for use by higher layers. Meta[0] +** is read-only, the others are read/write. +** +** The schema layer numbers meta values differently. At the schema +** layer (and the SetCookie and ReadCookie opcodes) the number of +** free pages is not visible. So Cookie[0] is the same as Meta[1]. +** +** This routine treats Meta[BTREE_DATA_VERSION] as a special case. Instead +** of reading the value out of the header, it instead loads the "DataVersion" +** from the pager. The BTREE_DATA_VERSION value is not actually stored in the +** database file. It is a number computed by the pager. But its access +** pattern is the same as header meta values, and so it is convenient to +** read it from this routine. +*/ +void sqlite3BtreeGetMeta(Btree *p, int idx, u32 *pMeta){ + BtShared *pBt = p->pBt; + + sqlite3BtreeEnter(p); + assert( p->inTrans>TRANS_NONE ); + assert( SQLITE_OK==querySharedCacheTableLock(p, SCHEMA_ROOT, READ_LOCK) ); + assert( pBt->pPage1 ); + assert( idx>=0 && idx<=15 ); + + if( idx==BTREE_DATA_VERSION ){ + *pMeta = sqlite3PagerDataVersion(pBt->pPager) + p->iBDataVersion; + }else{ + *pMeta = get4byte(&pBt->pPage1->aData[36 + idx*4]); + } + + /* If auto-vacuum is disabled in this build and this is an auto-vacuum + ** database, mark the database as read-only. */ +#ifdef SQLITE_OMIT_AUTOVACUUM + if( idx==BTREE_LARGEST_ROOT_PAGE && *pMeta>0 ){ + pBt->btsFlags |= BTS_READ_ONLY; + } +#endif + + sqlite3BtreeLeave(p); +} + +/* +** Write meta-information back into the database. Meta[0] is +** read-only and may not be written. +*/ +int sqlite3BtreeUpdateMeta(Btree *p, int idx, u32 iMeta){ + BtShared *pBt = p->pBt; + unsigned char *pP1; + int rc; + assert( idx>=1 && idx<=15 ); + sqlite3BtreeEnter(p); + assert( p->inTrans==TRANS_WRITE ); + assert( pBt->pPage1!=0 ); + pP1 = pBt->pPage1->aData; + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( rc==SQLITE_OK ){ + put4byte(&pP1[36 + idx*4], iMeta); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( idx==BTREE_INCR_VACUUM ){ + assert( pBt->autoVacuum || iMeta==0 ); + assert( iMeta==0 || iMeta==1 ); + pBt->incrVacuum = (u8)iMeta; + } +#endif + } + sqlite3BtreeLeave(p); + return rc; +} + +/* +** The first argument, pCur, is a cursor opened on some b-tree. Count the +** number of entries in the b-tree and write the result to *pnEntry. +** +** SQLITE_OK is returned if the operation is successfully executed. +** Otherwise, if an error is encountered (i.e. an IO error or database +** corruption) an SQLite error code is returned. +*/ +int sqlite3BtreeCount(sqlite3 *db, BtCursor *pCur, i64 *pnEntry){ + i64 nEntry = 0; /* Value to return in *pnEntry */ + int rc; /* Return code */ + + rc = moveToRoot(pCur); + if( rc==SQLITE_EMPTY ){ + *pnEntry = 0; + return SQLITE_OK; + } + + /* Unless an error occurs, the following loop runs one iteration for each + ** page in the B-Tree structure (not including overflow pages). + */ + while( rc==SQLITE_OK && !AtomicLoad(&db->u1.isInterrupted) ){ + int iIdx; /* Index of child node in parent */ + MemPage *pPage; /* Current page of the b-tree */ + + /* If this is a leaf page or the tree is not an int-key tree, then + ** this page contains countable entries. Increment the entry counter + ** accordingly. + */ + pPage = pCur->pPage; + if( pPage->leaf || !pPage->intKey ){ + nEntry += pPage->nCell; + } + + /* pPage is a leaf node. This loop navigates the cursor so that it + ** points to the first interior cell that it points to the parent of + ** the next page in the tree that has not yet been visited. The + ** pCur->aiIdx[pCur->iPage] value is set to the index of the parent cell + ** of the page, or to the number of cells in the page if the next page + ** to visit is the right-child of its parent. + ** + ** If all pages in the tree have been visited, return SQLITE_OK to the + ** caller. + */ + if( pPage->leaf ){ + do { + if( pCur->iPage==0 ){ + /* All pages of the b-tree have been visited. Return successfully. */ + *pnEntry = nEntry; + return moveToRoot(pCur); + } + moveToParent(pCur); + }while ( pCur->ix>=pCur->pPage->nCell ); + + pCur->ix++; + pPage = pCur->pPage; + } + + /* Descend to the child node of the cell that the cursor currently + ** points at. This is the right-child if (iIdx==pPage->nCell). + */ + iIdx = pCur->ix; + if( iIdx==pPage->nCell ){ + rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8])); + }else{ + rc = moveToChild(pCur, get4byte(findCell(pPage, iIdx))); + } + } + + /* An error has occurred. Return an error code. */ + return rc; +} + +/* +** Return the pager associated with a BTree. This routine is used for +** testing and debugging only. +*/ +Pager *sqlite3BtreePager(Btree *p){ + return p->pBt->pPager; +} + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK +/* +** Append a message to the error message string. +*/ +static void checkAppendMsg( + IntegrityCk *pCheck, + const char *zFormat, + ... +){ + va_list ap; + if( !pCheck->mxErr ) return; + pCheck->mxErr--; + pCheck->nErr++; + va_start(ap, zFormat); + if( pCheck->errMsg.nChar ){ + sqlite3_str_append(&pCheck->errMsg, "\n", 1); + } + if( pCheck->zPfx ){ + sqlite3_str_appendf(&pCheck->errMsg, pCheck->zPfx, pCheck->v1, pCheck->v2); + } + sqlite3_str_vappendf(&pCheck->errMsg, zFormat, ap); + va_end(ap); + if( pCheck->errMsg.accError==SQLITE_NOMEM ){ + pCheck->bOomFault = 1; + } +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK + +/* +** Return non-zero if the bit in the IntegrityCk.aPgRef[] array that +** corresponds to page iPg is already set. +*/ +static int getPageReferenced(IntegrityCk *pCheck, Pgno iPg){ + assert( iPg<=pCheck->nPage && sizeof(pCheck->aPgRef[0])==1 ); + return (pCheck->aPgRef[iPg/8] & (1 << (iPg & 0x07))); +} + +/* +** Set the bit in the IntegrityCk.aPgRef[] array that corresponds to page iPg. +*/ +static void setPageReferenced(IntegrityCk *pCheck, Pgno iPg){ + assert( iPg<=pCheck->nPage && sizeof(pCheck->aPgRef[0])==1 ); + pCheck->aPgRef[iPg/8] |= (1 << (iPg & 0x07)); +} + + +/* +** Add 1 to the reference count for page iPage. If this is the second +** reference to the page, add an error message to pCheck->zErrMsg. +** Return 1 if there are 2 or more references to the page and 0 if +** if this is the first reference to the page. +** +** Also check that the page number is in bounds. +*/ +static int checkRef(IntegrityCk *pCheck, Pgno iPage){ + if( iPage>pCheck->nPage || iPage==0 ){ + checkAppendMsg(pCheck, "invalid page number %d", iPage); + return 1; + } + if( getPageReferenced(pCheck, iPage) ){ + checkAppendMsg(pCheck, "2nd reference to page %d", iPage); + return 1; + } + if( AtomicLoad(&pCheck->db->u1.isInterrupted) ) return 1; + setPageReferenced(pCheck, iPage); + return 0; +} + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** Check that the entry in the pointer-map for page iChild maps to +** page iParent, pointer type ptrType. If not, append an error message +** to pCheck. +*/ +static void checkPtrmap( + IntegrityCk *pCheck, /* Integrity check context */ + Pgno iChild, /* Child page number */ + u8 eType, /* Expected pointer map type */ + Pgno iParent /* Expected pointer map parent page number */ +){ + int rc; + u8 ePtrmapType; + Pgno iPtrmapParent; + + rc = ptrmapGet(pCheck->pBt, iChild, &ePtrmapType, &iPtrmapParent); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_NOMEM || rc==SQLITE_IOERR_NOMEM ) pCheck->bOomFault = 1; + checkAppendMsg(pCheck, "Failed to read ptrmap key=%d", iChild); + return; + } + + if( ePtrmapType!=eType || iPtrmapParent!=iParent ){ + checkAppendMsg(pCheck, + "Bad ptr map entry key=%d expected=(%d,%d) got=(%d,%d)", + iChild, eType, iParent, ePtrmapType, iPtrmapParent); + } +} +#endif + +/* +** Check the integrity of the freelist or of an overflow page list. +** Verify that the number of pages on the list is N. +*/ +static void checkList( + IntegrityCk *pCheck, /* Integrity checking context */ + int isFreeList, /* True for a freelist. False for overflow page list */ + Pgno iPage, /* Page number for first page in the list */ + u32 N /* Expected number of pages in the list */ +){ + int i; + u32 expected = N; + int nErrAtStart = pCheck->nErr; + while( iPage!=0 && pCheck->mxErr ){ + DbPage *pOvflPage; + unsigned char *pOvflData; + if( checkRef(pCheck, iPage) ) break; + N--; + if( sqlite3PagerGet(pCheck->pPager, (Pgno)iPage, &pOvflPage, 0) ){ + checkAppendMsg(pCheck, "failed to get page %d", iPage); + break; + } + pOvflData = (unsigned char *)sqlite3PagerGetData(pOvflPage); + if( isFreeList ){ + u32 n = (u32)get4byte(&pOvflData[4]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pCheck->pBt->autoVacuum ){ + checkPtrmap(pCheck, iPage, PTRMAP_FREEPAGE, 0); + } +#endif + if( n>pCheck->pBt->usableSize/4-2 ){ + checkAppendMsg(pCheck, + "freelist leaf count too big on page %d", iPage); + N--; + }else{ + for(i=0; i<(int)n; i++){ + Pgno iFreePage = get4byte(&pOvflData[8+i*4]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pCheck->pBt->autoVacuum ){ + checkPtrmap(pCheck, iFreePage, PTRMAP_FREEPAGE, 0); + } +#endif + checkRef(pCheck, iFreePage); + } + N -= n; + } + } +#ifndef SQLITE_OMIT_AUTOVACUUM + else{ + /* If this database supports auto-vacuum and iPage is not the last + ** page in this overflow list, check that the pointer-map entry for + ** the following page matches iPage. + */ + if( pCheck->pBt->autoVacuum && N>0 ){ + i = get4byte(pOvflData); + checkPtrmap(pCheck, i, PTRMAP_OVERFLOW2, iPage); + } + } +#endif + iPage = get4byte(pOvflData); + sqlite3PagerUnref(pOvflPage); + } + if( N && nErrAtStart==pCheck->nErr ){ + checkAppendMsg(pCheck, + "%s is %d but should be %d", + isFreeList ? "size" : "overflow list length", + expected-N, expected); + } +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +/* +** An implementation of a min-heap. +** +** aHeap[0] is the number of elements on the heap. aHeap[1] is the +** root element. The daughter nodes of aHeap[N] are aHeap[N*2] +** and aHeap[N*2+1]. +** +** The heap property is this: Every node is less than or equal to both +** of its daughter nodes. A consequence of the heap property is that the +** root node aHeap[1] is always the minimum value currently in the heap. +** +** The btreeHeapInsert() routine inserts an unsigned 32-bit number onto +** the heap, preserving the heap property. The btreeHeapPull() routine +** removes the root element from the heap (the minimum value in the heap) +** and then moves other nodes around as necessary to preserve the heap +** property. +** +** This heap is used for cell overlap and coverage testing. Each u32 +** entry represents the span of a cell or freeblock on a btree page. +** The upper 16 bits are the index of the first byte of a range and the +** lower 16 bits are the index of the last byte of that range. +*/ +static void btreeHeapInsert(u32 *aHeap, u32 x){ + u32 j, i = ++aHeap[0]; + aHeap[i] = x; + while( (j = i/2)>0 && aHeap[j]>aHeap[i] ){ + x = aHeap[j]; + aHeap[j] = aHeap[i]; + aHeap[i] = x; + i = j; + } +} +static int btreeHeapPull(u32 *aHeap, u32 *pOut){ + u32 j, i, x; + if( (x = aHeap[0])==0 ) return 0; + *pOut = aHeap[1]; + aHeap[1] = aHeap[x]; + aHeap[x] = 0xffffffff; + aHeap[0]--; + i = 1; + while( (j = i*2)<=aHeap[0] ){ + if( aHeap[j]>aHeap[j+1] ) j++; + if( aHeap[i]zPfx; + int saved_v1 = pCheck->v1; + int saved_v2 = pCheck->v2; + u8 savedIsInit = 0; + + /* Check that the page exists + */ + pBt = pCheck->pBt; + usableSize = pBt->usableSize; + if( iPage==0 ) return 0; + if( checkRef(pCheck, iPage) ) return 0; + pCheck->zPfx = "Page %u: "; + pCheck->v1 = iPage; + if( (rc = btreeGetPage(pBt, iPage, &pPage, 0))!=0 ){ + checkAppendMsg(pCheck, + "unable to get the page. error code=%d", rc); + goto end_of_check; + } + + /* Clear MemPage.isInit to make sure the corruption detection code in + ** btreeInitPage() is executed. */ + savedIsInit = pPage->isInit; + pPage->isInit = 0; + if( (rc = btreeInitPage(pPage))!=0 ){ + assert( rc==SQLITE_CORRUPT ); /* The only possible error from InitPage */ + checkAppendMsg(pCheck, + "btreeInitPage() returns error code %d", rc); + goto end_of_check; + } + if( (rc = btreeComputeFreeSpace(pPage))!=0 ){ + assert( rc==SQLITE_CORRUPT ); + checkAppendMsg(pCheck, "free space corruption", rc); + goto end_of_check; + } + data = pPage->aData; + hdr = pPage->hdrOffset; + + /* Set up for cell analysis */ + pCheck->zPfx = "On tree page %u cell %d: "; + contentOffset = get2byteNotZero(&data[hdr+5]); + assert( contentOffset<=usableSize ); /* Enforced by btreeInitPage() */ + + /* EVIDENCE-OF: R-37002-32774 The two-byte integer at offset 3 gives the + ** number of cells on the page. */ + nCell = get2byte(&data[hdr+3]); + assert( pPage->nCell==nCell ); + + /* EVIDENCE-OF: R-23882-45353 The cell pointer array of a b-tree page + ** immediately follows the b-tree page header. */ + cellStart = hdr + 12 - 4*pPage->leaf; + assert( pPage->aCellIdx==&data[cellStart] ); + pCellIdx = &data[cellStart + 2*(nCell-1)]; + + if( !pPage->leaf ){ + /* Analyze the right-child page of internal pages */ + pgno = get4byte(&data[hdr+8]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + pCheck->zPfx = "On page %u at right child: "; + checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage); + } +#endif + depth = checkTreePage(pCheck, pgno, &maxKey, maxKey); + keyCanBeEqual = 0; + }else{ + /* For leaf pages, the coverage check will occur in the same loop + ** as the other cell checks, so initialize the heap. */ + heap = pCheck->heap; + heap[0] = 0; + } + + /* EVIDENCE-OF: R-02776-14802 The cell pointer array consists of K 2-byte + ** integer offsets to the cell contents. */ + for(i=nCell-1; i>=0 && pCheck->mxErr; i--){ + CellInfo info; + + /* Check cell size */ + pCheck->v2 = i; + assert( pCellIdx==&data[cellStart + i*2] ); + pc = get2byteAligned(pCellIdx); + pCellIdx -= 2; + if( pcusableSize-4 ){ + checkAppendMsg(pCheck, "Offset %d out of range %d..%d", + pc, contentOffset, usableSize-4); + doCoverageCheck = 0; + continue; + } + pCell = &data[pc]; + pPage->xParseCell(pPage, pCell, &info); + if( pc+info.nSize>usableSize ){ + checkAppendMsg(pCheck, "Extends off end of page"); + doCoverageCheck = 0; + continue; + } + + /* Check for integer primary key out of range */ + if( pPage->intKey ){ + if( keyCanBeEqual ? (info.nKey > maxKey) : (info.nKey >= maxKey) ){ + checkAppendMsg(pCheck, "Rowid %lld out of order", info.nKey); + } + maxKey = info.nKey; + keyCanBeEqual = 0; /* Only the first key on the page may ==maxKey */ + } + + /* Check the content overflow list */ + if( info.nPayload>info.nLocal ){ + u32 nPage; /* Number of pages on the overflow chain */ + Pgno pgnoOvfl; /* First page of the overflow chain */ + assert( pc + info.nSize - 4 <= usableSize ); + nPage = (info.nPayload - info.nLocal + usableSize - 5)/(usableSize - 4); + pgnoOvfl = get4byte(&pCell[info.nSize - 4]); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + checkPtrmap(pCheck, pgnoOvfl, PTRMAP_OVERFLOW1, iPage); + } +#endif + checkList(pCheck, 0, pgnoOvfl, nPage); + } + + if( !pPage->leaf ){ + /* Check sanity of left child page for internal pages */ + pgno = get4byte(pCell); +#ifndef SQLITE_OMIT_AUTOVACUUM + if( pBt->autoVacuum ){ + checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage); + } +#endif + d2 = checkTreePage(pCheck, pgno, &maxKey, maxKey); + keyCanBeEqual = 0; + if( d2!=depth ){ + checkAppendMsg(pCheck, "Child page depth differs"); + depth = d2; + } + }else{ + /* Populate the coverage-checking heap for leaf pages */ + btreeHeapInsert(heap, (pc<<16)|(pc+info.nSize-1)); + } + } + *piMinKey = maxKey; + + /* Check for complete coverage of the page + */ + pCheck->zPfx = 0; + if( doCoverageCheck && pCheck->mxErr>0 ){ + /* For leaf pages, the min-heap has already been initialized and the + ** cells have already been inserted. But for internal pages, that has + ** not yet been done, so do it now */ + if( !pPage->leaf ){ + heap = pCheck->heap; + heap[0] = 0; + for(i=nCell-1; i>=0; i--){ + u32 size; + pc = get2byteAligned(&data[cellStart+i*2]); + size = pPage->xCellSize(pPage, &data[pc]); + btreeHeapInsert(heap, (pc<<16)|(pc+size-1)); + } + } + /* Add the freeblocks to the min-heap + ** + ** EVIDENCE-OF: R-20690-50594 The second field of the b-tree page header + ** is the offset of the first freeblock, or zero if there are no + ** freeblocks on the page. + */ + i = get2byte(&data[hdr+1]); + while( i>0 ){ + int size, j; + assert( (u32)i<=usableSize-4 ); /* Enforced by btreeComputeFreeSpace() */ + size = get2byte(&data[i+2]); + assert( (u32)(i+size)<=usableSize ); /* due to btreeComputeFreeSpace() */ + btreeHeapInsert(heap, (((u32)i)<<16)|(i+size-1)); + /* EVIDENCE-OF: R-58208-19414 The first 2 bytes of a freeblock are a + ** big-endian integer which is the offset in the b-tree page of the next + ** freeblock in the chain, or zero if the freeblock is the last on the + ** chain. */ + j = get2byte(&data[i]); + /* EVIDENCE-OF: R-06866-39125 Freeblocks are always connected in order of + ** increasing offset. */ + assert( j==0 || j>i+size ); /* Enforced by btreeComputeFreeSpace() */ + assert( (u32)j<=usableSize-4 ); /* Enforced by btreeComputeFreeSpace() */ + i = j; + } + /* Analyze the min-heap looking for overlap between cells and/or + ** freeblocks, and counting the number of untracked bytes in nFrag. + ** + ** Each min-heap entry is of the form: (start_address<<16)|end_address. + ** There is an implied first entry the covers the page header, the cell + ** pointer index, and the gap between the cell pointer index and the start + ** of cell content. + ** + ** The loop below pulls entries from the min-heap in order and compares + ** the start_address against the previous end_address. If there is an + ** overlap, that means bytes are used multiple times. If there is a gap, + ** that gap is added to the fragmentation count. + */ + nFrag = 0; + prev = contentOffset - 1; /* Implied first min-heap entry */ + while( btreeHeapPull(heap,&x) ){ + if( (prev&0xffff)>=(x>>16) ){ + checkAppendMsg(pCheck, + "Multiple uses for byte %u of page %u", x>>16, iPage); + break; + }else{ + nFrag += (x>>16) - (prev&0xffff) - 1; + prev = x; + } + } + nFrag += usableSize - (prev&0xffff) - 1; + /* EVIDENCE-OF: R-43263-13491 The total number of bytes in all fragments + ** is stored in the fifth field of the b-tree page header. + ** EVIDENCE-OF: R-07161-27322 The one-byte integer at offset 7 gives the + ** number of fragmented free bytes within the cell content area. + */ + if( heap[0]==0 && nFrag!=data[hdr+7] ){ + checkAppendMsg(pCheck, + "Fragmentation of %d bytes reported as %d on page %u", + nFrag, data[hdr+7], iPage); + } + } + +end_of_check: + if( !doCoverageCheck ) pPage->isInit = savedIsInit; + releasePage(pPage); + pCheck->zPfx = saved_zPfx; + pCheck->v1 = saved_v1; + pCheck->v2 = saved_v2; + return depth+1; +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +#ifndef SQLITE_OMIT_INTEGRITY_CHECK +/* +** This routine does a complete check of the given BTree file. aRoot[] is +** an array of pages numbers were each page number is the root page of +** a table. nRoot is the number of entries in aRoot. +** +** A read-only or read-write transaction must be opened before calling +** this function. +** +** Write the number of error seen in *pnErr. Except for some memory +** allocation errors, an error message held in memory obtained from +** malloc is returned if *pnErr is non-zero. If *pnErr==0 then NULL is +** returned. If a memory allocation error occurs, NULL is returned. +** +** If the first entry in aRoot[] is 0, that indicates that the list of +** root pages is incomplete. This is a "partial integrity-check". This +** happens when performing an integrity check on a single table. The +** zero is skipped, of course. But in addition, the freelist checks +** and the checks to make sure every page is referenced are also skipped, +** since obviously it is not possible to know which pages are covered by +** the unverified btrees. Except, if aRoot[1] is 1, then the freelist +** checks are still performed. +*/ +char *sqlite3BtreeIntegrityCheck( + sqlite3 *db, /* Database connection that is running the check */ + Btree *p, /* The btree to be checked */ + Pgno *aRoot, /* An array of root pages numbers for individual trees */ + int nRoot, /* Number of entries in aRoot[] */ + int mxErr, /* Stop reporting errors after this many */ + int *pnErr /* Write number of errors seen to this variable */ +){ + Pgno i; + IntegrityCk sCheck; + BtShared *pBt = p->pBt; + u64 savedDbFlags = pBt->db->flags; + char zErr[100]; + int bPartial = 0; /* True if not checking all btrees */ + int bCkFreelist = 1; /* True to scan the freelist */ + VVA_ONLY( int nRef ); + assert( nRoot>0 ); + + /* aRoot[0]==0 means this is a partial check */ + if( aRoot[0]==0 ){ + assert( nRoot>1 ); + bPartial = 1; + if( aRoot[1]!=1 ) bCkFreelist = 0; + } + + sqlite3BtreeEnter(p); + assert( p->inTrans>TRANS_NONE && pBt->inTransaction>TRANS_NONE ); + VVA_ONLY( nRef = sqlite3PagerRefcount(pBt->pPager) ); + assert( nRef>=0 ); + sCheck.db = db; + sCheck.pBt = pBt; + sCheck.pPager = pBt->pPager; + sCheck.nPage = btreePagecount(sCheck.pBt); + sCheck.mxErr = mxErr; + sCheck.nErr = 0; + sCheck.bOomFault = 0; + sCheck.zPfx = 0; + sCheck.v1 = 0; + sCheck.v2 = 0; + sCheck.aPgRef = 0; + sCheck.heap = 0; + sqlite3StrAccumInit(&sCheck.errMsg, 0, zErr, sizeof(zErr), SQLITE_MAX_LENGTH); + sCheck.errMsg.printfFlags = SQLITE_PRINTF_INTERNAL; + if( sCheck.nPage==0 ){ + goto integrity_ck_cleanup; + } + + sCheck.aPgRef = sqlite3MallocZero((sCheck.nPage / 8)+ 1); + if( !sCheck.aPgRef ){ + sCheck.bOomFault = 1; + goto integrity_ck_cleanup; + } + sCheck.heap = (u32*)sqlite3PageMalloc( pBt->pageSize ); + if( sCheck.heap==0 ){ + sCheck.bOomFault = 1; + goto integrity_ck_cleanup; + } + + i = PENDING_BYTE_PAGE(pBt); + if( i<=sCheck.nPage ) setPageReferenced(&sCheck, i); + + /* Check the integrity of the freelist + */ + if( bCkFreelist ){ + sCheck.zPfx = "Main freelist: "; + checkList(&sCheck, 1, get4byte(&pBt->pPage1->aData[32]), + get4byte(&pBt->pPage1->aData[36])); + sCheck.zPfx = 0; + } + + /* Check all the tables. + */ +#ifndef SQLITE_OMIT_AUTOVACUUM + if( !bPartial ){ + if( pBt->autoVacuum ){ + Pgno mx = 0; + Pgno mxInHdr; + for(i=0; (int)ipPage1->aData[52]); + if( mx!=mxInHdr ){ + checkAppendMsg(&sCheck, + "max rootpage (%d) disagrees with header (%d)", + mx, mxInHdr + ); + } + }else if( get4byte(&pBt->pPage1->aData[64])!=0 ){ + checkAppendMsg(&sCheck, + "incremental_vacuum enabled with a max rootpage of zero" + ); + } + } +#endif + testcase( pBt->db->flags & SQLITE_CellSizeCk ); + pBt->db->flags &= ~(u64)SQLITE_CellSizeCk; + for(i=0; (int)iautoVacuum && aRoot[i]>1 && !bPartial ){ + checkPtrmap(&sCheck, aRoot[i], PTRMAP_ROOTPAGE, 0); + } +#endif + checkTreePage(&sCheck, aRoot[i], ¬Used, LARGEST_INT64); + } + pBt->db->flags = savedDbFlags; + + /* Make sure every page in the file is referenced + */ + if( !bPartial ){ + for(i=1; i<=sCheck.nPage && sCheck.mxErr; i++){ +#ifdef SQLITE_OMIT_AUTOVACUUM + if( getPageReferenced(&sCheck, i)==0 ){ + checkAppendMsg(&sCheck, "Page %d is never used", i); + } +#else + /* If the database supports auto-vacuum, make sure no tables contain + ** references to pointer-map pages. + */ + if( getPageReferenced(&sCheck, i)==0 && + (PTRMAP_PAGENO(pBt, i)!=i || !pBt->autoVacuum) ){ + checkAppendMsg(&sCheck, "Page %d is never used", i); + } + if( getPageReferenced(&sCheck, i)!=0 && + (PTRMAP_PAGENO(pBt, i)==i && pBt->autoVacuum) ){ + checkAppendMsg(&sCheck, "Pointer map page %d is referenced", i); + } +#endif + } + } + + /* Clean up and report errors. + */ +integrity_ck_cleanup: + sqlite3PageFree(sCheck.heap); + sqlite3_free(sCheck.aPgRef); + if( sCheck.bOomFault ){ + sqlite3_str_reset(&sCheck.errMsg); + sCheck.nErr++; + } + *pnErr = sCheck.nErr; + if( sCheck.nErr==0 ) sqlite3_str_reset(&sCheck.errMsg); + /* Make sure this analysis did not leave any unref() pages. */ + assert( nRef==sqlite3PagerRefcount(pBt->pPager) ); + sqlite3BtreeLeave(p); + return sqlite3StrAccumFinish(&sCheck.errMsg); +} +#endif /* SQLITE_OMIT_INTEGRITY_CHECK */ + +/* +** Return the full pathname of the underlying database file. Return +** an empty string if the database is in-memory or a TEMP database. +** +** The pager filename is invariant as long as the pager is +** open so it is safe to access without the BtShared mutex. +*/ +const char *sqlite3BtreeGetFilename(Btree *p){ + assert( p->pBt->pPager!=0 ); + return sqlite3PagerFilename(p->pBt->pPager, 1); +} + +/* +** Return the pathname of the journal file for this database. The return +** value of this routine is the same regardless of whether the journal file +** has been created or not. +** +** The pager journal filename is invariant as long as the pager is +** open so it is safe to access without the BtShared mutex. +*/ +const char *sqlite3BtreeGetJournalname(Btree *p){ + assert( p->pBt->pPager!=0 ); + return sqlite3PagerJournalname(p->pBt->pPager); +} + +/* +** Return one of SQLITE_TXN_NONE, SQLITE_TXN_READ, or SQLITE_TXN_WRITE +** to describe the current transaction state of Btree p. +*/ +int sqlite3BtreeTxnState(Btree *p){ + assert( p==0 || sqlite3_mutex_held(p->db->mutex) ); + return p ? p->inTrans : 0; +} + +#ifndef SQLITE_OMIT_WAL +/* +** Run a checkpoint on the Btree passed as the first argument. +** +** Return SQLITE_LOCKED if this or any other connection has an open +** transaction on the shared-cache the argument Btree is connected to. +** +** Parameter eMode is one of SQLITE_CHECKPOINT_PASSIVE, FULL or RESTART. +*/ +int sqlite3BtreeCheckpoint(Btree *p, int eMode, int *pnLog, int *pnCkpt){ + int rc = SQLITE_OK; + if( p ){ + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + if( pBt->inTransaction!=TRANS_NONE ){ + rc = SQLITE_LOCKED; + }else{ + rc = sqlite3PagerCheckpoint(pBt->pPager, p->db, eMode, pnLog, pnCkpt); + } + sqlite3BtreeLeave(p); + } + return rc; +} +#endif + +/* +** Return true if there is currently a backup running on Btree p. +*/ +int sqlite3BtreeIsInBackup(Btree *p){ + assert( p ); + assert( sqlite3_mutex_held(p->db->mutex) ); + return p->nBackup!=0; +} + +/* +** This function returns a pointer to a blob of memory associated with +** a single shared-btree. The memory is used by client code for its own +** purposes (for example, to store a high-level schema associated with +** the shared-btree). The btree layer manages reference counting issues. +** +** The first time this is called on a shared-btree, nBytes bytes of memory +** are allocated, zeroed, and returned to the caller. For each subsequent +** call the nBytes parameter is ignored and a pointer to the same blob +** of memory returned. +** +** If the nBytes parameter is 0 and the blob of memory has not yet been +** allocated, a null pointer is returned. If the blob has already been +** allocated, it is returned as normal. +** +** Just before the shared-btree is closed, the function passed as the +** xFree argument when the memory allocation was made is invoked on the +** blob of allocated memory. The xFree function should not call sqlite3_free() +** on the memory, the btree layer does that. +*/ +void *sqlite3BtreeSchema(Btree *p, int nBytes, void(*xFree)(void *)){ + BtShared *pBt = p->pBt; + sqlite3BtreeEnter(p); + if( !pBt->pSchema && nBytes ){ + pBt->pSchema = sqlite3DbMallocZero(0, nBytes); + pBt->xFreeSchema = xFree; + } + sqlite3BtreeLeave(p); + return pBt->pSchema; +} + +/* +** Return SQLITE_LOCKED_SHAREDCACHE if another user of the same shared +** btree as the argument handle holds an exclusive lock on the +** sqlite_schema table. Otherwise SQLITE_OK. +*/ +int sqlite3BtreeSchemaLocked(Btree *p){ + int rc; + assert( sqlite3_mutex_held(p->db->mutex) ); + sqlite3BtreeEnter(p); + rc = querySharedCacheTableLock(p, SCHEMA_ROOT, READ_LOCK); + assert( rc==SQLITE_OK || rc==SQLITE_LOCKED_SHAREDCACHE ); + sqlite3BtreeLeave(p); + return rc; +} + + +#ifndef SQLITE_OMIT_SHARED_CACHE +/* +** Obtain a lock on the table whose root page is iTab. The +** lock is a write lock if isWritelock is true or a read lock +** if it is false. +*/ +int sqlite3BtreeLockTable(Btree *p, int iTab, u8 isWriteLock){ + int rc = SQLITE_OK; + assert( p->inTrans!=TRANS_NONE ); + if( p->sharable ){ + u8 lockType = READ_LOCK + isWriteLock; + assert( READ_LOCK+1==WRITE_LOCK ); + assert( isWriteLock==0 || isWriteLock==1 ); + + sqlite3BtreeEnter(p); + rc = querySharedCacheTableLock(p, iTab, lockType); + if( rc==SQLITE_OK ){ + rc = setSharedCacheTableLock(p, iTab, lockType); + } + sqlite3BtreeLeave(p); + } + return rc; +} +#endif + +#ifndef SQLITE_OMIT_INCRBLOB +/* +** Argument pCsr must be a cursor opened for writing on an +** INTKEY table currently pointing at a valid table entry. +** This function modifies the data stored as part of that entry. +** +** Only the data content may only be modified, it is not possible to +** change the length of the data stored. If this function is called with +** parameters that attempt to write past the end of the existing data, +** no modifications are made and SQLITE_CORRUPT is returned. +*/ +int sqlite3BtreePutData(BtCursor *pCsr, u32 offset, u32 amt, void *z){ + int rc; + assert( cursorOwnsBtShared(pCsr) ); + assert( sqlite3_mutex_held(pCsr->pBtree->db->mutex) ); + assert( pCsr->curFlags & BTCF_Incrblob ); + + rc = restoreCursorPosition(pCsr); + if( rc!=SQLITE_OK ){ + return rc; + } + assert( pCsr->eState!=CURSOR_REQUIRESEEK ); + if( pCsr->eState!=CURSOR_VALID ){ + return SQLITE_ABORT; + } + + /* Save the positions of all other cursors open on this table. This is + ** required in case any of them are holding references to an xFetch + ** version of the b-tree page modified by the accessPayload call below. + ** + ** Note that pCsr must be open on a INTKEY table and saveCursorPosition() + ** and hence saveAllCursors() cannot fail on a BTREE_INTKEY table, hence + ** saveAllCursors can only return SQLITE_OK. + */ + VVA_ONLY(rc =) saveAllCursors(pCsr->pBt, pCsr->pgnoRoot, pCsr); + assert( rc==SQLITE_OK ); + + /* Check some assumptions: + ** (a) the cursor is open for writing, + ** (b) there is a read/write transaction open, + ** (c) the connection holds a write-lock on the table (if required), + ** (d) there are no conflicting read-locks, and + ** (e) the cursor points at a valid row of an intKey table. + */ + if( (pCsr->curFlags & BTCF_WriteFlag)==0 ){ + return SQLITE_READONLY; + } + assert( (pCsr->pBt->btsFlags & BTS_READ_ONLY)==0 + && pCsr->pBt->inTransaction==TRANS_WRITE ); + assert( hasSharedCacheTableLock(pCsr->pBtree, pCsr->pgnoRoot, 0, 2) ); + assert( !hasReadConflicts(pCsr->pBtree, pCsr->pgnoRoot) ); + assert( pCsr->pPage->intKey ); + + return accessPayload(pCsr, offset, amt, (unsigned char *)z, 1); +} + +/* +** Mark this cursor as an incremental blob cursor. +*/ +void sqlite3BtreeIncrblobCursor(BtCursor *pCur){ + pCur->curFlags |= BTCF_Incrblob; + pCur->pBtree->hasIncrblobCur = 1; +} +#endif + +/* +** Set both the "read version" (single byte at byte offset 18) and +** "write version" (single byte at byte offset 19) fields in the database +** header to iVersion. +*/ +int sqlite3BtreeSetVersion(Btree *pBtree, int iVersion){ + BtShared *pBt = pBtree->pBt; + int rc; /* Return code */ + + assert( iVersion==1 || iVersion==2 ); + + /* If setting the version fields to 1, do not automatically open the + ** WAL connection, even if the version fields are currently set to 2. + */ + pBt->btsFlags &= ~BTS_NO_WAL; + if( iVersion==1 ) pBt->btsFlags |= BTS_NO_WAL; + + rc = sqlite3BtreeBeginTrans(pBtree, 0, 0); + if( rc==SQLITE_OK ){ + u8 *aData = pBt->pPage1->aData; + if( aData[18]!=(u8)iVersion || aData[19]!=(u8)iVersion ){ + rc = sqlite3BtreeBeginTrans(pBtree, 2, 0); + if( rc==SQLITE_OK ){ + rc = sqlite3PagerWrite(pBt->pPage1->pDbPage); + if( rc==SQLITE_OK ){ + aData[18] = (u8)iVersion; + aData[19] = (u8)iVersion; + } + } + } + } + + pBt->btsFlags &= ~BTS_NO_WAL; + return rc; +} + +/* +** Return true if the cursor has a hint specified. This routine is +** only used from within assert() statements +*/ +int sqlite3BtreeCursorHasHint(BtCursor *pCsr, unsigned int mask){ + return (pCsr->hints & mask)!=0; +} + +/* +** Return true if the given Btree is read-only. +*/ +int sqlite3BtreeIsReadonly(Btree *p){ + return (p->pBt->btsFlags & BTS_READ_ONLY)!=0; +} + +/* +** Return the size of the header added to each page by this module. +*/ +int sqlite3HeaderSizeBtree(void){ return ROUND8(sizeof(MemPage)); } + +#if !defined(SQLITE_OMIT_SHARED_CACHE) +/* +** Return true if the Btree passed as the only argument is sharable. +*/ +int sqlite3BtreeSharable(Btree *p){ + return p->sharable; +} + +/* +** Return the number of connections to the BtShared object accessed by +** the Btree handle passed as the only argument. For private caches +** this is always 1. For shared caches it may be 1 or greater. +*/ +int sqlite3BtreeConnectionCount(Btree *p){ + testcase( p->sharable ); + return p->pBt->nRef; +} +#endif diff --git a/source/libs/tdb/src/sqlite/pager.c b/source/libs/tdb/src/sqlite/pager.c new file mode 100644 index 0000000000..0ed5848343 --- /dev/null +++ b/source/libs/tdb/src/sqlite/pager.c @@ -0,0 +1,7712 @@ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This is the implementation of the page cache subsystem or "pager". +** +** The pager is used to access a database disk file. It implements +** atomic commit and rollback through the use of a journal file that +** is separate from the database file. The pager also implements file +** locking to prevent two processes from writing the same database +** file simultaneously, or one process from reading the database while +** another is writing. +*/ +#ifndef SQLITE_OMIT_DISKIO +#include "sqliteInt.h" +#include "wal.h" + + +/******************* NOTES ON THE DESIGN OF THE PAGER ************************ +** +** This comment block describes invariants that hold when using a rollback +** journal. These invariants do not apply for journal_mode=WAL, +** journal_mode=MEMORY, or journal_mode=OFF. +** +** Within this comment block, a page is deemed to have been synced +** automatically as soon as it is written when PRAGMA synchronous=OFF. +** Otherwise, the page is not synced until the xSync method of the VFS +** is called successfully on the file containing the page. +** +** Definition: A page of the database file is said to be "overwriteable" if +** one or more of the following are true about the page: +** +** (a) The original content of the page as it was at the beginning of +** the transaction has been written into the rollback journal and +** synced. +** +** (b) The page was a freelist leaf page at the start of the transaction. +** +** (c) The page number is greater than the largest page that existed in +** the database file at the start of the transaction. +** +** (1) A page of the database file is never overwritten unless one of the +** following are true: +** +** (a) The page and all other pages on the same sector are overwriteable. +** +** (b) The atomic page write optimization is enabled, and the entire +** transaction other than the update of the transaction sequence +** number consists of a single page change. +** +** (2) The content of a page written into the rollback journal exactly matches +** both the content in the database when the rollback journal was written +** and the content in the database at the beginning of the current +** transaction. +** +** (3) Writes to the database file are an integer multiple of the page size +** in length and are aligned on a page boundary. +** +** (4) Reads from the database file are either aligned on a page boundary and +** an integer multiple of the page size in length or are taken from the +** first 100 bytes of the database file. +** +** (5) All writes to the database file are synced prior to the rollback journal +** being deleted, truncated, or zeroed. +** +** (6) If a super-journal file is used, then all writes to the database file +** are synced prior to the super-journal being deleted. +** +** Definition: Two databases (or the same database at two points it time) +** are said to be "logically equivalent" if they give the same answer to +** all queries. Note in particular the content of freelist leaf +** pages can be changed arbitrarily without affecting the logical equivalence +** of the database. +** +** (7) At any time, if any subset, including the empty set and the total set, +** of the unsynced changes to a rollback journal are removed and the +** journal is rolled back, the resulting database file will be logically +** equivalent to the database file at the beginning of the transaction. +** +** (8) When a transaction is rolled back, the xTruncate method of the VFS +** is called to restore the database file to the same size it was at +** the beginning of the transaction. (In some VFSes, the xTruncate +** method is a no-op, but that does not change the fact the SQLite will +** invoke it.) +** +** (9) Whenever the database file is modified, at least one bit in the range +** of bytes from 24 through 39 inclusive will be changed prior to releasing +** the EXCLUSIVE lock, thus signaling other connections on the same +** database to flush their caches. +** +** (10) The pattern of bits in bytes 24 through 39 shall not repeat in less +** than one billion transactions. +** +** (11) A database file is well-formed at the beginning and at the conclusion +** of every transaction. +** +** (12) An EXCLUSIVE lock is held on the database file when writing to +** the database file. +** +** (13) A SHARED lock is held on the database file while reading any +** content out of the database file. +** +******************************************************************************/ + +/* +** Macros for troubleshooting. Normally turned off +*/ +#if 0 +int sqlite3PagerTrace=1; /* True to enable tracing */ +#define sqlite3DebugPrintf printf +#define PAGERTRACE(X) if( sqlite3PagerTrace ){ sqlite3DebugPrintf X; } +#else +#define PAGERTRACE(X) +#endif + +/* +** The following two macros are used within the PAGERTRACE() macros above +** to print out file-descriptors. +** +** PAGERID() takes a pointer to a Pager struct as its argument. The +** associated file-descriptor is returned. FILEHANDLEID() takes an sqlite3_file +** struct as its argument. +*/ +#define PAGERID(p) (SQLITE_PTR_TO_INT(p->fd)) +#define FILEHANDLEID(fd) (SQLITE_PTR_TO_INT(fd)) + +/* +** The Pager.eState variable stores the current 'state' of a pager. A +** pager may be in any one of the seven states shown in the following +** state diagram. +** +** OPEN <------+------+ +** | | | +** V | | +** +---------> READER-------+ | +** | | | +** | V | +** |<-------WRITER_LOCKED------> ERROR +** | | ^ +** | V | +** |<------WRITER_CACHEMOD-------->| +** | | | +** | V | +** |<-------WRITER_DBMOD---------->| +** | | | +** | V | +** +<------WRITER_FINISHED-------->+ +** +** +** List of state transitions and the C [function] that performs each: +** +** OPEN -> READER [sqlite3PagerSharedLock] +** READER -> OPEN [pager_unlock] +** +** READER -> WRITER_LOCKED [sqlite3PagerBegin] +** WRITER_LOCKED -> WRITER_CACHEMOD [pager_open_journal] +** WRITER_CACHEMOD -> WRITER_DBMOD [syncJournal] +** WRITER_DBMOD -> WRITER_FINISHED [sqlite3PagerCommitPhaseOne] +** WRITER_*** -> READER [pager_end_transaction] +** +** WRITER_*** -> ERROR [pager_error] +** ERROR -> OPEN [pager_unlock] +** +** +** OPEN: +** +** The pager starts up in this state. Nothing is guaranteed in this +** state - the file may or may not be locked and the database size is +** unknown. The database may not be read or written. +** +** * No read or write transaction is active. +** * Any lock, or no lock at all, may be held on the database file. +** * The dbSize, dbOrigSize and dbFileSize variables may not be trusted. +** +** READER: +** +** In this state all the requirements for reading the database in +** rollback (non-WAL) mode are met. Unless the pager is (or recently +** was) in exclusive-locking mode, a user-level read transaction is +** open. The database size is known in this state. +** +** A connection running with locking_mode=normal enters this state when +** it opens a read-transaction on the database and returns to state +** OPEN after the read-transaction is completed. However a connection +** running in locking_mode=exclusive (including temp databases) remains in +** this state even after the read-transaction is closed. The only way +** a locking_mode=exclusive connection can transition from READER to OPEN +** is via the ERROR state (see below). +** +** * A read transaction may be active (but a write-transaction cannot). +** * A SHARED or greater lock is held on the database file. +** * The dbSize variable may be trusted (even if a user-level read +** transaction is not active). The dbOrigSize and dbFileSize variables +** may not be trusted at this point. +** * If the database is a WAL database, then the WAL connection is open. +** * Even if a read-transaction is not open, it is guaranteed that +** there is no hot-journal in the file-system. +** +** WRITER_LOCKED: +** +** The pager moves to this state from READER when a write-transaction +** is first opened on the database. In WRITER_LOCKED state, all locks +** required to start a write-transaction are held, but no actual +** modifications to the cache or database have taken place. +** +** In rollback mode, a RESERVED or (if the transaction was opened with +** BEGIN EXCLUSIVE) EXCLUSIVE lock is obtained on the database file when +** moving to this state, but the journal file is not written to or opened +** to in this state. If the transaction is committed or rolled back while +** in WRITER_LOCKED state, all that is required is to unlock the database +** file. +** +** IN WAL mode, WalBeginWriteTransaction() is called to lock the log file. +** If the connection is running with locking_mode=exclusive, an attempt +** is made to obtain an EXCLUSIVE lock on the database file. +** +** * A write transaction is active. +** * If the connection is open in rollback-mode, a RESERVED or greater +** lock is held on the database file. +** * If the connection is open in WAL-mode, a WAL write transaction +** is open (i.e. sqlite3WalBeginWriteTransaction() has been successfully +** called). +** * The dbSize, dbOrigSize and dbFileSize variables are all valid. +** * The contents of the pager cache have not been modified. +** * The journal file may or may not be open. +** * Nothing (not even the first header) has been written to the journal. +** +** WRITER_CACHEMOD: +** +** A pager moves from WRITER_LOCKED state to this state when a page is +** first modified by the upper layer. In rollback mode the journal file +** is opened (if it is not already open) and a header written to the +** start of it. The database file on disk has not been modified. +** +** * A write transaction is active. +** * A RESERVED or greater lock is held on the database file. +** * The journal file is open and the first header has been written +** to it, but the header has not been synced to disk. +** * The contents of the page cache have been modified. +** +** WRITER_DBMOD: +** +** The pager transitions from WRITER_CACHEMOD into WRITER_DBMOD state +** when it modifies the contents of the database file. WAL connections +** never enter this state (since they do not modify the database file, +** just the log file). +** +** * A write transaction is active. +** * An EXCLUSIVE or greater lock is held on the database file. +** * The journal file is open and the first header has been written +** and synced to disk. +** * The contents of the page cache have been modified (and possibly +** written to disk). +** +** WRITER_FINISHED: +** +** It is not possible for a WAL connection to enter this state. +** +** A rollback-mode pager changes to WRITER_FINISHED state from WRITER_DBMOD +** state after the entire transaction has been successfully written into the +** database file. In this state the transaction may be committed simply +** by finalizing the journal file. Once in WRITER_FINISHED state, it is +** not possible to modify the database further. At this point, the upper +** layer must either commit or rollback the transaction. +** +** * A write transaction is active. +** * An EXCLUSIVE or greater lock is held on the database file. +** * All writing and syncing of journal and database data has finished. +** If no error occurred, all that remains is to finalize the journal to +** commit the transaction. If an error did occur, the caller will need +** to rollback the transaction. +** +** ERROR: +** +** The ERROR state is entered when an IO or disk-full error (including +** SQLITE_IOERR_NOMEM) occurs at a point in the code that makes it +** difficult to be sure that the in-memory pager state (cache contents, +** db size etc.) are consistent with the contents of the file-system. +** +** Temporary pager files may enter the ERROR state, but in-memory pagers +** cannot. +** +** For example, if an IO error occurs while performing a rollback, +** the contents of the page-cache may be left in an inconsistent state. +** At this point it would be dangerous to change back to READER state +** (as usually happens after a rollback). Any subsequent readers might +** report database corruption (due to the inconsistent cache), and if +** they upgrade to writers, they may inadvertently corrupt the database +** file. To avoid this hazard, the pager switches into the ERROR state +** instead of READER following such an error. +** +** Once it has entered the ERROR state, any attempt to use the pager +** to read or write data returns an error. Eventually, once all +** outstanding transactions have been abandoned, the pager is able to +** transition back to OPEN state, discarding the contents of the +** page-cache and any other in-memory state at the same time. Everything +** is reloaded from disk (and, if necessary, hot-journal rollback peformed) +** when a read-transaction is next opened on the pager (transitioning +** the pager into READER state). At that point the system has recovered +** from the error. +** +** Specifically, the pager jumps into the ERROR state if: +** +** 1. An error occurs while attempting a rollback. This happens in +** function sqlite3PagerRollback(). +** +** 2. An error occurs while attempting to finalize a journal file +** following a commit in function sqlite3PagerCommitPhaseTwo(). +** +** 3. An error occurs while attempting to write to the journal or +** database file in function pagerStress() in order to free up +** memory. +** +** In other cases, the error is returned to the b-tree layer. The b-tree +** layer then attempts a rollback operation. If the error condition +** persists, the pager enters the ERROR state via condition (1) above. +** +** Condition (3) is necessary because it can be triggered by a read-only +** statement executed within a transaction. In this case, if the error +** code were simply returned to the user, the b-tree layer would not +** automatically attempt a rollback, as it assumes that an error in a +** read-only statement cannot leave the pager in an internally inconsistent +** state. +** +** * The Pager.errCode variable is set to something other than SQLITE_OK. +** * There are one or more outstanding references to pages (after the +** last reference is dropped the pager should move back to OPEN state). +** * The pager is not an in-memory pager. +** +** +** Notes: +** +** * A pager is never in WRITER_DBMOD or WRITER_FINISHED state if the +** connection is open in WAL mode. A WAL connection is always in one +** of the first four states. +** +** * Normally, a connection open in exclusive mode is never in PAGER_OPEN +** state. There are two exceptions: immediately after exclusive-mode has +** been turned on (and before any read or write transactions are +** executed), and when the pager is leaving the "error state". +** +** * See also: assert_pager_state(). +*/ +#define PAGER_OPEN 0 +#define PAGER_READER 1 +#define PAGER_WRITER_LOCKED 2 +#define PAGER_WRITER_CACHEMOD 3 +#define PAGER_WRITER_DBMOD 4 +#define PAGER_WRITER_FINISHED 5 +#define PAGER_ERROR 6 + +/* +** The Pager.eLock variable is almost always set to one of the +** following locking-states, according to the lock currently held on +** the database file: NO_LOCK, SHARED_LOCK, RESERVED_LOCK or EXCLUSIVE_LOCK. +** This variable is kept up to date as locks are taken and released by +** the pagerLockDb() and pagerUnlockDb() wrappers. +** +** If the VFS xLock() or xUnlock() returns an error other than SQLITE_BUSY +** (i.e. one of the SQLITE_IOERR subtypes), it is not clear whether or not +** the operation was successful. In these circumstances pagerLockDb() and +** pagerUnlockDb() take a conservative approach - eLock is always updated +** when unlocking the file, and only updated when locking the file if the +** VFS call is successful. This way, the Pager.eLock variable may be set +** to a less exclusive (lower) value than the lock that is actually held +** at the system level, but it is never set to a more exclusive value. +** +** This is usually safe. If an xUnlock fails or appears to fail, there may +** be a few redundant xLock() calls or a lock may be held for longer than +** required, but nothing really goes wrong. +** +** The exception is when the database file is unlocked as the pager moves +** from ERROR to OPEN state. At this point there may be a hot-journal file +** in the file-system that needs to be rolled back (as part of an OPEN->SHARED +** transition, by the same pager or any other). If the call to xUnlock() +** fails at this point and the pager is left holding an EXCLUSIVE lock, this +** can confuse the call to xCheckReservedLock() call made later as part +** of hot-journal detection. +** +** xCheckReservedLock() is defined as returning true "if there is a RESERVED +** lock held by this process or any others". So xCheckReservedLock may +** return true because the caller itself is holding an EXCLUSIVE lock (but +** doesn't know it because of a previous error in xUnlock). If this happens +** a hot-journal may be mistaken for a journal being created by an active +** transaction in another process, causing SQLite to read from the database +** without rolling it back. +** +** To work around this, if a call to xUnlock() fails when unlocking the +** database in the ERROR state, Pager.eLock is set to UNKNOWN_LOCK. It +** is only changed back to a real locking state after a successful call +** to xLock(EXCLUSIVE). Also, the code to do the OPEN->SHARED state transition +** omits the check for a hot-journal if Pager.eLock is set to UNKNOWN_LOCK +** lock. Instead, it assumes a hot-journal exists and obtains an EXCLUSIVE +** lock on the database file before attempting to roll it back. See function +** PagerSharedLock() for more detail. +** +** Pager.eLock may only be set to UNKNOWN_LOCK when the pager is in +** PAGER_OPEN state. +*/ +#define UNKNOWN_LOCK (EXCLUSIVE_LOCK+1) + +/* +** The maximum allowed sector size. 64KiB. If the xSectorsize() method +** returns a value larger than this, then MAX_SECTOR_SIZE is used instead. +** This could conceivably cause corruption following a power failure on +** such a system. This is currently an undocumented limit. +*/ +#define MAX_SECTOR_SIZE 0x10000 + + +/* +** An instance of the following structure is allocated for each active +** savepoint and statement transaction in the system. All such structures +** are stored in the Pager.aSavepoint[] array, which is allocated and +** resized using sqlite3Realloc(). +** +** When a savepoint is created, the PagerSavepoint.iHdrOffset field is +** set to 0. If a journal-header is written into the main journal while +** the savepoint is active, then iHdrOffset is set to the byte offset +** immediately following the last journal record written into the main +** journal before the journal-header. This is required during savepoint +** rollback (see pagerPlaybackSavepoint()). +*/ +typedef struct PagerSavepoint PagerSavepoint; +struct PagerSavepoint { + i64 iOffset; /* Starting offset in main journal */ + i64 iHdrOffset; /* See above */ + Bitvec *pInSavepoint; /* Set of pages in this savepoint */ + Pgno nOrig; /* Original number of pages in file */ + Pgno iSubRec; /* Index of first record in sub-journal */ + int bTruncateOnRelease; /* If stmt journal may be truncated on RELEASE */ +#ifndef SQLITE_OMIT_WAL + u32 aWalData[WAL_SAVEPOINT_NDATA]; /* WAL savepoint context */ +#endif +}; + +/* +** Bits of the Pager.doNotSpill flag. See further description below. +*/ +#define SPILLFLAG_OFF 0x01 /* Never spill cache. Set via pragma */ +#define SPILLFLAG_ROLLBACK 0x02 /* Current rolling back, so do not spill */ +#define SPILLFLAG_NOSYNC 0x04 /* Spill is ok, but do not sync */ + +/* +** An open page cache is an instance of struct Pager. A description of +** some of the more important member variables follows: +** +** eState +** +** The current 'state' of the pager object. See the comment and state +** diagram above for a description of the pager state. +** +** eLock +** +** For a real on-disk database, the current lock held on the database file - +** NO_LOCK, SHARED_LOCK, RESERVED_LOCK or EXCLUSIVE_LOCK. +** +** For a temporary or in-memory database (neither of which require any +** locks), this variable is always set to EXCLUSIVE_LOCK. Since such +** databases always have Pager.exclusiveMode==1, this tricks the pager +** logic into thinking that it already has all the locks it will ever +** need (and no reason to release them). +** +** In some (obscure) circumstances, this variable may also be set to +** UNKNOWN_LOCK. See the comment above the #define of UNKNOWN_LOCK for +** details. +** +** changeCountDone +** +** This boolean variable is used to make sure that the change-counter +** (the 4-byte header field at byte offset 24 of the database file) is +** not updated more often than necessary. +** +** It is set to true when the change-counter field is updated, which +** can only happen if an exclusive lock is held on the database file. +** It is cleared (set to false) whenever an exclusive lock is +** relinquished on the database file. Each time a transaction is committed, +** The changeCountDone flag is inspected. If it is true, the work of +** updating the change-counter is omitted for the current transaction. +** +** This mechanism means that when running in exclusive mode, a connection +** need only update the change-counter once, for the first transaction +** committed. +** +** setSuper +** +** When PagerCommitPhaseOne() is called to commit a transaction, it may +** (or may not) specify a super-journal name to be written into the +** journal file before it is synced to disk. +** +** Whether or not a journal file contains a super-journal pointer affects +** the way in which the journal file is finalized after the transaction is +** committed or rolled back when running in "journal_mode=PERSIST" mode. +** If a journal file does not contain a super-journal pointer, it is +** finalized by overwriting the first journal header with zeroes. If +** it does contain a super-journal pointer the journal file is finalized +** by truncating it to zero bytes, just as if the connection were +** running in "journal_mode=truncate" mode. +** +** Journal files that contain super-journal pointers cannot be finalized +** simply by overwriting the first journal-header with zeroes, as the +** super-journal pointer could interfere with hot-journal rollback of any +** subsequently interrupted transaction that reuses the journal file. +** +** The flag is cleared as soon as the journal file is finalized (either +** by PagerCommitPhaseTwo or PagerRollback). If an IO error prevents the +** journal file from being successfully finalized, the setSuper flag +** is cleared anyway (and the pager will move to ERROR state). +** +** doNotSpill +** +** This variables control the behavior of cache-spills (calls made by +** the pcache module to the pagerStress() routine to write cached data +** to the file-system in order to free up memory). +** +** When bits SPILLFLAG_OFF or SPILLFLAG_ROLLBACK of doNotSpill are set, +** writing to the database from pagerStress() is disabled altogether. +** The SPILLFLAG_ROLLBACK case is done in a very obscure case that +** comes up during savepoint rollback that requires the pcache module +** to allocate a new page to prevent the journal file from being written +** while it is being traversed by code in pager_playback(). The SPILLFLAG_OFF +** case is a user preference. +** +** If the SPILLFLAG_NOSYNC bit is set, writing to the database from +** pagerStress() is permitted, but syncing the journal file is not. +** This flag is set by sqlite3PagerWrite() when the file-system sector-size +** is larger than the database page-size in order to prevent a journal sync +** from happening in between the journalling of two pages on the same sector. +** +** subjInMemory +** +** This is a boolean variable. If true, then any required sub-journal +** is opened as an in-memory journal file. If false, then in-memory +** sub-journals are only used for in-memory pager files. +** +** This variable is updated by the upper layer each time a new +** write-transaction is opened. +** +** dbSize, dbOrigSize, dbFileSize +** +** Variable dbSize is set to the number of pages in the database file. +** It is valid in PAGER_READER and higher states (all states except for +** OPEN and ERROR). +** +** dbSize is set based on the size of the database file, which may be +** larger than the size of the database (the value stored at offset +** 28 of the database header by the btree). If the size of the file +** is not an integer multiple of the page-size, the value stored in +** dbSize is rounded down (i.e. a 5KB file with 2K page-size has dbSize==2). +** Except, any file that is greater than 0 bytes in size is considered +** to have at least one page. (i.e. a 1KB file with 2K page-size leads +** to dbSize==1). +** +** During a write-transaction, if pages with page-numbers greater than +** dbSize are modified in the cache, dbSize is updated accordingly. +** Similarly, if the database is truncated using PagerTruncateImage(), +** dbSize is updated. +** +** Variables dbOrigSize and dbFileSize are valid in states +** PAGER_WRITER_LOCKED and higher. dbOrigSize is a copy of the dbSize +** variable at the start of the transaction. It is used during rollback, +** and to determine whether or not pages need to be journalled before +** being modified. +** +** Throughout a write-transaction, dbFileSize contains the size of +** the file on disk in pages. It is set to a copy of dbSize when the +** write-transaction is first opened, and updated when VFS calls are made +** to write or truncate the database file on disk. +** +** The only reason the dbFileSize variable is required is to suppress +** unnecessary calls to xTruncate() after committing a transaction. If, +** when a transaction is committed, the dbFileSize variable indicates +** that the database file is larger than the database image (Pager.dbSize), +** pager_truncate() is called. The pager_truncate() call uses xFilesize() +** to measure the database file on disk, and then truncates it if required. +** dbFileSize is not used when rolling back a transaction. In this case +** pager_truncate() is called unconditionally (which means there may be +** a call to xFilesize() that is not strictly required). In either case, +** pager_truncate() may cause the file to become smaller or larger. +** +** dbHintSize +** +** The dbHintSize variable is used to limit the number of calls made to +** the VFS xFileControl(FCNTL_SIZE_HINT) method. +** +** dbHintSize is set to a copy of the dbSize variable when a +** write-transaction is opened (at the same time as dbFileSize and +** dbOrigSize). If the xFileControl(FCNTL_SIZE_HINT) method is called, +** dbHintSize is increased to the number of pages that correspond to the +** size-hint passed to the method call. See pager_write_pagelist() for +** details. +** +** errCode +** +** The Pager.errCode variable is only ever used in PAGER_ERROR state. It +** is set to zero in all other states. In PAGER_ERROR state, Pager.errCode +** is always set to SQLITE_FULL, SQLITE_IOERR or one of the SQLITE_IOERR_XXX +** sub-codes. +** +** syncFlags, walSyncFlags +** +** syncFlags is either SQLITE_SYNC_NORMAL (0x02) or SQLITE_SYNC_FULL (0x03). +** syncFlags is used for rollback mode. walSyncFlags is used for WAL mode +** and contains the flags used to sync the checkpoint operations in the +** lower two bits, and sync flags used for transaction commits in the WAL +** file in bits 0x04 and 0x08. In other words, to get the correct sync flags +** for checkpoint operations, use (walSyncFlags&0x03) and to get the correct +** sync flags for transaction commit, use ((walSyncFlags>>2)&0x03). Note +** that with synchronous=NORMAL in WAL mode, transaction commit is not synced +** meaning that the 0x04 and 0x08 bits are both zero. +*/ +struct Pager { + sqlite3_vfs *pVfs; /* OS functions to use for IO */ + u8 exclusiveMode; /* Boolean. True if locking_mode==EXCLUSIVE */ + u8 journalMode; /* One of the PAGER_JOURNALMODE_* values */ + u8 useJournal; /* Use a rollback journal on this file */ + u8 noSync; /* Do not sync the journal if true */ + u8 fullSync; /* Do extra syncs of the journal for robustness */ + u8 extraSync; /* sync directory after journal delete */ + u8 syncFlags; /* SYNC_NORMAL or SYNC_FULL otherwise */ + u8 walSyncFlags; /* See description above */ + u8 tempFile; /* zFilename is a temporary or immutable file */ + u8 noLock; /* Do not lock (except in WAL mode) */ + u8 readOnly; /* True for a read-only database */ + u8 memDb; /* True to inhibit all file I/O */ + u8 memVfs; /* VFS-implemented memory database */ + + /************************************************************************** + ** The following block contains those class members that change during + ** routine operation. Class members not in this block are either fixed + ** when the pager is first created or else only change when there is a + ** significant mode change (such as changing the page_size, locking_mode, + ** or the journal_mode). From another view, these class members describe + ** the "state" of the pager, while other class members describe the + ** "configuration" of the pager. + */ + u8 eState; /* Pager state (OPEN, READER, WRITER_LOCKED..) */ + u8 eLock; /* Current lock held on database file */ + u8 changeCountDone; /* Set after incrementing the change-counter */ + u8 setSuper; /* Super-jrnl name is written into jrnl */ + u8 doNotSpill; /* Do not spill the cache when non-zero */ + u8 subjInMemory; /* True to use in-memory sub-journals */ + u8 bUseFetch; /* True to use xFetch() */ + u8 hasHeldSharedLock; /* True if a shared lock has ever been held */ + Pgno dbSize; /* Number of pages in the database */ + Pgno dbOrigSize; /* dbSize before the current transaction */ + Pgno dbFileSize; /* Number of pages in the database file */ + Pgno dbHintSize; /* Value passed to FCNTL_SIZE_HINT call */ + int errCode; /* One of several kinds of errors */ + int nRec; /* Pages journalled since last j-header written */ + u32 cksumInit; /* Quasi-random value added to every checksum */ + u32 nSubRec; /* Number of records written to sub-journal */ + Bitvec *pInJournal; /* One bit for each page in the database file */ + sqlite3_file *fd; /* File descriptor for database */ + sqlite3_file *jfd; /* File descriptor for main journal */ + sqlite3_file *sjfd; /* File descriptor for sub-journal */ + i64 journalOff; /* Current write offset in the journal file */ + i64 journalHdr; /* Byte offset to previous journal header */ + sqlite3_backup *pBackup; /* Pointer to list of ongoing backup processes */ + PagerSavepoint *aSavepoint; /* Array of active savepoints */ + int nSavepoint; /* Number of elements in aSavepoint[] */ + u32 iDataVersion; /* Changes whenever database content changes */ + char dbFileVers[16]; /* Changes whenever database file changes */ + + int nMmapOut; /* Number of mmap pages currently outstanding */ + sqlite3_int64 szMmap; /* Desired maximum mmap size */ + PgHdr *pMmapFreelist; /* List of free mmap page headers (pDirty) */ + /* + ** End of the routinely-changing class members + ***************************************************************************/ + + u16 nExtra; /* Add this many bytes to each in-memory page */ + i16 nReserve; /* Number of unused bytes at end of each page */ + u32 vfsFlags; /* Flags for sqlite3_vfs.xOpen() */ + u32 sectorSize; /* Assumed sector size during rollback */ + Pgno mxPgno; /* Maximum allowed size of the database */ + i64 pageSize; /* Number of bytes in a page */ + i64 journalSizeLimit; /* Size limit for persistent journal files */ + char *zFilename; /* Name of the database file */ + char *zJournal; /* Name of the journal file */ + int (*xBusyHandler)(void*); /* Function to call when busy */ + void *pBusyHandlerArg; /* Context argument for xBusyHandler */ + int aStat[4]; /* Total cache hits, misses, writes, spills */ +#ifdef SQLITE_TEST + int nRead; /* Database pages read */ +#endif + void (*xReiniter)(DbPage*); /* Call this routine when reloading pages */ + int (*xGet)(Pager*,Pgno,DbPage**,int); /* Routine to fetch a patch */ + char *pTmpSpace; /* Pager.pageSize bytes of space for tmp use */ + PCache *pPCache; /* Pointer to page cache object */ +#ifndef SQLITE_OMIT_WAL + Wal *pWal; /* Write-ahead log used by "journal_mode=wal" */ + char *zWal; /* File name for write-ahead log */ +#endif +}; + +/* +** Indexes for use with Pager.aStat[]. The Pager.aStat[] array contains +** the values accessed by passing SQLITE_DBSTATUS_CACHE_HIT, CACHE_MISS +** or CACHE_WRITE to sqlite3_db_status(). +*/ +#define PAGER_STAT_HIT 0 +#define PAGER_STAT_MISS 1 +#define PAGER_STAT_WRITE 2 +#define PAGER_STAT_SPILL 3 + +/* +** The following global variables hold counters used for +** testing purposes only. These variables do not exist in +** a non-testing build. These variables are not thread-safe. +*/ +#ifdef SQLITE_TEST +int sqlite3_pager_readdb_count = 0; /* Number of full pages read from DB */ +int sqlite3_pager_writedb_count = 0; /* Number of full pages written to DB */ +int sqlite3_pager_writej_count = 0; /* Number of pages written to journal */ +# define PAGER_INCR(v) v++ +#else +# define PAGER_INCR(v) +#endif + + + +/* +** Journal files begin with the following magic string. The data +** was obtained from /dev/random. It is used only as a sanity check. +** +** Since version 2.8.0, the journal format contains additional sanity +** checking information. If the power fails while the journal is being +** written, semi-random garbage data might appear in the journal +** file after power is restored. If an attempt is then made +** to roll the journal back, the database could be corrupted. The additional +** sanity checking data is an attempt to discover the garbage in the +** journal and ignore it. +** +** The sanity checking information for the new journal format consists +** of a 32-bit checksum on each page of data. The checksum covers both +** the page number and the pPager->pageSize bytes of data for the page. +** This cksum is initialized to a 32-bit random value that appears in the +** journal file right after the header. The random initializer is important, +** because garbage data that appears at the end of a journal is likely +** data that was once in other files that have now been deleted. If the +** garbage data came from an obsolete journal file, the checksums might +** be correct. But by initializing the checksum to random value which +** is different for every journal, we minimize that risk. +*/ +static const unsigned char aJournalMagic[] = { + 0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7, +}; + +/* +** The size of the of each page record in the journal is given by +** the following macro. +*/ +#define JOURNAL_PG_SZ(pPager) ((pPager->pageSize) + 8) + +/* +** The journal header size for this pager. This is usually the same +** size as a single disk sector. See also setSectorSize(). +*/ +#define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize) + +/* +** The macro MEMDB is true if we are dealing with an in-memory database. +** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set, +** the value of MEMDB will be a constant and the compiler will optimize +** out code that would never execute. +*/ +#ifdef SQLITE_OMIT_MEMORYDB +# define MEMDB 0 +#else +# define MEMDB pPager->memDb +#endif + +/* +** The macro USEFETCH is true if we are allowed to use the xFetch and xUnfetch +** interfaces to access the database using memory-mapped I/O. +*/ +#if SQLITE_MAX_MMAP_SIZE>0 +# define USEFETCH(x) ((x)->bUseFetch) +#else +# define USEFETCH(x) 0 +#endif + +/* +** The argument to this macro is a file descriptor (type sqlite3_file*). +** Return 0 if it is not open, or non-zero (but not 1) if it is. +** +** This is so that expressions can be written as: +** +** if( isOpen(pPager->jfd) ){ ... +** +** instead of +** +** if( pPager->jfd->pMethods ){ ... +*/ +#define isOpen(pFd) ((pFd)->pMethods!=0) + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +/* +** Return true if page pgno can be read directly from the database file +** by the b-tree layer. This is the case if: +** +** * the database file is open, +** * there are no dirty pages in the cache, and +** * the desired page is not currently in the wal file. +*/ +int sqlite3PagerDirectReadOk(Pager *pPager, Pgno pgno){ + if( pPager->fd->pMethods==0 ) return 0; + if( sqlite3PCacheIsDirty(pPager->pPCache) ) return 0; +#ifndef SQLITE_OMIT_WAL + if( pPager->pWal ){ + u32 iRead = 0; + int rc; + rc = sqlite3WalFindFrame(pPager->pWal, pgno, &iRead); + return (rc==SQLITE_OK && iRead==0); + } +#endif + return 1; +} +#endif + +#ifndef SQLITE_OMIT_WAL +# define pagerUseWal(x) ((x)->pWal!=0) +#else +# define pagerUseWal(x) 0 +# define pagerRollbackWal(x) 0 +# define pagerWalFrames(v,w,x,y) 0 +# define pagerOpenWalIfPresent(z) SQLITE_OK +# define pagerBeginReadTransaction(z) SQLITE_OK +#endif + +#ifndef NDEBUG +/* +** Usage: +** +** assert( assert_pager_state(pPager) ); +** +** This function runs many asserts to try to find inconsistencies in +** the internal state of the Pager object. +*/ +static int assert_pager_state(Pager *p){ + Pager *pPager = p; + + /* State must be valid. */ + assert( p->eState==PAGER_OPEN + || p->eState==PAGER_READER + || p->eState==PAGER_WRITER_LOCKED + || p->eState==PAGER_WRITER_CACHEMOD + || p->eState==PAGER_WRITER_DBMOD + || p->eState==PAGER_WRITER_FINISHED + || p->eState==PAGER_ERROR + ); + + /* Regardless of the current state, a temp-file connection always behaves + ** as if it has an exclusive lock on the database file. It never updates + ** the change-counter field, so the changeCountDone flag is always set. + */ + assert( p->tempFile==0 || p->eLock==EXCLUSIVE_LOCK ); + assert( p->tempFile==0 || pPager->changeCountDone ); + + /* If the useJournal flag is clear, the journal-mode must be "OFF". + ** And if the journal-mode is "OFF", the journal file must not be open. + */ + assert( p->journalMode==PAGER_JOURNALMODE_OFF || p->useJournal ); + assert( p->journalMode!=PAGER_JOURNALMODE_OFF || !isOpen(p->jfd) ); + + /* Check that MEMDB implies noSync. And an in-memory journal. Since + ** this means an in-memory pager performs no IO at all, it cannot encounter + ** either SQLITE_IOERR or SQLITE_FULL during rollback or while finalizing + ** a journal file. (although the in-memory journal implementation may + ** return SQLITE_IOERR_NOMEM while the journal file is being written). It + ** is therefore not possible for an in-memory pager to enter the ERROR + ** state. + */ + if( MEMDB ){ + assert( !isOpen(p->fd) ); + assert( p->noSync ); + assert( p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_MEMORY + ); + assert( p->eState!=PAGER_ERROR && p->eState!=PAGER_OPEN ); + assert( pagerUseWal(p)==0 ); + } + + /* If changeCountDone is set, a RESERVED lock or greater must be held + ** on the file. + */ + assert( pPager->changeCountDone==0 || pPager->eLock>=RESERVED_LOCK ); + assert( p->eLock!=PENDING_LOCK ); + + switch( p->eState ){ + case PAGER_OPEN: + assert( !MEMDB ); + assert( pPager->errCode==SQLITE_OK ); + assert( sqlite3PcacheRefCount(pPager->pPCache)==0 || pPager->tempFile ); + break; + + case PAGER_READER: + assert( pPager->errCode==SQLITE_OK ); + assert( p->eLock!=UNKNOWN_LOCK ); + assert( p->eLock>=SHARED_LOCK ); + break; + + case PAGER_WRITER_LOCKED: + assert( p->eLock!=UNKNOWN_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + if( !pagerUseWal(pPager) ){ + assert( p->eLock>=RESERVED_LOCK ); + } + assert( pPager->dbSize==pPager->dbOrigSize ); + assert( pPager->dbOrigSize==pPager->dbFileSize ); + assert( pPager->dbOrigSize==pPager->dbHintSize ); + assert( pPager->setSuper==0 ); + break; + + case PAGER_WRITER_CACHEMOD: + assert( p->eLock!=UNKNOWN_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + if( !pagerUseWal(pPager) ){ + /* It is possible that if journal_mode=wal here that neither the + ** journal file nor the WAL file are open. This happens during + ** a rollback transaction that switches from journal_mode=off + ** to journal_mode=wal. + */ + assert( p->eLock>=RESERVED_LOCK ); + assert( isOpen(p->jfd) + || p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_WAL + ); + } + assert( pPager->dbOrigSize==pPager->dbFileSize ); + assert( pPager->dbOrigSize==pPager->dbHintSize ); + break; + + case PAGER_WRITER_DBMOD: + assert( p->eLock==EXCLUSIVE_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + assert( !pagerUseWal(pPager) ); + assert( p->eLock>=EXCLUSIVE_LOCK ); + assert( isOpen(p->jfd) + || p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_WAL + || (sqlite3OsDeviceCharacteristics(p->fd)&SQLITE_IOCAP_BATCH_ATOMIC) + ); + assert( pPager->dbOrigSize<=pPager->dbHintSize ); + break; + + case PAGER_WRITER_FINISHED: + assert( p->eLock==EXCLUSIVE_LOCK ); + assert( pPager->errCode==SQLITE_OK ); + assert( !pagerUseWal(pPager) ); + assert( isOpen(p->jfd) + || p->journalMode==PAGER_JOURNALMODE_OFF + || p->journalMode==PAGER_JOURNALMODE_WAL + || (sqlite3OsDeviceCharacteristics(p->fd)&SQLITE_IOCAP_BATCH_ATOMIC) + ); + break; + + case PAGER_ERROR: + /* There must be at least one outstanding reference to the pager if + ** in ERROR state. Otherwise the pager should have already dropped + ** back to OPEN state. + */ + assert( pPager->errCode!=SQLITE_OK ); + assert( sqlite3PcacheRefCount(pPager->pPCache)>0 || pPager->tempFile ); + break; + } + + return 1; +} +#endif /* ifndef NDEBUG */ + +#ifdef SQLITE_DEBUG +/* +** Return a pointer to a human readable string in a static buffer +** containing the state of the Pager object passed as an argument. This +** is intended to be used within debuggers. For example, as an alternative +** to "print *pPager" in gdb: +** +** (gdb) printf "%s", print_pager_state(pPager) +** +** This routine has external linkage in order to suppress compiler warnings +** about an unused function. It is enclosed within SQLITE_DEBUG and so does +** not appear in normal builds. +*/ +char *print_pager_state(Pager *p){ + static char zRet[1024]; + + sqlite3_snprintf(1024, zRet, + "Filename: %s\n" + "State: %s errCode=%d\n" + "Lock: %s\n" + "Locking mode: locking_mode=%s\n" + "Journal mode: journal_mode=%s\n" + "Backing store: tempFile=%d memDb=%d useJournal=%d\n" + "Journal: journalOff=%lld journalHdr=%lld\n" + "Size: dbsize=%d dbOrigSize=%d dbFileSize=%d\n" + , p->zFilename + , p->eState==PAGER_OPEN ? "OPEN" : + p->eState==PAGER_READER ? "READER" : + p->eState==PAGER_WRITER_LOCKED ? "WRITER_LOCKED" : + p->eState==PAGER_WRITER_CACHEMOD ? "WRITER_CACHEMOD" : + p->eState==PAGER_WRITER_DBMOD ? "WRITER_DBMOD" : + p->eState==PAGER_WRITER_FINISHED ? "WRITER_FINISHED" : + p->eState==PAGER_ERROR ? "ERROR" : "?error?" + , (int)p->errCode + , p->eLock==NO_LOCK ? "NO_LOCK" : + p->eLock==RESERVED_LOCK ? "RESERVED" : + p->eLock==EXCLUSIVE_LOCK ? "EXCLUSIVE" : + p->eLock==SHARED_LOCK ? "SHARED" : + p->eLock==UNKNOWN_LOCK ? "UNKNOWN" : "?error?" + , p->exclusiveMode ? "exclusive" : "normal" + , p->journalMode==PAGER_JOURNALMODE_MEMORY ? "memory" : + p->journalMode==PAGER_JOURNALMODE_OFF ? "off" : + p->journalMode==PAGER_JOURNALMODE_DELETE ? "delete" : + p->journalMode==PAGER_JOURNALMODE_PERSIST ? "persist" : + p->journalMode==PAGER_JOURNALMODE_TRUNCATE ? "truncate" : + p->journalMode==PAGER_JOURNALMODE_WAL ? "wal" : "?error?" + , (int)p->tempFile, (int)p->memDb, (int)p->useJournal + , p->journalOff, p->journalHdr + , (int)p->dbSize, (int)p->dbOrigSize, (int)p->dbFileSize + ); + + return zRet; +} +#endif + +/* Forward references to the various page getters */ +static int getPageNormal(Pager*,Pgno,DbPage**,int); +static int getPageError(Pager*,Pgno,DbPage**,int); +#if SQLITE_MAX_MMAP_SIZE>0 +static int getPageMMap(Pager*,Pgno,DbPage**,int); +#endif + +/* +** Set the Pager.xGet method for the appropriate routine used to fetch +** content from the pager. +*/ +static void setGetterMethod(Pager *pPager){ + if( pPager->errCode ){ + pPager->xGet = getPageError; +#if SQLITE_MAX_MMAP_SIZE>0 + }else if( USEFETCH(pPager) ){ + pPager->xGet = getPageMMap; +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + }else{ + pPager->xGet = getPageNormal; + } +} + +/* +** Return true if it is necessary to write page *pPg into the sub-journal. +** A page needs to be written into the sub-journal if there exists one +** or more open savepoints for which: +** +** * The page-number is less than or equal to PagerSavepoint.nOrig, and +** * The bit corresponding to the page-number is not set in +** PagerSavepoint.pInSavepoint. +*/ +static int subjRequiresPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + PagerSavepoint *p; + Pgno pgno = pPg->pgno; + int i; + for(i=0; inSavepoint; i++){ + p = &pPager->aSavepoint[i]; + if( p->nOrig>=pgno && 0==sqlite3BitvecTestNotNull(p->pInSavepoint, pgno) ){ + for(i=i+1; inSavepoint; i++){ + pPager->aSavepoint[i].bTruncateOnRelease = 0; + } + return 1; + } + } + return 0; +} + +#ifdef SQLITE_DEBUG +/* +** Return true if the page is already in the journal file. +*/ +static int pageInJournal(Pager *pPager, PgHdr *pPg){ + return sqlite3BitvecTest(pPager->pInJournal, pPg->pgno); +} +#endif + +/* +** Read a 32-bit integer from the given file descriptor. Store the integer +** that is read in *pRes. Return SQLITE_OK if everything worked, or an +** error code is something goes wrong. +** +** All values are stored on disk as big-endian. +*/ +static int read32bits(sqlite3_file *fd, i64 offset, u32 *pRes){ + unsigned char ac[4]; + int rc = sqlite3OsRead(fd, ac, sizeof(ac), offset); + if( rc==SQLITE_OK ){ + *pRes = sqlite3Get4byte(ac); + } + return rc; +} + +/* +** Write a 32-bit integer into a string buffer in big-endian byte order. +*/ +#define put32bits(A,B) sqlite3Put4byte((u8*)A,B) + + +/* +** Write a 32-bit integer into the given file descriptor. Return SQLITE_OK +** on success or an error code is something goes wrong. +*/ +static int write32bits(sqlite3_file *fd, i64 offset, u32 val){ + char ac[4]; + put32bits(ac, val); + return sqlite3OsWrite(fd, ac, 4, offset); +} + +/* +** Unlock the database file to level eLock, which must be either NO_LOCK +** or SHARED_LOCK. Regardless of whether or not the call to xUnlock() +** succeeds, set the Pager.eLock variable to match the (attempted) new lock. +** +** Except, if Pager.eLock is set to UNKNOWN_LOCK when this function is +** called, do not modify it. See the comment above the #define of +** UNKNOWN_LOCK for an explanation of this. +*/ +static int pagerUnlockDb(Pager *pPager, int eLock){ + int rc = SQLITE_OK; + + assert( !pPager->exclusiveMode || pPager->eLock==eLock ); + assert( eLock==NO_LOCK || eLock==SHARED_LOCK ); + assert( eLock!=NO_LOCK || pagerUseWal(pPager)==0 ); + if( isOpen(pPager->fd) ){ + assert( pPager->eLock>=eLock ); + rc = pPager->noLock ? SQLITE_OK : sqlite3OsUnlock(pPager->fd, eLock); + if( pPager->eLock!=UNKNOWN_LOCK ){ + pPager->eLock = (u8)eLock; + } + IOTRACE(("UNLOCK %p %d\n", pPager, eLock)) + } + pPager->changeCountDone = pPager->tempFile; /* ticket fb3b3024ea238d5c */ + return rc; +} + +/* +** Lock the database file to level eLock, which must be either SHARED_LOCK, +** RESERVED_LOCK or EXCLUSIVE_LOCK. If the caller is successful, set the +** Pager.eLock variable to the new locking state. +** +** Except, if Pager.eLock is set to UNKNOWN_LOCK when this function is +** called, do not modify it unless the new locking state is EXCLUSIVE_LOCK. +** See the comment above the #define of UNKNOWN_LOCK for an explanation +** of this. +*/ +static int pagerLockDb(Pager *pPager, int eLock){ + int rc = SQLITE_OK; + + assert( eLock==SHARED_LOCK || eLock==RESERVED_LOCK || eLock==EXCLUSIVE_LOCK ); + if( pPager->eLockeLock==UNKNOWN_LOCK ){ + rc = pPager->noLock ? SQLITE_OK : sqlite3OsLock(pPager->fd, eLock); + if( rc==SQLITE_OK && (pPager->eLock!=UNKNOWN_LOCK||eLock==EXCLUSIVE_LOCK) ){ + pPager->eLock = (u8)eLock; + IOTRACE(("LOCK %p %d\n", pPager, eLock)) + } + } + return rc; +} + +/* +** This function determines whether or not the atomic-write or +** atomic-batch-write optimizations can be used with this pager. The +** atomic-write optimization can be used if: +** +** (a) the value returned by OsDeviceCharacteristics() indicates that +** a database page may be written atomically, and +** (b) the value returned by OsSectorSize() is less than or equal +** to the page size. +** +** If it can be used, then the value returned is the size of the journal +** file when it contains rollback data for exactly one page. +** +** The atomic-batch-write optimization can be used if OsDeviceCharacteristics() +** returns a value with the SQLITE_IOCAP_BATCH_ATOMIC bit set. -1 is +** returned in this case. +** +** If neither optimization can be used, 0 is returned. +*/ +static int jrnlBufferSize(Pager *pPager){ + assert( !MEMDB ); + +#if defined(SQLITE_ENABLE_ATOMIC_WRITE) \ + || defined(SQLITE_ENABLE_BATCH_ATOMIC_WRITE) + int dc; /* Device characteristics */ + + assert( isOpen(pPager->fd) ); + dc = sqlite3OsDeviceCharacteristics(pPager->fd); +#else + UNUSED_PARAMETER(pPager); +#endif + +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( pPager->dbSize>0 && (dc&SQLITE_IOCAP_BATCH_ATOMIC) ){ + return -1; + } +#endif + +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + { + int nSector = pPager->sectorSize; + int szPage = pPager->pageSize; + + assert(SQLITE_IOCAP_ATOMIC512==(512>>8)); + assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8)); + if( 0==(dc&(SQLITE_IOCAP_ATOMIC|(szPage>>8)) || nSector>szPage) ){ + return 0; + } + } + + return JOURNAL_HDR_SZ(pPager) + JOURNAL_PG_SZ(pPager); +#endif + + return 0; +} + +/* +** If SQLITE_CHECK_PAGES is defined then we do some sanity checking +** on the cache using a hash function. This is used for testing +** and debugging only. +*/ +#ifdef SQLITE_CHECK_PAGES +/* +** Return a 32-bit hash of the page data for pPage. +*/ +static u32 pager_datahash(int nByte, unsigned char *pData){ + u32 hash = 0; + int i; + for(i=0; ipPager->pageSize, (unsigned char *)pPage->pData); +} +static void pager_set_pagehash(PgHdr *pPage){ + pPage->pageHash = pager_pagehash(pPage); +} + +/* +** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES +** is defined, and NDEBUG is not defined, an assert() statement checks +** that the page is either dirty or still matches the calculated page-hash. +*/ +#define CHECK_PAGE(x) checkPage(x) +static void checkPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + assert( pPager->eState!=PAGER_ERROR ); + assert( (pPg->flags&PGHDR_DIRTY) || pPg->pageHash==pager_pagehash(pPg) ); +} + +#else +#define pager_datahash(X,Y) 0 +#define pager_pagehash(X) 0 +#define pager_set_pagehash(X) +#define CHECK_PAGE(x) +#endif /* SQLITE_CHECK_PAGES */ + +/* +** When this is called the journal file for pager pPager must be open. +** This function attempts to read a super-journal file name from the +** end of the file and, if successful, copies it into memory supplied +** by the caller. See comments above writeSuperJournal() for the format +** used to store a super-journal file name at the end of a journal file. +** +** zSuper must point to a buffer of at least nSuper bytes allocated by +** the caller. This should be sqlite3_vfs.mxPathname+1 (to ensure there is +** enough space to write the super-journal name). If the super-journal +** name in the journal is longer than nSuper bytes (including a +** nul-terminator), then this is handled as if no super-journal name +** were present in the journal. +** +** If a super-journal file name is present at the end of the journal +** file, then it is copied into the buffer pointed to by zSuper. A +** nul-terminator byte is appended to the buffer following the +** super-journal file name. +** +** If it is determined that no super-journal file name is present +** zSuper[0] is set to 0 and SQLITE_OK returned. +** +** If an error occurs while reading from the journal file, an SQLite +** error code is returned. +*/ +static int readSuperJournal(sqlite3_file *pJrnl, char *zSuper, u32 nSuper){ + int rc; /* Return code */ + u32 len; /* Length in bytes of super-journal name */ + i64 szJ; /* Total size in bytes of journal file pJrnl */ + u32 cksum; /* MJ checksum value read from journal */ + u32 u; /* Unsigned loop counter */ + unsigned char aMagic[8]; /* A buffer to hold the magic header */ + zSuper[0] = '\0'; + + if( SQLITE_OK!=(rc = sqlite3OsFileSize(pJrnl, &szJ)) + || szJ<16 + || SQLITE_OK!=(rc = read32bits(pJrnl, szJ-16, &len)) + || len>=nSuper + || len>szJ-16 + || len==0 + || SQLITE_OK!=(rc = read32bits(pJrnl, szJ-12, &cksum)) + || SQLITE_OK!=(rc = sqlite3OsRead(pJrnl, aMagic, 8, szJ-8)) + || memcmp(aMagic, aJournalMagic, 8) + || SQLITE_OK!=(rc = sqlite3OsRead(pJrnl, zSuper, len, szJ-16-len)) + ){ + return rc; + } + + /* See if the checksum matches the super-journal name */ + for(u=0; ujournalOff, assuming a sector +** size of pPager->sectorSize bytes. +** +** i.e for a sector size of 512: +** +** Pager.journalOff Return value +** --------------------------------------- +** 0 0 +** 512 512 +** 100 512 +** 2000 2048 +** +*/ +static i64 journalHdrOffset(Pager *pPager){ + i64 offset = 0; + i64 c = pPager->journalOff; + if( c ){ + offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager); + } + assert( offset%JOURNAL_HDR_SZ(pPager)==0 ); + assert( offset>=c ); + assert( (offset-c)jfd) ); + assert( !sqlite3JournalIsInMemory(pPager->jfd) ); + if( pPager->journalOff ){ + const i64 iLimit = pPager->journalSizeLimit; /* Local cache of jsl */ + + IOTRACE(("JZEROHDR %p\n", pPager)) + if( doTruncate || iLimit==0 ){ + rc = sqlite3OsTruncate(pPager->jfd, 0); + }else{ + static const char zeroHdr[28] = {0}; + rc = sqlite3OsWrite(pPager->jfd, zeroHdr, sizeof(zeroHdr), 0); + } + if( rc==SQLITE_OK && !pPager->noSync ){ + rc = sqlite3OsSync(pPager->jfd, SQLITE_SYNC_DATAONLY|pPager->syncFlags); + } + + /* At this point the transaction is committed but the write lock + ** is still held on the file. If there is a size limit configured for + ** the persistent journal and the journal file currently consumes more + ** space than that limit allows for, truncate it now. There is no need + ** to sync the file following this operation. + */ + if( rc==SQLITE_OK && iLimit>0 ){ + i64 sz; + rc = sqlite3OsFileSize(pPager->jfd, &sz); + if( rc==SQLITE_OK && sz>iLimit ){ + rc = sqlite3OsTruncate(pPager->jfd, iLimit); + } + } + } + return rc; +} + +/* +** The journal file must be open when this routine is called. A journal +** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the +** current location. +** +** The format for the journal header is as follows: +** - 8 bytes: Magic identifying journal format. +** - 4 bytes: Number of records in journal, or -1 no-sync mode is on. +** - 4 bytes: Random number used for page hash. +** - 4 bytes: Initial database page count. +** - 4 bytes: Sector size used by the process that wrote this journal. +** - 4 bytes: Database page size. +** +** Followed by (JOURNAL_HDR_SZ - 28) bytes of unused space. +*/ +static int writeJournalHdr(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + char *zHeader = pPager->pTmpSpace; /* Temporary space used to build header */ + u32 nHeader = (u32)pPager->pageSize;/* Size of buffer pointed to by zHeader */ + u32 nWrite; /* Bytes of header sector written */ + int ii; /* Loop counter */ + + assert( isOpen(pPager->jfd) ); /* Journal file must be open. */ + + if( nHeader>JOURNAL_HDR_SZ(pPager) ){ + nHeader = JOURNAL_HDR_SZ(pPager); + } + + /* If there are active savepoints and any of them were created + ** since the most recent journal header was written, update the + ** PagerSavepoint.iHdrOffset fields now. + */ + for(ii=0; iinSavepoint; ii++){ + if( pPager->aSavepoint[ii].iHdrOffset==0 ){ + pPager->aSavepoint[ii].iHdrOffset = pPager->journalOff; + } + } + + pPager->journalHdr = pPager->journalOff = journalHdrOffset(pPager); + + /* + ** Write the nRec Field - the number of page records that follow this + ** journal header. Normally, zero is written to this value at this time. + ** After the records are added to the journal (and the journal synced, + ** if in full-sync mode), the zero is overwritten with the true number + ** of records (see syncJournal()). + ** + ** A faster alternative is to write 0xFFFFFFFF to the nRec field. When + ** reading the journal this value tells SQLite to assume that the + ** rest of the journal file contains valid page records. This assumption + ** is dangerous, as if a failure occurred whilst writing to the journal + ** file it may contain some garbage data. There are two scenarios + ** where this risk can be ignored: + ** + ** * When the pager is in no-sync mode. Corruption can follow a + ** power failure in this case anyway. + ** + ** * When the SQLITE_IOCAP_SAFE_APPEND flag is set. This guarantees + ** that garbage data is never appended to the journal file. + */ + assert( isOpen(pPager->fd) || pPager->noSync ); + if( pPager->noSync || (pPager->journalMode==PAGER_JOURNALMODE_MEMORY) + || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_SAFE_APPEND) + ){ + memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic)); + put32bits(&zHeader[sizeof(aJournalMagic)], 0xffffffff); + }else{ + memset(zHeader, 0, sizeof(aJournalMagic)+4); + } + + /* The random check-hash initializer */ + sqlite3_randomness(sizeof(pPager->cksumInit), &pPager->cksumInit); + put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit); + /* The initial database size */ + put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbOrigSize); + /* The assumed sector size for this process */ + put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize); + + /* The page size */ + put32bits(&zHeader[sizeof(aJournalMagic)+16], pPager->pageSize); + + /* Initializing the tail of the buffer is not necessary. Everything + ** works find if the following memset() is omitted. But initializing + ** the memory prevents valgrind from complaining, so we are willing to + ** take the performance hit. + */ + memset(&zHeader[sizeof(aJournalMagic)+20], 0, + nHeader-(sizeof(aJournalMagic)+20)); + + /* In theory, it is only necessary to write the 28 bytes that the + ** journal header consumes to the journal file here. Then increment the + ** Pager.journalOff variable by JOURNAL_HDR_SZ so that the next + ** record is written to the following sector (leaving a gap in the file + ** that will be implicitly filled in by the OS). + ** + ** However it has been discovered that on some systems this pattern can + ** be significantly slower than contiguously writing data to the file, + ** even if that means explicitly writing data to the block of + ** (JOURNAL_HDR_SZ - 28) bytes that will not be used. So that is what + ** is done. + ** + ** The loop is required here in case the sector-size is larger than the + ** database page size. Since the zHeader buffer is only Pager.pageSize + ** bytes in size, more than one call to sqlite3OsWrite() may be required + ** to populate the entire journal header sector. + */ + for(nWrite=0; rc==SQLITE_OK&&nWritejournalHdr, nHeader)) + rc = sqlite3OsWrite(pPager->jfd, zHeader, nHeader, pPager->journalOff); + assert( pPager->journalHdr <= pPager->journalOff ); + pPager->journalOff += nHeader; + } + + return rc; +} + +/* +** The journal file must be open when this is called. A journal header file +** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal +** file. The current location in the journal file is given by +** pPager->journalOff. See comments above function writeJournalHdr() for +** a description of the journal header format. +** +** If the header is read successfully, *pNRec is set to the number of +** page records following this header and *pDbSize is set to the size of the +** database before the transaction began, in pages. Also, pPager->cksumInit +** is set to the value read from the journal header. SQLITE_OK is returned +** in this case. +** +** If the journal header file appears to be corrupted, SQLITE_DONE is +** returned and *pNRec and *PDbSize are undefined. If JOURNAL_HDR_SZ bytes +** cannot be read from the journal file an error code is returned. +*/ +static int readJournalHdr( + Pager *pPager, /* Pager object */ + int isHot, + i64 journalSize, /* Size of the open journal file in bytes */ + u32 *pNRec, /* OUT: Value read from the nRec field */ + u32 *pDbSize /* OUT: Value of original database size field */ +){ + int rc; /* Return code */ + unsigned char aMagic[8]; /* A buffer to hold the magic header */ + i64 iHdrOff; /* Offset of journal header being read */ + + assert( isOpen(pPager->jfd) ); /* Journal file must be open. */ + + /* Advance Pager.journalOff to the start of the next sector. If the + ** journal file is too small for there to be a header stored at this + ** point, return SQLITE_DONE. + */ + pPager->journalOff = journalHdrOffset(pPager); + if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){ + return SQLITE_DONE; + } + iHdrOff = pPager->journalOff; + + /* Read in the first 8 bytes of the journal header. If they do not match + ** the magic string found at the start of each journal header, return + ** SQLITE_DONE. If an IO error occurs, return an error code. Otherwise, + ** proceed. + */ + if( isHot || iHdrOff!=pPager->journalHdr ){ + rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic), iHdrOff); + if( rc ){ + return rc; + } + if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){ + return SQLITE_DONE; + } + } + + /* Read the first three 32-bit fields of the journal header: The nRec + ** field, the checksum-initializer and the database size at the start + ** of the transaction. Return an error code if anything goes wrong. + */ + if( SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+8, pNRec)) + || SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+12, &pPager->cksumInit)) + || SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+16, pDbSize)) + ){ + return rc; + } + + if( pPager->journalOff==0 ){ + u32 iPageSize; /* Page-size field of journal header */ + u32 iSectorSize; /* Sector-size field of journal header */ + + /* Read the page-size and sector-size journal header fields. */ + if( SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+20, &iSectorSize)) + || SQLITE_OK!=(rc = read32bits(pPager->jfd, iHdrOff+24, &iPageSize)) + ){ + return rc; + } + + /* Versions of SQLite prior to 3.5.8 set the page-size field of the + ** journal header to zero. In this case, assume that the Pager.pageSize + ** variable is already set to the correct page size. + */ + if( iPageSize==0 ){ + iPageSize = pPager->pageSize; + } + + /* Check that the values read from the page-size and sector-size fields + ** are within range. To be 'in range', both values need to be a power + ** of two greater than or equal to 512 or 32, and not greater than their + ** respective compile time maximum limits. + */ + if( iPageSize<512 || iSectorSize<32 + || iPageSize>SQLITE_MAX_PAGE_SIZE || iSectorSize>MAX_SECTOR_SIZE + || ((iPageSize-1)&iPageSize)!=0 || ((iSectorSize-1)&iSectorSize)!=0 + ){ + /* If the either the page-size or sector-size in the journal-header is + ** invalid, then the process that wrote the journal-header must have + ** crashed before the header was synced. In this case stop reading + ** the journal file here. + */ + return SQLITE_DONE; + } + + /* Update the page-size to match the value read from the journal. + ** Use a testcase() macro to make sure that malloc failure within + ** PagerSetPagesize() is tested. + */ + rc = sqlite3PagerSetPagesize(pPager, &iPageSize, -1); + testcase( rc!=SQLITE_OK ); + + /* Update the assumed sector-size to match the value used by + ** the process that created this journal. If this journal was + ** created by a process other than this one, then this routine + ** is being called from within pager_playback(). The local value + ** of Pager.sectorSize is restored at the end of that routine. + */ + pPager->sectorSize = iSectorSize; + } + + pPager->journalOff += JOURNAL_HDR_SZ(pPager); + return rc; +} + + +/* +** Write the supplied super-journal name into the journal file for pager +** pPager at the current location. The super-journal name must be the last +** thing written to a journal file. If the pager is in full-sync mode, the +** journal file descriptor is advanced to the next sector boundary before +** anything is written. The format is: +** +** + 4 bytes: PAGER_MJ_PGNO. +** + N bytes: super-journal filename in utf-8. +** + 4 bytes: N (length of super-journal name in bytes, no nul-terminator). +** + 4 bytes: super-journal name checksum. +** + 8 bytes: aJournalMagic[]. +** +** The super-journal page checksum is the sum of the bytes in thesuper-journal +** name, where each byte is interpreted as a signed 8-bit integer. +** +** If zSuper is a NULL pointer (occurs for a single database transaction), +** this call is a no-op. +*/ +static int writeSuperJournal(Pager *pPager, const char *zSuper){ + int rc; /* Return code */ + int nSuper; /* Length of string zSuper */ + i64 iHdrOff; /* Offset of header in journal file */ + i64 jrnlSize; /* Size of journal file on disk */ + u32 cksum = 0; /* Checksum of string zSuper */ + + assert( pPager->setSuper==0 ); + assert( !pagerUseWal(pPager) ); + + if( !zSuper + || pPager->journalMode==PAGER_JOURNALMODE_MEMORY + || !isOpen(pPager->jfd) + ){ + return SQLITE_OK; + } + pPager->setSuper = 1; + assert( pPager->journalHdr <= pPager->journalOff ); + + /* Calculate the length in bytes and the checksum of zSuper */ + for(nSuper=0; zSuper[nSuper]; nSuper++){ + cksum += zSuper[nSuper]; + } + + /* If in full-sync mode, advance to the next disk sector before writing + ** the super-journal name. This is in case the previous page written to + ** the journal has already been synced. + */ + if( pPager->fullSync ){ + pPager->journalOff = journalHdrOffset(pPager); + } + iHdrOff = pPager->journalOff; + + /* Write the super-journal data to the end of the journal file. If + ** an error occurs, return the error code to the caller. + */ + if( (0 != (rc = write32bits(pPager->jfd, iHdrOff, PAGER_MJ_PGNO(pPager)))) + || (0 != (rc = sqlite3OsWrite(pPager->jfd, zSuper, nSuper, iHdrOff+4))) + || (0 != (rc = write32bits(pPager->jfd, iHdrOff+4+nSuper, nSuper))) + || (0 != (rc = write32bits(pPager->jfd, iHdrOff+4+nSuper+4, cksum))) + || (0 != (rc = sqlite3OsWrite(pPager->jfd, aJournalMagic, 8, + iHdrOff+4+nSuper+8))) + ){ + return rc; + } + pPager->journalOff += (nSuper+20); + + /* If the pager is in peristent-journal mode, then the physical + ** journal-file may extend past the end of the super-journal name + ** and 8 bytes of magic data just written to the file. This is + ** dangerous because the code to rollback a hot-journal file + ** will not be able to find the super-journal name to determine + ** whether or not the journal is hot. + ** + ** Easiest thing to do in this scenario is to truncate the journal + ** file to the required size. + */ + if( SQLITE_OK==(rc = sqlite3OsFileSize(pPager->jfd, &jrnlSize)) + && jrnlSize>pPager->journalOff + ){ + rc = sqlite3OsTruncate(pPager->jfd, pPager->journalOff); + } + return rc; +} + +/* +** Discard the entire contents of the in-memory page-cache. +*/ +static void pager_reset(Pager *pPager){ + pPager->iDataVersion++; + sqlite3BackupRestart(pPager->pBackup); + sqlite3PcacheClear(pPager->pPCache); +} + +/* +** Return the pPager->iDataVersion value +*/ +u32 sqlite3PagerDataVersion(Pager *pPager){ + return pPager->iDataVersion; +} + +/* +** Free all structures in the Pager.aSavepoint[] array and set both +** Pager.aSavepoint and Pager.nSavepoint to zero. Close the sub-journal +** if it is open and the pager is not in exclusive mode. +*/ +static void releaseAllSavepoints(Pager *pPager){ + int ii; /* Iterator for looping through Pager.aSavepoint */ + for(ii=0; iinSavepoint; ii++){ + sqlite3BitvecDestroy(pPager->aSavepoint[ii].pInSavepoint); + } + if( !pPager->exclusiveMode || sqlite3JournalIsInMemory(pPager->sjfd) ){ + sqlite3OsClose(pPager->sjfd); + } + sqlite3_free(pPager->aSavepoint); + pPager->aSavepoint = 0; + pPager->nSavepoint = 0; + pPager->nSubRec = 0; +} + +/* +** Set the bit number pgno in the PagerSavepoint.pInSavepoint +** bitvecs of all open savepoints. Return SQLITE_OK if successful +** or SQLITE_NOMEM if a malloc failure occurs. +*/ +static int addToSavepointBitvecs(Pager *pPager, Pgno pgno){ + int ii; /* Loop counter */ + int rc = SQLITE_OK; /* Result code */ + + for(ii=0; iinSavepoint; ii++){ + PagerSavepoint *p = &pPager->aSavepoint[ii]; + if( pgno<=p->nOrig ){ + rc |= sqlite3BitvecSet(p->pInSavepoint, pgno); + testcase( rc==SQLITE_NOMEM ); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + } + } + return rc; +} + +/* +** This function is a no-op if the pager is in exclusive mode and not +** in the ERROR state. Otherwise, it switches the pager to PAGER_OPEN +** state. +** +** If the pager is not in exclusive-access mode, the database file is +** completely unlocked. If the file is unlocked and the file-system does +** not exhibit the UNDELETABLE_WHEN_OPEN property, the journal file is +** closed (if it is open). +** +** If the pager is in ERROR state when this function is called, the +** contents of the pager cache are discarded before switching back to +** the OPEN state. Regardless of whether the pager is in exclusive-mode +** or not, any journal file left in the file-system will be treated +** as a hot-journal and rolled back the next time a read-transaction +** is opened (by this or by any other connection). +*/ +static void pager_unlock(Pager *pPager){ + + assert( pPager->eState==PAGER_READER + || pPager->eState==PAGER_OPEN + || pPager->eState==PAGER_ERROR + ); + + sqlite3BitvecDestroy(pPager->pInJournal); + pPager->pInJournal = 0; + releaseAllSavepoints(pPager); + + if( pagerUseWal(pPager) ){ + assert( !isOpen(pPager->jfd) ); + sqlite3WalEndReadTransaction(pPager->pWal); + pPager->eState = PAGER_OPEN; + }else if( !pPager->exclusiveMode ){ + int rc; /* Error code returned by pagerUnlockDb() */ + int iDc = isOpen(pPager->fd)?sqlite3OsDeviceCharacteristics(pPager->fd):0; + + /* If the operating system support deletion of open files, then + ** close the journal file when dropping the database lock. Otherwise + ** another connection with journal_mode=delete might delete the file + ** out from under us. + */ + assert( (PAGER_JOURNALMODE_MEMORY & 5)!=1 ); + assert( (PAGER_JOURNALMODE_OFF & 5)!=1 ); + assert( (PAGER_JOURNALMODE_WAL & 5)!=1 ); + assert( (PAGER_JOURNALMODE_DELETE & 5)!=1 ); + assert( (PAGER_JOURNALMODE_TRUNCATE & 5)==1 ); + assert( (PAGER_JOURNALMODE_PERSIST & 5)==1 ); + if( 0==(iDc & SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN) + || 1!=(pPager->journalMode & 5) + ){ + sqlite3OsClose(pPager->jfd); + } + + /* If the pager is in the ERROR state and the call to unlock the database + ** file fails, set the current lock to UNKNOWN_LOCK. See the comment + ** above the #define for UNKNOWN_LOCK for an explanation of why this + ** is necessary. + */ + rc = pagerUnlockDb(pPager, NO_LOCK); + if( rc!=SQLITE_OK && pPager->eState==PAGER_ERROR ){ + pPager->eLock = UNKNOWN_LOCK; + } + + /* The pager state may be changed from PAGER_ERROR to PAGER_OPEN here + ** without clearing the error code. This is intentional - the error + ** code is cleared and the cache reset in the block below. + */ + assert( pPager->errCode || pPager->eState!=PAGER_ERROR ); + pPager->eState = PAGER_OPEN; + } + + /* If Pager.errCode is set, the contents of the pager cache cannot be + ** trusted. Now that there are no outstanding references to the pager, + ** it can safely move back to PAGER_OPEN state. This happens in both + ** normal and exclusive-locking mode. + */ + assert( pPager->errCode==SQLITE_OK || !MEMDB ); + if( pPager->errCode ){ + if( pPager->tempFile==0 ){ + pager_reset(pPager); + pPager->changeCountDone = 0; + pPager->eState = PAGER_OPEN; + }else{ + pPager->eState = (isOpen(pPager->jfd) ? PAGER_OPEN : PAGER_READER); + } + if( USEFETCH(pPager) ) sqlite3OsUnfetch(pPager->fd, 0, 0); + pPager->errCode = SQLITE_OK; + setGetterMethod(pPager); + } + + pPager->journalOff = 0; + pPager->journalHdr = 0; + pPager->setSuper = 0; +} + +/* +** This function is called whenever an IOERR or FULL error that requires +** the pager to transition into the ERROR state may ahve occurred. +** The first argument is a pointer to the pager structure, the second +** the error-code about to be returned by a pager API function. The +** value returned is a copy of the second argument to this function. +** +** If the second argument is SQLITE_FULL, SQLITE_IOERR or one of the +** IOERR sub-codes, the pager enters the ERROR state and the error code +** is stored in Pager.errCode. While the pager remains in the ERROR state, +** all major API calls on the Pager will immediately return Pager.errCode. +** +** The ERROR state indicates that the contents of the pager-cache +** cannot be trusted. This state can be cleared by completely discarding +** the contents of the pager-cache. If a transaction was active when +** the persistent error occurred, then the rollback journal may need +** to be replayed to restore the contents of the database file (as if +** it were a hot-journal). +*/ +static int pager_error(Pager *pPager, int rc){ + int rc2 = rc & 0xff; + assert( rc==SQLITE_OK || !MEMDB ); + assert( + pPager->errCode==SQLITE_FULL || + pPager->errCode==SQLITE_OK || + (pPager->errCode & 0xff)==SQLITE_IOERR + ); + if( rc2==SQLITE_FULL || rc2==SQLITE_IOERR ){ + pPager->errCode = rc; + pPager->eState = PAGER_ERROR; + setGetterMethod(pPager); + } + return rc; +} + +static int pager_truncate(Pager *pPager, Pgno nPage); + +/* +** The write transaction open on pPager is being committed (bCommit==1) +** or rolled back (bCommit==0). +** +** Return TRUE if and only if all dirty pages should be flushed to disk. +** +** Rules: +** +** * For non-TEMP databases, always sync to disk. This is necessary +** for transactions to be durable. +** +** * Sync TEMP database only on a COMMIT (not a ROLLBACK) when the backing +** file has been created already (via a spill on pagerStress()) and +** when the number of dirty pages in memory exceeds 25% of the total +** cache size. +*/ +static int pagerFlushOnCommit(Pager *pPager, int bCommit){ + if( pPager->tempFile==0 ) return 1; + if( !bCommit ) return 0; + if( !isOpen(pPager->fd) ) return 0; + return (sqlite3PCachePercentDirty(pPager->pPCache)>=25); +} + +/* +** This routine ends a transaction. A transaction is usually ended by +** either a COMMIT or a ROLLBACK operation. This routine may be called +** after rollback of a hot-journal, or if an error occurs while opening +** the journal file or writing the very first journal-header of a +** database transaction. +** +** This routine is never called in PAGER_ERROR state. If it is called +** in PAGER_NONE or PAGER_SHARED state and the lock held is less +** exclusive than a RESERVED lock, it is a no-op. +** +** Otherwise, any active savepoints are released. +** +** If the journal file is open, then it is "finalized". Once a journal +** file has been finalized it is not possible to use it to roll back a +** transaction. Nor will it be considered to be a hot-journal by this +** or any other database connection. Exactly how a journal is finalized +** depends on whether or not the pager is running in exclusive mode and +** the current journal-mode (Pager.journalMode value), as follows: +** +** journalMode==MEMORY +** Journal file descriptor is simply closed. This destroys an +** in-memory journal. +** +** journalMode==TRUNCATE +** Journal file is truncated to zero bytes in size. +** +** journalMode==PERSIST +** The first 28 bytes of the journal file are zeroed. This invalidates +** the first journal header in the file, and hence the entire journal +** file. An invalid journal file cannot be rolled back. +** +** journalMode==DELETE +** The journal file is closed and deleted using sqlite3OsDelete(). +** +** If the pager is running in exclusive mode, this method of finalizing +** the journal file is never used. Instead, if the journalMode is +** DELETE and the pager is in exclusive mode, the method described under +** journalMode==PERSIST is used instead. +** +** After the journal is finalized, the pager moves to PAGER_READER state. +** If running in non-exclusive rollback mode, the lock on the file is +** downgraded to a SHARED_LOCK. +** +** SQLITE_OK is returned if no error occurs. If an error occurs during +** any of the IO operations to finalize the journal file or unlock the +** database then the IO error code is returned to the user. If the +** operation to finalize the journal file fails, then the code still +** tries to unlock the database file if not in exclusive mode. If the +** unlock operation fails as well, then the first error code related +** to the first error encountered (the journal finalization one) is +** returned. +*/ +static int pager_end_transaction(Pager *pPager, int hasSuper, int bCommit){ + int rc = SQLITE_OK; /* Error code from journal finalization operation */ + int rc2 = SQLITE_OK; /* Error code from db file unlock operation */ + + /* Do nothing if the pager does not have an open write transaction + ** or at least a RESERVED lock. This function may be called when there + ** is no write-transaction active but a RESERVED or greater lock is + ** held under two circumstances: + ** + ** 1. After a successful hot-journal rollback, it is called with + ** eState==PAGER_NONE and eLock==EXCLUSIVE_LOCK. + ** + ** 2. If a connection with locking_mode=exclusive holding an EXCLUSIVE + ** lock switches back to locking_mode=normal and then executes a + ** read-transaction, this function is called with eState==PAGER_READER + ** and eLock==EXCLUSIVE_LOCK when the read-transaction is closed. + */ + assert( assert_pager_state(pPager) ); + assert( pPager->eState!=PAGER_ERROR ); + if( pPager->eStateeLockjfd) || pPager->pInJournal==0 + || (sqlite3OsDeviceCharacteristics(pPager->fd)&SQLITE_IOCAP_BATCH_ATOMIC) + ); + if( isOpen(pPager->jfd) ){ + assert( !pagerUseWal(pPager) ); + + /* Finalize the journal file. */ + if( sqlite3JournalIsInMemory(pPager->jfd) ){ + /* assert( pPager->journalMode==PAGER_JOURNALMODE_MEMORY ); */ + sqlite3OsClose(pPager->jfd); + }else if( pPager->journalMode==PAGER_JOURNALMODE_TRUNCATE ){ + if( pPager->journalOff==0 ){ + rc = SQLITE_OK; + }else{ + rc = sqlite3OsTruncate(pPager->jfd, 0); + if( rc==SQLITE_OK && pPager->fullSync ){ + /* Make sure the new file size is written into the inode right away. + ** Otherwise the journal might resurrect following a power loss and + ** cause the last transaction to roll back. See + ** https://bugzilla.mozilla.org/show_bug.cgi?id=1072773 + */ + rc = sqlite3OsSync(pPager->jfd, pPager->syncFlags); + } + } + pPager->journalOff = 0; + }else if( pPager->journalMode==PAGER_JOURNALMODE_PERSIST + || (pPager->exclusiveMode && pPager->journalMode!=PAGER_JOURNALMODE_WAL) + ){ + rc = zeroJournalHdr(pPager, hasSuper||pPager->tempFile); + pPager->journalOff = 0; + }else{ + /* This branch may be executed with Pager.journalMode==MEMORY if + ** a hot-journal was just rolled back. In this case the journal + ** file should be closed and deleted. If this connection writes to + ** the database file, it will do so using an in-memory journal. + */ + int bDelete = !pPager->tempFile; + assert( sqlite3JournalIsInMemory(pPager->jfd)==0 ); + assert( pPager->journalMode==PAGER_JOURNALMODE_DELETE + || pPager->journalMode==PAGER_JOURNALMODE_MEMORY + || pPager->journalMode==PAGER_JOURNALMODE_WAL + ); + sqlite3OsClose(pPager->jfd); + if( bDelete ){ + rc = sqlite3OsDelete(pPager->pVfs, pPager->zJournal, pPager->extraSync); + } + } + } + +#ifdef SQLITE_CHECK_PAGES + sqlite3PcacheIterateDirty(pPager->pPCache, pager_set_pagehash); + if( pPager->dbSize==0 && sqlite3PcacheRefCount(pPager->pPCache)>0 ){ + PgHdr *p = sqlite3PagerLookup(pPager, 1); + if( p ){ + p->pageHash = 0; + sqlite3PagerUnrefNotNull(p); + } + } +#endif + + sqlite3BitvecDestroy(pPager->pInJournal); + pPager->pInJournal = 0; + pPager->nRec = 0; + if( rc==SQLITE_OK ){ + if( MEMDB || pagerFlushOnCommit(pPager, bCommit) ){ + sqlite3PcacheCleanAll(pPager->pPCache); + }else{ + sqlite3PcacheClearWritable(pPager->pPCache); + } + sqlite3PcacheTruncate(pPager->pPCache, pPager->dbSize); + } + + if( pagerUseWal(pPager) ){ + /* Drop the WAL write-lock, if any. Also, if the connection was in + ** locking_mode=exclusive mode but is no longer, drop the EXCLUSIVE + ** lock held on the database file. + */ + rc2 = sqlite3WalEndWriteTransaction(pPager->pWal); + assert( rc2==SQLITE_OK ); + }else if( rc==SQLITE_OK && bCommit && pPager->dbFileSize>pPager->dbSize ){ + /* This branch is taken when committing a transaction in rollback-journal + ** mode if the database file on disk is larger than the database image. + ** At this point the journal has been finalized and the transaction + ** successfully committed, but the EXCLUSIVE lock is still held on the + ** file. So it is safe to truncate the database file to its minimum + ** required size. */ + assert( pPager->eLock==EXCLUSIVE_LOCK ); + rc = pager_truncate(pPager, pPager->dbSize); + } + + if( rc==SQLITE_OK && bCommit ){ + rc = sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_COMMIT_PHASETWO, 0); + if( rc==SQLITE_NOTFOUND ) rc = SQLITE_OK; + } + + if( !pPager->exclusiveMode + && (!pagerUseWal(pPager) || sqlite3WalExclusiveMode(pPager->pWal, 0)) + ){ + rc2 = pagerUnlockDb(pPager, SHARED_LOCK); + } + pPager->eState = PAGER_READER; + pPager->setSuper = 0; + + return (rc==SQLITE_OK?rc2:rc); +} + +/* +** Execute a rollback if a transaction is active and unlock the +** database file. +** +** If the pager has already entered the ERROR state, do not attempt +** the rollback at this time. Instead, pager_unlock() is called. The +** call to pager_unlock() will discard all in-memory pages, unlock +** the database file and move the pager back to OPEN state. If this +** means that there is a hot-journal left in the file-system, the next +** connection to obtain a shared lock on the pager (which may be this one) +** will roll it back. +** +** If the pager has not already entered the ERROR state, but an IO or +** malloc error occurs during a rollback, then this will itself cause +** the pager to enter the ERROR state. Which will be cleared by the +** call to pager_unlock(), as described above. +*/ +static void pagerUnlockAndRollback(Pager *pPager){ + if( pPager->eState!=PAGER_ERROR && pPager->eState!=PAGER_OPEN ){ + assert( assert_pager_state(pPager) ); + if( pPager->eState>=PAGER_WRITER_LOCKED ){ + sqlite3BeginBenignMalloc(); + sqlite3PagerRollback(pPager); + sqlite3EndBenignMalloc(); + }else if( !pPager->exclusiveMode ){ + assert( pPager->eState==PAGER_READER ); + pager_end_transaction(pPager, 0, 0); + } + } + pager_unlock(pPager); +} + +/* +** Parameter aData must point to a buffer of pPager->pageSize bytes +** of data. Compute and return a checksum based ont the contents of the +** page of data and the current value of pPager->cksumInit. +** +** This is not a real checksum. It is really just the sum of the +** random initial value (pPager->cksumInit) and every 200th byte +** of the page data, starting with byte offset (pPager->pageSize%200). +** Each byte is interpreted as an 8-bit unsigned integer. +** +** Changing the formula used to compute this checksum results in an +** incompatible journal file format. +** +** If journal corruption occurs due to a power failure, the most likely +** scenario is that one end or the other of the record will be changed. +** It is much less likely that the two ends of the journal record will be +** correct and the middle be corrupt. Thus, this "checksum" scheme, +** though fast and simple, catches the mostly likely kind of corruption. +*/ +static u32 pager_cksum(Pager *pPager, const u8 *aData){ + u32 cksum = pPager->cksumInit; /* Checksum value to return */ + int i = pPager->pageSize-200; /* Loop counter */ + while( i>0 ){ + cksum += aData[i]; + i -= 200; + } + return cksum; +} + +/* +** Read a single page from either the journal file (if isMainJrnl==1) or +** from the sub-journal (if isMainJrnl==0) and playback that page. +** The page begins at offset *pOffset into the file. The *pOffset +** value is increased to the start of the next page in the journal. +** +** The main rollback journal uses checksums - the statement journal does +** not. +** +** If the page number of the page record read from the (sub-)journal file +** is greater than the current value of Pager.dbSize, then playback is +** skipped and SQLITE_OK is returned. +** +** If pDone is not NULL, then it is a record of pages that have already +** been played back. If the page at *pOffset has already been played back +** (if the corresponding pDone bit is set) then skip the playback. +** Make sure the pDone bit corresponding to the *pOffset page is set +** prior to returning. +** +** If the page record is successfully read from the (sub-)journal file +** and played back, then SQLITE_OK is returned. If an IO error occurs +** while reading the record from the (sub-)journal file or while writing +** to the database file, then the IO error code is returned. If data +** is successfully read from the (sub-)journal file but appears to be +** corrupted, SQLITE_DONE is returned. Data is considered corrupted in +** two circumstances: +** +** * If the record page-number is illegal (0 or PAGER_MJ_PGNO), or +** * If the record is being rolled back from the main journal file +** and the checksum field does not match the record content. +** +** Neither of these two scenarios are possible during a savepoint rollback. +** +** If this is a savepoint rollback, then memory may have to be dynamically +** allocated by this function. If this is the case and an allocation fails, +** SQLITE_NOMEM is returned. +*/ +static int pager_playback_one_page( + Pager *pPager, /* The pager being played back */ + i64 *pOffset, /* Offset of record to playback */ + Bitvec *pDone, /* Bitvec of pages already played back */ + int isMainJrnl, /* 1 -> main journal. 0 -> sub-journal. */ + int isSavepnt /* True for a savepoint rollback */ +){ + int rc; + PgHdr *pPg; /* An existing page in the cache */ + Pgno pgno; /* The page number of a page in journal */ + u32 cksum; /* Checksum used for sanity checking */ + char *aData; /* Temporary storage for the page */ + sqlite3_file *jfd; /* The file descriptor for the journal file */ + int isSynced; /* True if journal page is synced */ + + assert( (isMainJrnl&~1)==0 ); /* isMainJrnl is 0 or 1 */ + assert( (isSavepnt&~1)==0 ); /* isSavepnt is 0 or 1 */ + assert( isMainJrnl || pDone ); /* pDone always used on sub-journals */ + assert( isSavepnt || pDone==0 ); /* pDone never used on non-savepoint */ + + aData = pPager->pTmpSpace; + assert( aData ); /* Temp storage must have already been allocated */ + assert( pagerUseWal(pPager)==0 || (!isMainJrnl && isSavepnt) ); + + /* Either the state is greater than PAGER_WRITER_CACHEMOD (a transaction + ** or savepoint rollback done at the request of the caller) or this is + ** a hot-journal rollback. If it is a hot-journal rollback, the pager + ** is in state OPEN and holds an EXCLUSIVE lock. Hot-journal rollback + ** only reads from the main journal, not the sub-journal. + */ + assert( pPager->eState>=PAGER_WRITER_CACHEMOD + || (pPager->eState==PAGER_OPEN && pPager->eLock==EXCLUSIVE_LOCK) + ); + assert( pPager->eState>=PAGER_WRITER_CACHEMOD || isMainJrnl ); + + /* Read the page number and page data from the journal or sub-journal + ** file. Return an error code to the caller if an IO error occurs. + */ + jfd = isMainJrnl ? pPager->jfd : pPager->sjfd; + rc = read32bits(jfd, *pOffset, &pgno); + if( rc!=SQLITE_OK ) return rc; + rc = sqlite3OsRead(jfd, (u8*)aData, pPager->pageSize, (*pOffset)+4); + if( rc!=SQLITE_OK ) return rc; + *pOffset += pPager->pageSize + 4 + isMainJrnl*4; + + /* Sanity checking on the page. This is more important that I originally + ** thought. If a power failure occurs while the journal is being written, + ** it could cause invalid data to be written into the journal. We need to + ** detect this invalid data (with high probability) and ignore it. + */ + if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){ + assert( !isSavepnt ); + return SQLITE_DONE; + } + if( pgno>(Pgno)pPager->dbSize || sqlite3BitvecTest(pDone, pgno) ){ + return SQLITE_OK; + } + if( isMainJrnl ){ + rc = read32bits(jfd, (*pOffset)-4, &cksum); + if( rc ) return rc; + if( !isSavepnt && pager_cksum(pPager, (u8*)aData)!=cksum ){ + return SQLITE_DONE; + } + } + + /* If this page has already been played back before during the current + ** rollback, then don't bother to play it back again. + */ + if( pDone && (rc = sqlite3BitvecSet(pDone, pgno))!=SQLITE_OK ){ + return rc; + } + + /* When playing back page 1, restore the nReserve setting + */ + if( pgno==1 && pPager->nReserve!=((u8*)aData)[20] ){ + pPager->nReserve = ((u8*)aData)[20]; + } + + /* If the pager is in CACHEMOD state, then there must be a copy of this + ** page in the pager cache. In this case just update the pager cache, + ** not the database file. The page is left marked dirty in this case. + ** + ** An exception to the above rule: If the database is in no-sync mode + ** and a page is moved during an incremental vacuum then the page may + ** not be in the pager cache. Later: if a malloc() or IO error occurs + ** during a Movepage() call, then the page may not be in the cache + ** either. So the condition described in the above paragraph is not + ** assert()able. + ** + ** If in WRITER_DBMOD, WRITER_FINISHED or OPEN state, then we update the + ** pager cache if it exists and the main file. The page is then marked + ** not dirty. Since this code is only executed in PAGER_OPEN state for + ** a hot-journal rollback, it is guaranteed that the page-cache is empty + ** if the pager is in OPEN state. + ** + ** Ticket #1171: The statement journal might contain page content that is + ** different from the page content at the start of the transaction. + ** This occurs when a page is changed prior to the start of a statement + ** then changed again within the statement. When rolling back such a + ** statement we must not write to the original database unless we know + ** for certain that original page contents are synced into the main rollback + ** journal. Otherwise, a power loss might leave modified data in the + ** database file without an entry in the rollback journal that can + ** restore the database to its original form. Two conditions must be + ** met before writing to the database files. (1) the database must be + ** locked. (2) we know that the original page content is fully synced + ** in the main journal either because the page is not in cache or else + ** the page is marked as needSync==0. + ** + ** 2008-04-14: When attempting to vacuum a corrupt database file, it + ** is possible to fail a statement on a database that does not yet exist. + ** Do not attempt to write if database file has never been opened. + */ + if( pagerUseWal(pPager) ){ + pPg = 0; + }else{ + pPg = sqlite3PagerLookup(pPager, pgno); + } + assert( pPg || !MEMDB ); + assert( pPager->eState!=PAGER_OPEN || pPg==0 || pPager->tempFile ); + PAGERTRACE(("PLAYBACK %d page %d hash(%08x) %s\n", + PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, (u8*)aData), + (isMainJrnl?"main-journal":"sub-journal") + )); + if( isMainJrnl ){ + isSynced = pPager->noSync || (*pOffset <= pPager->journalHdr); + }else{ + isSynced = (pPg==0 || 0==(pPg->flags & PGHDR_NEED_SYNC)); + } + if( isOpen(pPager->fd) + && (pPager->eState>=PAGER_WRITER_DBMOD || pPager->eState==PAGER_OPEN) + && isSynced + ){ + i64 ofst = (pgno-1)*(i64)pPager->pageSize; + testcase( !isSavepnt && pPg!=0 && (pPg->flags&PGHDR_NEED_SYNC)!=0 ); + assert( !pagerUseWal(pPager) ); + + /* Write the data read from the journal back into the database file. + ** This is usually safe even for an encrypted database - as the data + ** was encrypted before it was written to the journal file. The exception + ** is if the data was just read from an in-memory sub-journal. In that + ** case it must be encrypted here before it is copied into the database + ** file. */ + rc = sqlite3OsWrite(pPager->fd, (u8 *)aData, pPager->pageSize, ofst); + + if( pgno>pPager->dbFileSize ){ + pPager->dbFileSize = pgno; + } + if( pPager->pBackup ){ + sqlite3BackupUpdate(pPager->pBackup, pgno, (u8*)aData); + } + }else if( !isMainJrnl && pPg==0 ){ + /* If this is a rollback of a savepoint and data was not written to + ** the database and the page is not in-memory, there is a potential + ** problem. When the page is next fetched by the b-tree layer, it + ** will be read from the database file, which may or may not be + ** current. + ** + ** There are a couple of different ways this can happen. All are quite + ** obscure. When running in synchronous mode, this can only happen + ** if the page is on the free-list at the start of the transaction, then + ** populated, then moved using sqlite3PagerMovepage(). + ** + ** The solution is to add an in-memory page to the cache containing + ** the data just read from the sub-journal. Mark the page as dirty + ** and if the pager requires a journal-sync, then mark the page as + ** requiring a journal-sync before it is written. + */ + assert( isSavepnt ); + assert( (pPager->doNotSpill & SPILLFLAG_ROLLBACK)==0 ); + pPager->doNotSpill |= SPILLFLAG_ROLLBACK; + rc = sqlite3PagerGet(pPager, pgno, &pPg, 1); + assert( (pPager->doNotSpill & SPILLFLAG_ROLLBACK)!=0 ); + pPager->doNotSpill &= ~SPILLFLAG_ROLLBACK; + if( rc!=SQLITE_OK ) return rc; + sqlite3PcacheMakeDirty(pPg); + } + if( pPg ){ + /* No page should ever be explicitly rolled back that is in use, except + ** for page 1 which is held in use in order to keep the lock on the + ** database active. However such a page may be rolled back as a result + ** of an internal error resulting in an automatic call to + ** sqlite3PagerRollback(). + */ + void *pData; + pData = pPg->pData; + memcpy(pData, (u8*)aData, pPager->pageSize); + pPager->xReiniter(pPg); + /* It used to be that sqlite3PcacheMakeClean(pPg) was called here. But + ** that call was dangerous and had no detectable benefit since the cache + ** is normally cleaned by sqlite3PcacheCleanAll() after rollback and so + ** has been removed. */ + pager_set_pagehash(pPg); + + /* If this was page 1, then restore the value of Pager.dbFileVers. + ** Do this before any decoding. */ + if( pgno==1 ){ + memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers)); + } + sqlite3PcacheRelease(pPg); + } + return rc; +} + +/* +** Parameter zSuper is the name of a super-journal file. A single journal +** file that referred to the super-journal file has just been rolled back. +** This routine checks if it is possible to delete the super-journal file, +** and does so if it is. +** +** Argument zSuper may point to Pager.pTmpSpace. So that buffer is not +** available for use within this function. +** +** When a super-journal file is created, it is populated with the names +** of all of its child journals, one after another, formatted as utf-8 +** encoded text. The end of each child journal file is marked with a +** nul-terminator byte (0x00). i.e. the entire contents of a super-journal +** file for a transaction involving two databases might be: +** +** "/home/bill/a.db-journal\x00/home/bill/b.db-journal\x00" +** +** A super-journal file may only be deleted once all of its child +** journals have been rolled back. +** +** This function reads the contents of the super-journal file into +** memory and loops through each of the child journal names. For +** each child journal, it checks if: +** +** * if the child journal exists, and if so +** * if the child journal contains a reference to super-journal +** file zSuper +** +** If a child journal can be found that matches both of the criteria +** above, this function returns without doing anything. Otherwise, if +** no such child journal can be found, file zSuper is deleted from +** the file-system using sqlite3OsDelete(). +** +** If an IO error within this function, an error code is returned. This +** function allocates memory by calling sqlite3Malloc(). If an allocation +** fails, SQLITE_NOMEM is returned. Otherwise, if no IO or malloc errors +** occur, SQLITE_OK is returned. +** +** TODO: This function allocates a single block of memory to load +** the entire contents of the super-journal file. This could be +** a couple of kilobytes or so - potentially larger than the page +** size. +*/ +static int pager_delsuper(Pager *pPager, const char *zSuper){ + sqlite3_vfs *pVfs = pPager->pVfs; + int rc; /* Return code */ + sqlite3_file *pSuper; /* Malloc'd super-journal file descriptor */ + sqlite3_file *pJournal; /* Malloc'd child-journal file descriptor */ + char *zSuperJournal = 0; /* Contents of super-journal file */ + i64 nSuperJournal; /* Size of super-journal file */ + char *zJournal; /* Pointer to one journal within MJ file */ + char *zSuperPtr; /* Space to hold super-journal filename */ + char *zFree = 0; /* Free this buffer */ + int nSuperPtr; /* Amount of space allocated to zSuperPtr[] */ + + /* Allocate space for both the pJournal and pSuper file descriptors. + ** If successful, open the super-journal file for reading. + */ + pSuper = (sqlite3_file *)sqlite3MallocZero(pVfs->szOsFile * 2); + if( !pSuper ){ + rc = SQLITE_NOMEM_BKPT; + pJournal = 0; + }else{ + const int flags = (SQLITE_OPEN_READONLY|SQLITE_OPEN_SUPER_JOURNAL); + rc = sqlite3OsOpen(pVfs, zSuper, pSuper, flags, 0); + pJournal = (sqlite3_file *)(((u8 *)pSuper) + pVfs->szOsFile); + } + if( rc!=SQLITE_OK ) goto delsuper_out; + + /* Load the entire super-journal file into space obtained from + ** sqlite3_malloc() and pointed to by zSuperJournal. Also obtain + ** sufficient space (in zSuperPtr) to hold the names of super-journal + ** files extracted from regular rollback-journals. + */ + rc = sqlite3OsFileSize(pSuper, &nSuperJournal); + if( rc!=SQLITE_OK ) goto delsuper_out; + nSuperPtr = pVfs->mxPathname+1; + zFree = sqlite3Malloc(4 + nSuperJournal + nSuperPtr + 2); + if( !zFree ){ + rc = SQLITE_NOMEM_BKPT; + goto delsuper_out; + } + zFree[0] = zFree[1] = zFree[2] = zFree[3] = 0; + zSuperJournal = &zFree[4]; + zSuperPtr = &zSuperJournal[nSuperJournal+2]; + rc = sqlite3OsRead(pSuper, zSuperJournal, (int)nSuperJournal, 0); + if( rc!=SQLITE_OK ) goto delsuper_out; + zSuperJournal[nSuperJournal] = 0; + zSuperJournal[nSuperJournal+1] = 0; + + zJournal = zSuperJournal; + while( (zJournal-zSuperJournal)pageSize bytes). +** If the file on disk is currently larger than nPage pages, then use the VFS +** xTruncate() method to truncate it. +** +** Or, it might be the case that the file on disk is smaller than +** nPage pages. Some operating system implementations can get confused if +** you try to truncate a file to some size that is larger than it +** currently is, so detect this case and write a single zero byte to +** the end of the new file instead. +** +** If successful, return SQLITE_OK. If an IO error occurs while modifying +** the database file, return the error code to the caller. +*/ +static int pager_truncate(Pager *pPager, Pgno nPage){ + int rc = SQLITE_OK; + assert( pPager->eState!=PAGER_ERROR ); + assert( pPager->eState!=PAGER_READER ); + + if( isOpen(pPager->fd) + && (pPager->eState>=PAGER_WRITER_DBMOD || pPager->eState==PAGER_OPEN) + ){ + i64 currentSize, newSize; + int szPage = pPager->pageSize; + assert( pPager->eLock==EXCLUSIVE_LOCK ); + /* TODO: Is it safe to use Pager.dbFileSize here? */ + rc = sqlite3OsFileSize(pPager->fd, ¤tSize); + newSize = szPage*(i64)nPage; + if( rc==SQLITE_OK && currentSize!=newSize ){ + if( currentSize>newSize ){ + rc = sqlite3OsTruncate(pPager->fd, newSize); + }else if( (currentSize+szPage)<=newSize ){ + char *pTmp = pPager->pTmpSpace; + memset(pTmp, 0, szPage); + testcase( (newSize-szPage) == currentSize ); + testcase( (newSize-szPage) > currentSize ); + rc = sqlite3OsWrite(pPager->fd, pTmp, szPage, newSize-szPage); + } + if( rc==SQLITE_OK ){ + pPager->dbFileSize = nPage; + } + } + } + return rc; +} + +/* +** Return a sanitized version of the sector-size of OS file pFile. The +** return value is guaranteed to lie between 32 and MAX_SECTOR_SIZE. +*/ +int sqlite3SectorSize(sqlite3_file *pFile){ + int iRet = sqlite3OsSectorSize(pFile); + if( iRet<32 ){ + iRet = 512; + }else if( iRet>MAX_SECTOR_SIZE ){ + assert( MAX_SECTOR_SIZE>=512 ); + iRet = MAX_SECTOR_SIZE; + } + return iRet; +} + +/* +** Set the value of the Pager.sectorSize variable for the given +** pager based on the value returned by the xSectorSize method +** of the open database file. The sector size will be used +** to determine the size and alignment of journal header and +** super-journal pointers within created journal files. +** +** For temporary files the effective sector size is always 512 bytes. +** +** Otherwise, for non-temporary files, the effective sector size is +** the value returned by the xSectorSize() method rounded up to 32 if +** it is less than 32, or rounded down to MAX_SECTOR_SIZE if it +** is greater than MAX_SECTOR_SIZE. +** +** If the file has the SQLITE_IOCAP_POWERSAFE_OVERWRITE property, then set +** the effective sector size to its minimum value (512). The purpose of +** pPager->sectorSize is to define the "blast radius" of bytes that +** might change if a crash occurs while writing to a single byte in +** that range. But with POWERSAFE_OVERWRITE, the blast radius is zero +** (that is what POWERSAFE_OVERWRITE means), so we minimize the sector +** size. For backwards compatibility of the rollback journal file format, +** we cannot reduce the effective sector size below 512. +*/ +static void setSectorSize(Pager *pPager){ + assert( isOpen(pPager->fd) || pPager->tempFile ); + + if( pPager->tempFile + || (sqlite3OsDeviceCharacteristics(pPager->fd) & + SQLITE_IOCAP_POWERSAFE_OVERWRITE)!=0 + ){ + /* Sector size doesn't matter for temporary files. Also, the file + ** may not have been opened yet, in which case the OsSectorSize() + ** call will segfault. */ + pPager->sectorSize = 512; + }else{ + pPager->sectorSize = sqlite3SectorSize(pPager->fd); + } +} + +/* +** Playback the journal and thus restore the database file to +** the state it was in before we started making changes. +** +** The journal file format is as follows: +** +** (1) 8 byte prefix. A copy of aJournalMagic[]. +** (2) 4 byte big-endian integer which is the number of valid page records +** in the journal. If this value is 0xffffffff, then compute the +** number of page records from the journal size. +** (3) 4 byte big-endian integer which is the initial value for the +** sanity checksum. +** (4) 4 byte integer which is the number of pages to truncate the +** database to during a rollback. +** (5) 4 byte big-endian integer which is the sector size. The header +** is this many bytes in size. +** (6) 4 byte big-endian integer which is the page size. +** (7) zero padding out to the next sector size. +** (8) Zero or more pages instances, each as follows: +** + 4 byte page number. +** + pPager->pageSize bytes of data. +** + 4 byte checksum +** +** When we speak of the journal header, we mean the first 7 items above. +** Each entry in the journal is an instance of the 8th item. +** +** Call the value from the second bullet "nRec". nRec is the number of +** valid page entries in the journal. In most cases, you can compute the +** value of nRec from the size of the journal file. But if a power +** failure occurred while the journal was being written, it could be the +** case that the size of the journal file had already been increased but +** the extra entries had not yet made it safely to disk. In such a case, +** the value of nRec computed from the file size would be too large. For +** that reason, we always use the nRec value in the header. +** +** If the nRec value is 0xffffffff it means that nRec should be computed +** from the file size. This value is used when the user selects the +** no-sync option for the journal. A power failure could lead to corruption +** in this case. But for things like temporary table (which will be +** deleted when the power is restored) we don't care. +** +** If the file opened as the journal file is not a well-formed +** journal file then all pages up to the first corrupted page are rolled +** back (or no pages if the journal header is corrupted). The journal file +** is then deleted and SQLITE_OK returned, just as if no corruption had +** been encountered. +** +** If an I/O or malloc() error occurs, the journal-file is not deleted +** and an error code is returned. +** +** The isHot parameter indicates that we are trying to rollback a journal +** that might be a hot journal. Or, it could be that the journal is +** preserved because of JOURNALMODE_PERSIST or JOURNALMODE_TRUNCATE. +** If the journal really is hot, reset the pager cache prior rolling +** back any content. If the journal is merely persistent, no reset is +** needed. +*/ +static int pager_playback(Pager *pPager, int isHot){ + sqlite3_vfs *pVfs = pPager->pVfs; + i64 szJ; /* Size of the journal file in bytes */ + u32 nRec; /* Number of Records in the journal */ + u32 u; /* Unsigned loop counter */ + Pgno mxPg = 0; /* Size of the original file in pages */ + int rc; /* Result code of a subroutine */ + int res = 1; /* Value returned by sqlite3OsAccess() */ + char *zSuper = 0; /* Name of super-journal file if any */ + int needPagerReset; /* True to reset page prior to first page rollback */ + int nPlayback = 0; /* Total number of pages restored from journal */ + u32 savedPageSize = pPager->pageSize; + + /* Figure out how many records are in the journal. Abort early if + ** the journal is empty. + */ + assert( isOpen(pPager->jfd) ); + rc = sqlite3OsFileSize(pPager->jfd, &szJ); + if( rc!=SQLITE_OK ){ + goto end_playback; + } + + /* Read the super-journal name from the journal, if it is present. + ** If a super-journal file name is specified, but the file is not + ** present on disk, then the journal is not hot and does not need to be + ** played back. + ** + ** TODO: Technically the following is an error because it assumes that + ** buffer Pager.pTmpSpace is (mxPathname+1) bytes or larger. i.e. that + ** (pPager->pageSize >= pPager->pVfs->mxPathname+1). Using os_unix.c, + ** mxPathname is 512, which is the same as the minimum allowable value + ** for pageSize. + */ + zSuper = pPager->pTmpSpace; + rc = readSuperJournal(pPager->jfd, zSuper, pPager->pVfs->mxPathname+1); + if( rc==SQLITE_OK && zSuper[0] ){ + rc = sqlite3OsAccess(pVfs, zSuper, SQLITE_ACCESS_EXISTS, &res); + } + zSuper = 0; + if( rc!=SQLITE_OK || !res ){ + goto end_playback; + } + pPager->journalOff = 0; + needPagerReset = isHot; + + /* This loop terminates either when a readJournalHdr() or + ** pager_playback_one_page() call returns SQLITE_DONE or an IO error + ** occurs. + */ + while( 1 ){ + /* Read the next journal header from the journal file. If there are + ** not enough bytes left in the journal file for a complete header, or + ** it is corrupted, then a process must have failed while writing it. + ** This indicates nothing more needs to be rolled back. + */ + rc = readJournalHdr(pPager, isHot, szJ, &nRec, &mxPg); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_DONE ){ + rc = SQLITE_OK; + } + goto end_playback; + } + + /* If nRec is 0xffffffff, then this journal was created by a process + ** working in no-sync mode. This means that the rest of the journal + ** file consists of pages, there are no more journal headers. Compute + ** the value of nRec based on this assumption. + */ + if( nRec==0xffffffff ){ + assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ); + nRec = (int)((szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager)); + } + + /* If nRec is 0 and this rollback is of a transaction created by this + ** process and if this is the final header in the journal, then it means + ** that this part of the journal was being filled but has not yet been + ** synced to disk. Compute the number of pages based on the remaining + ** size of the file. + ** + ** The third term of the test was added to fix ticket #2565. + ** When rolling back a hot journal, nRec==0 always means that the next + ** chunk of the journal contains zero pages to be rolled back. But + ** when doing a ROLLBACK and the nRec==0 chunk is the last chunk in + ** the journal, it means that the journal might contain additional + ** pages that need to be rolled back and that the number of pages + ** should be computed based on the journal file size. + */ + if( nRec==0 && !isHot && + pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff ){ + nRec = (int)((szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager)); + } + + /* If this is the first header read from the journal, truncate the + ** database file back to its original size. + */ + if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){ + rc = pager_truncate(pPager, mxPg); + if( rc!=SQLITE_OK ){ + goto end_playback; + } + pPager->dbSize = mxPg; + } + + /* Copy original pages out of the journal and back into the + ** database file and/or page cache. + */ + for(u=0; ujournalOff,0,1,0); + if( rc==SQLITE_OK ){ + nPlayback++; + }else{ + if( rc==SQLITE_DONE ){ + pPager->journalOff = szJ; + break; + }else if( rc==SQLITE_IOERR_SHORT_READ ){ + /* If the journal has been truncated, simply stop reading and + ** processing the journal. This might happen if the journal was + ** not completely written and synced prior to a crash. In that + ** case, the database should have never been written in the + ** first place so it is OK to simply abandon the rollback. */ + rc = SQLITE_OK; + goto end_playback; + }else{ + /* If we are unable to rollback, quit and return the error + ** code. This will cause the pager to enter the error state + ** so that no further harm will be done. Perhaps the next + ** process to come along will be able to rollback the database. + */ + goto end_playback; + } + } + } + } + /*NOTREACHED*/ + assert( 0 ); + +end_playback: + if( rc==SQLITE_OK ){ + rc = sqlite3PagerSetPagesize(pPager, &savedPageSize, -1); + } + /* Following a rollback, the database file should be back in its original + ** state prior to the start of the transaction, so invoke the + ** SQLITE_FCNTL_DB_UNCHANGED file-control method to disable the + ** assertion that the transaction counter was modified. + */ +#ifdef SQLITE_DEBUG + sqlite3OsFileControlHint(pPager->fd,SQLITE_FCNTL_DB_UNCHANGED,0); +#endif + + /* If this playback is happening automatically as a result of an IO or + ** malloc error that occurred after the change-counter was updated but + ** before the transaction was committed, then the change-counter + ** modification may just have been reverted. If this happens in exclusive + ** mode, then subsequent transactions performed by the connection will not + ** update the change-counter at all. This may lead to cache inconsistency + ** problems for other processes at some point in the future. So, just + ** in case this has happened, clear the changeCountDone flag now. + */ + pPager->changeCountDone = pPager->tempFile; + + if( rc==SQLITE_OK ){ + /* Leave 4 bytes of space before the super-journal filename in memory. + ** This is because it may end up being passed to sqlite3OsOpen(), in + ** which case it requires 4 0x00 bytes in memory immediately before + ** the filename. */ + zSuper = &pPager->pTmpSpace[4]; + rc = readSuperJournal(pPager->jfd, zSuper, pPager->pVfs->mxPathname+1); + testcase( rc!=SQLITE_OK ); + } + if( rc==SQLITE_OK + && (pPager->eState>=PAGER_WRITER_DBMOD || pPager->eState==PAGER_OPEN) + ){ + rc = sqlite3PagerSync(pPager, 0); + } + if( rc==SQLITE_OK ){ + rc = pager_end_transaction(pPager, zSuper[0]!='\0', 0); + testcase( rc!=SQLITE_OK ); + } + if( rc==SQLITE_OK && zSuper[0] && res ){ + /* If there was a super-journal and this routine will return success, + ** see if it is possible to delete the super-journal. + */ + assert( zSuper==&pPager->pTmpSpace[4] ); + memset(&zSuper[-4], 0, 4); + rc = pager_delsuper(pPager, zSuper); + testcase( rc!=SQLITE_OK ); + } + if( isHot && nPlayback ){ + sqlite3_log(SQLITE_NOTICE_RECOVER_ROLLBACK, "recovered %d pages from %s", + nPlayback, pPager->zJournal); + } + + /* The Pager.sectorSize variable may have been updated while rolling + ** back a journal created by a process with a different sector size + ** value. Reset it to the correct value for this process. + */ + setSectorSize(pPager); + return rc; +} + + +/* +** Read the content for page pPg out of the database file (or out of +** the WAL if that is where the most recent copy if found) into +** pPg->pData. A shared lock or greater must be held on the database +** file before this function is called. +** +** If page 1 is read, then the value of Pager.dbFileVers[] is set to +** the value read from the database file. +** +** If an IO error occurs, then the IO error is returned to the caller. +** Otherwise, SQLITE_OK is returned. +*/ +static int readDbPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; /* Pager object associated with page pPg */ + int rc = SQLITE_OK; /* Return code */ + +#ifndef SQLITE_OMIT_WAL + u32 iFrame = 0; /* Frame of WAL containing pgno */ + + assert( pPager->eState>=PAGER_READER && !MEMDB ); + assert( isOpen(pPager->fd) ); + + if( pagerUseWal(pPager) ){ + rc = sqlite3WalFindFrame(pPager->pWal, pPg->pgno, &iFrame); + if( rc ) return rc; + } + if( iFrame ){ + rc = sqlite3WalReadFrame(pPager->pWal, iFrame,pPager->pageSize,pPg->pData); + }else +#endif + { + i64 iOffset = (pPg->pgno-1)*(i64)pPager->pageSize; + rc = sqlite3OsRead(pPager->fd, pPg->pData, pPager->pageSize, iOffset); + if( rc==SQLITE_IOERR_SHORT_READ ){ + rc = SQLITE_OK; + } + } + + if( pPg->pgno==1 ){ + if( rc ){ + /* If the read is unsuccessful, set the dbFileVers[] to something + ** that will never be a valid file version. dbFileVers[] is a copy + ** of bytes 24..39 of the database. Bytes 28..31 should always be + ** zero or the size of the database in page. Bytes 32..35 and 35..39 + ** should be page numbers which are never 0xffffffff. So filling + ** pPager->dbFileVers[] with all 0xff bytes should suffice. + ** + ** For an encrypted database, the situation is more complex: bytes + ** 24..39 of the database are white noise. But the probability of + ** white noise equaling 16 bytes of 0xff is vanishingly small so + ** we should still be ok. + */ + memset(pPager->dbFileVers, 0xff, sizeof(pPager->dbFileVers)); + }else{ + u8 *dbFileVers = &((u8*)pPg->pData)[24]; + memcpy(&pPager->dbFileVers, dbFileVers, sizeof(pPager->dbFileVers)); + } + } + PAGER_INCR(sqlite3_pager_readdb_count); + PAGER_INCR(pPager->nRead); + IOTRACE(("PGIN %p %d\n", pPager, pPg->pgno)); + PAGERTRACE(("FETCH %d page %d hash(%08x)\n", + PAGERID(pPager), pPg->pgno, pager_pagehash(pPg))); + + return rc; +} + +/* +** Update the value of the change-counter at offsets 24 and 92 in +** the header and the sqlite version number at offset 96. +** +** This is an unconditional update. See also the pager_incr_changecounter() +** routine which only updates the change-counter if the update is actually +** needed, as determined by the pPager->changeCountDone state variable. +*/ +static void pager_write_changecounter(PgHdr *pPg){ + u32 change_counter; + if( NEVER(pPg==0) ) return; + + /* Increment the value just read and write it back to byte 24. */ + change_counter = sqlite3Get4byte((u8*)pPg->pPager->dbFileVers)+1; + put32bits(((char*)pPg->pData)+24, change_counter); + + /* Also store the SQLite version number in bytes 96..99 and in + ** bytes 92..95 store the change counter for which the version number + ** is valid. */ + put32bits(((char*)pPg->pData)+92, change_counter); + put32bits(((char*)pPg->pData)+96, SQLITE_VERSION_NUMBER); +} + +#ifndef SQLITE_OMIT_WAL +/* +** This function is invoked once for each page that has already been +** written into the log file when a WAL transaction is rolled back. +** Parameter iPg is the page number of said page. The pCtx argument +** is actually a pointer to the Pager structure. +** +** If page iPg is present in the cache, and has no outstanding references, +** it is discarded. Otherwise, if there are one or more outstanding +** references, the page content is reloaded from the database. If the +** attempt to reload content from the database is required and fails, +** return an SQLite error code. Otherwise, SQLITE_OK. +*/ +static int pagerUndoCallback(void *pCtx, Pgno iPg){ + int rc = SQLITE_OK; + Pager *pPager = (Pager *)pCtx; + PgHdr *pPg; + + assert( pagerUseWal(pPager) ); + pPg = sqlite3PagerLookup(pPager, iPg); + if( pPg ){ + if( sqlite3PcachePageRefcount(pPg)==1 ){ + sqlite3PcacheDrop(pPg); + }else{ + rc = readDbPage(pPg); + if( rc==SQLITE_OK ){ + pPager->xReiniter(pPg); + } + sqlite3PagerUnrefNotNull(pPg); + } + } + + /* Normally, if a transaction is rolled back, any backup processes are + ** updated as data is copied out of the rollback journal and into the + ** database. This is not generally possible with a WAL database, as + ** rollback involves simply truncating the log file. Therefore, if one + ** or more frames have already been written to the log (and therefore + ** also copied into the backup databases) as part of this transaction, + ** the backups must be restarted. + */ + sqlite3BackupRestart(pPager->pBackup); + + return rc; +} + +/* +** This function is called to rollback a transaction on a WAL database. +*/ +static int pagerRollbackWal(Pager *pPager){ + int rc; /* Return Code */ + PgHdr *pList; /* List of dirty pages to revert */ + + /* For all pages in the cache that are currently dirty or have already + ** been written (but not committed) to the log file, do one of the + ** following: + ** + ** + Discard the cached page (if refcount==0), or + ** + Reload page content from the database (if refcount>0). + */ + pPager->dbSize = pPager->dbOrigSize; + rc = sqlite3WalUndo(pPager->pWal, pagerUndoCallback, (void *)pPager); + pList = sqlite3PcacheDirtyList(pPager->pPCache); + while( pList && rc==SQLITE_OK ){ + PgHdr *pNext = pList->pDirty; + rc = pagerUndoCallback((void *)pPager, pList->pgno); + pList = pNext; + } + + return rc; +} + +/* +** This function is a wrapper around sqlite3WalFrames(). As well as logging +** the contents of the list of pages headed by pList (connected by pDirty), +** this function notifies any active backup processes that the pages have +** changed. +** +** The list of pages passed into this routine is always sorted by page number. +** Hence, if page 1 appears anywhere on the list, it will be the first page. +*/ +static int pagerWalFrames( + Pager *pPager, /* Pager object */ + PgHdr *pList, /* List of frames to log */ + Pgno nTruncate, /* Database size after this commit */ + int isCommit /* True if this is a commit */ +){ + int rc; /* Return code */ + int nList; /* Number of pages in pList */ + PgHdr *p; /* For looping over pages */ + + assert( pPager->pWal ); + assert( pList ); +#ifdef SQLITE_DEBUG + /* Verify that the page list is in accending order */ + for(p=pList; p && p->pDirty; p=p->pDirty){ + assert( p->pgno < p->pDirty->pgno ); + } +#endif + + assert( pList->pDirty==0 || isCommit ); + if( isCommit ){ + /* If a WAL transaction is being committed, there is no point in writing + ** any pages with page numbers greater than nTruncate into the WAL file. + ** They will never be read by any client. So remove them from the pDirty + ** list here. */ + PgHdr **ppNext = &pList; + nList = 0; + for(p=pList; (*ppNext = p)!=0; p=p->pDirty){ + if( p->pgno<=nTruncate ){ + ppNext = &p->pDirty; + nList++; + } + } + assert( pList ); + }else{ + nList = 1; + } + pPager->aStat[PAGER_STAT_WRITE] += nList; + + if( pList->pgno==1 ) pager_write_changecounter(pList); + rc = sqlite3WalFrames(pPager->pWal, + pPager->pageSize, pList, nTruncate, isCommit, pPager->walSyncFlags + ); + if( rc==SQLITE_OK && pPager->pBackup ){ + for(p=pList; p; p=p->pDirty){ + sqlite3BackupUpdate(pPager->pBackup, p->pgno, (u8 *)p->pData); + } + } + +#ifdef SQLITE_CHECK_PAGES + pList = sqlite3PcacheDirtyList(pPager->pPCache); + for(p=pList; p; p=p->pDirty){ + pager_set_pagehash(p); + } +#endif + + return rc; +} + +/* +** Begin a read transaction on the WAL. +** +** This routine used to be called "pagerOpenSnapshot()" because it essentially +** makes a snapshot of the database at the current point in time and preserves +** that snapshot for use by the reader in spite of concurrently changes by +** other writers or checkpointers. +*/ +static int pagerBeginReadTransaction(Pager *pPager){ + int rc; /* Return code */ + int changed = 0; /* True if cache must be reset */ + + assert( pagerUseWal(pPager) ); + assert( pPager->eState==PAGER_OPEN || pPager->eState==PAGER_READER ); + + /* sqlite3WalEndReadTransaction() was not called for the previous + ** transaction in locking_mode=EXCLUSIVE. So call it now. If we + ** are in locking_mode=NORMAL and EndRead() was previously called, + ** the duplicate call is harmless. + */ + sqlite3WalEndReadTransaction(pPager->pWal); + + rc = sqlite3WalBeginReadTransaction(pPager->pWal, &changed); + if( rc!=SQLITE_OK || changed ){ + pager_reset(pPager); + if( USEFETCH(pPager) ) sqlite3OsUnfetch(pPager->fd, 0, 0); + } + + return rc; +} +#endif + +/* +** This function is called as part of the transition from PAGER_OPEN +** to PAGER_READER state to determine the size of the database file +** in pages (assuming the page size currently stored in Pager.pageSize). +** +** If no error occurs, SQLITE_OK is returned and the size of the database +** in pages is stored in *pnPage. Otherwise, an error code (perhaps +** SQLITE_IOERR_FSTAT) is returned and *pnPage is left unmodified. +*/ +static int pagerPagecount(Pager *pPager, Pgno *pnPage){ + Pgno nPage; /* Value to return via *pnPage */ + + /* Query the WAL sub-system for the database size. The WalDbsize() + ** function returns zero if the WAL is not open (i.e. Pager.pWal==0), or + ** if the database size is not available. The database size is not + ** available from the WAL sub-system if the log file is empty or + ** contains no valid committed transactions. + */ + assert( pPager->eState==PAGER_OPEN ); + assert( pPager->eLock>=SHARED_LOCK ); + assert( isOpen(pPager->fd) ); + assert( pPager->tempFile==0 ); + nPage = sqlite3WalDbsize(pPager->pWal); + + /* If the number of pages in the database is not available from the + ** WAL sub-system, determine the page count based on the size of + ** the database file. If the size of the database file is not an + ** integer multiple of the page-size, round up the result. + */ + if( nPage==0 && ALWAYS(isOpen(pPager->fd)) ){ + i64 n = 0; /* Size of db file in bytes */ + int rc = sqlite3OsFileSize(pPager->fd, &n); + if( rc!=SQLITE_OK ){ + return rc; + } + nPage = (Pgno)((n+pPager->pageSize-1) / pPager->pageSize); + } + + /* If the current number of pages in the file is greater than the + ** configured maximum pager number, increase the allowed limit so + ** that the file can be read. + */ + if( nPage>pPager->mxPgno ){ + pPager->mxPgno = (Pgno)nPage; + } + + *pnPage = nPage; + return SQLITE_OK; +} + +#ifndef SQLITE_OMIT_WAL +/* +** Check if the *-wal file that corresponds to the database opened by pPager +** exists if the database is not empy, or verify that the *-wal file does +** not exist (by deleting it) if the database file is empty. +** +** If the database is not empty and the *-wal file exists, open the pager +** in WAL mode. If the database is empty or if no *-wal file exists and +** if no error occurs, make sure Pager.journalMode is not set to +** PAGER_JOURNALMODE_WAL. +** +** Return SQLITE_OK or an error code. +** +** The caller must hold a SHARED lock on the database file to call this +** function. Because an EXCLUSIVE lock on the db file is required to delete +** a WAL on a none-empty database, this ensures there is no race condition +** between the xAccess() below and an xDelete() being executed by some +** other connection. +*/ +static int pagerOpenWalIfPresent(Pager *pPager){ + int rc = SQLITE_OK; + assert( pPager->eState==PAGER_OPEN ); + assert( pPager->eLock>=SHARED_LOCK ); + + if( !pPager->tempFile ){ + int isWal; /* True if WAL file exists */ + rc = sqlite3OsAccess( + pPager->pVfs, pPager->zWal, SQLITE_ACCESS_EXISTS, &isWal + ); + if( rc==SQLITE_OK ){ + if( isWal ){ + Pgno nPage; /* Size of the database file */ + + rc = pagerPagecount(pPager, &nPage); + if( rc ) return rc; + if( nPage==0 ){ + rc = sqlite3OsDelete(pPager->pVfs, pPager->zWal, 0); + }else{ + testcase( sqlite3PcachePagecount(pPager->pPCache)==0 ); + rc = sqlite3PagerOpenWal(pPager, 0); + } + }else if( pPager->journalMode==PAGER_JOURNALMODE_WAL ){ + pPager->journalMode = PAGER_JOURNALMODE_DELETE; + } + } + } + return rc; +} +#endif + +/* +** Playback savepoint pSavepoint. Or, if pSavepoint==NULL, then playback +** the entire super-journal file. The case pSavepoint==NULL occurs when +** a ROLLBACK TO command is invoked on a SAVEPOINT that is a transaction +** savepoint. +** +** When pSavepoint is not NULL (meaning a non-transaction savepoint is +** being rolled back), then the rollback consists of up to three stages, +** performed in the order specified: +** +** * Pages are played back from the main journal starting at byte +** offset PagerSavepoint.iOffset and continuing to +** PagerSavepoint.iHdrOffset, or to the end of the main journal +** file if PagerSavepoint.iHdrOffset is zero. +** +** * If PagerSavepoint.iHdrOffset is not zero, then pages are played +** back starting from the journal header immediately following +** PagerSavepoint.iHdrOffset to the end of the main journal file. +** +** * Pages are then played back from the sub-journal file, starting +** with the PagerSavepoint.iSubRec and continuing to the end of +** the journal file. +** +** Throughout the rollback process, each time a page is rolled back, the +** corresponding bit is set in a bitvec structure (variable pDone in the +** implementation below). This is used to ensure that a page is only +** rolled back the first time it is encountered in either journal. +** +** If pSavepoint is NULL, then pages are only played back from the main +** journal file. There is no need for a bitvec in this case. +** +** In either case, before playback commences the Pager.dbSize variable +** is reset to the value that it held at the start of the savepoint +** (or transaction). No page with a page-number greater than this value +** is played back. If one is encountered it is simply skipped. +*/ +static int pagerPlaybackSavepoint(Pager *pPager, PagerSavepoint *pSavepoint){ + i64 szJ; /* Effective size of the main journal */ + i64 iHdrOff; /* End of first segment of main-journal records */ + int rc = SQLITE_OK; /* Return code */ + Bitvec *pDone = 0; /* Bitvec to ensure pages played back only once */ + + assert( pPager->eState!=PAGER_ERROR ); + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + + /* Allocate a bitvec to use to store the set of pages rolled back */ + if( pSavepoint ){ + pDone = sqlite3BitvecCreate(pSavepoint->nOrig); + if( !pDone ){ + return SQLITE_NOMEM_BKPT; + } + } + + /* Set the database size back to the value it was before the savepoint + ** being reverted was opened. + */ + pPager->dbSize = pSavepoint ? pSavepoint->nOrig : pPager->dbOrigSize; + pPager->changeCountDone = pPager->tempFile; + + if( !pSavepoint && pagerUseWal(pPager) ){ + return pagerRollbackWal(pPager); + } + + /* Use pPager->journalOff as the effective size of the main rollback + ** journal. The actual file might be larger than this in + ** PAGER_JOURNALMODE_TRUNCATE or PAGER_JOURNALMODE_PERSIST. But anything + ** past pPager->journalOff is off-limits to us. + */ + szJ = pPager->journalOff; + assert( pagerUseWal(pPager)==0 || szJ==0 ); + + /* Begin by rolling back records from the main journal starting at + ** PagerSavepoint.iOffset and continuing to the next journal header. + ** There might be records in the main journal that have a page number + ** greater than the current database size (pPager->dbSize) but those + ** will be skipped automatically. Pages are added to pDone as they + ** are played back. + */ + if( pSavepoint && !pagerUseWal(pPager) ){ + iHdrOff = pSavepoint->iHdrOffset ? pSavepoint->iHdrOffset : szJ; + pPager->journalOff = pSavepoint->iOffset; + while( rc==SQLITE_OK && pPager->journalOffjournalOff, pDone, 1, 1); + } + assert( rc!=SQLITE_DONE ); + }else{ + pPager->journalOff = 0; + } + + /* Continue rolling back records out of the main journal starting at + ** the first journal header seen and continuing until the effective end + ** of the main journal file. Continue to skip out-of-range pages and + ** continue adding pages rolled back to pDone. + */ + while( rc==SQLITE_OK && pPager->journalOffjournalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff" + ** test is related to ticket #2565. See the discussion in the + ** pager_playback() function for additional information. + */ + if( nJRec==0 + && pPager->journalHdr+JOURNAL_HDR_SZ(pPager)==pPager->journalOff + ){ + nJRec = (u32)((szJ - pPager->journalOff)/JOURNAL_PG_SZ(pPager)); + } + for(ii=0; rc==SQLITE_OK && iijournalOffjournalOff, pDone, 1, 1); + } + assert( rc!=SQLITE_DONE ); + } + assert( rc!=SQLITE_OK || pPager->journalOff>=szJ ); + + /* Finally, rollback pages from the sub-journal. Page that were + ** previously rolled back out of the main journal (and are hence in pDone) + ** will be skipped. Out-of-range pages are also skipped. + */ + if( pSavepoint ){ + u32 ii; /* Loop counter */ + i64 offset = (i64)pSavepoint->iSubRec*(4+pPager->pageSize); + + if( pagerUseWal(pPager) ){ + rc = sqlite3WalSavepointUndo(pPager->pWal, pSavepoint->aWalData); + } + for(ii=pSavepoint->iSubRec; rc==SQLITE_OK && iinSubRec; ii++){ + assert( offset==(i64)ii*(4+pPager->pageSize) ); + rc = pager_playback_one_page(pPager, &offset, pDone, 0, 1); + } + assert( rc!=SQLITE_DONE ); + } + + sqlite3BitvecDestroy(pDone); + if( rc==SQLITE_OK ){ + pPager->journalOff = szJ; + } + + return rc; +} + +/* +** Change the maximum number of in-memory pages that are allowed +** before attempting to recycle clean and unused pages. +*/ +void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){ + sqlite3PcacheSetCachesize(pPager->pPCache, mxPage); +} + +/* +** Change the maximum number of in-memory pages that are allowed +** before attempting to spill pages to journal. +*/ +int sqlite3PagerSetSpillsize(Pager *pPager, int mxPage){ + return sqlite3PcacheSetSpillsize(pPager->pPCache, mxPage); +} + +/* +** Invoke SQLITE_FCNTL_MMAP_SIZE based on the current value of szMmap. +*/ +static void pagerFixMaplimit(Pager *pPager){ +#if SQLITE_MAX_MMAP_SIZE>0 + sqlite3_file *fd = pPager->fd; + if( isOpen(fd) && fd->pMethods->iVersion>=3 ){ + sqlite3_int64 sz; + sz = pPager->szMmap; + pPager->bUseFetch = (sz>0); + setGetterMethod(pPager); + sqlite3OsFileControlHint(pPager->fd, SQLITE_FCNTL_MMAP_SIZE, &sz); + } +#endif +} + +/* +** Change the maximum size of any memory mapping made of the database file. +*/ +void sqlite3PagerSetMmapLimit(Pager *pPager, sqlite3_int64 szMmap){ + pPager->szMmap = szMmap; + pagerFixMaplimit(pPager); +} + +/* +** Free as much memory as possible from the pager. +*/ +void sqlite3PagerShrink(Pager *pPager){ + sqlite3PcacheShrink(pPager->pPCache); +} + +/* +** Adjust settings of the pager to those specified in the pgFlags parameter. +** +** The "level" in pgFlags & PAGER_SYNCHRONOUS_MASK sets the robustness +** of the database to damage due to OS crashes or power failures by +** changing the number of syncs()s when writing the journals. +** There are four levels: +** +** OFF sqlite3OsSync() is never called. This is the default +** for temporary and transient files. +** +** NORMAL The journal is synced once before writes begin on the +** database. This is normally adequate protection, but +** it is theoretically possible, though very unlikely, +** that an inopertune power failure could leave the journal +** in a state which would cause damage to the database +** when it is rolled back. +** +** FULL The journal is synced twice before writes begin on the +** database (with some additional information - the nRec field +** of the journal header - being written in between the two +** syncs). If we assume that writing a +** single disk sector is atomic, then this mode provides +** assurance that the journal will not be corrupted to the +** point of causing damage to the database during rollback. +** +** EXTRA This is like FULL except that is also syncs the directory +** that contains the rollback journal after the rollback +** journal is unlinked. +** +** The above is for a rollback-journal mode. For WAL mode, OFF continues +** to mean that no syncs ever occur. NORMAL means that the WAL is synced +** prior to the start of checkpoint and that the database file is synced +** at the conclusion of the checkpoint if the entire content of the WAL +** was written back into the database. But no sync operations occur for +** an ordinary commit in NORMAL mode with WAL. FULL means that the WAL +** file is synced following each commit operation, in addition to the +** syncs associated with NORMAL. There is no difference between FULL +** and EXTRA for WAL mode. +** +** Do not confuse synchronous=FULL with SQLITE_SYNC_FULL. The +** SQLITE_SYNC_FULL macro means to use the MacOSX-style full-fsync +** using fcntl(F_FULLFSYNC). SQLITE_SYNC_NORMAL means to do an +** ordinary fsync() call. There is no difference between SQLITE_SYNC_FULL +** and SQLITE_SYNC_NORMAL on platforms other than MacOSX. But the +** synchronous=FULL versus synchronous=NORMAL setting determines when +** the xSync primitive is called and is relevant to all platforms. +** +** Numeric values associated with these states are OFF==1, NORMAL=2, +** and FULL=3. +*/ +#ifndef SQLITE_OMIT_PAGER_PRAGMAS +void sqlite3PagerSetFlags( + Pager *pPager, /* The pager to set safety level for */ + unsigned pgFlags /* Various flags */ +){ + unsigned level = pgFlags & PAGER_SYNCHRONOUS_MASK; + if( pPager->tempFile ){ + pPager->noSync = 1; + pPager->fullSync = 0; + pPager->extraSync = 0; + }else{ + pPager->noSync = level==PAGER_SYNCHRONOUS_OFF ?1:0; + pPager->fullSync = level>=PAGER_SYNCHRONOUS_FULL ?1:0; + pPager->extraSync = level==PAGER_SYNCHRONOUS_EXTRA ?1:0; + } + if( pPager->noSync ){ + pPager->syncFlags = 0; + }else if( pgFlags & PAGER_FULLFSYNC ){ + pPager->syncFlags = SQLITE_SYNC_FULL; + }else{ + pPager->syncFlags = SQLITE_SYNC_NORMAL; + } + pPager->walSyncFlags = (pPager->syncFlags<<2); + if( pPager->fullSync ){ + pPager->walSyncFlags |= pPager->syncFlags; + } + if( (pgFlags & PAGER_CKPT_FULLFSYNC) && !pPager->noSync ){ + pPager->walSyncFlags |= (SQLITE_SYNC_FULL<<2); + } + if( pgFlags & PAGER_CACHESPILL ){ + pPager->doNotSpill &= ~SPILLFLAG_OFF; + }else{ + pPager->doNotSpill |= SPILLFLAG_OFF; + } +} +#endif + +/* +** The following global variable is incremented whenever the library +** attempts to open a temporary file. This information is used for +** testing and analysis only. +*/ +#ifdef SQLITE_TEST +int sqlite3_opentemp_count = 0; +#endif + +/* +** Open a temporary file. +** +** Write the file descriptor into *pFile. Return SQLITE_OK on success +** or some other error code if we fail. The OS will automatically +** delete the temporary file when it is closed. +** +** The flags passed to the VFS layer xOpen() call are those specified +** by parameter vfsFlags ORed with the following: +** +** SQLITE_OPEN_READWRITE +** SQLITE_OPEN_CREATE +** SQLITE_OPEN_EXCLUSIVE +** SQLITE_OPEN_DELETEONCLOSE +*/ +static int pagerOpentemp( + Pager *pPager, /* The pager object */ + sqlite3_file *pFile, /* Write the file descriptor here */ + int vfsFlags /* Flags passed through to the VFS */ +){ + int rc; /* Return code */ + +#ifdef SQLITE_TEST + sqlite3_opentemp_count++; /* Used for testing and analysis only */ +#endif + + vfsFlags |= SQLITE_OPEN_READWRITE | SQLITE_OPEN_CREATE | + SQLITE_OPEN_EXCLUSIVE | SQLITE_OPEN_DELETEONCLOSE; + rc = sqlite3OsOpen(pPager->pVfs, 0, pFile, vfsFlags, 0); + assert( rc!=SQLITE_OK || isOpen(pFile) ); + return rc; +} + +/* +** Set the busy handler function. +** +** The pager invokes the busy-handler if sqlite3OsLock() returns +** SQLITE_BUSY when trying to upgrade from no-lock to a SHARED lock, +** or when trying to upgrade from a RESERVED lock to an EXCLUSIVE +** lock. It does *not* invoke the busy handler when upgrading from +** SHARED to RESERVED, or when upgrading from SHARED to EXCLUSIVE +** (which occurs during hot-journal rollback). Summary: +** +** Transition | Invokes xBusyHandler +** -------------------------------------------------------- +** NO_LOCK -> SHARED_LOCK | Yes +** SHARED_LOCK -> RESERVED_LOCK | No +** SHARED_LOCK -> EXCLUSIVE_LOCK | No +** RESERVED_LOCK -> EXCLUSIVE_LOCK | Yes +** +** If the busy-handler callback returns non-zero, the lock is +** retried. If it returns zero, then the SQLITE_BUSY error is +** returned to the caller of the pager API function. +*/ +void sqlite3PagerSetBusyHandler( + Pager *pPager, /* Pager object */ + int (*xBusyHandler)(void *), /* Pointer to busy-handler function */ + void *pBusyHandlerArg /* Argument to pass to xBusyHandler */ +){ + void **ap; + pPager->xBusyHandler = xBusyHandler; + pPager->pBusyHandlerArg = pBusyHandlerArg; + ap = (void **)&pPager->xBusyHandler; + assert( ((int(*)(void *))(ap[0]))==xBusyHandler ); + assert( ap[1]==pBusyHandlerArg ); + sqlite3OsFileControlHint(pPager->fd, SQLITE_FCNTL_BUSYHANDLER, (void *)ap); +} + +/* +** Change the page size used by the Pager object. The new page size +** is passed in *pPageSize. +** +** If the pager is in the error state when this function is called, it +** is a no-op. The value returned is the error state error code (i.e. +** one of SQLITE_IOERR, an SQLITE_IOERR_xxx sub-code or SQLITE_FULL). +** +** Otherwise, if all of the following are true: +** +** * the new page size (value of *pPageSize) is valid (a power +** of two between 512 and SQLITE_MAX_PAGE_SIZE, inclusive), and +** +** * there are no outstanding page references, and +** +** * the database is either not an in-memory database or it is +** an in-memory database that currently consists of zero pages. +** +** then the pager object page size is set to *pPageSize. +** +** If the page size is changed, then this function uses sqlite3PagerMalloc() +** to obtain a new Pager.pTmpSpace buffer. If this allocation attempt +** fails, SQLITE_NOMEM is returned and the page size remains unchanged. +** In all other cases, SQLITE_OK is returned. +** +** If the page size is not changed, either because one of the enumerated +** conditions above is not true, the pager was in error state when this +** function was called, or because the memory allocation attempt failed, +** then *pPageSize is set to the old, retained page size before returning. +*/ +int sqlite3PagerSetPagesize(Pager *pPager, u32 *pPageSize, int nReserve){ + int rc = SQLITE_OK; + + /* It is not possible to do a full assert_pager_state() here, as this + ** function may be called from within PagerOpen(), before the state + ** of the Pager object is internally consistent. + ** + ** At one point this function returned an error if the pager was in + ** PAGER_ERROR state. But since PAGER_ERROR state guarantees that + ** there is at least one outstanding page reference, this function + ** is a no-op for that case anyhow. + */ + + u32 pageSize = *pPageSize; + assert( pageSize==0 || (pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE) ); + if( (pPager->memDb==0 || pPager->dbSize==0) + && sqlite3PcacheRefCount(pPager->pPCache)==0 + && pageSize && pageSize!=(u32)pPager->pageSize + ){ + char *pNew = NULL; /* New temp space */ + i64 nByte = 0; + + if( pPager->eState>PAGER_OPEN && isOpen(pPager->fd) ){ + rc = sqlite3OsFileSize(pPager->fd, &nByte); + } + if( rc==SQLITE_OK ){ + /* 8 bytes of zeroed overrun space is sufficient so that the b-tree + * cell header parser will never run off the end of the allocation */ + pNew = (char *)sqlite3PageMalloc(pageSize+8); + if( !pNew ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + memset(pNew+pageSize, 0, 8); + } + } + + if( rc==SQLITE_OK ){ + pager_reset(pPager); + rc = sqlite3PcacheSetPageSize(pPager->pPCache, pageSize); + } + if( rc==SQLITE_OK ){ + sqlite3PageFree(pPager->pTmpSpace); + pPager->pTmpSpace = pNew; + pPager->dbSize = (Pgno)((nByte+pageSize-1)/pageSize); + pPager->pageSize = pageSize; + }else{ + sqlite3PageFree(pNew); + } + } + + *pPageSize = pPager->pageSize; + if( rc==SQLITE_OK ){ + if( nReserve<0 ) nReserve = pPager->nReserve; + assert( nReserve>=0 && nReserve<1000 ); + pPager->nReserve = (i16)nReserve; + pagerFixMaplimit(pPager); + } + return rc; +} + +/* +** Return a pointer to the "temporary page" buffer held internally +** by the pager. This is a buffer that is big enough to hold the +** entire content of a database page. This buffer is used internally +** during rollback and will be overwritten whenever a rollback +** occurs. But other modules are free to use it too, as long as +** no rollbacks are happening. +*/ +void *sqlite3PagerTempSpace(Pager *pPager){ + return pPager->pTmpSpace; +} + +/* +** Attempt to set the maximum database page count if mxPage is positive. +** Make no changes if mxPage is zero or negative. And never reduce the +** maximum page count below the current size of the database. +** +** Regardless of mxPage, return the current maximum page count. +*/ +Pgno sqlite3PagerMaxPageCount(Pager *pPager, Pgno mxPage){ + if( mxPage>0 ){ + pPager->mxPgno = mxPage; + } + assert( pPager->eState!=PAGER_OPEN ); /* Called only by OP_MaxPgcnt */ + /* assert( pPager->mxPgno>=pPager->dbSize ); */ + /* OP_MaxPgcnt ensures that the parameter passed to this function is not + ** less than the total number of valid pages in the database. But this + ** may be less than Pager.dbSize, and so the assert() above is not valid */ + return pPager->mxPgno; +} + +/* +** The following set of routines are used to disable the simulated +** I/O error mechanism. These routines are used to avoid simulated +** errors in places where we do not care about errors. +** +** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops +** and generate no code. +*/ +#ifdef SQLITE_TEST +extern int sqlite3_io_error_pending; +extern int sqlite3_io_error_hit; +static int saved_cnt; +void disable_simulated_io_errors(void){ + saved_cnt = sqlite3_io_error_pending; + sqlite3_io_error_pending = -1; +} +void enable_simulated_io_errors(void){ + sqlite3_io_error_pending = saved_cnt; +} +#else +# define disable_simulated_io_errors() +# define enable_simulated_io_errors() +#endif + +/* +** Read the first N bytes from the beginning of the file into memory +** that pDest points to. +** +** If the pager was opened on a transient file (zFilename==""), or +** opened on a file less than N bytes in size, the output buffer is +** zeroed and SQLITE_OK returned. The rationale for this is that this +** function is used to read database headers, and a new transient or +** zero sized database has a header than consists entirely of zeroes. +** +** If any IO error apart from SQLITE_IOERR_SHORT_READ is encountered, +** the error code is returned to the caller and the contents of the +** output buffer undefined. +*/ +int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){ + int rc = SQLITE_OK; + memset(pDest, 0, N); + assert( isOpen(pPager->fd) || pPager->tempFile ); + + /* This routine is only called by btree immediately after creating + ** the Pager object. There has not been an opportunity to transition + ** to WAL mode yet. + */ + assert( !pagerUseWal(pPager) ); + + if( isOpen(pPager->fd) ){ + IOTRACE(("DBHDR %p 0 %d\n", pPager, N)) + rc = sqlite3OsRead(pPager->fd, pDest, N, 0); + if( rc==SQLITE_IOERR_SHORT_READ ){ + rc = SQLITE_OK; + } + } + return rc; +} + +/* +** This function may only be called when a read-transaction is open on +** the pager. It returns the total number of pages in the database. +** +** However, if the file is between 1 and bytes in size, then +** this is considered a 1 page file. +*/ +void sqlite3PagerPagecount(Pager *pPager, int *pnPage){ + assert( pPager->eState>=PAGER_READER ); + assert( pPager->eState!=PAGER_WRITER_FINISHED ); + *pnPage = (int)pPager->dbSize; +} + + +/* +** Try to obtain a lock of type locktype on the database file. If +** a similar or greater lock is already held, this function is a no-op +** (returning SQLITE_OK immediately). +** +** Otherwise, attempt to obtain the lock using sqlite3OsLock(). Invoke +** the busy callback if the lock is currently not available. Repeat +** until the busy callback returns false or until the attempt to +** obtain the lock succeeds. +** +** Return SQLITE_OK on success and an error code if we cannot obtain +** the lock. If the lock is obtained successfully, set the Pager.state +** variable to locktype before returning. +*/ +static int pager_wait_on_lock(Pager *pPager, int locktype){ + int rc; /* Return code */ + + /* Check that this is either a no-op (because the requested lock is + ** already held), or one of the transitions that the busy-handler + ** may be invoked during, according to the comment above + ** sqlite3PagerSetBusyhandler(). + */ + assert( (pPager->eLock>=locktype) + || (pPager->eLock==NO_LOCK && locktype==SHARED_LOCK) + || (pPager->eLock==RESERVED_LOCK && locktype==EXCLUSIVE_LOCK) + ); + + do { + rc = pagerLockDb(pPager, locktype); + }while( rc==SQLITE_BUSY && pPager->xBusyHandler(pPager->pBusyHandlerArg) ); + return rc; +} + +/* +** Function assertTruncateConstraint(pPager) checks that one of the +** following is true for all dirty pages currently in the page-cache: +** +** a) The page number is less than or equal to the size of the +** current database image, in pages, OR +** +** b) if the page content were written at this time, it would not +** be necessary to write the current content out to the sub-journal +** (as determined by function subjRequiresPage()). +** +** If the condition asserted by this function were not true, and the +** dirty page were to be discarded from the cache via the pagerStress() +** routine, pagerStress() would not write the current page content to +** the database file. If a savepoint transaction were rolled back after +** this happened, the correct behavior would be to restore the current +** content of the page. However, since this content is not present in either +** the database file or the portion of the rollback journal and +** sub-journal rolled back the content could not be restored and the +** database image would become corrupt. It is therefore fortunate that +** this circumstance cannot arise. +*/ +#if defined(SQLITE_DEBUG) +static void assertTruncateConstraintCb(PgHdr *pPg){ + assert( pPg->flags&PGHDR_DIRTY ); + assert( !subjRequiresPage(pPg) || pPg->pgno<=pPg->pPager->dbSize ); +} +static void assertTruncateConstraint(Pager *pPager){ + sqlite3PcacheIterateDirty(pPager->pPCache, assertTruncateConstraintCb); +} +#else +# define assertTruncateConstraint(pPager) +#endif + +/* +** Truncate the in-memory database file image to nPage pages. This +** function does not actually modify the database file on disk. It +** just sets the internal state of the pager object so that the +** truncation will be done when the current transaction is committed. +** +** This function is only called right before committing a transaction. +** Once this function has been called, the transaction must either be +** rolled back or committed. It is not safe to call this function and +** then continue writing to the database. +*/ +void sqlite3PagerTruncateImage(Pager *pPager, Pgno nPage){ + assert( pPager->dbSize>=nPage || CORRUPT_DB ); + testcase( pPager->dbSizeeState>=PAGER_WRITER_CACHEMOD ); + pPager->dbSize = nPage; + + /* At one point the code here called assertTruncateConstraint() to + ** ensure that all pages being truncated away by this operation are, + ** if one or more savepoints are open, present in the savepoint + ** journal so that they can be restored if the savepoint is rolled + ** back. This is no longer necessary as this function is now only + ** called right before committing a transaction. So although the + ** Pager object may still have open savepoints (Pager.nSavepoint!=0), + ** they cannot be rolled back. So the assertTruncateConstraint() call + ** is no longer correct. */ +} + + +/* +** This function is called before attempting a hot-journal rollback. It +** syncs the journal file to disk, then sets pPager->journalHdr to the +** size of the journal file so that the pager_playback() routine knows +** that the entire journal file has been synced. +** +** Syncing a hot-journal to disk before attempting to roll it back ensures +** that if a power-failure occurs during the rollback, the process that +** attempts rollback following system recovery sees the same journal +** content as this process. +** +** If everything goes as planned, SQLITE_OK is returned. Otherwise, +** an SQLite error code. +*/ +static int pagerSyncHotJournal(Pager *pPager){ + int rc = SQLITE_OK; + if( !pPager->noSync ){ + rc = sqlite3OsSync(pPager->jfd, SQLITE_SYNC_NORMAL); + } + if( rc==SQLITE_OK ){ + rc = sqlite3OsFileSize(pPager->jfd, &pPager->journalHdr); + } + return rc; +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* +** Obtain a reference to a memory mapped page object for page number pgno. +** The new object will use the pointer pData, obtained from xFetch(). +** If successful, set *ppPage to point to the new page reference +** and return SQLITE_OK. Otherwise, return an SQLite error code and set +** *ppPage to zero. +** +** Page references obtained by calling this function should be released +** by calling pagerReleaseMapPage(). +*/ +static int pagerAcquireMapPage( + Pager *pPager, /* Pager object */ + Pgno pgno, /* Page number */ + void *pData, /* xFetch()'d data for this page */ + PgHdr **ppPage /* OUT: Acquired page object */ +){ + PgHdr *p; /* Memory mapped page to return */ + + if( pPager->pMmapFreelist ){ + *ppPage = p = pPager->pMmapFreelist; + pPager->pMmapFreelist = p->pDirty; + p->pDirty = 0; + assert( pPager->nExtra>=8 ); + memset(p->pExtra, 0, 8); + }else{ + *ppPage = p = (PgHdr *)sqlite3MallocZero(sizeof(PgHdr) + pPager->nExtra); + if( p==0 ){ + sqlite3OsUnfetch(pPager->fd, (i64)(pgno-1) * pPager->pageSize, pData); + return SQLITE_NOMEM_BKPT; + } + p->pExtra = (void *)&p[1]; + p->flags = PGHDR_MMAP; + p->nRef = 1; + p->pPager = pPager; + } + + assert( p->pExtra==(void *)&p[1] ); + assert( p->pPage==0 ); + assert( p->flags==PGHDR_MMAP ); + assert( p->pPager==pPager ); + assert( p->nRef==1 ); + + p->pgno = pgno; + p->pData = pData; + pPager->nMmapOut++; + + return SQLITE_OK; +} +#endif + +/* +** Release a reference to page pPg. pPg must have been returned by an +** earlier call to pagerAcquireMapPage(). +*/ +static void pagerReleaseMapPage(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + pPager->nMmapOut--; + pPg->pDirty = pPager->pMmapFreelist; + pPager->pMmapFreelist = pPg; + + assert( pPager->fd->pMethods->iVersion>=3 ); + sqlite3OsUnfetch(pPager->fd, (i64)(pPg->pgno-1)*pPager->pageSize, pPg->pData); +} + +/* +** Free all PgHdr objects stored in the Pager.pMmapFreelist list. +*/ +static void pagerFreeMapHdrs(Pager *pPager){ + PgHdr *p; + PgHdr *pNext; + for(p=pPager->pMmapFreelist; p; p=pNext){ + pNext = p->pDirty; + sqlite3_free(p); + } +} + +/* Verify that the database file has not be deleted or renamed out from +** under the pager. Return SQLITE_OK if the database is still where it ought +** to be on disk. Return non-zero (SQLITE_READONLY_DBMOVED or some other error +** code from sqlite3OsAccess()) if the database has gone missing. +*/ +static int databaseIsUnmoved(Pager *pPager){ + int bHasMoved = 0; + int rc; + + if( pPager->tempFile ) return SQLITE_OK; + if( pPager->dbSize==0 ) return SQLITE_OK; + assert( pPager->zFilename && pPager->zFilename[0] ); + rc = sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_HAS_MOVED, &bHasMoved); + if( rc==SQLITE_NOTFOUND ){ + /* If the HAS_MOVED file-control is unimplemented, assume that the file + ** has not been moved. That is the historical behavior of SQLite: prior to + ** version 3.8.3, it never checked */ + rc = SQLITE_OK; + }else if( rc==SQLITE_OK && bHasMoved ){ + rc = SQLITE_READONLY_DBMOVED; + } + return rc; +} + + +/* +** Shutdown the page cache. Free all memory and close all files. +** +** If a transaction was in progress when this routine is called, that +** transaction is rolled back. All outstanding pages are invalidated +** and their memory is freed. Any attempt to use a page associated +** with this page cache after this function returns will likely +** result in a coredump. +** +** This function always succeeds. If a transaction is active an attempt +** is made to roll it back. If an error occurs during the rollback +** a hot journal may be left in the filesystem but no error is returned +** to the caller. +*/ +int sqlite3PagerClose(Pager *pPager, sqlite3 *db){ + u8 *pTmp = (u8*)pPager->pTmpSpace; + assert( db || pagerUseWal(pPager)==0 ); + assert( assert_pager_state(pPager) ); + disable_simulated_io_errors(); + sqlite3BeginBenignMalloc(); + pagerFreeMapHdrs(pPager); + /* pPager->errCode = 0; */ + pPager->exclusiveMode = 0; +#ifndef SQLITE_OMIT_WAL + { + u8 *a = 0; + assert( db || pPager->pWal==0 ); + if( db && 0==(db->flags & SQLITE_NoCkptOnClose) + && SQLITE_OK==databaseIsUnmoved(pPager) + ){ + a = pTmp; + } + sqlite3WalClose(pPager->pWal, db, pPager->walSyncFlags, pPager->pageSize,a); + pPager->pWal = 0; + } +#endif + pager_reset(pPager); + if( MEMDB ){ + pager_unlock(pPager); + }else{ + /* If it is open, sync the journal file before calling UnlockAndRollback. + ** If this is not done, then an unsynced portion of the open journal + ** file may be played back into the database. If a power failure occurs + ** while this is happening, the database could become corrupt. + ** + ** If an error occurs while trying to sync the journal, shift the pager + ** into the ERROR state. This causes UnlockAndRollback to unlock the + ** database and close the journal file without attempting to roll it + ** back or finalize it. The next database user will have to do hot-journal + ** rollback before accessing the database file. + */ + if( isOpen(pPager->jfd) ){ + pager_error(pPager, pagerSyncHotJournal(pPager)); + } + pagerUnlockAndRollback(pPager); + } + sqlite3EndBenignMalloc(); + enable_simulated_io_errors(); + PAGERTRACE(("CLOSE %d\n", PAGERID(pPager))); + IOTRACE(("CLOSE %p\n", pPager)) + sqlite3OsClose(pPager->jfd); + sqlite3OsClose(pPager->fd); + sqlite3PageFree(pTmp); + sqlite3PcacheClose(pPager->pPCache); + assert( !pPager->aSavepoint && !pPager->pInJournal ); + assert( !isOpen(pPager->jfd) && !isOpen(pPager->sjfd) ); + + sqlite3_free(pPager); + return SQLITE_OK; +} + +#if !defined(NDEBUG) || defined(SQLITE_TEST) +/* +** Return the page number for page pPg. +*/ +Pgno sqlite3PagerPagenumber(DbPage *pPg){ + return pPg->pgno; +} +#endif + +/* +** Increment the reference count for page pPg. +*/ +void sqlite3PagerRef(DbPage *pPg){ + sqlite3PcacheRef(pPg); +} + +/* +** Sync the journal. In other words, make sure all the pages that have +** been written to the journal have actually reached the surface of the +** disk and can be restored in the event of a hot-journal rollback. +** +** If the Pager.noSync flag is set, then this function is a no-op. +** Otherwise, the actions required depend on the journal-mode and the +** device characteristics of the file-system, as follows: +** +** * If the journal file is an in-memory journal file, no action need +** be taken. +** +** * Otherwise, if the device does not support the SAFE_APPEND property, +** then the nRec field of the most recently written journal header +** is updated to contain the number of journal records that have +** been written following it. If the pager is operating in full-sync +** mode, then the journal file is synced before this field is updated. +** +** * If the device does not support the SEQUENTIAL property, then +** journal file is synced. +** +** Or, in pseudo-code: +** +** if( NOT ){ +** if( NOT SAFE_APPEND ){ +** if( ) xSync(); +** +** } +** if( NOT SEQUENTIAL ) xSync(); +** } +** +** If successful, this routine clears the PGHDR_NEED_SYNC flag of every +** page currently held in memory before returning SQLITE_OK. If an IO +** error is encountered, then the IO error code is returned to the caller. +*/ +static int syncJournal(Pager *pPager, int newHdr){ + int rc; /* Return code */ + + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + assert( !pagerUseWal(pPager) ); + + rc = sqlite3PagerExclusiveLock(pPager); + if( rc!=SQLITE_OK ) return rc; + + if( !pPager->noSync ){ + assert( !pPager->tempFile ); + if( isOpen(pPager->jfd) && pPager->journalMode!=PAGER_JOURNALMODE_MEMORY ){ + const int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); + assert( isOpen(pPager->jfd) ); + + if( 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){ + /* This block deals with an obscure problem. If the last connection + ** that wrote to this database was operating in persistent-journal + ** mode, then the journal file may at this point actually be larger + ** than Pager.journalOff bytes. If the next thing in the journal + ** file happens to be a journal-header (written as part of the + ** previous connection's transaction), and a crash or power-failure + ** occurs after nRec is updated but before this connection writes + ** anything else to the journal file (or commits/rolls back its + ** transaction), then SQLite may become confused when doing the + ** hot-journal rollback following recovery. It may roll back all + ** of this connections data, then proceed to rolling back the old, + ** out-of-date data that follows it. Database corruption. + ** + ** To work around this, if the journal file does appear to contain + ** a valid header following Pager.journalOff, then write a 0x00 + ** byte to the start of it to prevent it from being recognized. + ** + ** Variable iNextHdrOffset is set to the offset at which this + ** problematic header will occur, if it exists. aMagic is used + ** as a temporary buffer to inspect the first couple of bytes of + ** the potential journal header. + */ + i64 iNextHdrOffset; + u8 aMagic[8]; + u8 zHeader[sizeof(aJournalMagic)+4]; + + memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic)); + put32bits(&zHeader[sizeof(aJournalMagic)], pPager->nRec); + + iNextHdrOffset = journalHdrOffset(pPager); + rc = sqlite3OsRead(pPager->jfd, aMagic, 8, iNextHdrOffset); + if( rc==SQLITE_OK && 0==memcmp(aMagic, aJournalMagic, 8) ){ + static const u8 zerobyte = 0; + rc = sqlite3OsWrite(pPager->jfd, &zerobyte, 1, iNextHdrOffset); + } + if( rc!=SQLITE_OK && rc!=SQLITE_IOERR_SHORT_READ ){ + return rc; + } + + /* Write the nRec value into the journal file header. If in + ** full-synchronous mode, sync the journal first. This ensures that + ** all data has really hit the disk before nRec is updated to mark + ** it as a candidate for rollback. + ** + ** This is not required if the persistent media supports the + ** SAFE_APPEND property. Because in this case it is not possible + ** for garbage data to be appended to the file, the nRec field + ** is populated with 0xFFFFFFFF when the journal header is written + ** and never needs to be updated. + */ + if( pPager->fullSync && 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){ + PAGERTRACE(("SYNC journal of %d\n", PAGERID(pPager))); + IOTRACE(("JSYNC %p\n", pPager)) + rc = sqlite3OsSync(pPager->jfd, pPager->syncFlags); + if( rc!=SQLITE_OK ) return rc; + } + IOTRACE(("JHDR %p %lld\n", pPager, pPager->journalHdr)); + rc = sqlite3OsWrite( + pPager->jfd, zHeader, sizeof(zHeader), pPager->journalHdr + ); + if( rc!=SQLITE_OK ) return rc; + } + if( 0==(iDc&SQLITE_IOCAP_SEQUENTIAL) ){ + PAGERTRACE(("SYNC journal of %d\n", PAGERID(pPager))); + IOTRACE(("JSYNC %p\n", pPager)) + rc = sqlite3OsSync(pPager->jfd, pPager->syncFlags| + (pPager->syncFlags==SQLITE_SYNC_FULL?SQLITE_SYNC_DATAONLY:0) + ); + if( rc!=SQLITE_OK ) return rc; + } + + pPager->journalHdr = pPager->journalOff; + if( newHdr && 0==(iDc&SQLITE_IOCAP_SAFE_APPEND) ){ + pPager->nRec = 0; + rc = writeJournalHdr(pPager); + if( rc!=SQLITE_OK ) return rc; + } + }else{ + pPager->journalHdr = pPager->journalOff; + } + } + + /* Unless the pager is in noSync mode, the journal file was just + ** successfully synced. Either way, clear the PGHDR_NEED_SYNC flag on + ** all pages. + */ + sqlite3PcacheClearSyncFlags(pPager->pPCache); + pPager->eState = PAGER_WRITER_DBMOD; + assert( assert_pager_state(pPager) ); + return SQLITE_OK; +} + +/* +** The argument is the first in a linked list of dirty pages connected +** by the PgHdr.pDirty pointer. This function writes each one of the +** in-memory pages in the list to the database file. The argument may +** be NULL, representing an empty list. In this case this function is +** a no-op. +** +** The pager must hold at least a RESERVED lock when this function +** is called. Before writing anything to the database file, this lock +** is upgraded to an EXCLUSIVE lock. If the lock cannot be obtained, +** SQLITE_BUSY is returned and no data is written to the database file. +** +** If the pager is a temp-file pager and the actual file-system file +** is not yet open, it is created and opened before any data is +** written out. +** +** Once the lock has been upgraded and, if necessary, the file opened, +** the pages are written out to the database file in list order. Writing +** a page is skipped if it meets either of the following criteria: +** +** * The page number is greater than Pager.dbSize, or +** * The PGHDR_DONT_WRITE flag is set on the page. +** +** If writing out a page causes the database file to grow, Pager.dbFileSize +** is updated accordingly. If page 1 is written out, then the value cached +** in Pager.dbFileVers[] is updated to match the new value stored in +** the database file. +** +** If everything is successful, SQLITE_OK is returned. If an IO error +** occurs, an IO error code is returned. Or, if the EXCLUSIVE lock cannot +** be obtained, SQLITE_BUSY is returned. +*/ +static int pager_write_pagelist(Pager *pPager, PgHdr *pList){ + int rc = SQLITE_OK; /* Return code */ + + /* This function is only called for rollback pagers in WRITER_DBMOD state. */ + assert( !pagerUseWal(pPager) ); + assert( pPager->tempFile || pPager->eState==PAGER_WRITER_DBMOD ); + assert( pPager->eLock==EXCLUSIVE_LOCK ); + assert( isOpen(pPager->fd) || pList->pDirty==0 ); + + /* If the file is a temp-file has not yet been opened, open it now. It + ** is not possible for rc to be other than SQLITE_OK if this branch + ** is taken, as pager_wait_on_lock() is a no-op for temp-files. + */ + if( !isOpen(pPager->fd) ){ + assert( pPager->tempFile && rc==SQLITE_OK ); + rc = pagerOpentemp(pPager, pPager->fd, pPager->vfsFlags); + } + + /* Before the first write, give the VFS a hint of what the final + ** file size will be. + */ + assert( rc!=SQLITE_OK || isOpen(pPager->fd) ); + if( rc==SQLITE_OK + && pPager->dbHintSizedbSize + && (pList->pDirty || pList->pgno>pPager->dbHintSize) + ){ + sqlite3_int64 szFile = pPager->pageSize * (sqlite3_int64)pPager->dbSize; + sqlite3OsFileControlHint(pPager->fd, SQLITE_FCNTL_SIZE_HINT, &szFile); + pPager->dbHintSize = pPager->dbSize; + } + + while( rc==SQLITE_OK && pList ){ + Pgno pgno = pList->pgno; + + /* If there are dirty pages in the page cache with page numbers greater + ** than Pager.dbSize, this means sqlite3PagerTruncateImage() was called to + ** make the file smaller (presumably by auto-vacuum code). Do not write + ** any such pages to the file. + ** + ** Also, do not write out any page that has the PGHDR_DONT_WRITE flag + ** set (set by sqlite3PagerDontWrite()). + */ + if( pgno<=pPager->dbSize && 0==(pList->flags&PGHDR_DONT_WRITE) ){ + i64 offset = (pgno-1)*(i64)pPager->pageSize; /* Offset to write */ + char *pData; /* Data to write */ + + assert( (pList->flags&PGHDR_NEED_SYNC)==0 ); + if( pList->pgno==1 ) pager_write_changecounter(pList); + + pData = pList->pData; + + /* Write out the page data. */ + rc = sqlite3OsWrite(pPager->fd, pData, pPager->pageSize, offset); + + /* If page 1 was just written, update Pager.dbFileVers to match + ** the value now stored in the database file. If writing this + ** page caused the database file to grow, update dbFileSize. + */ + if( pgno==1 ){ + memcpy(&pPager->dbFileVers, &pData[24], sizeof(pPager->dbFileVers)); + } + if( pgno>pPager->dbFileSize ){ + pPager->dbFileSize = pgno; + } + pPager->aStat[PAGER_STAT_WRITE]++; + + /* Update any backup objects copying the contents of this pager. */ + sqlite3BackupUpdate(pPager->pBackup, pgno, (u8*)pList->pData); + + PAGERTRACE(("STORE %d page %d hash(%08x)\n", + PAGERID(pPager), pgno, pager_pagehash(pList))); + IOTRACE(("PGOUT %p %d\n", pPager, pgno)); + PAGER_INCR(sqlite3_pager_writedb_count); + }else{ + PAGERTRACE(("NOSTORE %d page %d\n", PAGERID(pPager), pgno)); + } + pager_set_pagehash(pList); + pList = pList->pDirty; + } + + return rc; +} + +/* +** Ensure that the sub-journal file is open. If it is already open, this +** function is a no-op. +** +** SQLITE_OK is returned if everything goes according to plan. An +** SQLITE_IOERR_XXX error code is returned if a call to sqlite3OsOpen() +** fails. +*/ +static int openSubJournal(Pager *pPager){ + int rc = SQLITE_OK; + if( !isOpen(pPager->sjfd) ){ + const int flags = SQLITE_OPEN_SUBJOURNAL | SQLITE_OPEN_READWRITE + | SQLITE_OPEN_CREATE | SQLITE_OPEN_EXCLUSIVE + | SQLITE_OPEN_DELETEONCLOSE; + int nStmtSpill = sqlite3Config.nStmtSpill; + if( pPager->journalMode==PAGER_JOURNALMODE_MEMORY || pPager->subjInMemory ){ + nStmtSpill = -1; + } + rc = sqlite3JournalOpen(pPager->pVfs, 0, pPager->sjfd, flags, nStmtSpill); + } + return rc; +} + +/* +** Append a record of the current state of page pPg to the sub-journal. +** +** If successful, set the bit corresponding to pPg->pgno in the bitvecs +** for all open savepoints before returning. +** +** This function returns SQLITE_OK if everything is successful, an IO +** error code if the attempt to write to the sub-journal fails, or +** SQLITE_NOMEM if a malloc fails while setting a bit in a savepoint +** bitvec. +*/ +static int subjournalPage(PgHdr *pPg){ + int rc = SQLITE_OK; + Pager *pPager = pPg->pPager; + if( pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ + + /* Open the sub-journal, if it has not already been opened */ + assert( pPager->useJournal ); + assert( isOpen(pPager->jfd) || pagerUseWal(pPager) ); + assert( isOpen(pPager->sjfd) || pPager->nSubRec==0 ); + assert( pagerUseWal(pPager) + || pageInJournal(pPager, pPg) + || pPg->pgno>pPager->dbOrigSize + ); + rc = openSubJournal(pPager); + + /* If the sub-journal was opened successfully (or was already open), + ** write the journal record into the file. */ + if( rc==SQLITE_OK ){ + void *pData = pPg->pData; + i64 offset = (i64)pPager->nSubRec*(4+pPager->pageSize); + char *pData2; + pData2 = pData; + PAGERTRACE(("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno)); + rc = write32bits(pPager->sjfd, offset, pPg->pgno); + if( rc==SQLITE_OK ){ + rc = sqlite3OsWrite(pPager->sjfd, pData2, pPager->pageSize, offset+4); + } + } + } + if( rc==SQLITE_OK ){ + pPager->nSubRec++; + assert( pPager->nSavepoint>0 ); + rc = addToSavepointBitvecs(pPager, pPg->pgno); + } + return rc; +} +static int subjournalPageIfRequired(PgHdr *pPg){ + if( subjRequiresPage(pPg) ){ + return subjournalPage(pPg); + }else{ + return SQLITE_OK; + } +} + +/* +** This function is called by the pcache layer when it has reached some +** soft memory limit. The first argument is a pointer to a Pager object +** (cast as a void*). The pager is always 'purgeable' (not an in-memory +** database). The second argument is a reference to a page that is +** currently dirty but has no outstanding references. The page +** is always associated with the Pager object passed as the first +** argument. +** +** The job of this function is to make pPg clean by writing its contents +** out to the database file, if possible. This may involve syncing the +** journal file. +** +** If successful, sqlite3PcacheMakeClean() is called on the page and +** SQLITE_OK returned. If an IO error occurs while trying to make the +** page clean, the IO error code is returned. If the page cannot be +** made clean for some other reason, but no error occurs, then SQLITE_OK +** is returned by sqlite3PcacheMakeClean() is not called. +*/ +static int pagerStress(void *p, PgHdr *pPg){ + Pager *pPager = (Pager *)p; + int rc = SQLITE_OK; + + assert( pPg->pPager==pPager ); + assert( pPg->flags&PGHDR_DIRTY ); + + /* The doNotSpill NOSYNC bit is set during times when doing a sync of + ** journal (and adding a new header) is not allowed. This occurs + ** during calls to sqlite3PagerWrite() while trying to journal multiple + ** pages belonging to the same sector. + ** + ** The doNotSpill ROLLBACK and OFF bits inhibits all cache spilling + ** regardless of whether or not a sync is required. This is set during + ** a rollback or by user request, respectively. + ** + ** Spilling is also prohibited when in an error state since that could + ** lead to database corruption. In the current implementation it + ** is impossible for sqlite3PcacheFetch() to be called with createFlag==3 + ** while in the error state, hence it is impossible for this routine to + ** be called in the error state. Nevertheless, we include a NEVER() + ** test for the error state as a safeguard against future changes. + */ + if( NEVER(pPager->errCode) ) return SQLITE_OK; + testcase( pPager->doNotSpill & SPILLFLAG_ROLLBACK ); + testcase( pPager->doNotSpill & SPILLFLAG_OFF ); + testcase( pPager->doNotSpill & SPILLFLAG_NOSYNC ); + if( pPager->doNotSpill + && ((pPager->doNotSpill & (SPILLFLAG_ROLLBACK|SPILLFLAG_OFF))!=0 + || (pPg->flags & PGHDR_NEED_SYNC)!=0) + ){ + return SQLITE_OK; + } + + pPager->aStat[PAGER_STAT_SPILL]++; + pPg->pDirty = 0; + if( pagerUseWal(pPager) ){ + /* Write a single frame for this page to the log. */ + rc = subjournalPageIfRequired(pPg); + if( rc==SQLITE_OK ){ + rc = pagerWalFrames(pPager, pPg, 0, 0); + } + }else{ + +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( pPager->tempFile==0 ){ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc!=SQLITE_OK ) return pager_error(pPager, rc); + } +#endif + + /* Sync the journal file if required. */ + if( pPg->flags&PGHDR_NEED_SYNC + || pPager->eState==PAGER_WRITER_CACHEMOD + ){ + rc = syncJournal(pPager, 1); + } + + /* Write the contents of the page out to the database file. */ + if( rc==SQLITE_OK ){ + assert( (pPg->flags&PGHDR_NEED_SYNC)==0 ); + rc = pager_write_pagelist(pPager, pPg); + } + } + + /* Mark the page as clean. */ + if( rc==SQLITE_OK ){ + PAGERTRACE(("STRESS %d page %d\n", PAGERID(pPager), pPg->pgno)); + sqlite3PcacheMakeClean(pPg); + } + + return pager_error(pPager, rc); +} + +/* +** Flush all unreferenced dirty pages to disk. +*/ +int sqlite3PagerFlush(Pager *pPager){ + int rc = pPager->errCode; + if( !MEMDB ){ + PgHdr *pList = sqlite3PcacheDirtyList(pPager->pPCache); + assert( assert_pager_state(pPager) ); + while( rc==SQLITE_OK && pList ){ + PgHdr *pNext = pList->pDirty; + if( pList->nRef==0 ){ + rc = pagerStress((void*)pPager, pList); + } + pList = pNext; + } + } + + return rc; +} + +/* +** Allocate and initialize a new Pager object and put a pointer to it +** in *ppPager. The pager should eventually be freed by passing it +** to sqlite3PagerClose(). +** +** The zFilename argument is the path to the database file to open. +** If zFilename is NULL then a randomly-named temporary file is created +** and used as the file to be cached. Temporary files are be deleted +** automatically when they are closed. If zFilename is ":memory:" then +** all information is held in cache. It is never written to disk. +** This can be used to implement an in-memory database. +** +** The nExtra parameter specifies the number of bytes of space allocated +** along with each page reference. This space is available to the user +** via the sqlite3PagerGetExtra() API. When a new page is allocated, the +** first 8 bytes of this space are zeroed but the remainder is uninitialized. +** (The extra space is used by btree as the MemPage object.) +** +** The flags argument is used to specify properties that affect the +** operation of the pager. It should be passed some bitwise combination +** of the PAGER_* flags. +** +** The vfsFlags parameter is a bitmask to pass to the flags parameter +** of the xOpen() method of the supplied VFS when opening files. +** +** If the pager object is allocated and the specified file opened +** successfully, SQLITE_OK is returned and *ppPager set to point to +** the new pager object. If an error occurs, *ppPager is set to NULL +** and error code returned. This function may return SQLITE_NOMEM +** (sqlite3Malloc() is used to allocate memory), SQLITE_CANTOPEN or +** various SQLITE_IO_XXX errors. +*/ +int sqlite3PagerOpen( + sqlite3_vfs *pVfs, /* The virtual file system to use */ + Pager **ppPager, /* OUT: Return the Pager structure here */ + const char *zFilename, /* Name of the database file to open */ + int nExtra, /* Extra bytes append to each in-memory page */ + int flags, /* flags controlling this file */ + int vfsFlags, /* flags passed through to sqlite3_vfs.xOpen() */ + void (*xReinit)(DbPage*) /* Function to reinitialize pages */ +){ + u8 *pPtr; + Pager *pPager = 0; /* Pager object to allocate and return */ + int rc = SQLITE_OK; /* Return code */ + int tempFile = 0; /* True for temp files (incl. in-memory files) */ + int memDb = 0; /* True if this is an in-memory file */ +#ifndef SQLITE_OMIT_DESERIALIZE + int memJM = 0; /* Memory journal mode */ +#else +# define memJM 0 +#endif + int readOnly = 0; /* True if this is a read-only file */ + int journalFileSize; /* Bytes to allocate for each journal fd */ + char *zPathname = 0; /* Full path to database file */ + int nPathname = 0; /* Number of bytes in zPathname */ + int useJournal = (flags & PAGER_OMIT_JOURNAL)==0; /* False to omit journal */ + int pcacheSize = sqlite3PcacheSize(); /* Bytes to allocate for PCache */ + u32 szPageDflt = SQLITE_DEFAULT_PAGE_SIZE; /* Default page size */ + const char *zUri = 0; /* URI args to copy */ + int nUriByte = 1; /* Number of bytes of URI args at *zUri */ + int nUri = 0; /* Number of URI parameters */ + + /* Figure out how much space is required for each journal file-handle + ** (there are two of them, the main journal and the sub-journal). */ + journalFileSize = ROUND8(sqlite3JournalSize(pVfs)); + + /* Set the output variable to NULL in case an error occurs. */ + *ppPager = 0; + +#ifndef SQLITE_OMIT_MEMORYDB + if( flags & PAGER_MEMORY ){ + memDb = 1; + if( zFilename && zFilename[0] ){ + zPathname = sqlite3DbStrDup(0, zFilename); + if( zPathname==0 ) return SQLITE_NOMEM_BKPT; + nPathname = sqlite3Strlen30(zPathname); + zFilename = 0; + } + } +#endif + + /* Compute and store the full pathname in an allocated buffer pointed + ** to by zPathname, length nPathname. Or, if this is a temporary file, + ** leave both nPathname and zPathname set to 0. + */ + if( zFilename && zFilename[0] ){ + const char *z; + nPathname = pVfs->mxPathname+1; + zPathname = sqlite3DbMallocRaw(0, nPathname*2); + if( zPathname==0 ){ + return SQLITE_NOMEM_BKPT; + } + zPathname[0] = 0; /* Make sure initialized even if FullPathname() fails */ + rc = sqlite3OsFullPathname(pVfs, zFilename, nPathname, zPathname); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_OK_SYMLINK ){ + if( vfsFlags & SQLITE_OPEN_NOFOLLOW ){ + rc = SQLITE_CANTOPEN_SYMLINK; + }else{ + rc = SQLITE_OK; + } + } + } + nPathname = sqlite3Strlen30(zPathname); + z = zUri = &zFilename[sqlite3Strlen30(zFilename)+1]; + while( *z ){ + z += strlen(z)+1; + z += strlen(z)+1; + nUri++; + } + nUriByte = (int)(&z[1] - zUri); + assert( nUriByte>=1 ); + if( rc==SQLITE_OK && nPathname+8>pVfs->mxPathname ){ + /* This branch is taken when the journal path required by + ** the database being opened will be more than pVfs->mxPathname + ** bytes in length. This means the database cannot be opened, + ** as it will not be possible to open the journal file or even + ** check for a hot-journal before reading. + */ + rc = SQLITE_CANTOPEN_BKPT; + } + if( rc!=SQLITE_OK ){ + sqlite3DbFree(0, zPathname); + return rc; + } + } + + /* Allocate memory for the Pager structure, PCache object, the + ** three file descriptors, the database file name and the journal + ** file name. The layout in memory is as follows: + ** + ** Pager object (sizeof(Pager) bytes) + ** PCache object (sqlite3PcacheSize() bytes) + ** Database file handle (pVfs->szOsFile bytes) + ** Sub-journal file handle (journalFileSize bytes) + ** Main journal file handle (journalFileSize bytes) + ** Ptr back to the Pager (sizeof(Pager*) bytes) + ** \0\0\0\0 database prefix (4 bytes) + ** Database file name (nPathname+1 bytes) + ** URI query parameters (nUriByte bytes) + ** Journal filename (nPathname+8+1 bytes) + ** WAL filename (nPathname+4+1 bytes) + ** \0\0\0 terminator (3 bytes) + ** + ** Some 3rd-party software, over which we have no control, depends on + ** the specific order of the filenames and the \0 separators between them + ** so that it can (for example) find the database filename given the WAL + ** filename without using the sqlite3_filename_database() API. This is a + ** misuse of SQLite and a bug in the 3rd-party software, but the 3rd-party + ** software is in widespread use, so we try to avoid changing the filename + ** order and formatting if possible. In particular, the details of the + ** filename format expected by 3rd-party software should be as follows: + ** + ** - Main Database Path + ** - \0 + ** - Multiple URI components consisting of: + ** - Key + ** - \0 + ** - Value + ** - \0 + ** - \0 + ** - Journal Path + ** - \0 + ** - WAL Path (zWALName) + ** - \0 + ** + ** The sqlite3_create_filename() interface and the databaseFilename() utility + ** that is used by sqlite3_filename_database() and kin also depend on the + ** specific formatting and order of the various filenames, so if the format + ** changes here, be sure to change it there as well. + */ + pPtr = (u8 *)sqlite3MallocZero( + ROUND8(sizeof(*pPager)) + /* Pager structure */ + ROUND8(pcacheSize) + /* PCache object */ + ROUND8(pVfs->szOsFile) + /* The main db file */ + journalFileSize * 2 + /* The two journal files */ + sizeof(pPager) + /* Space to hold a pointer */ + 4 + /* Database prefix */ + nPathname + 1 + /* database filename */ + nUriByte + /* query parameters */ + nPathname + 8 + 1 + /* Journal filename */ +#ifndef SQLITE_OMIT_WAL + nPathname + 4 + 1 + /* WAL filename */ +#endif + 3 /* Terminator */ + ); + assert( EIGHT_BYTE_ALIGNMENT(SQLITE_INT_TO_PTR(journalFileSize)) ); + if( !pPtr ){ + sqlite3DbFree(0, zPathname); + return SQLITE_NOMEM_BKPT; + } + pPager = (Pager*)pPtr; pPtr += ROUND8(sizeof(*pPager)); + pPager->pPCache = (PCache*)pPtr; pPtr += ROUND8(pcacheSize); + pPager->fd = (sqlite3_file*)pPtr; pPtr += ROUND8(pVfs->szOsFile); + pPager->sjfd = (sqlite3_file*)pPtr; pPtr += journalFileSize; + pPager->jfd = (sqlite3_file*)pPtr; pPtr += journalFileSize; + assert( EIGHT_BYTE_ALIGNMENT(pPager->jfd) ); + memcpy(pPtr, &pPager, sizeof(pPager)); pPtr += sizeof(pPager); + + /* Fill in the Pager.zFilename and pPager.zQueryParam fields */ + pPtr += 4; /* Skip zero prefix */ + pPager->zFilename = (char*)pPtr; + if( nPathname>0 ){ + memcpy(pPtr, zPathname, nPathname); pPtr += nPathname + 1; + if( zUri ){ + memcpy(pPtr, zUri, nUriByte); pPtr += nUriByte; + }else{ + pPtr++; + } + } + + + /* Fill in Pager.zJournal */ + if( nPathname>0 ){ + pPager->zJournal = (char*)pPtr; + memcpy(pPtr, zPathname, nPathname); pPtr += nPathname; + memcpy(pPtr, "-journal",8); pPtr += 8 + 1; +#ifdef SQLITE_ENABLE_8_3_NAMES + sqlite3FileSuffix3(zFilename,pPager->zJournal); + pPtr = (u8*)(pPager->zJournal + sqlite3Strlen30(pPager->zJournal)+1); +#endif + }else{ + pPager->zJournal = 0; + } + +#ifndef SQLITE_OMIT_WAL + /* Fill in Pager.zWal */ + if( nPathname>0 ){ + pPager->zWal = (char*)pPtr; + memcpy(pPtr, zPathname, nPathname); pPtr += nPathname; + memcpy(pPtr, "-wal", 4); pPtr += 4 + 1; +#ifdef SQLITE_ENABLE_8_3_NAMES + sqlite3FileSuffix3(zFilename, pPager->zWal); + pPtr = (u8*)(pPager->zWal + sqlite3Strlen30(pPager->zWal)+1); +#endif + }else{ + pPager->zWal = 0; + } +#endif + (void)pPtr; /* Suppress warning about unused pPtr value */ + + if( nPathname ) sqlite3DbFree(0, zPathname); + pPager->pVfs = pVfs; + pPager->vfsFlags = vfsFlags; + + /* Open the pager file. + */ + if( zFilename && zFilename[0] ){ + int fout = 0; /* VFS flags returned by xOpen() */ + rc = sqlite3OsOpen(pVfs, pPager->zFilename, pPager->fd, vfsFlags, &fout); + assert( !memDb ); +#ifndef SQLITE_OMIT_DESERIALIZE + pPager->memVfs = memJM = (fout&SQLITE_OPEN_MEMORY)!=0; +#endif + readOnly = (fout&SQLITE_OPEN_READONLY)!=0; + + /* If the file was successfully opened for read/write access, + ** choose a default page size in case we have to create the + ** database file. The default page size is the maximum of: + ** + ** + SQLITE_DEFAULT_PAGE_SIZE, + ** + The value returned by sqlite3OsSectorSize() + ** + The largest page size that can be written atomically. + */ + if( rc==SQLITE_OK ){ + int iDc = sqlite3OsDeviceCharacteristics(pPager->fd); + if( !readOnly ){ + setSectorSize(pPager); + assert(SQLITE_DEFAULT_PAGE_SIZE<=SQLITE_MAX_DEFAULT_PAGE_SIZE); + if( szPageDfltsectorSize ){ + if( pPager->sectorSize>SQLITE_MAX_DEFAULT_PAGE_SIZE ){ + szPageDflt = SQLITE_MAX_DEFAULT_PAGE_SIZE; + }else{ + szPageDflt = (u32)pPager->sectorSize; + } + } +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + { + int ii; + assert(SQLITE_IOCAP_ATOMIC512==(512>>8)); + assert(SQLITE_IOCAP_ATOMIC64K==(65536>>8)); + assert(SQLITE_MAX_DEFAULT_PAGE_SIZE<=65536); + for(ii=szPageDflt; ii<=SQLITE_MAX_DEFAULT_PAGE_SIZE; ii=ii*2){ + if( iDc&(SQLITE_IOCAP_ATOMIC|(ii>>8)) ){ + szPageDflt = ii; + } + } + } +#endif + } + pPager->noLock = sqlite3_uri_boolean(pPager->zFilename, "nolock", 0); + if( (iDc & SQLITE_IOCAP_IMMUTABLE)!=0 + || sqlite3_uri_boolean(pPager->zFilename, "immutable", 0) ){ + vfsFlags |= SQLITE_OPEN_READONLY; + goto act_like_temp_file; + } + } + }else{ + /* If a temporary file is requested, it is not opened immediately. + ** In this case we accept the default page size and delay actually + ** opening the file until the first call to OsWrite(). + ** + ** This branch is also run for an in-memory database. An in-memory + ** database is the same as a temp-file that is never written out to + ** disk and uses an in-memory rollback journal. + ** + ** This branch also runs for files marked as immutable. + */ +act_like_temp_file: + tempFile = 1; + pPager->eState = PAGER_READER; /* Pretend we already have a lock */ + pPager->eLock = EXCLUSIVE_LOCK; /* Pretend we are in EXCLUSIVE mode */ + pPager->noLock = 1; /* Do no locking */ + readOnly = (vfsFlags&SQLITE_OPEN_READONLY); + } + + /* The following call to PagerSetPagesize() serves to set the value of + ** Pager.pageSize and to allocate the Pager.pTmpSpace buffer. + */ + if( rc==SQLITE_OK ){ + assert( pPager->memDb==0 ); + rc = sqlite3PagerSetPagesize(pPager, &szPageDflt, -1); + testcase( rc!=SQLITE_OK ); + } + + /* Initialize the PCache object. */ + if( rc==SQLITE_OK ){ + nExtra = ROUND8(nExtra); + assert( nExtra>=8 && nExtra<1000 ); + rc = sqlite3PcacheOpen(szPageDflt, nExtra, !memDb, + !memDb?pagerStress:0, (void *)pPager, pPager->pPCache); + } + + /* If an error occurred above, free the Pager structure and close the file. + */ + if( rc!=SQLITE_OK ){ + sqlite3OsClose(pPager->fd); + sqlite3PageFree(pPager->pTmpSpace); + sqlite3_free(pPager); + return rc; + } + + PAGERTRACE(("OPEN %d %s\n", FILEHANDLEID(pPager->fd), pPager->zFilename)); + IOTRACE(("OPEN %p %s\n", pPager, pPager->zFilename)) + + pPager->useJournal = (u8)useJournal; + /* pPager->stmtOpen = 0; */ + /* pPager->stmtInUse = 0; */ + /* pPager->nRef = 0; */ + /* pPager->stmtSize = 0; */ + /* pPager->stmtJSize = 0; */ + /* pPager->nPage = 0; */ + pPager->mxPgno = SQLITE_MAX_PAGE_COUNT; + /* pPager->state = PAGER_UNLOCK; */ + /* pPager->errMask = 0; */ + pPager->tempFile = (u8)tempFile; + assert( tempFile==PAGER_LOCKINGMODE_NORMAL + || tempFile==PAGER_LOCKINGMODE_EXCLUSIVE ); + assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 ); + pPager->exclusiveMode = (u8)tempFile; + pPager->changeCountDone = pPager->tempFile; + pPager->memDb = (u8)memDb; + pPager->readOnly = (u8)readOnly; + assert( useJournal || pPager->tempFile ); + pPager->noSync = pPager->tempFile; + if( pPager->noSync ){ + assert( pPager->fullSync==0 ); + assert( pPager->extraSync==0 ); + assert( pPager->syncFlags==0 ); + assert( pPager->walSyncFlags==0 ); + }else{ + pPager->fullSync = 1; + pPager->extraSync = 0; + pPager->syncFlags = SQLITE_SYNC_NORMAL; + pPager->walSyncFlags = SQLITE_SYNC_NORMAL | (SQLITE_SYNC_NORMAL<<2); + } + /* pPager->pFirst = 0; */ + /* pPager->pFirstSynced = 0; */ + /* pPager->pLast = 0; */ + pPager->nExtra = (u16)nExtra; + pPager->journalSizeLimit = SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT; + assert( isOpen(pPager->fd) || tempFile ); + setSectorSize(pPager); + if( !useJournal ){ + pPager->journalMode = PAGER_JOURNALMODE_OFF; + }else if( memDb || memJM ){ + pPager->journalMode = PAGER_JOURNALMODE_MEMORY; + } + /* pPager->xBusyHandler = 0; */ + /* pPager->pBusyHandlerArg = 0; */ + pPager->xReiniter = xReinit; + setGetterMethod(pPager); + /* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */ + /* pPager->szMmap = SQLITE_DEFAULT_MMAP_SIZE // will be set by btree.c */ + + *ppPager = pPager; + return SQLITE_OK; +} + +/* +** Return the sqlite3_file for the main database given the name +** of the corresonding WAL or Journal name as passed into +** xOpen. +*/ +sqlite3_file *sqlite3_database_file_object(const char *zName){ + Pager *pPager; + while( zName[-1]!=0 || zName[-2]!=0 || zName[-3]!=0 || zName[-4]!=0 ){ + zName--; + } + pPager = *(Pager**)(zName - 4 - sizeof(Pager*)); + return pPager->fd; +} + + +/* +** This function is called after transitioning from PAGER_UNLOCK to +** PAGER_SHARED state. It tests if there is a hot journal present in +** the file-system for the given pager. A hot journal is one that +** needs to be played back. According to this function, a hot-journal +** file exists if the following criteria are met: +** +** * The journal file exists in the file system, and +** * No process holds a RESERVED or greater lock on the database file, and +** * The database file itself is greater than 0 bytes in size, and +** * The first byte of the journal file exists and is not 0x00. +** +** If the current size of the database file is 0 but a journal file +** exists, that is probably an old journal left over from a prior +** database with the same name. In this case the journal file is +** just deleted using OsDelete, *pExists is set to 0 and SQLITE_OK +** is returned. +** +** This routine does not check if there is a super-journal filename +** at the end of the file. If there is, and that super-journal file +** does not exist, then the journal file is not really hot. In this +** case this routine will return a false-positive. The pager_playback() +** routine will discover that the journal file is not really hot and +** will not roll it back. +** +** If a hot-journal file is found to exist, *pExists is set to 1 and +** SQLITE_OK returned. If no hot-journal file is present, *pExists is +** set to 0 and SQLITE_OK returned. If an IO error occurs while trying +** to determine whether or not a hot-journal file exists, the IO error +** code is returned and the value of *pExists is undefined. +*/ +static int hasHotJournal(Pager *pPager, int *pExists){ + sqlite3_vfs * const pVfs = pPager->pVfs; + int rc = SQLITE_OK; /* Return code */ + int exists = 1; /* True if a journal file is present */ + int jrnlOpen = !!isOpen(pPager->jfd); + + assert( pPager->useJournal ); + assert( isOpen(pPager->fd) ); + assert( pPager->eState==PAGER_OPEN ); + + assert( jrnlOpen==0 || ( sqlite3OsDeviceCharacteristics(pPager->jfd) & + SQLITE_IOCAP_UNDELETABLE_WHEN_OPEN + )); + + *pExists = 0; + if( !jrnlOpen ){ + rc = sqlite3OsAccess(pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS, &exists); + } + if( rc==SQLITE_OK && exists ){ + int locked = 0; /* True if some process holds a RESERVED lock */ + + /* Race condition here: Another process might have been holding the + ** the RESERVED lock and have a journal open at the sqlite3OsAccess() + ** call above, but then delete the journal and drop the lock before + ** we get to the following sqlite3OsCheckReservedLock() call. If that + ** is the case, this routine might think there is a hot journal when + ** in fact there is none. This results in a false-positive which will + ** be dealt with by the playback routine. Ticket #3883. + */ + rc = sqlite3OsCheckReservedLock(pPager->fd, &locked); + if( rc==SQLITE_OK && !locked ){ + Pgno nPage; /* Number of pages in database file */ + + assert( pPager->tempFile==0 ); + rc = pagerPagecount(pPager, &nPage); + if( rc==SQLITE_OK ){ + /* If the database is zero pages in size, that means that either (1) the + ** journal is a remnant from a prior database with the same name where + ** the database file but not the journal was deleted, or (2) the initial + ** transaction that populates a new database is being rolled back. + ** In either case, the journal file can be deleted. However, take care + ** not to delete the journal file if it is already open due to + ** journal_mode=PERSIST. + */ + if( nPage==0 && !jrnlOpen ){ + sqlite3BeginBenignMalloc(); + if( pagerLockDb(pPager, RESERVED_LOCK)==SQLITE_OK ){ + sqlite3OsDelete(pVfs, pPager->zJournal, 0); + if( !pPager->exclusiveMode ) pagerUnlockDb(pPager, SHARED_LOCK); + } + sqlite3EndBenignMalloc(); + }else{ + /* The journal file exists and no other connection has a reserved + ** or greater lock on the database file. Now check that there is + ** at least one non-zero bytes at the start of the journal file. + ** If there is, then we consider this journal to be hot. If not, + ** it can be ignored. + */ + if( !jrnlOpen ){ + int f = SQLITE_OPEN_READONLY|SQLITE_OPEN_MAIN_JOURNAL; + rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, f, &f); + } + if( rc==SQLITE_OK ){ + u8 first = 0; + rc = sqlite3OsRead(pPager->jfd, (void *)&first, 1, 0); + if( rc==SQLITE_IOERR_SHORT_READ ){ + rc = SQLITE_OK; + } + if( !jrnlOpen ){ + sqlite3OsClose(pPager->jfd); + } + *pExists = (first!=0); + }else if( rc==SQLITE_CANTOPEN ){ + /* If we cannot open the rollback journal file in order to see if + ** it has a zero header, that might be due to an I/O error, or + ** it might be due to the race condition described above and in + ** ticket #3883. Either way, assume that the journal is hot. + ** This might be a false positive. But if it is, then the + ** automatic journal playback and recovery mechanism will deal + ** with it under an EXCLUSIVE lock where we do not need to + ** worry so much with race conditions. + */ + *pExists = 1; + rc = SQLITE_OK; + } + } + } + } + } + + return rc; +} + +/* +** This function is called to obtain a shared lock on the database file. +** It is illegal to call sqlite3PagerGet() until after this function +** has been successfully called. If a shared-lock is already held when +** this function is called, it is a no-op. +** +** The following operations are also performed by this function. +** +** 1) If the pager is currently in PAGER_OPEN state (no lock held +** on the database file), then an attempt is made to obtain a +** SHARED lock on the database file. Immediately after obtaining +** the SHARED lock, the file-system is checked for a hot-journal, +** which is played back if present. Following any hot-journal +** rollback, the contents of the cache are validated by checking +** the 'change-counter' field of the database file header and +** discarded if they are found to be invalid. +** +** 2) If the pager is running in exclusive-mode, and there are currently +** no outstanding references to any pages, and is in the error state, +** then an attempt is made to clear the error state by discarding +** the contents of the page cache and rolling back any open journal +** file. +** +** If everything is successful, SQLITE_OK is returned. If an IO error +** occurs while locking the database, checking for a hot-journal file or +** rolling back a journal file, the IO error code is returned. +*/ +int sqlite3PagerSharedLock(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + + /* This routine is only called from b-tree and only when there are no + ** outstanding pages. This implies that the pager state should either + ** be OPEN or READER. READER is only possible if the pager is or was in + ** exclusive access mode. */ + assert( sqlite3PcacheRefCount(pPager->pPCache)==0 ); + assert( assert_pager_state(pPager) ); + assert( pPager->eState==PAGER_OPEN || pPager->eState==PAGER_READER ); + assert( pPager->errCode==SQLITE_OK ); + + if( !pagerUseWal(pPager) && pPager->eState==PAGER_OPEN ){ + int bHotJournal = 1; /* True if there exists a hot journal-file */ + + assert( !MEMDB ); + assert( pPager->tempFile==0 || pPager->eLock==EXCLUSIVE_LOCK ); + + rc = pager_wait_on_lock(pPager, SHARED_LOCK); + if( rc!=SQLITE_OK ){ + assert( pPager->eLock==NO_LOCK || pPager->eLock==UNKNOWN_LOCK ); + goto failed; + } + + /* If a journal file exists, and there is no RESERVED lock on the + ** database file, then it either needs to be played back or deleted. + */ + if( pPager->eLock<=SHARED_LOCK ){ + rc = hasHotJournal(pPager, &bHotJournal); + } + if( rc!=SQLITE_OK ){ + goto failed; + } + if( bHotJournal ){ + if( pPager->readOnly ){ + rc = SQLITE_READONLY_ROLLBACK; + goto failed; + } + + /* Get an EXCLUSIVE lock on the database file. At this point it is + ** important that a RESERVED lock is not obtained on the way to the + ** EXCLUSIVE lock. If it were, another process might open the + ** database file, detect the RESERVED lock, and conclude that the + ** database is safe to read while this process is still rolling the + ** hot-journal back. + ** + ** Because the intermediate RESERVED lock is not requested, any + ** other process attempting to access the database file will get to + ** this point in the code and fail to obtain its own EXCLUSIVE lock + ** on the database file. + ** + ** Unless the pager is in locking_mode=exclusive mode, the lock is + ** downgraded to SHARED_LOCK before this function returns. + */ + rc = pagerLockDb(pPager, EXCLUSIVE_LOCK); + if( rc!=SQLITE_OK ){ + goto failed; + } + + /* If it is not already open and the file exists on disk, open the + ** journal for read/write access. Write access is required because + ** in exclusive-access mode the file descriptor will be kept open + ** and possibly used for a transaction later on. Also, write-access + ** is usually required to finalize the journal in journal_mode=persist + ** mode (and also for journal_mode=truncate on some systems). + ** + ** If the journal does not exist, it usually means that some + ** other connection managed to get in and roll it back before + ** this connection obtained the exclusive lock above. Or, it + ** may mean that the pager was in the error-state when this + ** function was called and the journal file does not exist. + */ + if( !isOpen(pPager->jfd) ){ + sqlite3_vfs * const pVfs = pPager->pVfs; + int bExists; /* True if journal file exists */ + rc = sqlite3OsAccess( + pVfs, pPager->zJournal, SQLITE_ACCESS_EXISTS, &bExists); + if( rc==SQLITE_OK && bExists ){ + int fout = 0; + int f = SQLITE_OPEN_READWRITE|SQLITE_OPEN_MAIN_JOURNAL; + assert( !pPager->tempFile ); + rc = sqlite3OsOpen(pVfs, pPager->zJournal, pPager->jfd, f, &fout); + assert( rc!=SQLITE_OK || isOpen(pPager->jfd) ); + if( rc==SQLITE_OK && fout&SQLITE_OPEN_READONLY ){ + rc = SQLITE_CANTOPEN_BKPT; + sqlite3OsClose(pPager->jfd); + } + } + } + + /* Playback and delete the journal. Drop the database write + ** lock and reacquire the read lock. Purge the cache before + ** playing back the hot-journal so that we don't end up with + ** an inconsistent cache. Sync the hot journal before playing + ** it back since the process that crashed and left the hot journal + ** probably did not sync it and we are required to always sync + ** the journal before playing it back. + */ + if( isOpen(pPager->jfd) ){ + assert( rc==SQLITE_OK ); + rc = pagerSyncHotJournal(pPager); + if( rc==SQLITE_OK ){ + rc = pager_playback(pPager, !pPager->tempFile); + pPager->eState = PAGER_OPEN; + } + }else if( !pPager->exclusiveMode ){ + pagerUnlockDb(pPager, SHARED_LOCK); + } + + if( rc!=SQLITE_OK ){ + /* This branch is taken if an error occurs while trying to open + ** or roll back a hot-journal while holding an EXCLUSIVE lock. The + ** pager_unlock() routine will be called before returning to unlock + ** the file. If the unlock attempt fails, then Pager.eLock must be + ** set to UNKNOWN_LOCK (see the comment above the #define for + ** UNKNOWN_LOCK above for an explanation). + ** + ** In order to get pager_unlock() to do this, set Pager.eState to + ** PAGER_ERROR now. This is not actually counted as a transition + ** to ERROR state in the state diagram at the top of this file, + ** since we know that the same call to pager_unlock() will very + ** shortly transition the pager object to the OPEN state. Calling + ** assert_pager_state() would fail now, as it should not be possible + ** to be in ERROR state when there are zero outstanding page + ** references. + */ + pager_error(pPager, rc); + goto failed; + } + + assert( pPager->eState==PAGER_OPEN ); + assert( (pPager->eLock==SHARED_LOCK) + || (pPager->exclusiveMode && pPager->eLock>SHARED_LOCK) + ); + } + + if( !pPager->tempFile && pPager->hasHeldSharedLock ){ + /* The shared-lock has just been acquired then check to + ** see if the database has been modified. If the database has changed, + ** flush the cache. The hasHeldSharedLock flag prevents this from + ** occurring on the very first access to a file, in order to save a + ** single unnecessary sqlite3OsRead() call at the start-up. + ** + ** Database changes are detected by looking at 15 bytes beginning + ** at offset 24 into the file. The first 4 of these 16 bytes are + ** a 32-bit counter that is incremented with each change. The + ** other bytes change randomly with each file change when + ** a codec is in use. + ** + ** There is a vanishingly small chance that a change will not be + ** detected. The chance of an undetected change is so small that + ** it can be neglected. + */ + char dbFileVers[sizeof(pPager->dbFileVers)]; + + IOTRACE(("CKVERS %p %d\n", pPager, sizeof(dbFileVers))); + rc = sqlite3OsRead(pPager->fd, &dbFileVers, sizeof(dbFileVers), 24); + if( rc!=SQLITE_OK ){ + if( rc!=SQLITE_IOERR_SHORT_READ ){ + goto failed; + } + memset(dbFileVers, 0, sizeof(dbFileVers)); + } + + if( memcmp(pPager->dbFileVers, dbFileVers, sizeof(dbFileVers))!=0 ){ + pager_reset(pPager); + + /* Unmap the database file. It is possible that external processes + ** may have truncated the database file and then extended it back + ** to its original size while this process was not holding a lock. + ** In this case there may exist a Pager.pMap mapping that appears + ** to be the right size but is not actually valid. Avoid this + ** possibility by unmapping the db here. */ + if( USEFETCH(pPager) ){ + sqlite3OsUnfetch(pPager->fd, 0, 0); + } + } + } + + /* If there is a WAL file in the file-system, open this database in WAL + ** mode. Otherwise, the following function call is a no-op. + */ + rc = pagerOpenWalIfPresent(pPager); +#ifndef SQLITE_OMIT_WAL + assert( pPager->pWal==0 || rc==SQLITE_OK ); +#endif + } + + if( pagerUseWal(pPager) ){ + assert( rc==SQLITE_OK ); + rc = pagerBeginReadTransaction(pPager); + } + + if( pPager->tempFile==0 && pPager->eState==PAGER_OPEN && rc==SQLITE_OK ){ + rc = pagerPagecount(pPager, &pPager->dbSize); + } + + failed: + if( rc!=SQLITE_OK ){ + assert( !MEMDB ); + pager_unlock(pPager); + assert( pPager->eState==PAGER_OPEN ); + }else{ + pPager->eState = PAGER_READER; + pPager->hasHeldSharedLock = 1; + } + return rc; +} + +/* +** If the reference count has reached zero, rollback any active +** transaction and unlock the pager. +** +** Except, in locking_mode=EXCLUSIVE when there is nothing to in +** the rollback journal, the unlock is not performed and there is +** nothing to rollback, so this routine is a no-op. +*/ +static void pagerUnlockIfUnused(Pager *pPager){ + if( sqlite3PcacheRefCount(pPager->pPCache)==0 ){ + assert( pPager->nMmapOut==0 ); /* because page1 is never memory mapped */ + pagerUnlockAndRollback(pPager); + } +} + +/* +** The page getter methods each try to acquire a reference to a +** page with page number pgno. If the requested reference is +** successfully obtained, it is copied to *ppPage and SQLITE_OK returned. +** +** There are different implementations of the getter method depending +** on the current state of the pager. +** +** getPageNormal() -- The normal getter +** getPageError() -- Used if the pager is in an error state +** getPageMmap() -- Used if memory-mapped I/O is enabled +** +** If the requested page is already in the cache, it is returned. +** Otherwise, a new page object is allocated and populated with data +** read from the database file. In some cases, the pcache module may +** choose not to allocate a new page object and may reuse an existing +** object with no outstanding references. +** +** The extra data appended to a page is always initialized to zeros the +** first time a page is loaded into memory. If the page requested is +** already in the cache when this function is called, then the extra +** data is left as it was when the page object was last used. +** +** If the database image is smaller than the requested page or if +** the flags parameter contains the PAGER_GET_NOCONTENT bit and the +** requested page is not already stored in the cache, then no +** actual disk read occurs. In this case the memory image of the +** page is initialized to all zeros. +** +** If PAGER_GET_NOCONTENT is true, it means that we do not care about +** the contents of the page. This occurs in two scenarios: +** +** a) When reading a free-list leaf page from the database, and +** +** b) When a savepoint is being rolled back and we need to load +** a new page into the cache to be filled with the data read +** from the savepoint journal. +** +** If PAGER_GET_NOCONTENT is true, then the data returned is zeroed instead +** of being read from the database. Additionally, the bits corresponding +** to pgno in Pager.pInJournal (bitvec of pages already written to the +** journal file) and the PagerSavepoint.pInSavepoint bitvecs of any open +** savepoints are set. This means if the page is made writable at any +** point in the future, using a call to sqlite3PagerWrite(), its contents +** will not be journaled. This saves IO. +** +** The acquisition might fail for several reasons. In all cases, +** an appropriate error code is returned and *ppPage is set to NULL. +** +** See also sqlite3PagerLookup(). Both this routine and Lookup() attempt +** to find a page in the in-memory cache first. If the page is not already +** in memory, this routine goes to disk to read it in whereas Lookup() +** just returns 0. This routine acquires a read-lock the first time it +** has to go to disk, and could also playback an old journal if necessary. +** Since Lookup() never goes to disk, it never has to deal with locks +** or journal files. +*/ +static int getPageNormal( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + int rc = SQLITE_OK; + PgHdr *pPg; + u8 noContent; /* True if PAGER_GET_NOCONTENT is set */ + sqlite3_pcache_page *pBase; + + assert( pPager->errCode==SQLITE_OK ); + assert( pPager->eState>=PAGER_READER ); + assert( assert_pager_state(pPager) ); + assert( pPager->hasHeldSharedLock==1 ); + + if( pgno==0 ) return SQLITE_CORRUPT_BKPT; + pBase = sqlite3PcacheFetch(pPager->pPCache, pgno, 3); + if( pBase==0 ){ + pPg = 0; + rc = sqlite3PcacheFetchStress(pPager->pPCache, pgno, &pBase); + if( rc!=SQLITE_OK ) goto pager_acquire_err; + if( pBase==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto pager_acquire_err; + } + } + pPg = *ppPage = sqlite3PcacheFetchFinish(pPager->pPCache, pgno, pBase); + assert( pPg==(*ppPage) ); + assert( pPg->pgno==pgno ); + assert( pPg->pPager==pPager || pPg->pPager==0 ); + + noContent = (flags & PAGER_GET_NOCONTENT)!=0; + if( pPg->pPager && !noContent ){ + /* In this case the pcache already contains an initialized copy of + ** the page. Return without further ado. */ + assert( pgno!=PAGER_MJ_PGNO(pPager) ); + pPager->aStat[PAGER_STAT_HIT]++; + return SQLITE_OK; + + }else{ + /* The pager cache has created a new page. Its content needs to + ** be initialized. But first some error checks: + ** + ** (*) obsolete. Was: maximum page number is 2^31 + ** (2) Never try to fetch the locking page + */ + if( pgno==PAGER_MJ_PGNO(pPager) ){ + rc = SQLITE_CORRUPT_BKPT; + goto pager_acquire_err; + } + + pPg->pPager = pPager; + + assert( !isOpen(pPager->fd) || !MEMDB ); + if( !isOpen(pPager->fd) || pPager->dbSizepPager->mxPgno ){ + rc = SQLITE_FULL; + goto pager_acquire_err; + } + if( noContent ){ + /* Failure to set the bits in the InJournal bit-vectors is benign. + ** It merely means that we might do some extra work to journal a + ** page that does not need to be journaled. Nevertheless, be sure + ** to test the case where a malloc error occurs while trying to set + ** a bit in a bit vector. + */ + sqlite3BeginBenignMalloc(); + if( pgno<=pPager->dbOrigSize ){ + TESTONLY( rc = ) sqlite3BitvecSet(pPager->pInJournal, pgno); + testcase( rc==SQLITE_NOMEM ); + } + TESTONLY( rc = ) addToSavepointBitvecs(pPager, pgno); + testcase( rc==SQLITE_NOMEM ); + sqlite3EndBenignMalloc(); + } + memset(pPg->pData, 0, pPager->pageSize); + IOTRACE(("ZERO %p %d\n", pPager, pgno)); + }else{ + assert( pPg->pPager==pPager ); + pPager->aStat[PAGER_STAT_MISS]++; + rc = readDbPage(pPg); + if( rc!=SQLITE_OK ){ + goto pager_acquire_err; + } + } + pager_set_pagehash(pPg); + } + return SQLITE_OK; + +pager_acquire_err: + assert( rc!=SQLITE_OK ); + if( pPg ){ + sqlite3PcacheDrop(pPg); + } + pagerUnlockIfUnused(pPager); + *ppPage = 0; + return rc; +} + +#if SQLITE_MAX_MMAP_SIZE>0 +/* The page getter for when memory-mapped I/O is enabled */ +static int getPageMMap( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + int rc = SQLITE_OK; + PgHdr *pPg = 0; + u32 iFrame = 0; /* Frame to read from WAL file */ + + /* It is acceptable to use a read-only (mmap) page for any page except + ** page 1 if there is no write-transaction open or the ACQUIRE_READONLY + ** flag was specified by the caller. And so long as the db is not a + ** temporary or in-memory database. */ + const int bMmapOk = (pgno>1 + && (pPager->eState==PAGER_READER || (flags & PAGER_GET_READONLY)) + ); + + assert( USEFETCH(pPager) ); + + /* Optimization note: Adding the "pgno<=1" term before "pgno==0" here + ** allows the compiler optimizer to reuse the results of the "pgno>1" + ** test in the previous statement, and avoid testing pgno==0 in the + ** common case where pgno is large. */ + if( pgno<=1 && pgno==0 ){ + return SQLITE_CORRUPT_BKPT; + } + assert( pPager->eState>=PAGER_READER ); + assert( assert_pager_state(pPager) ); + assert( pPager->hasHeldSharedLock==1 ); + assert( pPager->errCode==SQLITE_OK ); + + if( bMmapOk && pagerUseWal(pPager) ){ + rc = sqlite3WalFindFrame(pPager->pWal, pgno, &iFrame); + if( rc!=SQLITE_OK ){ + *ppPage = 0; + return rc; + } + } + if( bMmapOk && iFrame==0 ){ + void *pData = 0; + rc = sqlite3OsFetch(pPager->fd, + (i64)(pgno-1) * pPager->pageSize, pPager->pageSize, &pData + ); + if( rc==SQLITE_OK && pData ){ + if( pPager->eState>PAGER_READER || pPager->tempFile ){ + pPg = sqlite3PagerLookup(pPager, pgno); + } + if( pPg==0 ){ + rc = pagerAcquireMapPage(pPager, pgno, pData, &pPg); + }else{ + sqlite3OsUnfetch(pPager->fd, (i64)(pgno-1)*pPager->pageSize, pData); + } + if( pPg ){ + assert( rc==SQLITE_OK ); + *ppPage = pPg; + return SQLITE_OK; + } + } + if( rc!=SQLITE_OK ){ + *ppPage = 0; + return rc; + } + } + return getPageNormal(pPager, pgno, ppPage, flags); +} +#endif /* SQLITE_MAX_MMAP_SIZE>0 */ + +/* The page getter method for when the pager is an error state */ +static int getPageError( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + UNUSED_PARAMETER(pgno); + UNUSED_PARAMETER(flags); + assert( pPager->errCode!=SQLITE_OK ); + *ppPage = 0; + return pPager->errCode; +} + + +/* Dispatch all page fetch requests to the appropriate getter method. +*/ +int sqlite3PagerGet( + Pager *pPager, /* The pager open on the database file */ + Pgno pgno, /* Page number to fetch */ + DbPage **ppPage, /* Write a pointer to the page here */ + int flags /* PAGER_GET_XXX flags */ +){ + return pPager->xGet(pPager, pgno, ppPage, flags); +} + +/* +** Acquire a page if it is already in the in-memory cache. Do +** not read the page from disk. Return a pointer to the page, +** or 0 if the page is not in cache. +** +** See also sqlite3PagerGet(). The difference between this routine +** and sqlite3PagerGet() is that _get() will go to the disk and read +** in the page if the page is not already in cache. This routine +** returns NULL if the page is not in cache or if a disk I/O error +** has ever happened. +*/ +DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno){ + sqlite3_pcache_page *pPage; + assert( pPager!=0 ); + assert( pgno!=0 ); + assert( pPager->pPCache!=0 ); + pPage = sqlite3PcacheFetch(pPager->pPCache, pgno, 0); + assert( pPage==0 || pPager->hasHeldSharedLock ); + if( pPage==0 ) return 0; + return sqlite3PcacheFetchFinish(pPager->pPCache, pgno, pPage); +} + +/* +** Release a page reference. +** +** The sqlite3PagerUnref() and sqlite3PagerUnrefNotNull() may only be +** used if we know that the page being released is not the last page. +** The btree layer always holds page1 open until the end, so these first +** to routines can be used to release any page other than BtShared.pPage1. +** +** Use sqlite3PagerUnrefPageOne() to release page1. This latter routine +** checks the total number of outstanding pages and if the number of +** pages reaches zero it drops the database lock. +*/ +void sqlite3PagerUnrefNotNull(DbPage *pPg){ + TESTONLY( Pager *pPager = pPg->pPager; ) + assert( pPg!=0 ); + if( pPg->flags & PGHDR_MMAP ){ + assert( pPg->pgno!=1 ); /* Page1 is never memory mapped */ + pagerReleaseMapPage(pPg); + }else{ + sqlite3PcacheRelease(pPg); + } + /* Do not use this routine to release the last reference to page1 */ + assert( sqlite3PcacheRefCount(pPager->pPCache)>0 ); +} +void sqlite3PagerUnref(DbPage *pPg){ + if( pPg ) sqlite3PagerUnrefNotNull(pPg); +} +void sqlite3PagerUnrefPageOne(DbPage *pPg){ + Pager *pPager; + assert( pPg!=0 ); + assert( pPg->pgno==1 ); + assert( (pPg->flags & PGHDR_MMAP)==0 ); /* Page1 is never memory mapped */ + pPager = pPg->pPager; + sqlite3PcacheRelease(pPg); + pagerUnlockIfUnused(pPager); +} + +/* +** This function is called at the start of every write transaction. +** There must already be a RESERVED or EXCLUSIVE lock on the database +** file when this routine is called. +** +** Open the journal file for pager pPager and write a journal header +** to the start of it. If there are active savepoints, open the sub-journal +** as well. This function is only used when the journal file is being +** opened to write a rollback log for a transaction. It is not used +** when opening a hot journal file to roll it back. +** +** If the journal file is already open (as it may be in exclusive mode), +** then this function just writes a journal header to the start of the +** already open file. +** +** Whether or not the journal file is opened by this function, the +** Pager.pInJournal bitvec structure is allocated. +** +** Return SQLITE_OK if everything is successful. Otherwise, return +** SQLITE_NOMEM if the attempt to allocate Pager.pInJournal fails, or +** an IO error code if opening or writing the journal file fails. +*/ +static int pager_open_journal(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + sqlite3_vfs * const pVfs = pPager->pVfs; /* Local cache of vfs pointer */ + + assert( pPager->eState==PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + assert( pPager->pInJournal==0 ); + + /* If already in the error state, this function is a no-op. But on + ** the other hand, this routine is never called if we are already in + ** an error state. */ + if( NEVER(pPager->errCode) ) return pPager->errCode; + + if( !pagerUseWal(pPager) && pPager->journalMode!=PAGER_JOURNALMODE_OFF ){ + pPager->pInJournal = sqlite3BitvecCreate(pPager->dbSize); + if( pPager->pInJournal==0 ){ + return SQLITE_NOMEM_BKPT; + } + + /* Open the journal file if it is not already open. */ + if( !isOpen(pPager->jfd) ){ + if( pPager->journalMode==PAGER_JOURNALMODE_MEMORY ){ + sqlite3MemJournalOpen(pPager->jfd); + }else{ + int flags = SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE; + int nSpill; + + if( pPager->tempFile ){ + flags |= (SQLITE_OPEN_DELETEONCLOSE|SQLITE_OPEN_TEMP_JOURNAL); + nSpill = sqlite3Config.nStmtSpill; + }else{ + flags |= SQLITE_OPEN_MAIN_JOURNAL; + nSpill = jrnlBufferSize(pPager); + } + + /* Verify that the database still has the same name as it did when + ** it was originally opened. */ + rc = databaseIsUnmoved(pPager); + if( rc==SQLITE_OK ){ + rc = sqlite3JournalOpen ( + pVfs, pPager->zJournal, pPager->jfd, flags, nSpill + ); + } + } + assert( rc!=SQLITE_OK || isOpen(pPager->jfd) ); + } + + + /* Write the first journal header to the journal file and open + ** the sub-journal if necessary. + */ + if( rc==SQLITE_OK ){ + /* TODO: Check if all of these are really required. */ + pPager->nRec = 0; + pPager->journalOff = 0; + pPager->setSuper = 0; + pPager->journalHdr = 0; + rc = writeJournalHdr(pPager); + } + } + + if( rc!=SQLITE_OK ){ + sqlite3BitvecDestroy(pPager->pInJournal); + pPager->pInJournal = 0; + }else{ + assert( pPager->eState==PAGER_WRITER_LOCKED ); + pPager->eState = PAGER_WRITER_CACHEMOD; + } + + return rc; +} + +/* +** Begin a write-transaction on the specified pager object. If a +** write-transaction has already been opened, this function is a no-op. +** +** If the exFlag argument is false, then acquire at least a RESERVED +** lock on the database file. If exFlag is true, then acquire at least +** an EXCLUSIVE lock. If such a lock is already held, no locking +** functions need be called. +** +** If the subjInMemory argument is non-zero, then any sub-journal opened +** within this transaction will be opened as an in-memory file. This +** has no effect if the sub-journal is already opened (as it may be when +** running in exclusive mode) or if the transaction does not require a +** sub-journal. If the subjInMemory argument is zero, then any required +** sub-journal is implemented in-memory if pPager is an in-memory database, +** or using a temporary file otherwise. +*/ +int sqlite3PagerBegin(Pager *pPager, int exFlag, int subjInMemory){ + int rc = SQLITE_OK; + + if( pPager->errCode ) return pPager->errCode; + assert( pPager->eState>=PAGER_READER && pPager->eStatesubjInMemory = (u8)subjInMemory; + + if( pPager->eState==PAGER_READER ){ + assert( pPager->pInJournal==0 ); + + if( pagerUseWal(pPager) ){ + /* If the pager is configured to use locking_mode=exclusive, and an + ** exclusive lock on the database is not already held, obtain it now. + */ + if( pPager->exclusiveMode && sqlite3WalExclusiveMode(pPager->pWal, -1) ){ + rc = pagerLockDb(pPager, EXCLUSIVE_LOCK); + if( rc!=SQLITE_OK ){ + return rc; + } + (void)sqlite3WalExclusiveMode(pPager->pWal, 1); + } + + /* Grab the write lock on the log file. If successful, upgrade to + ** PAGER_RESERVED state. Otherwise, return an error code to the caller. + ** The busy-handler is not invoked if another connection already + ** holds the write-lock. If possible, the upper layer will call it. + */ + rc = sqlite3WalBeginWriteTransaction(pPager->pWal); + }else{ + /* Obtain a RESERVED lock on the database file. If the exFlag parameter + ** is true, then immediately upgrade this to an EXCLUSIVE lock. The + ** busy-handler callback can be used when upgrading to the EXCLUSIVE + ** lock, but not when obtaining the RESERVED lock. + */ + rc = pagerLockDb(pPager, RESERVED_LOCK); + if( rc==SQLITE_OK && exFlag ){ + rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); + } + } + + if( rc==SQLITE_OK ){ + /* Change to WRITER_LOCKED state. + ** + ** WAL mode sets Pager.eState to PAGER_WRITER_LOCKED or CACHEMOD + ** when it has an open transaction, but never to DBMOD or FINISHED. + ** This is because in those states the code to roll back savepoint + ** transactions may copy data from the sub-journal into the database + ** file as well as into the page cache. Which would be incorrect in + ** WAL mode. + */ + pPager->eState = PAGER_WRITER_LOCKED; + pPager->dbHintSize = pPager->dbSize; + pPager->dbFileSize = pPager->dbSize; + pPager->dbOrigSize = pPager->dbSize; + pPager->journalOff = 0; + } + + assert( rc==SQLITE_OK || pPager->eState==PAGER_READER ); + assert( rc!=SQLITE_OK || pPager->eState==PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + } + + PAGERTRACE(("TRANSACTION %d\n", PAGERID(pPager))); + return rc; +} + +/* +** Write page pPg onto the end of the rollback journal. +*/ +static SQLITE_NOINLINE int pagerAddPageToRollbackJournal(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + int rc; + u32 cksum; + char *pData2; + i64 iOff = pPager->journalOff; + + /* We should never write to the journal file the page that + ** contains the database locks. The following assert verifies + ** that we do not. */ + assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) ); + + assert( pPager->journalHdr<=pPager->journalOff ); + pData2 = pPg->pData; + cksum = pager_cksum(pPager, (u8*)pData2); + + /* Even if an IO or diskfull error occurs while journalling the + ** page in the block above, set the need-sync flag for the page. + ** Otherwise, when the transaction is rolled back, the logic in + ** playback_one_page() will think that the page needs to be restored + ** in the database file. And if an IO error occurs while doing so, + ** then corruption may follow. + */ + pPg->flags |= PGHDR_NEED_SYNC; + + rc = write32bits(pPager->jfd, iOff, pPg->pgno); + if( rc!=SQLITE_OK ) return rc; + rc = sqlite3OsWrite(pPager->jfd, pData2, pPager->pageSize, iOff+4); + if( rc!=SQLITE_OK ) return rc; + rc = write32bits(pPager->jfd, iOff+pPager->pageSize+4, cksum); + if( rc!=SQLITE_OK ) return rc; + + IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno, + pPager->journalOff, pPager->pageSize)); + PAGER_INCR(sqlite3_pager_writej_count); + PAGERTRACE(("JOURNAL %d page %d needSync=%d hash(%08x)\n", + PAGERID(pPager), pPg->pgno, + ((pPg->flags&PGHDR_NEED_SYNC)?1:0), pager_pagehash(pPg))); + + pPager->journalOff += 8 + pPager->pageSize; + pPager->nRec++; + assert( pPager->pInJournal!=0 ); + rc = sqlite3BitvecSet(pPager->pInJournal, pPg->pgno); + testcase( rc==SQLITE_NOMEM ); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + rc |= addToSavepointBitvecs(pPager, pPg->pgno); + assert( rc==SQLITE_OK || rc==SQLITE_NOMEM ); + return rc; +} + +/* +** Mark a single data page as writeable. The page is written into the +** main journal or sub-journal as required. If the page is written into +** one of the journals, the corresponding bit is set in the +** Pager.pInJournal bitvec and the PagerSavepoint.pInSavepoint bitvecs +** of any open savepoints as appropriate. +*/ +static int pager_write(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + int rc = SQLITE_OK; + + /* This routine is not called unless a write-transaction has already + ** been started. The journal file may or may not be open at this point. + ** It is never called in the ERROR state. + */ + assert( pPager->eState==PAGER_WRITER_LOCKED + || pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + assert( pPager->errCode==0 ); + assert( pPager->readOnly==0 ); + CHECK_PAGE(pPg); + + /* The journal file needs to be opened. Higher level routines have already + ** obtained the necessary locks to begin the write-transaction, but the + ** rollback journal might not yet be open. Open it now if this is the case. + ** + ** This is done before calling sqlite3PcacheMakeDirty() on the page. + ** Otherwise, if it were done after calling sqlite3PcacheMakeDirty(), then + ** an error might occur and the pager would end up in WRITER_LOCKED state + ** with pages marked as dirty in the cache. + */ + if( pPager->eState==PAGER_WRITER_LOCKED ){ + rc = pager_open_journal(pPager); + if( rc!=SQLITE_OK ) return rc; + } + assert( pPager->eState>=PAGER_WRITER_CACHEMOD ); + assert( assert_pager_state(pPager) ); + + /* Mark the page that is about to be modified as dirty. */ + sqlite3PcacheMakeDirty(pPg); + + /* If a rollback journal is in use, them make sure the page that is about + ** to change is in the rollback journal, or if the page is a new page off + ** then end of the file, make sure it is marked as PGHDR_NEED_SYNC. + */ + assert( (pPager->pInJournal!=0) == isOpen(pPager->jfd) ); + if( pPager->pInJournal!=0 + && sqlite3BitvecTestNotNull(pPager->pInJournal, pPg->pgno)==0 + ){ + assert( pagerUseWal(pPager)==0 ); + if( pPg->pgno<=pPager->dbOrigSize ){ + rc = pagerAddPageToRollbackJournal(pPg); + if( rc!=SQLITE_OK ){ + return rc; + } + }else{ + if( pPager->eState!=PAGER_WRITER_DBMOD ){ + pPg->flags |= PGHDR_NEED_SYNC; + } + PAGERTRACE(("APPEND %d page %d needSync=%d\n", + PAGERID(pPager), pPg->pgno, + ((pPg->flags&PGHDR_NEED_SYNC)?1:0))); + } + } + + /* The PGHDR_DIRTY bit is set above when the page was added to the dirty-list + ** and before writing the page into the rollback journal. Wait until now, + ** after the page has been successfully journalled, before setting the + ** PGHDR_WRITEABLE bit that indicates that the page can be safely modified. + */ + pPg->flags |= PGHDR_WRITEABLE; + + /* If the statement journal is open and the page is not in it, + ** then write the page into the statement journal. + */ + if( pPager->nSavepoint>0 ){ + rc = subjournalPageIfRequired(pPg); + } + + /* Update the database size and return. */ + if( pPager->dbSizepgno ){ + pPager->dbSize = pPg->pgno; + } + return rc; +} + +/* +** This is a variant of sqlite3PagerWrite() that runs when the sector size +** is larger than the page size. SQLite makes the (reasonable) assumption that +** all bytes of a sector are written together by hardware. Hence, all bytes of +** a sector need to be journalled in case of a power loss in the middle of +** a write. +** +** Usually, the sector size is less than or equal to the page size, in which +** case pages can be individually written. This routine only runs in the +** exceptional case where the page size is smaller than the sector size. +*/ +static SQLITE_NOINLINE int pagerWriteLargeSector(PgHdr *pPg){ + int rc = SQLITE_OK; /* Return code */ + Pgno nPageCount; /* Total number of pages in database file */ + Pgno pg1; /* First page of the sector pPg is located on. */ + int nPage = 0; /* Number of pages starting at pg1 to journal */ + int ii; /* Loop counter */ + int needSync = 0; /* True if any page has PGHDR_NEED_SYNC */ + Pager *pPager = pPg->pPager; /* The pager that owns pPg */ + Pgno nPagePerSector = (pPager->sectorSize/pPager->pageSize); + + /* Set the doNotSpill NOSYNC bit to 1. This is because we cannot allow + ** a journal header to be written between the pages journaled by + ** this function. + */ + assert( !MEMDB ); + assert( (pPager->doNotSpill & SPILLFLAG_NOSYNC)==0 ); + pPager->doNotSpill |= SPILLFLAG_NOSYNC; + + /* This trick assumes that both the page-size and sector-size are + ** an integer power of 2. It sets variable pg1 to the identifier + ** of the first page of the sector pPg is located on. + */ + pg1 = ((pPg->pgno-1) & ~(nPagePerSector-1)) + 1; + + nPageCount = pPager->dbSize; + if( pPg->pgno>nPageCount ){ + nPage = (pPg->pgno - pg1)+1; + }else if( (pg1+nPagePerSector-1)>nPageCount ){ + nPage = nPageCount+1-pg1; + }else{ + nPage = nPagePerSector; + } + assert(nPage>0); + assert(pg1<=pPg->pgno); + assert((pg1+nPage)>pPg->pgno); + + for(ii=0; iipgno || !sqlite3BitvecTest(pPager->pInJournal, pg) ){ + if( pg!=PAGER_MJ_PGNO(pPager) ){ + rc = sqlite3PagerGet(pPager, pg, &pPage, 0); + if( rc==SQLITE_OK ){ + rc = pager_write(pPage); + if( pPage->flags&PGHDR_NEED_SYNC ){ + needSync = 1; + } + sqlite3PagerUnrefNotNull(pPage); + } + } + }else if( (pPage = sqlite3PagerLookup(pPager, pg))!=0 ){ + if( pPage->flags&PGHDR_NEED_SYNC ){ + needSync = 1; + } + sqlite3PagerUnrefNotNull(pPage); + } + } + + /* If the PGHDR_NEED_SYNC flag is set for any of the nPage pages + ** starting at pg1, then it needs to be set for all of them. Because + ** writing to any of these nPage pages may damage the others, the + ** journal file must contain sync()ed copies of all of them + ** before any of them can be written out to the database file. + */ + if( rc==SQLITE_OK && needSync ){ + assert( !MEMDB ); + for(ii=0; iiflags |= PGHDR_NEED_SYNC; + sqlite3PagerUnrefNotNull(pPage); + } + } + } + + assert( (pPager->doNotSpill & SPILLFLAG_NOSYNC)!=0 ); + pPager->doNotSpill &= ~SPILLFLAG_NOSYNC; + return rc; +} + +/* +** Mark a data page as writeable. This routine must be called before +** making changes to a page. The caller must check the return value +** of this function and be careful not to change any page data unless +** this routine returns SQLITE_OK. +** +** The difference between this function and pager_write() is that this +** function also deals with the special case where 2 or more pages +** fit on a single disk sector. In this case all co-resident pages +** must have been written to the journal file before returning. +** +** If an error occurs, SQLITE_NOMEM or an IO error code is returned +** as appropriate. Otherwise, SQLITE_OK. +*/ +int sqlite3PagerWrite(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + assert( (pPg->flags & PGHDR_MMAP)==0 ); + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + if( (pPg->flags & PGHDR_WRITEABLE)!=0 && pPager->dbSize>=pPg->pgno ){ + if( pPager->nSavepoint ) return subjournalPageIfRequired(pPg); + return SQLITE_OK; + }else if( pPager->errCode ){ + return pPager->errCode; + }else if( pPager->sectorSize > (u32)pPager->pageSize ){ + assert( pPager->tempFile==0 ); + return pagerWriteLargeSector(pPg); + }else{ + return pager_write(pPg); + } +} + +/* +** Return TRUE if the page given in the argument was previously passed +** to sqlite3PagerWrite(). In other words, return TRUE if it is ok +** to change the content of the page. +*/ +#ifndef NDEBUG +int sqlite3PagerIswriteable(DbPage *pPg){ + return pPg->flags & PGHDR_WRITEABLE; +} +#endif + +/* +** A call to this routine tells the pager that it is not necessary to +** write the information on page pPg back to the disk, even though +** that page might be marked as dirty. This happens, for example, when +** the page has been added as a leaf of the freelist and so its +** content no longer matters. +** +** The overlying software layer calls this routine when all of the data +** on the given page is unused. The pager marks the page as clean so +** that it does not get written to disk. +** +** Tests show that this optimization can quadruple the speed of large +** DELETE operations. +** +** This optimization cannot be used with a temp-file, as the page may +** have been dirty at the start of the transaction. In that case, if +** memory pressure forces page pPg out of the cache, the data does need +** to be written out to disk so that it may be read back in if the +** current transaction is rolled back. +*/ +void sqlite3PagerDontWrite(PgHdr *pPg){ + Pager *pPager = pPg->pPager; + if( !pPager->tempFile && (pPg->flags&PGHDR_DIRTY) && pPager->nSavepoint==0 ){ + PAGERTRACE(("DONT_WRITE page %d of %d\n", pPg->pgno, PAGERID(pPager))); + IOTRACE(("CLEAN %p %d\n", pPager, pPg->pgno)) + pPg->flags |= PGHDR_DONT_WRITE; + pPg->flags &= ~PGHDR_WRITEABLE; + testcase( pPg->flags & PGHDR_NEED_SYNC ); + pager_set_pagehash(pPg); + } +} + +/* +** This routine is called to increment the value of the database file +** change-counter, stored as a 4-byte big-endian integer starting at +** byte offset 24 of the pager file. The secondary change counter at +** 92 is also updated, as is the SQLite version number at offset 96. +** +** But this only happens if the pPager->changeCountDone flag is false. +** To avoid excess churning of page 1, the update only happens once. +** See also the pager_write_changecounter() routine that does an +** unconditional update of the change counters. +** +** If the isDirectMode flag is zero, then this is done by calling +** sqlite3PagerWrite() on page 1, then modifying the contents of the +** page data. In this case the file will be updated when the current +** transaction is committed. +** +** The isDirectMode flag may only be non-zero if the library was compiled +** with the SQLITE_ENABLE_ATOMIC_WRITE macro defined. In this case, +** if isDirect is non-zero, then the database file is updated directly +** by writing an updated version of page 1 using a call to the +** sqlite3OsWrite() function. +*/ +static int pager_incr_changecounter(Pager *pPager, int isDirectMode){ + int rc = SQLITE_OK; + + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + + /* Declare and initialize constant integer 'isDirect'. If the + ** atomic-write optimization is enabled in this build, then isDirect + ** is initialized to the value passed as the isDirectMode parameter + ** to this function. Otherwise, it is always set to zero. + ** + ** The idea is that if the atomic-write optimization is not + ** enabled at compile time, the compiler can omit the tests of + ** 'isDirect' below, as well as the block enclosed in the + ** "if( isDirect )" condition. + */ +#ifndef SQLITE_ENABLE_ATOMIC_WRITE +# define DIRECT_MODE 0 + assert( isDirectMode==0 ); + UNUSED_PARAMETER(isDirectMode); +#else +# define DIRECT_MODE isDirectMode +#endif + + if( !pPager->changeCountDone && ALWAYS(pPager->dbSize>0) ){ + PgHdr *pPgHdr; /* Reference to page 1 */ + + assert( !pPager->tempFile && isOpen(pPager->fd) ); + + /* Open page 1 of the file for writing. */ + rc = sqlite3PagerGet(pPager, 1, &pPgHdr, 0); + assert( pPgHdr==0 || rc==SQLITE_OK ); + + /* If page one was fetched successfully, and this function is not + ** operating in direct-mode, make page 1 writable. When not in + ** direct mode, page 1 is always held in cache and hence the PagerGet() + ** above is always successful - hence the ALWAYS on rc==SQLITE_OK. + */ + if( !DIRECT_MODE && ALWAYS(rc==SQLITE_OK) ){ + rc = sqlite3PagerWrite(pPgHdr); + } + + if( rc==SQLITE_OK ){ + /* Actually do the update of the change counter */ + pager_write_changecounter(pPgHdr); + + /* If running in direct mode, write the contents of page 1 to the file. */ + if( DIRECT_MODE ){ + const void *zBuf; + assert( pPager->dbFileSize>0 ); + zBuf = pPgHdr->pData; + if( rc==SQLITE_OK ){ + rc = sqlite3OsWrite(pPager->fd, zBuf, pPager->pageSize, 0); + pPager->aStat[PAGER_STAT_WRITE]++; + } + if( rc==SQLITE_OK ){ + /* Update the pager's copy of the change-counter. Otherwise, the + ** next time a read transaction is opened the cache will be + ** flushed (as the change-counter values will not match). */ + const void *pCopy = (const void *)&((const char *)zBuf)[24]; + memcpy(&pPager->dbFileVers, pCopy, sizeof(pPager->dbFileVers)); + pPager->changeCountDone = 1; + } + }else{ + pPager->changeCountDone = 1; + } + } + + /* Release the page reference. */ + sqlite3PagerUnref(pPgHdr); + } + return rc; +} + +/* +** Sync the database file to disk. This is a no-op for in-memory databases +** or pages with the Pager.noSync flag set. +** +** If successful, or if called on a pager for which it is a no-op, this +** function returns SQLITE_OK. Otherwise, an IO error code is returned. +*/ +int sqlite3PagerSync(Pager *pPager, const char *zSuper){ + int rc = SQLITE_OK; + void *pArg = (void*)zSuper; + rc = sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_SYNC, pArg); + if( rc==SQLITE_NOTFOUND ) rc = SQLITE_OK; + if( rc==SQLITE_OK && !pPager->noSync ){ + assert( !MEMDB ); + rc = sqlite3OsSync(pPager->fd, pPager->syncFlags); + } + return rc; +} + +/* +** This function may only be called while a write-transaction is active in +** rollback. If the connection is in WAL mode, this call is a no-op. +** Otherwise, if the connection does not already have an EXCLUSIVE lock on +** the database file, an attempt is made to obtain one. +** +** If the EXCLUSIVE lock is already held or the attempt to obtain it is +** successful, or the connection is in WAL mode, SQLITE_OK is returned. +** Otherwise, either SQLITE_BUSY or an SQLITE_IOERR_XXX error code is +** returned. +*/ +int sqlite3PagerExclusiveLock(Pager *pPager){ + int rc = pPager->errCode; + assert( assert_pager_state(pPager) ); + if( rc==SQLITE_OK ){ + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + || pPager->eState==PAGER_WRITER_LOCKED + ); + assert( assert_pager_state(pPager) ); + if( 0==pagerUseWal(pPager) ){ + rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK); + } + } + return rc; +} + +/* +** Sync the database file for the pager pPager. zSuper points to the name +** of a super-journal file that should be written into the individual +** journal file. zSuper may be NULL, which is interpreted as no +** super-journal (a single database transaction). +** +** This routine ensures that: +** +** * The database file change-counter is updated, +** * the journal is synced (unless the atomic-write optimization is used), +** * all dirty pages are written to the database file, +** * the database file is truncated (if required), and +** * the database file synced. +** +** The only thing that remains to commit the transaction is to finalize +** (delete, truncate or zero the first part of) the journal file (or +** delete the super-journal file if specified). +** +** Note that if zSuper==NULL, this does not overwrite a previous value +** passed to an sqlite3PagerCommitPhaseOne() call. +** +** If the final parameter - noSync - is true, then the database file itself +** is not synced. The caller must call sqlite3PagerSync() directly to +** sync the database file before calling CommitPhaseTwo() to delete the +** journal file in this case. +*/ +int sqlite3PagerCommitPhaseOne( + Pager *pPager, /* Pager object */ + const char *zSuper, /* If not NULL, the super-journal name */ + int noSync /* True to omit the xSync on the db file */ +){ + int rc = SQLITE_OK; /* Return code */ + + assert( pPager->eState==PAGER_WRITER_LOCKED + || pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + || pPager->eState==PAGER_ERROR + ); + assert( assert_pager_state(pPager) ); + + /* If a prior error occurred, report that error again. */ + if( NEVER(pPager->errCode) ) return pPager->errCode; + + /* Provide the ability to easily simulate an I/O error during testing */ + if( sqlite3FaultSim(400) ) return SQLITE_IOERR; + + PAGERTRACE(("DATABASE SYNC: File=%s zSuper=%s nSize=%d\n", + pPager->zFilename, zSuper, pPager->dbSize)); + + /* If no database changes have been made, return early. */ + if( pPager->eStatetempFile ); + assert( isOpen(pPager->fd) || pPager->tempFile ); + if( 0==pagerFlushOnCommit(pPager, 1) ){ + /* If this is an in-memory db, or no pages have been written to, or this + ** function has already been called, it is mostly a no-op. However, any + ** backup in progress needs to be restarted. */ + sqlite3BackupRestart(pPager->pBackup); + }else{ + PgHdr *pList; + if( pagerUseWal(pPager) ){ + PgHdr *pPageOne = 0; + pList = sqlite3PcacheDirtyList(pPager->pPCache); + if( pList==0 ){ + /* Must have at least one page for the WAL commit flag. + ** Ticket [2d1a5c67dfc2363e44f29d9bbd57f] 2011-05-18 */ + rc = sqlite3PagerGet(pPager, 1, &pPageOne, 0); + pList = pPageOne; + pList->pDirty = 0; + } + assert( rc==SQLITE_OK ); + if( ALWAYS(pList) ){ + rc = pagerWalFrames(pPager, pList, pPager->dbSize, 1); + } + sqlite3PagerUnref(pPageOne); + if( rc==SQLITE_OK ){ + sqlite3PcacheCleanAll(pPager->pPCache); + } + }else{ + /* The bBatch boolean is true if the batch-atomic-write commit method + ** should be used. No rollback journal is created if batch-atomic-write + ** is enabled. + */ +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + sqlite3_file *fd = pPager->fd; + int bBatch = zSuper==0 /* An SQLITE_IOCAP_BATCH_ATOMIC commit */ + && (sqlite3OsDeviceCharacteristics(fd) & SQLITE_IOCAP_BATCH_ATOMIC) + && !pPager->noSync + && sqlite3JournalIsInMemory(pPager->jfd); +#else +# define bBatch 0 +#endif + +#ifdef SQLITE_ENABLE_ATOMIC_WRITE + /* The following block updates the change-counter. Exactly how it + ** does this depends on whether or not the atomic-update optimization + ** was enabled at compile time, and if this transaction meets the + ** runtime criteria to use the operation: + ** + ** * The file-system supports the atomic-write property for + ** blocks of size page-size, and + ** * This commit is not part of a multi-file transaction, and + ** * Exactly one page has been modified and store in the journal file. + ** + ** If the optimization was not enabled at compile time, then the + ** pager_incr_changecounter() function is called to update the change + ** counter in 'indirect-mode'. If the optimization is compiled in but + ** is not applicable to this transaction, call sqlite3JournalCreate() + ** to make sure the journal file has actually been created, then call + ** pager_incr_changecounter() to update the change-counter in indirect + ** mode. + ** + ** Otherwise, if the optimization is both enabled and applicable, + ** then call pager_incr_changecounter() to update the change-counter + ** in 'direct' mode. In this case the journal file will never be + ** created for this transaction. + */ + if( bBatch==0 ){ + PgHdr *pPg; + assert( isOpen(pPager->jfd) + || pPager->journalMode==PAGER_JOURNALMODE_OFF + || pPager->journalMode==PAGER_JOURNALMODE_WAL + ); + if( !zSuper && isOpen(pPager->jfd) + && pPager->journalOff==jrnlBufferSize(pPager) + && pPager->dbSize>=pPager->dbOrigSize + && (!(pPg = sqlite3PcacheDirtyList(pPager->pPCache)) || 0==pPg->pDirty) + ){ + /* Update the db file change counter via the direct-write method. The + ** following call will modify the in-memory representation of page 1 + ** to include the updated change counter and then write page 1 + ** directly to the database file. Because of the atomic-write + ** property of the host file-system, this is safe. + */ + rc = pager_incr_changecounter(pPager, 1); + }else{ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc==SQLITE_OK ){ + rc = pager_incr_changecounter(pPager, 0); + } + } + } +#else /* SQLITE_ENABLE_ATOMIC_WRITE */ +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( zSuper ){ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + assert( bBatch==0 ); + } +#endif + rc = pager_incr_changecounter(pPager, 0); +#endif /* !SQLITE_ENABLE_ATOMIC_WRITE */ + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + + /* Write the super-journal name into the journal file. If a + ** super-journal file name has already been written to the journal file, + ** or if zSuper is NULL (no super-journal), then this call is a no-op. + */ + rc = writeSuperJournal(pPager, zSuper); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + + /* Sync the journal file and write all dirty pages to the database. + ** If the atomic-update optimization is being used, this sync will not + ** create the journal file or perform any real IO. + ** + ** Because the change-counter page was just modified, unless the + ** atomic-update optimization is used it is almost certain that the + ** journal requires a sync here. However, in locking_mode=exclusive + ** on a system under memory pressure it is just possible that this is + ** not the case. In this case it is likely enough that the redundant + ** xSync() call will be changed to a no-op by the OS anyhow. + */ + rc = syncJournal(pPager, 0); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + + pList = sqlite3PcacheDirtyList(pPager->pPCache); +#ifdef SQLITE_ENABLE_BATCH_ATOMIC_WRITE + if( bBatch ){ + rc = sqlite3OsFileControl(fd, SQLITE_FCNTL_BEGIN_ATOMIC_WRITE, 0); + if( rc==SQLITE_OK ){ + rc = pager_write_pagelist(pPager, pList); + if( rc==SQLITE_OK ){ + rc = sqlite3OsFileControl(fd, SQLITE_FCNTL_COMMIT_ATOMIC_WRITE, 0); + } + if( rc!=SQLITE_OK ){ + sqlite3OsFileControlHint(fd, SQLITE_FCNTL_ROLLBACK_ATOMIC_WRITE, 0); + } + } + + if( (rc&0xFF)==SQLITE_IOERR && rc!=SQLITE_IOERR_NOMEM ){ + rc = sqlite3JournalCreate(pPager->jfd); + if( rc!=SQLITE_OK ){ + sqlite3OsClose(pPager->jfd); + goto commit_phase_one_exit; + } + bBatch = 0; + }else{ + sqlite3OsClose(pPager->jfd); + } + } +#endif /* SQLITE_ENABLE_BATCH_ATOMIC_WRITE */ + + if( bBatch==0 ){ + rc = pager_write_pagelist(pPager, pList); + } + if( rc!=SQLITE_OK ){ + assert( rc!=SQLITE_IOERR_BLOCKED ); + goto commit_phase_one_exit; + } + sqlite3PcacheCleanAll(pPager->pPCache); + + /* If the file on disk is smaller than the database image, use + ** pager_truncate to grow the file here. This can happen if the database + ** image was extended as part of the current transaction and then the + ** last page in the db image moved to the free-list. In this case the + ** last page is never written out to disk, leaving the database file + ** undersized. Fix this now if it is the case. */ + if( pPager->dbSize>pPager->dbFileSize ){ + Pgno nNew = pPager->dbSize - (pPager->dbSize==PAGER_MJ_PGNO(pPager)); + assert( pPager->eState==PAGER_WRITER_DBMOD ); + rc = pager_truncate(pPager, nNew); + if( rc!=SQLITE_OK ) goto commit_phase_one_exit; + } + + /* Finally, sync the database file. */ + if( !noSync ){ + rc = sqlite3PagerSync(pPager, zSuper); + } + IOTRACE(("DBSYNC %p\n", pPager)) + } + } + +commit_phase_one_exit: + if( rc==SQLITE_OK && !pagerUseWal(pPager) ){ + pPager->eState = PAGER_WRITER_FINISHED; + } + return rc; +} + + +/* +** When this function is called, the database file has been completely +** updated to reflect the changes made by the current transaction and +** synced to disk. The journal file still exists in the file-system +** though, and if a failure occurs at this point it will eventually +** be used as a hot-journal and the current transaction rolled back. +** +** This function finalizes the journal file, either by deleting, +** truncating or partially zeroing it, so that it cannot be used +** for hot-journal rollback. Once this is done the transaction is +** irrevocably committed. +** +** If an error occurs, an IO error code is returned and the pager +** moves into the error state. Otherwise, SQLITE_OK is returned. +*/ +int sqlite3PagerCommitPhaseTwo(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + + /* This routine should not be called if a prior error has occurred. + ** But if (due to a coding error elsewhere in the system) it does get + ** called, just return the same error code without doing anything. */ + if( NEVER(pPager->errCode) ) return pPager->errCode; + pPager->iDataVersion++; + + assert( pPager->eState==PAGER_WRITER_LOCKED + || pPager->eState==PAGER_WRITER_FINISHED + || (pagerUseWal(pPager) && pPager->eState==PAGER_WRITER_CACHEMOD) + ); + assert( assert_pager_state(pPager) ); + + /* An optimization. If the database was not actually modified during + ** this transaction, the pager is running in exclusive-mode and is + ** using persistent journals, then this function is a no-op. + ** + ** The start of the journal file currently contains a single journal + ** header with the nRec field set to 0. If such a journal is used as + ** a hot-journal during hot-journal rollback, 0 changes will be made + ** to the database file. So there is no need to zero the journal + ** header. Since the pager is in exclusive mode, there is no need + ** to drop any locks either. + */ + if( pPager->eState==PAGER_WRITER_LOCKED + && pPager->exclusiveMode + && pPager->journalMode==PAGER_JOURNALMODE_PERSIST + ){ + assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) || !pPager->journalOff ); + pPager->eState = PAGER_READER; + return SQLITE_OK; + } + + PAGERTRACE(("COMMIT %d\n", PAGERID(pPager))); + rc = pager_end_transaction(pPager, pPager->setSuper, 1); + return pager_error(pPager, rc); +} + +/* +** If a write transaction is open, then all changes made within the +** transaction are reverted and the current write-transaction is closed. +** The pager falls back to PAGER_READER state if successful, or PAGER_ERROR +** state if an error occurs. +** +** If the pager is already in PAGER_ERROR state when this function is called, +** it returns Pager.errCode immediately. No work is performed in this case. +** +** Otherwise, in rollback mode, this function performs two functions: +** +** 1) It rolls back the journal file, restoring all database file and +** in-memory cache pages to the state they were in when the transaction +** was opened, and +** +** 2) It finalizes the journal file, so that it is not used for hot +** rollback at any point in the future. +** +** Finalization of the journal file (task 2) is only performed if the +** rollback is successful. +** +** In WAL mode, all cache-entries containing data modified within the +** current transaction are either expelled from the cache or reverted to +** their pre-transaction state by re-reading data from the database or +** WAL files. The WAL transaction is then closed. +*/ +int sqlite3PagerRollback(Pager *pPager){ + int rc = SQLITE_OK; /* Return code */ + PAGERTRACE(("ROLLBACK %d\n", PAGERID(pPager))); + + /* PagerRollback() is a no-op if called in READER or OPEN state. If + ** the pager is already in the ERROR state, the rollback is not + ** attempted here. Instead, the error code is returned to the caller. + */ + assert( assert_pager_state(pPager) ); + if( pPager->eState==PAGER_ERROR ) return pPager->errCode; + if( pPager->eState<=PAGER_READER ) return SQLITE_OK; + + if( pagerUseWal(pPager) ){ + int rc2; + rc = sqlite3PagerSavepoint(pPager, SAVEPOINT_ROLLBACK, -1); + rc2 = pager_end_transaction(pPager, pPager->setSuper, 0); + if( rc==SQLITE_OK ) rc = rc2; + }else if( !isOpen(pPager->jfd) || pPager->eState==PAGER_WRITER_LOCKED ){ + int eState = pPager->eState; + rc = pager_end_transaction(pPager, 0, 0); + if( !MEMDB && eState>PAGER_WRITER_LOCKED ){ + /* This can happen using journal_mode=off. Move the pager to the error + ** state to indicate that the contents of the cache may not be trusted. + ** Any active readers will get SQLITE_ABORT. + */ + pPager->errCode = SQLITE_ABORT; + pPager->eState = PAGER_ERROR; + setGetterMethod(pPager); + return rc; + } + }else{ + rc = pager_playback(pPager, 0); + } + + assert( pPager->eState==PAGER_READER || rc!=SQLITE_OK ); + assert( rc==SQLITE_OK || rc==SQLITE_FULL || rc==SQLITE_CORRUPT + || rc==SQLITE_NOMEM || (rc&0xFF)==SQLITE_IOERR + || rc==SQLITE_CANTOPEN + ); + + /* If an error occurs during a ROLLBACK, we can no longer trust the pager + ** cache. So call pager_error() on the way out to make any error persistent. + */ + return pager_error(pPager, rc); +} + +/* +** Return TRUE if the database file is opened read-only. Return FALSE +** if the database is (in theory) writable. +*/ +u8 sqlite3PagerIsreadonly(Pager *pPager){ + return pPager->readOnly; +} + +#ifdef SQLITE_DEBUG +/* +** Return the sum of the reference counts for all pages held by pPager. +*/ +int sqlite3PagerRefcount(Pager *pPager){ + return sqlite3PcacheRefCount(pPager->pPCache); +} +#endif + +/* +** Return the approximate number of bytes of memory currently +** used by the pager and its associated cache. +*/ +int sqlite3PagerMemUsed(Pager *pPager){ + int perPageSize = pPager->pageSize + pPager->nExtra + + (int)(sizeof(PgHdr) + 5*sizeof(void*)); + return perPageSize*sqlite3PcachePagecount(pPager->pPCache) + + sqlite3MallocSize(pPager) + + pPager->pageSize; +} + +/* +** Return the number of references to the specified page. +*/ +int sqlite3PagerPageRefcount(DbPage *pPage){ + return sqlite3PcachePageRefcount(pPage); +} + +#ifdef SQLITE_TEST +/* +** This routine is used for testing and analysis only. +*/ +int *sqlite3PagerStats(Pager *pPager){ + static int a[11]; + a[0] = sqlite3PcacheRefCount(pPager->pPCache); + a[1] = sqlite3PcachePagecount(pPager->pPCache); + a[2] = sqlite3PcacheGetCachesize(pPager->pPCache); + a[3] = pPager->eState==PAGER_OPEN ? -1 : (int) pPager->dbSize; + a[4] = pPager->eState; + a[5] = pPager->errCode; + a[6] = pPager->aStat[PAGER_STAT_HIT]; + a[7] = pPager->aStat[PAGER_STAT_MISS]; + a[8] = 0; /* Used to be pPager->nOvfl */ + a[9] = pPager->nRead; + a[10] = pPager->aStat[PAGER_STAT_WRITE]; + return a; +} +#endif + +/* +** Parameter eStat must be one of SQLITE_DBSTATUS_CACHE_HIT, _MISS, _WRITE, +** or _WRITE+1. The SQLITE_DBSTATUS_CACHE_WRITE+1 case is a translation +** of SQLITE_DBSTATUS_CACHE_SPILL. The _SPILL case is not contiguous because +** it was added later. +** +** Before returning, *pnVal is incremented by the +** current cache hit or miss count, according to the value of eStat. If the +** reset parameter is non-zero, the cache hit or miss count is zeroed before +** returning. +*/ +void sqlite3PagerCacheStat(Pager *pPager, int eStat, int reset, int *pnVal){ + + assert( eStat==SQLITE_DBSTATUS_CACHE_HIT + || eStat==SQLITE_DBSTATUS_CACHE_MISS + || eStat==SQLITE_DBSTATUS_CACHE_WRITE + || eStat==SQLITE_DBSTATUS_CACHE_WRITE+1 + ); + + assert( SQLITE_DBSTATUS_CACHE_HIT+1==SQLITE_DBSTATUS_CACHE_MISS ); + assert( SQLITE_DBSTATUS_CACHE_HIT+2==SQLITE_DBSTATUS_CACHE_WRITE ); + assert( PAGER_STAT_HIT==0 && PAGER_STAT_MISS==1 + && PAGER_STAT_WRITE==2 && PAGER_STAT_SPILL==3 ); + + eStat -= SQLITE_DBSTATUS_CACHE_HIT; + *pnVal += pPager->aStat[eStat]; + if( reset ){ + pPager->aStat[eStat] = 0; + } +} + +/* +** Return true if this is an in-memory or temp-file backed pager. +*/ +int sqlite3PagerIsMemdb(Pager *pPager){ + return pPager->tempFile || pPager->memVfs; +} + +/* +** Check that there are at least nSavepoint savepoints open. If there are +** currently less than nSavepoints open, then open one or more savepoints +** to make up the difference. If the number of savepoints is already +** equal to nSavepoint, then this function is a no-op. +** +** If a memory allocation fails, SQLITE_NOMEM is returned. If an error +** occurs while opening the sub-journal file, then an IO error code is +** returned. Otherwise, SQLITE_OK. +*/ +static SQLITE_NOINLINE int pagerOpenSavepoint(Pager *pPager, int nSavepoint){ + int rc = SQLITE_OK; /* Return code */ + int nCurrent = pPager->nSavepoint; /* Current number of savepoints */ + int ii; /* Iterator variable */ + PagerSavepoint *aNew; /* New Pager.aSavepoint array */ + + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + assert( nSavepoint>nCurrent && pPager->useJournal ); + + /* Grow the Pager.aSavepoint array using realloc(). Return SQLITE_NOMEM + ** if the allocation fails. Otherwise, zero the new portion in case a + ** malloc failure occurs while populating it in the for(...) loop below. + */ + aNew = (PagerSavepoint *)sqlite3Realloc( + pPager->aSavepoint, sizeof(PagerSavepoint)*nSavepoint + ); + if( !aNew ){ + return SQLITE_NOMEM_BKPT; + } + memset(&aNew[nCurrent], 0, (nSavepoint-nCurrent) * sizeof(PagerSavepoint)); + pPager->aSavepoint = aNew; + + /* Populate the PagerSavepoint structures just allocated. */ + for(ii=nCurrent; iidbSize; + if( isOpen(pPager->jfd) && pPager->journalOff>0 ){ + aNew[ii].iOffset = pPager->journalOff; + }else{ + aNew[ii].iOffset = JOURNAL_HDR_SZ(pPager); + } + aNew[ii].iSubRec = pPager->nSubRec; + aNew[ii].pInSavepoint = sqlite3BitvecCreate(pPager->dbSize); + aNew[ii].bTruncateOnRelease = 1; + if( !aNew[ii].pInSavepoint ){ + return SQLITE_NOMEM_BKPT; + } + if( pagerUseWal(pPager) ){ + sqlite3WalSavepoint(pPager->pWal, aNew[ii].aWalData); + } + pPager->nSavepoint = ii+1; + } + assert( pPager->nSavepoint==nSavepoint ); + assertTruncateConstraint(pPager); + return rc; +} +int sqlite3PagerOpenSavepoint(Pager *pPager, int nSavepoint){ + assert( pPager->eState>=PAGER_WRITER_LOCKED ); + assert( assert_pager_state(pPager) ); + + if( nSavepoint>pPager->nSavepoint && pPager->useJournal ){ + return pagerOpenSavepoint(pPager, nSavepoint); + }else{ + return SQLITE_OK; + } +} + + +/* +** This function is called to rollback or release (commit) a savepoint. +** The savepoint to release or rollback need not be the most recently +** created savepoint. +** +** Parameter op is always either SAVEPOINT_ROLLBACK or SAVEPOINT_RELEASE. +** If it is SAVEPOINT_RELEASE, then release and destroy the savepoint with +** index iSavepoint. If it is SAVEPOINT_ROLLBACK, then rollback all changes +** that have occurred since the specified savepoint was created. +** +** The savepoint to rollback or release is identified by parameter +** iSavepoint. A value of 0 means to operate on the outermost savepoint +** (the first created). A value of (Pager.nSavepoint-1) means operate +** on the most recently created savepoint. If iSavepoint is greater than +** (Pager.nSavepoint-1), then this function is a no-op. +** +** If a negative value is passed to this function, then the current +** transaction is rolled back. This is different to calling +** sqlite3PagerRollback() because this function does not terminate +** the transaction or unlock the database, it just restores the +** contents of the database to its original state. +** +** In any case, all savepoints with an index greater than iSavepoint +** are destroyed. If this is a release operation (op==SAVEPOINT_RELEASE), +** then savepoint iSavepoint is also destroyed. +** +** This function may return SQLITE_NOMEM if a memory allocation fails, +** or an IO error code if an IO error occurs while rolling back a +** savepoint. If no errors occur, SQLITE_OK is returned. +*/ +int sqlite3PagerSavepoint(Pager *pPager, int op, int iSavepoint){ + int rc = pPager->errCode; + +#ifdef SQLITE_ENABLE_ZIPVFS + if( op==SAVEPOINT_RELEASE ) rc = SQLITE_OK; +#endif + + assert( op==SAVEPOINT_RELEASE || op==SAVEPOINT_ROLLBACK ); + assert( iSavepoint>=0 || op==SAVEPOINT_ROLLBACK ); + + if( rc==SQLITE_OK && iSavepointnSavepoint ){ + int ii; /* Iterator variable */ + int nNew; /* Number of remaining savepoints after this op. */ + + /* Figure out how many savepoints will still be active after this + ** operation. Store this value in nNew. Then free resources associated + ** with any savepoints that are destroyed by this operation. + */ + nNew = iSavepoint + (( op==SAVEPOINT_RELEASE ) ? 0 : 1); + for(ii=nNew; iinSavepoint; ii++){ + sqlite3BitvecDestroy(pPager->aSavepoint[ii].pInSavepoint); + } + pPager->nSavepoint = nNew; + + /* Truncate the sub-journal so that it only includes the parts + ** that are still in use. */ + if( op==SAVEPOINT_RELEASE ){ + PagerSavepoint *pRel = &pPager->aSavepoint[nNew]; + if( pRel->bTruncateOnRelease && isOpen(pPager->sjfd) ){ + /* Only truncate if it is an in-memory sub-journal. */ + if( sqlite3JournalIsInMemory(pPager->sjfd) ){ + i64 sz = (pPager->pageSize+4)*(i64)pRel->iSubRec; + rc = sqlite3OsTruncate(pPager->sjfd, sz); + assert( rc==SQLITE_OK ); + } + pPager->nSubRec = pRel->iSubRec; + } + } + /* Else this is a rollback operation, playback the specified savepoint. + ** If this is a temp-file, it is possible that the journal file has + ** not yet been opened. In this case there have been no changes to + ** the database file, so the playback operation can be skipped. + */ + else if( pagerUseWal(pPager) || isOpen(pPager->jfd) ){ + PagerSavepoint *pSavepoint = (nNew==0)?0:&pPager->aSavepoint[nNew-1]; + rc = pagerPlaybackSavepoint(pPager, pSavepoint); + assert(rc!=SQLITE_DONE); + } + +#ifdef SQLITE_ENABLE_ZIPVFS + /* If the cache has been modified but the savepoint cannot be rolled + ** back journal_mode=off, put the pager in the error state. This way, + ** if the VFS used by this pager includes ZipVFS, the entire transaction + ** can be rolled back at the ZipVFS level. */ + else if( + pPager->journalMode==PAGER_JOURNALMODE_OFF + && pPager->eState>=PAGER_WRITER_CACHEMOD + ){ + pPager->errCode = SQLITE_ABORT; + pPager->eState = PAGER_ERROR; + setGetterMethod(pPager); + } +#endif + } + + return rc; +} + +/* +** Return the full pathname of the database file. +** +** Except, if the pager is in-memory only, then return an empty string if +** nullIfMemDb is true. This routine is called with nullIfMemDb==1 when +** used to report the filename to the user, for compatibility with legacy +** behavior. But when the Btree needs to know the filename for matching to +** shared cache, it uses nullIfMemDb==0 so that in-memory databases can +** participate in shared-cache. +** +** The return value to this routine is always safe to use with +** sqlite3_uri_parameter() and sqlite3_filename_database() and friends. +*/ +const char *sqlite3PagerFilename(const Pager *pPager, int nullIfMemDb){ + static const char zFake[8] = { 0, 0, 0, 0, 0, 0, 0, 0 }; + return (nullIfMemDb && pPager->memDb) ? &zFake[4] : pPager->zFilename; +} + +/* +** Return the VFS structure for the pager. +*/ +sqlite3_vfs *sqlite3PagerVfs(Pager *pPager){ + return pPager->pVfs; +} + +/* +** Return the file handle for the database file associated +** with the pager. This might return NULL if the file has +** not yet been opened. +*/ +sqlite3_file *sqlite3PagerFile(Pager *pPager){ + return pPager->fd; +} + +/* +** Return the file handle for the journal file (if it exists). +** This will be either the rollback journal or the WAL file. +*/ +sqlite3_file *sqlite3PagerJrnlFile(Pager *pPager){ +#if SQLITE_OMIT_WAL + return pPager->jfd; +#else + return pPager->pWal ? sqlite3WalFile(pPager->pWal) : pPager->jfd; +#endif +} + +/* +** Return the full pathname of the journal file. +*/ +const char *sqlite3PagerJournalname(Pager *pPager){ + return pPager->zJournal; +} + +#ifndef SQLITE_OMIT_AUTOVACUUM +/* +** Move the page pPg to location pgno in the file. +** +** There must be no references to the page previously located at +** pgno (which we call pPgOld) though that page is allowed to be +** in cache. If the page previously located at pgno is not already +** in the rollback journal, it is not put there by by this routine. +** +** References to the page pPg remain valid. Updating any +** meta-data associated with pPg (i.e. data stored in the nExtra bytes +** allocated along with the page) is the responsibility of the caller. +** +** A transaction must be active when this routine is called. It used to be +** required that a statement transaction was not active, but this restriction +** has been removed (CREATE INDEX needs to move a page when a statement +** transaction is active). +** +** If the fourth argument, isCommit, is non-zero, then this page is being +** moved as part of a database reorganization just before the transaction +** is being committed. In this case, it is guaranteed that the database page +** pPg refers to will not be written to again within this transaction. +** +** This function may return SQLITE_NOMEM or an IO error code if an error +** occurs. Otherwise, it returns SQLITE_OK. +*/ +int sqlite3PagerMovepage(Pager *pPager, DbPage *pPg, Pgno pgno, int isCommit){ + PgHdr *pPgOld; /* The page being overwritten. */ + Pgno needSyncPgno = 0; /* Old value of pPg->pgno, if sync is required */ + int rc; /* Return code */ + Pgno origPgno; /* The original page number */ + + assert( pPg->nRef>0 ); + assert( pPager->eState==PAGER_WRITER_CACHEMOD + || pPager->eState==PAGER_WRITER_DBMOD + ); + assert( assert_pager_state(pPager) ); + + /* In order to be able to rollback, an in-memory database must journal + ** the page we are moving from. + */ + assert( pPager->tempFile || !MEMDB ); + if( pPager->tempFile ){ + rc = sqlite3PagerWrite(pPg); + if( rc ) return rc; + } + + /* If the page being moved is dirty and has not been saved by the latest + ** savepoint, then save the current contents of the page into the + ** sub-journal now. This is required to handle the following scenario: + ** + ** BEGIN; + ** + ** SAVEPOINT one; + ** + ** ROLLBACK TO one; + ** + ** If page X were not written to the sub-journal here, it would not + ** be possible to restore its contents when the "ROLLBACK TO one" + ** statement were is processed. + ** + ** subjournalPage() may need to allocate space to store pPg->pgno into + ** one or more savepoint bitvecs. This is the reason this function + ** may return SQLITE_NOMEM. + */ + if( (pPg->flags & PGHDR_DIRTY)!=0 + && SQLITE_OK!=(rc = subjournalPageIfRequired(pPg)) + ){ + return rc; + } + + PAGERTRACE(("MOVE %d page %d (needSync=%d) moves to %d\n", + PAGERID(pPager), pPg->pgno, (pPg->flags&PGHDR_NEED_SYNC)?1:0, pgno)); + IOTRACE(("MOVE %p %d %d\n", pPager, pPg->pgno, pgno)) + + /* If the journal needs to be sync()ed before page pPg->pgno can + ** be written to, store pPg->pgno in local variable needSyncPgno. + ** + ** If the isCommit flag is set, there is no need to remember that + ** the journal needs to be sync()ed before database page pPg->pgno + ** can be written to. The caller has already promised not to write to it. + */ + if( (pPg->flags&PGHDR_NEED_SYNC) && !isCommit ){ + needSyncPgno = pPg->pgno; + assert( pPager->journalMode==PAGER_JOURNALMODE_OFF || + pageInJournal(pPager, pPg) || pPg->pgno>pPager->dbOrigSize ); + assert( pPg->flags&PGHDR_DIRTY ); + } + + /* If the cache contains a page with page-number pgno, remove it + ** from its hash chain. Also, if the PGHDR_NEED_SYNC flag was set for + ** page pgno before the 'move' operation, it needs to be retained + ** for the page moved there. + */ + pPg->flags &= ~PGHDR_NEED_SYNC; + pPgOld = sqlite3PagerLookup(pPager, pgno); + assert( !pPgOld || pPgOld->nRef==1 || CORRUPT_DB ); + if( pPgOld ){ + if( NEVER(pPgOld->nRef>1) ){ + sqlite3PagerUnrefNotNull(pPgOld); + return SQLITE_CORRUPT_BKPT; + } + pPg->flags |= (pPgOld->flags&PGHDR_NEED_SYNC); + if( pPager->tempFile ){ + /* Do not discard pages from an in-memory database since we might + ** need to rollback later. Just move the page out of the way. */ + sqlite3PcacheMove(pPgOld, pPager->dbSize+1); + }else{ + sqlite3PcacheDrop(pPgOld); + } + } + + origPgno = pPg->pgno; + sqlite3PcacheMove(pPg, pgno); + sqlite3PcacheMakeDirty(pPg); + + /* For an in-memory database, make sure the original page continues + ** to exist, in case the transaction needs to roll back. Use pPgOld + ** as the original page since it has already been allocated. + */ + if( pPager->tempFile && pPgOld ){ + sqlite3PcacheMove(pPgOld, origPgno); + sqlite3PagerUnrefNotNull(pPgOld); + } + + if( needSyncPgno ){ + /* If needSyncPgno is non-zero, then the journal file needs to be + ** sync()ed before any data is written to database file page needSyncPgno. + ** Currently, no such page exists in the page-cache and the + ** "is journaled" bitvec flag has been set. This needs to be remedied by + ** loading the page into the pager-cache and setting the PGHDR_NEED_SYNC + ** flag. + ** + ** If the attempt to load the page into the page-cache fails, (due + ** to a malloc() or IO failure), clear the bit in the pInJournal[] + ** array. Otherwise, if the page is loaded and written again in + ** this transaction, it may be written to the database file before + ** it is synced into the journal file. This way, it may end up in + ** the journal file twice, but that is not a problem. + */ + PgHdr *pPgHdr; + rc = sqlite3PagerGet(pPager, needSyncPgno, &pPgHdr, 0); + if( rc!=SQLITE_OK ){ + if( needSyncPgno<=pPager->dbOrigSize ){ + assert( pPager->pTmpSpace!=0 ); + sqlite3BitvecClear(pPager->pInJournal, needSyncPgno, pPager->pTmpSpace); + } + return rc; + } + pPgHdr->flags |= PGHDR_NEED_SYNC; + sqlite3PcacheMakeDirty(pPgHdr); + sqlite3PagerUnrefNotNull(pPgHdr); + } + + return SQLITE_OK; +} +#endif + +/* +** The page handle passed as the first argument refers to a dirty page +** with a page number other than iNew. This function changes the page's +** page number to iNew and sets the value of the PgHdr.flags field to +** the value passed as the third parameter. +*/ +void sqlite3PagerRekey(DbPage *pPg, Pgno iNew, u16 flags){ + assert( pPg->pgno!=iNew ); + pPg->flags = flags; + sqlite3PcacheMove(pPg, iNew); +} + +/* +** Return a pointer to the data for the specified page. +*/ +void *sqlite3PagerGetData(DbPage *pPg){ + assert( pPg->nRef>0 || pPg->pPager->memDb ); + return pPg->pData; +} + +/* +** Return a pointer to the Pager.nExtra bytes of "extra" space +** allocated along with the specified page. +*/ +void *sqlite3PagerGetExtra(DbPage *pPg){ + return pPg->pExtra; +} + +/* +** Get/set the locking-mode for this pager. Parameter eMode must be one +** of PAGER_LOCKINGMODE_QUERY, PAGER_LOCKINGMODE_NORMAL or +** PAGER_LOCKINGMODE_EXCLUSIVE. If the parameter is not _QUERY, then +** the locking-mode is set to the value specified. +** +** The returned value is either PAGER_LOCKINGMODE_NORMAL or +** PAGER_LOCKINGMODE_EXCLUSIVE, indicating the current (possibly updated) +** locking-mode. +*/ +int sqlite3PagerLockingMode(Pager *pPager, int eMode){ + assert( eMode==PAGER_LOCKINGMODE_QUERY + || eMode==PAGER_LOCKINGMODE_NORMAL + || eMode==PAGER_LOCKINGMODE_EXCLUSIVE ); + assert( PAGER_LOCKINGMODE_QUERY<0 ); + assert( PAGER_LOCKINGMODE_NORMAL>=0 && PAGER_LOCKINGMODE_EXCLUSIVE>=0 ); + assert( pPager->exclusiveMode || 0==sqlite3WalHeapMemory(pPager->pWal) ); + if( eMode>=0 && !pPager->tempFile && !sqlite3WalHeapMemory(pPager->pWal) ){ + pPager->exclusiveMode = (u8)eMode; + } + return (int)pPager->exclusiveMode; +} + +/* +** Set the journal-mode for this pager. Parameter eMode must be one of: +** +** PAGER_JOURNALMODE_DELETE +** PAGER_JOURNALMODE_TRUNCATE +** PAGER_JOURNALMODE_PERSIST +** PAGER_JOURNALMODE_OFF +** PAGER_JOURNALMODE_MEMORY +** PAGER_JOURNALMODE_WAL +** +** The journalmode is set to the value specified if the change is allowed. +** The change may be disallowed for the following reasons: +** +** * An in-memory database can only have its journal_mode set to _OFF +** or _MEMORY. +** +** * Temporary databases cannot have _WAL journalmode. +** +** The returned indicate the current (possibly updated) journal-mode. +*/ +int sqlite3PagerSetJournalMode(Pager *pPager, int eMode){ + u8 eOld = pPager->journalMode; /* Prior journalmode */ + + /* The eMode parameter is always valid */ + assert( eMode==PAGER_JOURNALMODE_DELETE + || eMode==PAGER_JOURNALMODE_TRUNCATE + || eMode==PAGER_JOURNALMODE_PERSIST + || eMode==PAGER_JOURNALMODE_OFF + || eMode==PAGER_JOURNALMODE_WAL + || eMode==PAGER_JOURNALMODE_MEMORY ); + + /* This routine is only called from the OP_JournalMode opcode, and + ** the logic there will never allow a temporary file to be changed + ** to WAL mode. + */ + assert( pPager->tempFile==0 || eMode!=PAGER_JOURNALMODE_WAL ); + + /* Do allow the journalmode of an in-memory database to be set to + ** anything other than MEMORY or OFF + */ + if( MEMDB ){ + assert( eOld==PAGER_JOURNALMODE_MEMORY || eOld==PAGER_JOURNALMODE_OFF ); + if( eMode!=PAGER_JOURNALMODE_MEMORY && eMode!=PAGER_JOURNALMODE_OFF ){ + eMode = eOld; + } + } + + if( eMode!=eOld ){ + + /* Change the journal mode. */ + assert( pPager->eState!=PAGER_ERROR ); + pPager->journalMode = (u8)eMode; + + /* When transistioning from TRUNCATE or PERSIST to any other journal + ** mode except WAL, unless the pager is in locking_mode=exclusive mode, + ** delete the journal file. + */ + assert( (PAGER_JOURNALMODE_TRUNCATE & 5)==1 ); + assert( (PAGER_JOURNALMODE_PERSIST & 5)==1 ); + assert( (PAGER_JOURNALMODE_DELETE & 5)==0 ); + assert( (PAGER_JOURNALMODE_MEMORY & 5)==4 ); + assert( (PAGER_JOURNALMODE_OFF & 5)==0 ); + assert( (PAGER_JOURNALMODE_WAL & 5)==5 ); + + assert( isOpen(pPager->fd) || pPager->exclusiveMode ); + if( !pPager->exclusiveMode && (eOld & 5)==1 && (eMode & 1)==0 ){ + + /* In this case we would like to delete the journal file. If it is + ** not possible, then that is not a problem. Deleting the journal file + ** here is an optimization only. + ** + ** Before deleting the journal file, obtain a RESERVED lock on the + ** database file. This ensures that the journal file is not deleted + ** while it is in use by some other client. + */ + sqlite3OsClose(pPager->jfd); + if( pPager->eLock>=RESERVED_LOCK ){ + sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0); + }else{ + int rc = SQLITE_OK; + int state = pPager->eState; + assert( state==PAGER_OPEN || state==PAGER_READER ); + if( state==PAGER_OPEN ){ + rc = sqlite3PagerSharedLock(pPager); + } + if( pPager->eState==PAGER_READER ){ + assert( rc==SQLITE_OK ); + rc = pagerLockDb(pPager, RESERVED_LOCK); + } + if( rc==SQLITE_OK ){ + sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0); + } + if( rc==SQLITE_OK && state==PAGER_READER ){ + pagerUnlockDb(pPager, SHARED_LOCK); + }else if( state==PAGER_OPEN ){ + pager_unlock(pPager); + } + assert( state==pPager->eState ); + } + }else if( eMode==PAGER_JOURNALMODE_OFF ){ + sqlite3OsClose(pPager->jfd); + } + } + + /* Return the new journal mode */ + return (int)pPager->journalMode; +} + +/* +** Return the current journal mode. +*/ +int sqlite3PagerGetJournalMode(Pager *pPager){ + return (int)pPager->journalMode; +} + +/* +** Return TRUE if the pager is in a state where it is OK to change the +** journalmode. Journalmode changes can only happen when the database +** is unmodified. +*/ +int sqlite3PagerOkToChangeJournalMode(Pager *pPager){ + assert( assert_pager_state(pPager) ); + if( pPager->eState>=PAGER_WRITER_CACHEMOD ) return 0; + if( NEVER(isOpen(pPager->jfd) && pPager->journalOff>0) ) return 0; + return 1; +} + +/* +** Get/set the size-limit used for persistent journal files. +** +** Setting the size limit to -1 means no limit is enforced. +** An attempt to set a limit smaller than -1 is a no-op. +*/ +i64 sqlite3PagerJournalSizeLimit(Pager *pPager, i64 iLimit){ + if( iLimit>=-1 ){ + pPager->journalSizeLimit = iLimit; + sqlite3WalLimit(pPager->pWal, iLimit); + } + return pPager->journalSizeLimit; +} + +/* +** Return a pointer to the pPager->pBackup variable. The backup module +** in backup.c maintains the content of this variable. This module +** uses it opaquely as an argument to sqlite3BackupRestart() and +** sqlite3BackupUpdate() only. +*/ +sqlite3_backup **sqlite3PagerBackupPtr(Pager *pPager){ + return &pPager->pBackup; +} + +#ifndef SQLITE_OMIT_VACUUM +/* +** Unless this is an in-memory or temporary database, clear the pager cache. +*/ +void sqlite3PagerClearCache(Pager *pPager){ + assert( MEMDB==0 || pPager->tempFile ); + if( pPager->tempFile==0 ) pager_reset(pPager); +} +#endif + + +#ifndef SQLITE_OMIT_WAL +/* +** This function is called when the user invokes "PRAGMA wal_checkpoint", +** "PRAGMA wal_blocking_checkpoint" or calls the sqlite3_wal_checkpoint() +** or wal_blocking_checkpoint() API functions. +** +** Parameter eMode is one of SQLITE_CHECKPOINT_PASSIVE, FULL or RESTART. +*/ +int sqlite3PagerCheckpoint( + Pager *pPager, /* Checkpoint on this pager */ + sqlite3 *db, /* Db handle used to check for interrupts */ + int eMode, /* Type of checkpoint */ + int *pnLog, /* OUT: Final number of frames in log */ + int *pnCkpt /* OUT: Final number of checkpointed frames */ +){ + int rc = SQLITE_OK; + if( pPager->pWal ){ + rc = sqlite3WalCheckpoint(pPager->pWal, db, eMode, + (eMode==SQLITE_CHECKPOINT_PASSIVE ? 0 : pPager->xBusyHandler), + pPager->pBusyHandlerArg, + pPager->walSyncFlags, pPager->pageSize, (u8 *)pPager->pTmpSpace, + pnLog, pnCkpt + ); + } + return rc; +} + +int sqlite3PagerWalCallback(Pager *pPager){ + return sqlite3WalCallback(pPager->pWal); +} + +/* +** Return true if the underlying VFS for the given pager supports the +** primitives necessary for write-ahead logging. +*/ +int sqlite3PagerWalSupported(Pager *pPager){ + const sqlite3_io_methods *pMethods = pPager->fd->pMethods; + if( pPager->noLock ) return 0; + return pPager->exclusiveMode || (pMethods->iVersion>=2 && pMethods->xShmMap); +} + +/* +** Attempt to take an exclusive lock on the database file. If a PENDING lock +** is obtained instead, immediately release it. +*/ +static int pagerExclusiveLock(Pager *pPager){ + int rc; /* Return code */ + + assert( pPager->eLock==SHARED_LOCK || pPager->eLock==EXCLUSIVE_LOCK ); + rc = pagerLockDb(pPager, EXCLUSIVE_LOCK); + if( rc!=SQLITE_OK ){ + /* If the attempt to grab the exclusive lock failed, release the + ** pending lock that may have been obtained instead. */ + pagerUnlockDb(pPager, SHARED_LOCK); + } + + return rc; +} + +/* +** Call sqlite3WalOpen() to open the WAL handle. If the pager is in +** exclusive-locking mode when this function is called, take an EXCLUSIVE +** lock on the database file and use heap-memory to store the wal-index +** in. Otherwise, use the normal shared-memory. +*/ +static int pagerOpenWal(Pager *pPager){ + int rc = SQLITE_OK; + + assert( pPager->pWal==0 && pPager->tempFile==0 ); + assert( pPager->eLock==SHARED_LOCK || pPager->eLock==EXCLUSIVE_LOCK ); + + /* If the pager is already in exclusive-mode, the WAL module will use + ** heap-memory for the wal-index instead of the VFS shared-memory + ** implementation. Take the exclusive lock now, before opening the WAL + ** file, to make sure this is safe. + */ + if( pPager->exclusiveMode ){ + rc = pagerExclusiveLock(pPager); + } + + /* Open the connection to the log file. If this operation fails, + ** (e.g. due to malloc() failure), return an error code. + */ + if( rc==SQLITE_OK ){ + rc = sqlite3WalOpen(pPager->pVfs, + pPager->fd, pPager->zWal, pPager->exclusiveMode, + pPager->journalSizeLimit, &pPager->pWal + ); + } + pagerFixMaplimit(pPager); + + return rc; +} + + +/* +** The caller must be holding a SHARED lock on the database file to call +** this function. +** +** If the pager passed as the first argument is open on a real database +** file (not a temp file or an in-memory database), and the WAL file +** is not already open, make an attempt to open it now. If successful, +** return SQLITE_OK. If an error occurs or the VFS used by the pager does +** not support the xShmXXX() methods, return an error code. *pbOpen is +** not modified in either case. +** +** If the pager is open on a temp-file (or in-memory database), or if +** the WAL file is already open, set *pbOpen to 1 and return SQLITE_OK +** without doing anything. +*/ +int sqlite3PagerOpenWal( + Pager *pPager, /* Pager object */ + int *pbOpen /* OUT: Set to true if call is a no-op */ +){ + int rc = SQLITE_OK; /* Return code */ + + assert( assert_pager_state(pPager) ); + assert( pPager->eState==PAGER_OPEN || pbOpen ); + assert( pPager->eState==PAGER_READER || !pbOpen ); + assert( pbOpen==0 || *pbOpen==0 ); + assert( pbOpen!=0 || (!pPager->tempFile && !pPager->pWal) ); + + if( !pPager->tempFile && !pPager->pWal ){ + if( !sqlite3PagerWalSupported(pPager) ) return SQLITE_CANTOPEN; + + /* Close any rollback journal previously open */ + sqlite3OsClose(pPager->jfd); + + rc = pagerOpenWal(pPager); + if( rc==SQLITE_OK ){ + pPager->journalMode = PAGER_JOURNALMODE_WAL; + pPager->eState = PAGER_OPEN; + } + }else{ + *pbOpen = 1; + } + + return rc; +} + +/* +** This function is called to close the connection to the log file prior +** to switching from WAL to rollback mode. +** +** Before closing the log file, this function attempts to take an +** EXCLUSIVE lock on the database file. If this cannot be obtained, an +** error (SQLITE_BUSY) is returned and the log connection is not closed. +** If successful, the EXCLUSIVE lock is not released before returning. +*/ +int sqlite3PagerCloseWal(Pager *pPager, sqlite3 *db){ + int rc = SQLITE_OK; + + assert( pPager->journalMode==PAGER_JOURNALMODE_WAL ); + + /* If the log file is not already open, but does exist in the file-system, + ** it may need to be checkpointed before the connection can switch to + ** rollback mode. Open it now so this can happen. + */ + if( !pPager->pWal ){ + int logexists = 0; + rc = pagerLockDb(pPager, SHARED_LOCK); + if( rc==SQLITE_OK ){ + rc = sqlite3OsAccess( + pPager->pVfs, pPager->zWal, SQLITE_ACCESS_EXISTS, &logexists + ); + } + if( rc==SQLITE_OK && logexists ){ + rc = pagerOpenWal(pPager); + } + } + + /* Checkpoint and close the log. Because an EXCLUSIVE lock is held on + ** the database file, the log and log-summary files will be deleted. + */ + if( rc==SQLITE_OK && pPager->pWal ){ + rc = pagerExclusiveLock(pPager); + if( rc==SQLITE_OK ){ + rc = sqlite3WalClose(pPager->pWal, db, pPager->walSyncFlags, + pPager->pageSize, (u8*)pPager->pTmpSpace); + pPager->pWal = 0; + pagerFixMaplimit(pPager); + if( rc && !pPager->exclusiveMode ) pagerUnlockDb(pPager, SHARED_LOCK); + } + } + return rc; +} + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT +/* +** If pager pPager is a wal-mode database not in exclusive locking mode, +** invoke the sqlite3WalWriteLock() function on the associated Wal object +** with the same db and bLock parameters as were passed to this function. +** Return an SQLite error code if an error occurs, or SQLITE_OK otherwise. +*/ +int sqlite3PagerWalWriteLock(Pager *pPager, int bLock){ + int rc = SQLITE_OK; + if( pagerUseWal(pPager) && pPager->exclusiveMode==0 ){ + rc = sqlite3WalWriteLock(pPager->pWal, bLock); + } + return rc; +} + +/* +** Set the database handle used by the wal layer to determine if +** blocking locks are required. +*/ +void sqlite3PagerWalDb(Pager *pPager, sqlite3 *db){ + if( pagerUseWal(pPager) ){ + sqlite3WalDb(pPager->pWal, db); + } +} +#endif + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* +** If this is a WAL database, obtain a snapshot handle for the snapshot +** currently open. Otherwise, return an error. +*/ +int sqlite3PagerSnapshotGet(Pager *pPager, sqlite3_snapshot **ppSnapshot){ + int rc = SQLITE_ERROR; + if( pPager->pWal ){ + rc = sqlite3WalSnapshotGet(pPager->pWal, ppSnapshot); + } + return rc; +} + +/* +** If this is a WAL database, store a pointer to pSnapshot. Next time a +** read transaction is opened, attempt to read from the snapshot it +** identifies. If this is not a WAL database, return an error. +*/ +int sqlite3PagerSnapshotOpen( + Pager *pPager, + sqlite3_snapshot *pSnapshot +){ + int rc = SQLITE_OK; + if( pPager->pWal ){ + sqlite3WalSnapshotOpen(pPager->pWal, pSnapshot); + }else{ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** If this is a WAL database, call sqlite3WalSnapshotRecover(). If this +** is not a WAL database, return an error. +*/ +int sqlite3PagerSnapshotRecover(Pager *pPager){ + int rc; + if( pPager->pWal ){ + rc = sqlite3WalSnapshotRecover(pPager->pWal); + }else{ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** The caller currently has a read transaction open on the database. +** If this is not a WAL database, SQLITE_ERROR is returned. Otherwise, +** this function takes a SHARED lock on the CHECKPOINTER slot and then +** checks if the snapshot passed as the second argument is still +** available. If so, SQLITE_OK is returned. +** +** If the snapshot is not available, SQLITE_ERROR is returned. Or, if +** the CHECKPOINTER lock cannot be obtained, SQLITE_BUSY. If any error +** occurs (any value other than SQLITE_OK is returned), the CHECKPOINTER +** lock is released before returning. +*/ +int sqlite3PagerSnapshotCheck(Pager *pPager, sqlite3_snapshot *pSnapshot){ + int rc; + if( pPager->pWal ){ + rc = sqlite3WalSnapshotCheck(pPager->pWal, pSnapshot); + }else{ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** Release a lock obtained by an earlier successful call to +** sqlite3PagerSnapshotCheck(). +*/ +void sqlite3PagerSnapshotUnlock(Pager *pPager){ + assert( pPager->pWal ); + sqlite3WalSnapshotUnlock(pPager->pWal); +} + +#endif /* SQLITE_ENABLE_SNAPSHOT */ +#endif /* !SQLITE_OMIT_WAL */ + +#ifdef SQLITE_ENABLE_ZIPVFS +/* +** A read-lock must be held on the pager when this function is called. If +** the pager is in WAL mode and the WAL file currently contains one or more +** frames, return the size in bytes of the page images stored within the +** WAL frames. Otherwise, if this is not a WAL database or the WAL file +** is empty, return 0. +*/ +int sqlite3PagerWalFramesize(Pager *pPager){ + assert( pPager->eState>=PAGER_READER ); + return sqlite3WalFramesize(pPager->pWal); +} +#endif + +#endif /* SQLITE_OMIT_DISKIO */ diff --git a/source/libs/tdb/src/sqlite/pcache.c b/source/libs/tdb/src/sqlite/pcache.c new file mode 100644 index 0000000000..14d1e7cde0 --- /dev/null +++ b/source/libs/tdb/src/sqlite/pcache.c @@ -0,0 +1,885 @@ +/* +** 2008 August 05 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements that page cache. +*/ +#include "sqliteInt.h" + +/* +** A complete page cache is an instance of this structure. Every +** entry in the cache holds a single page of the database file. The +** btree layer only operates on the cached copy of the database pages. +** +** A page cache entry is "clean" if it exactly matches what is currently +** on disk. A page is "dirty" if it has been modified and needs to be +** persisted to disk. +** +** pDirty, pDirtyTail, pSynced: +** All dirty pages are linked into the doubly linked list using +** PgHdr.pDirtyNext and pDirtyPrev. The list is maintained in LRU order +** such that p was added to the list more recently than p->pDirtyNext. +** PCache.pDirty points to the first (newest) element in the list and +** pDirtyTail to the last (oldest). +** +** The PCache.pSynced variable is used to optimize searching for a dirty +** page to eject from the cache mid-transaction. It is better to eject +** a page that does not require a journal sync than one that does. +** Therefore, pSynced is maintained so that it *almost* always points +** to either the oldest page in the pDirty/pDirtyTail list that has a +** clear PGHDR_NEED_SYNC flag or to a page that is older than this one +** (so that the right page to eject can be found by following pDirtyPrev +** pointers). +*/ +struct PCache { + PgHdr *pDirty, *pDirtyTail; /* List of dirty pages in LRU order */ + PgHdr *pSynced; /* Last synced page in dirty page list */ + int nRefSum; /* Sum of ref counts over all pages */ + int szCache; /* Configured cache size */ + int szSpill; /* Size before spilling occurs */ + int szPage; /* Size of every page in this cache */ + int szExtra; /* Size of extra space for each page */ + u8 bPurgeable; /* True if pages are on backing store */ + u8 eCreate; /* eCreate value for for xFetch() */ + int (*xStress)(void*,PgHdr*); /* Call to try make a page clean */ + void *pStress; /* Argument to xStress */ + sqlite3_pcache *pCache; /* Pluggable cache module */ +}; + +/********************************** Test and Debug Logic **********************/ +/* +** Debug tracing macros. Enable by by changing the "0" to "1" and +** recompiling. +** +** When sqlite3PcacheTrace is 1, single line trace messages are issued. +** When sqlite3PcacheTrace is 2, a dump of the pcache showing all cache entries +** is displayed for many operations, resulting in a lot of output. +*/ +#if defined(SQLITE_DEBUG) && 0 + int sqlite3PcacheTrace = 2; /* 0: off 1: simple 2: cache dumps */ + int sqlite3PcacheMxDump = 9999; /* Max cache entries for pcacheDump() */ +# define pcacheTrace(X) if(sqlite3PcacheTrace){sqlite3DebugPrintf X;} + void pcacheDump(PCache *pCache){ + int N; + int i, j; + sqlite3_pcache_page *pLower; + PgHdr *pPg; + unsigned char *a; + + if( sqlite3PcacheTrace<2 ) return; + if( pCache->pCache==0 ) return; + N = sqlite3PcachePagecount(pCache); + if( N>sqlite3PcacheMxDump ) N = sqlite3PcacheMxDump; + for(i=1; i<=N; i++){ + pLower = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache, i, 0); + if( pLower==0 ) continue; + pPg = (PgHdr*)pLower->pExtra; + printf("%3d: nRef %2d flgs %02x data ", i, pPg->nRef, pPg->flags); + a = (unsigned char *)pLower->pBuf; + for(j=0; j<12; j++) printf("%02x", a[j]); + printf("\n"); + if( pPg->pPage==0 ){ + sqlite3GlobalConfig.pcache2.xUnpin(pCache->pCache, pLower, 0); + } + } + } + #else +# define pcacheTrace(X) +# define pcacheDump(X) +#endif + +/* +** Check invariants on a PgHdr entry. Return true if everything is OK. +** Return false if any invariant is violated. +** +** This routine is for use inside of assert() statements only. For +** example: +** +** assert( sqlite3PcachePageSanity(pPg) ); +*/ +#ifdef SQLITE_DEBUG +int sqlite3PcachePageSanity(PgHdr *pPg){ + PCache *pCache; + assert( pPg!=0 ); + assert( pPg->pgno>0 || pPg->pPager==0 ); /* Page number is 1 or more */ + pCache = pPg->pCache; + assert( pCache!=0 ); /* Every page has an associated PCache */ + if( pPg->flags & PGHDR_CLEAN ){ + assert( (pPg->flags & PGHDR_DIRTY)==0 );/* Cannot be both CLEAN and DIRTY */ + assert( pCache->pDirty!=pPg ); /* CLEAN pages not on dirty list */ + assert( pCache->pDirtyTail!=pPg ); + } + /* WRITEABLE pages must also be DIRTY */ + if( pPg->flags & PGHDR_WRITEABLE ){ + assert( pPg->flags & PGHDR_DIRTY ); /* WRITEABLE implies DIRTY */ + } + /* NEED_SYNC can be set independently of WRITEABLE. This can happen, + ** for example, when using the sqlite3PagerDontWrite() optimization: + ** (1) Page X is journalled, and gets WRITEABLE and NEED_SEEK. + ** (2) Page X moved to freelist, WRITEABLE is cleared + ** (3) Page X reused, WRITEABLE is set again + ** If NEED_SYNC had been cleared in step 2, then it would not be reset + ** in step 3, and page might be written into the database without first + ** syncing the rollback journal, which might cause corruption on a power + ** loss. + ** + ** Another example is when the database page size is smaller than the + ** disk sector size. When any page of a sector is journalled, all pages + ** in that sector are marked NEED_SYNC even if they are still CLEAN, just + ** in case they are later modified, since all pages in the same sector + ** must be journalled and synced before any of those pages can be safely + ** written. + */ + return 1; +} +#endif /* SQLITE_DEBUG */ + + +/********************************** Linked List Management ********************/ + +/* Allowed values for second argument to pcacheManageDirtyList() */ +#define PCACHE_DIRTYLIST_REMOVE 1 /* Remove pPage from dirty list */ +#define PCACHE_DIRTYLIST_ADD 2 /* Add pPage to the dirty list */ +#define PCACHE_DIRTYLIST_FRONT 3 /* Move pPage to the front of the list */ + +/* +** Manage pPage's participation on the dirty list. Bits of the addRemove +** argument determines what operation to do. The 0x01 bit means first +** remove pPage from the dirty list. The 0x02 means add pPage back to +** the dirty list. Doing both moves pPage to the front of the dirty list. +*/ +static void pcacheManageDirtyList(PgHdr *pPage, u8 addRemove){ + PCache *p = pPage->pCache; + + pcacheTrace(("%p.DIRTYLIST.%s %d\n", p, + addRemove==1 ? "REMOVE" : addRemove==2 ? "ADD" : "FRONT", + pPage->pgno)); + if( addRemove & PCACHE_DIRTYLIST_REMOVE ){ + assert( pPage->pDirtyNext || pPage==p->pDirtyTail ); + assert( pPage->pDirtyPrev || pPage==p->pDirty ); + + /* Update the PCache1.pSynced variable if necessary. */ + if( p->pSynced==pPage ){ + p->pSynced = pPage->pDirtyPrev; + } + + if( pPage->pDirtyNext ){ + pPage->pDirtyNext->pDirtyPrev = pPage->pDirtyPrev; + }else{ + assert( pPage==p->pDirtyTail ); + p->pDirtyTail = pPage->pDirtyPrev; + } + if( pPage->pDirtyPrev ){ + pPage->pDirtyPrev->pDirtyNext = pPage->pDirtyNext; + }else{ + /* If there are now no dirty pages in the cache, set eCreate to 2. + ** This is an optimization that allows sqlite3PcacheFetch() to skip + ** searching for a dirty page to eject from the cache when it might + ** otherwise have to. */ + assert( pPage==p->pDirty ); + p->pDirty = pPage->pDirtyNext; + assert( p->bPurgeable || p->eCreate==2 ); + if( p->pDirty==0 ){ /*OPTIMIZATION-IF-TRUE*/ + assert( p->bPurgeable==0 || p->eCreate==1 ); + p->eCreate = 2; + } + } + } + if( addRemove & PCACHE_DIRTYLIST_ADD ){ + pPage->pDirtyPrev = 0; + pPage->pDirtyNext = p->pDirty; + if( pPage->pDirtyNext ){ + assert( pPage->pDirtyNext->pDirtyPrev==0 ); + pPage->pDirtyNext->pDirtyPrev = pPage; + }else{ + p->pDirtyTail = pPage; + if( p->bPurgeable ){ + assert( p->eCreate==2 ); + p->eCreate = 1; + } + } + p->pDirty = pPage; + + /* If pSynced is NULL and this page has a clear NEED_SYNC flag, set + ** pSynced to point to it. Checking the NEED_SYNC flag is an + ** optimization, as if pSynced points to a page with the NEED_SYNC + ** flag set sqlite3PcacheFetchStress() searches through all newer + ** entries of the dirty-list for a page with NEED_SYNC clear anyway. */ + if( !p->pSynced + && 0==(pPage->flags&PGHDR_NEED_SYNC) /*OPTIMIZATION-IF-FALSE*/ + ){ + p->pSynced = pPage; + } + } + pcacheDump(p); +} + +/* +** Wrapper around the pluggable caches xUnpin method. If the cache is +** being used for an in-memory database, this function is a no-op. +*/ +static void pcacheUnpin(PgHdr *p){ + if( p->pCache->bPurgeable ){ + pcacheTrace(("%p.UNPIN %d\n", p->pCache, p->pgno)); + sqlite3GlobalConfig.pcache2.xUnpin(p->pCache->pCache, p->pPage, 0); + pcacheDump(p->pCache); + } +} + +/* +** Compute the number of pages of cache requested. p->szCache is the +** cache size requested by the "PRAGMA cache_size" statement. +*/ +static int numberOfCachePages(PCache *p){ + if( p->szCache>=0 ){ + /* IMPLEMENTATION-OF: R-42059-47211 If the argument N is positive then the + ** suggested cache size is set to N. */ + return p->szCache; + }else{ + i64 n; + /* IMPLEMANTATION-OF: R-59858-46238 If the argument N is negative, then the + ** number of cache pages is adjusted to be a number of pages that would + ** use approximately abs(N*1024) bytes of memory based on the current + ** page size. */ + n = ((-1024*(i64)p->szCache)/(p->szPage+p->szExtra)); + if( n>1000000000 ) n = 1000000000; + return (int)n; + } +} + +/*************************************************** General Interfaces ****** +** +** Initialize and shutdown the page cache subsystem. Neither of these +** functions are threadsafe. +*/ +int sqlite3PcacheInitialize(void){ + if( sqlite3GlobalConfig.pcache2.xInit==0 ){ + /* IMPLEMENTATION-OF: R-26801-64137 If the xInit() method is NULL, then the + ** built-in default page cache is used instead of the application defined + ** page cache. */ + sqlite3PCacheSetDefault(); + assert( sqlite3GlobalConfig.pcache2.xInit!=0 ); + } + return sqlite3GlobalConfig.pcache2.xInit(sqlite3GlobalConfig.pcache2.pArg); +} +void sqlite3PcacheShutdown(void){ + if( sqlite3GlobalConfig.pcache2.xShutdown ){ + /* IMPLEMENTATION-OF: R-26000-56589 The xShutdown() method may be NULL. */ + sqlite3GlobalConfig.pcache2.xShutdown(sqlite3GlobalConfig.pcache2.pArg); + } +} + +/* +** Return the size in bytes of a PCache object. +*/ +int sqlite3PcacheSize(void){ return sizeof(PCache); } + +/* +** Create a new PCache object. Storage space to hold the object +** has already been allocated and is passed in as the p pointer. +** The caller discovers how much space needs to be allocated by +** calling sqlite3PcacheSize(). +** +** szExtra is some extra space allocated for each page. The first +** 8 bytes of the extra space will be zeroed as the page is allocated, +** but remaining content will be uninitialized. Though it is opaque +** to this module, the extra space really ends up being the MemPage +** structure in the pager. +*/ +int sqlite3PcacheOpen( + int szPage, /* Size of every page */ + int szExtra, /* Extra space associated with each page */ + int bPurgeable, /* True if pages are on backing store */ + int (*xStress)(void*,PgHdr*),/* Call to try to make pages clean */ + void *pStress, /* Argument to xStress */ + PCache *p /* Preallocated space for the PCache */ +){ + memset(p, 0, sizeof(PCache)); + p->szPage = 1; + p->szExtra = szExtra; + assert( szExtra>=8 ); /* First 8 bytes will be zeroed */ + p->bPurgeable = bPurgeable; + p->eCreate = 2; + p->xStress = xStress; + p->pStress = pStress; + p->szCache = 100; + p->szSpill = 1; + pcacheTrace(("%p.OPEN szPage %d bPurgeable %d\n",p,szPage,bPurgeable)); + return sqlite3PcacheSetPageSize(p, szPage); +} + +/* +** Change the page size for PCache object. The caller must ensure that there +** are no outstanding page references when this function is called. +*/ +int sqlite3PcacheSetPageSize(PCache *pCache, int szPage){ + assert( pCache->nRefSum==0 && pCache->pDirty==0 ); + if( pCache->szPage ){ + sqlite3_pcache *pNew; + pNew = sqlite3GlobalConfig.pcache2.xCreate( + szPage, pCache->szExtra + ROUND8(sizeof(PgHdr)), + pCache->bPurgeable + ); + if( pNew==0 ) return SQLITE_NOMEM_BKPT; + sqlite3GlobalConfig.pcache2.xCachesize(pNew, numberOfCachePages(pCache)); + if( pCache->pCache ){ + sqlite3GlobalConfig.pcache2.xDestroy(pCache->pCache); + } + pCache->pCache = pNew; + pCache->szPage = szPage; + pcacheTrace(("%p.PAGESIZE %d\n",pCache,szPage)); + } + return SQLITE_OK; +} + +/* +** Try to obtain a page from the cache. +** +** This routine returns a pointer to an sqlite3_pcache_page object if +** such an object is already in cache, or if a new one is created. +** This routine returns a NULL pointer if the object was not in cache +** and could not be created. +** +** The createFlags should be 0 to check for existing pages and should +** be 3 (not 1, but 3) to try to create a new page. +** +** If the createFlag is 0, then NULL is always returned if the page +** is not already in the cache. If createFlag is 1, then a new page +** is created only if that can be done without spilling dirty pages +** and without exceeding the cache size limit. +** +** The caller needs to invoke sqlite3PcacheFetchFinish() to properly +** initialize the sqlite3_pcache_page object and convert it into a +** PgHdr object. The sqlite3PcacheFetch() and sqlite3PcacheFetchFinish() +** routines are split this way for performance reasons. When separated +** they can both (usually) operate without having to push values to +** the stack on entry and pop them back off on exit, which saves a +** lot of pushing and popping. +*/ +sqlite3_pcache_page *sqlite3PcacheFetch( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number to obtain */ + int createFlag /* If true, create page if it does not exist already */ +){ + int eCreate; + sqlite3_pcache_page *pRes; + + assert( pCache!=0 ); + assert( pCache->pCache!=0 ); + assert( createFlag==3 || createFlag==0 ); + assert( pCache->eCreate==((pCache->bPurgeable && pCache->pDirty) ? 1 : 2) ); + + /* eCreate defines what to do if the page does not exist. + ** 0 Do not allocate a new page. (createFlag==0) + ** 1 Allocate a new page if doing so is inexpensive. + ** (createFlag==1 AND bPurgeable AND pDirty) + ** 2 Allocate a new page even it doing so is difficult. + ** (createFlag==1 AND !(bPurgeable AND pDirty) + */ + eCreate = createFlag & pCache->eCreate; + assert( eCreate==0 || eCreate==1 || eCreate==2 ); + assert( createFlag==0 || pCache->eCreate==eCreate ); + assert( createFlag==0 || eCreate==1+(!pCache->bPurgeable||!pCache->pDirty) ); + pRes = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache, pgno, eCreate); + pcacheTrace(("%p.FETCH %d%s (result: %p)\n",pCache,pgno, + createFlag?" create":"",pRes)); + return pRes; +} + +/* +** If the sqlite3PcacheFetch() routine is unable to allocate a new +** page because no clean pages are available for reuse and the cache +** size limit has been reached, then this routine can be invoked to +** try harder to allocate a page. This routine might invoke the stress +** callback to spill dirty pages to the journal. It will then try to +** allocate the new page and will only fail to allocate a new page on +** an OOM error. +** +** This routine should be invoked only after sqlite3PcacheFetch() fails. +*/ +int sqlite3PcacheFetchStress( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number to obtain */ + sqlite3_pcache_page **ppPage /* Write result here */ +){ + PgHdr *pPg; + if( pCache->eCreate==2 ) return 0; + + if( sqlite3PcachePagecount(pCache)>pCache->szSpill ){ + /* Find a dirty page to write-out and recycle. First try to find a + ** page that does not require a journal-sync (one with PGHDR_NEED_SYNC + ** cleared), but if that is not possible settle for any other + ** unreferenced dirty page. + ** + ** If the LRU page in the dirty list that has a clear PGHDR_NEED_SYNC + ** flag is currently referenced, then the following may leave pSynced + ** set incorrectly (pointing to other than the LRU page with NEED_SYNC + ** cleared). This is Ok, as pSynced is just an optimization. */ + for(pPg=pCache->pSynced; + pPg && (pPg->nRef || (pPg->flags&PGHDR_NEED_SYNC)); + pPg=pPg->pDirtyPrev + ); + pCache->pSynced = pPg; + if( !pPg ){ + for(pPg=pCache->pDirtyTail; pPg && pPg->nRef; pPg=pPg->pDirtyPrev); + } + if( pPg ){ + int rc; +#ifdef SQLITE_LOG_CACHE_SPILL + sqlite3_log(SQLITE_FULL, + "spill page %d making room for %d - cache used: %d/%d", + pPg->pgno, pgno, + sqlite3GlobalConfig.pcache2.xPagecount(pCache->pCache), + numberOfCachePages(pCache)); +#endif + pcacheTrace(("%p.SPILL %d\n",pCache,pPg->pgno)); + rc = pCache->xStress(pCache->pStress, pPg); + pcacheDump(pCache); + if( rc!=SQLITE_OK && rc!=SQLITE_BUSY ){ + return rc; + } + } + } + *ppPage = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache, pgno, 2); + return *ppPage==0 ? SQLITE_NOMEM_BKPT : SQLITE_OK; +} + +/* +** This is a helper routine for sqlite3PcacheFetchFinish() +** +** In the uncommon case where the page being fetched has not been +** initialized, this routine is invoked to do the initialization. +** This routine is broken out into a separate function since it +** requires extra stack manipulation that can be avoided in the common +** case. +*/ +static SQLITE_NOINLINE PgHdr *pcacheFetchFinishWithInit( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number obtained */ + sqlite3_pcache_page *pPage /* Page obtained by prior PcacheFetch() call */ +){ + PgHdr *pPgHdr; + assert( pPage!=0 ); + pPgHdr = (PgHdr*)pPage->pExtra; + assert( pPgHdr->pPage==0 ); + memset(&pPgHdr->pDirty, 0, sizeof(PgHdr) - offsetof(PgHdr,pDirty)); + pPgHdr->pPage = pPage; + pPgHdr->pData = pPage->pBuf; + pPgHdr->pExtra = (void *)&pPgHdr[1]; + memset(pPgHdr->pExtra, 0, 8); + pPgHdr->pCache = pCache; + pPgHdr->pgno = pgno; + pPgHdr->flags = PGHDR_CLEAN; + return sqlite3PcacheFetchFinish(pCache,pgno,pPage); +} + +/* +** This routine converts the sqlite3_pcache_page object returned by +** sqlite3PcacheFetch() into an initialized PgHdr object. This routine +** must be called after sqlite3PcacheFetch() in order to get a usable +** result. +*/ +PgHdr *sqlite3PcacheFetchFinish( + PCache *pCache, /* Obtain the page from this cache */ + Pgno pgno, /* Page number obtained */ + sqlite3_pcache_page *pPage /* Page obtained by prior PcacheFetch() call */ +){ + PgHdr *pPgHdr; + + assert( pPage!=0 ); + pPgHdr = (PgHdr *)pPage->pExtra; + + if( !pPgHdr->pPage ){ + return pcacheFetchFinishWithInit(pCache, pgno, pPage); + } + pCache->nRefSum++; + pPgHdr->nRef++; + assert( sqlite3PcachePageSanity(pPgHdr) ); + return pPgHdr; +} + +/* +** Decrement the reference count on a page. If the page is clean and the +** reference count drops to 0, then it is made eligible for recycling. +*/ +void SQLITE_NOINLINE sqlite3PcacheRelease(PgHdr *p){ + assert( p->nRef>0 ); + p->pCache->nRefSum--; + if( (--p->nRef)==0 ){ + if( p->flags&PGHDR_CLEAN ){ + pcacheUnpin(p); + }else{ + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_FRONT); + } + } +} + +/* +** Increase the reference count of a supplied page by 1. +*/ +void sqlite3PcacheRef(PgHdr *p){ + assert(p->nRef>0); + assert( sqlite3PcachePageSanity(p) ); + p->nRef++; + p->pCache->nRefSum++; +} + +/* +** Drop a page from the cache. There must be exactly one reference to the +** page. This function deletes that reference, so after it returns the +** page pointed to by p is invalid. +*/ +void sqlite3PcacheDrop(PgHdr *p){ + assert( p->nRef==1 ); + assert( sqlite3PcachePageSanity(p) ); + if( p->flags&PGHDR_DIRTY ){ + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_REMOVE); + } + p->pCache->nRefSum--; + sqlite3GlobalConfig.pcache2.xUnpin(p->pCache->pCache, p->pPage, 1); +} + +/* +** Make sure the page is marked as dirty. If it isn't dirty already, +** make it so. +*/ +void sqlite3PcacheMakeDirty(PgHdr *p){ + assert( p->nRef>0 ); + assert( sqlite3PcachePageSanity(p) ); + if( p->flags & (PGHDR_CLEAN|PGHDR_DONT_WRITE) ){ /*OPTIMIZATION-IF-FALSE*/ + p->flags &= ~PGHDR_DONT_WRITE; + if( p->flags & PGHDR_CLEAN ){ + p->flags ^= (PGHDR_DIRTY|PGHDR_CLEAN); + pcacheTrace(("%p.DIRTY %d\n",p->pCache,p->pgno)); + assert( (p->flags & (PGHDR_DIRTY|PGHDR_CLEAN))==PGHDR_DIRTY ); + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_ADD); + } + assert( sqlite3PcachePageSanity(p) ); + } +} + +/* +** Make sure the page is marked as clean. If it isn't clean already, +** make it so. +*/ +void sqlite3PcacheMakeClean(PgHdr *p){ + assert( sqlite3PcachePageSanity(p) ); + assert( (p->flags & PGHDR_DIRTY)!=0 ); + assert( (p->flags & PGHDR_CLEAN)==0 ); + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_REMOVE); + p->flags &= ~(PGHDR_DIRTY|PGHDR_NEED_SYNC|PGHDR_WRITEABLE); + p->flags |= PGHDR_CLEAN; + pcacheTrace(("%p.CLEAN %d\n",p->pCache,p->pgno)); + assert( sqlite3PcachePageSanity(p) ); + if( p->nRef==0 ){ + pcacheUnpin(p); + } +} + +/* +** Make every page in the cache clean. +*/ +void sqlite3PcacheCleanAll(PCache *pCache){ + PgHdr *p; + pcacheTrace(("%p.CLEAN-ALL\n",pCache)); + while( (p = pCache->pDirty)!=0 ){ + sqlite3PcacheMakeClean(p); + } +} + +/* +** Clear the PGHDR_NEED_SYNC and PGHDR_WRITEABLE flag from all dirty pages. +*/ +void sqlite3PcacheClearWritable(PCache *pCache){ + PgHdr *p; + pcacheTrace(("%p.CLEAR-WRITEABLE\n",pCache)); + for(p=pCache->pDirty; p; p=p->pDirtyNext){ + p->flags &= ~(PGHDR_NEED_SYNC|PGHDR_WRITEABLE); + } + pCache->pSynced = pCache->pDirtyTail; +} + +/* +** Clear the PGHDR_NEED_SYNC flag from all dirty pages. +*/ +void sqlite3PcacheClearSyncFlags(PCache *pCache){ + PgHdr *p; + for(p=pCache->pDirty; p; p=p->pDirtyNext){ + p->flags &= ~PGHDR_NEED_SYNC; + } + pCache->pSynced = pCache->pDirtyTail; +} + +/* +** Change the page number of page p to newPgno. +*/ +void sqlite3PcacheMove(PgHdr *p, Pgno newPgno){ + PCache *pCache = p->pCache; + assert( p->nRef>0 ); + assert( newPgno>0 ); + assert( sqlite3PcachePageSanity(p) ); + pcacheTrace(("%p.MOVE %d -> %d\n",pCache,p->pgno,newPgno)); + sqlite3GlobalConfig.pcache2.xRekey(pCache->pCache, p->pPage, p->pgno,newPgno); + p->pgno = newPgno; + if( (p->flags&PGHDR_DIRTY) && (p->flags&PGHDR_NEED_SYNC) ){ + pcacheManageDirtyList(p, PCACHE_DIRTYLIST_FRONT); + } +} + +/* +** Drop every cache entry whose page number is greater than "pgno". The +** caller must ensure that there are no outstanding references to any pages +** other than page 1 with a page number greater than pgno. +** +** If there is a reference to page 1 and the pgno parameter passed to this +** function is 0, then the data area associated with page 1 is zeroed, but +** the page object is not dropped. +*/ +void sqlite3PcacheTruncate(PCache *pCache, Pgno pgno){ + if( pCache->pCache ){ + PgHdr *p; + PgHdr *pNext; + pcacheTrace(("%p.TRUNCATE %d\n",pCache,pgno)); + for(p=pCache->pDirty; p; p=pNext){ + pNext = p->pDirtyNext; + /* This routine never gets call with a positive pgno except right + ** after sqlite3PcacheCleanAll(). So if there are dirty pages, + ** it must be that pgno==0. + */ + assert( p->pgno>0 ); + if( p->pgno>pgno ){ + assert( p->flags&PGHDR_DIRTY ); + sqlite3PcacheMakeClean(p); + } + } + if( pgno==0 && pCache->nRefSum ){ + sqlite3_pcache_page *pPage1; + pPage1 = sqlite3GlobalConfig.pcache2.xFetch(pCache->pCache,1,0); + if( ALWAYS(pPage1) ){ /* Page 1 is always available in cache, because + ** pCache->nRefSum>0 */ + memset(pPage1->pBuf, 0, pCache->szPage); + pgno = 1; + } + } + sqlite3GlobalConfig.pcache2.xTruncate(pCache->pCache, pgno+1); + } +} + +/* +** Close a cache. +*/ +void sqlite3PcacheClose(PCache *pCache){ + assert( pCache->pCache!=0 ); + pcacheTrace(("%p.CLOSE\n",pCache)); + sqlite3GlobalConfig.pcache2.xDestroy(pCache->pCache); +} + +/* +** Discard the contents of the cache. +*/ +void sqlite3PcacheClear(PCache *pCache){ + sqlite3PcacheTruncate(pCache, 0); +} + +/* +** Merge two lists of pages connected by pDirty and in pgno order. +** Do not bother fixing the pDirtyPrev pointers. +*/ +static PgHdr *pcacheMergeDirtyList(PgHdr *pA, PgHdr *pB){ + PgHdr result, *pTail; + pTail = &result; + assert( pA!=0 && pB!=0 ); + for(;;){ + if( pA->pgnopgno ){ + pTail->pDirty = pA; + pTail = pA; + pA = pA->pDirty; + if( pA==0 ){ + pTail->pDirty = pB; + break; + } + }else{ + pTail->pDirty = pB; + pTail = pB; + pB = pB->pDirty; + if( pB==0 ){ + pTail->pDirty = pA; + break; + } + } + } + return result.pDirty; +} + +/* +** Sort the list of pages in accending order by pgno. Pages are +** connected by pDirty pointers. The pDirtyPrev pointers are +** corrupted by this sort. +** +** Since there cannot be more than 2^31 distinct pages in a database, +** there cannot be more than 31 buckets required by the merge sorter. +** One extra bucket is added to catch overflow in case something +** ever changes to make the previous sentence incorrect. +*/ +#define N_SORT_BUCKET 32 +static PgHdr *pcacheSortDirtyList(PgHdr *pIn){ + PgHdr *a[N_SORT_BUCKET], *p; + int i; + memset(a, 0, sizeof(a)); + while( pIn ){ + p = pIn; + pIn = p->pDirty; + p->pDirty = 0; + for(i=0; ALWAYS(ipDirty; p; p=p->pDirtyNext){ + p->pDirty = p->pDirtyNext; + } + return pcacheSortDirtyList(pCache->pDirty); +} + +/* +** Return the total number of references to all pages held by the cache. +** +** This is not the total number of pages referenced, but the sum of the +** reference count for all pages. +*/ +int sqlite3PcacheRefCount(PCache *pCache){ + return pCache->nRefSum; +} + +/* +** Return the number of references to the page supplied as an argument. +*/ +int sqlite3PcachePageRefcount(PgHdr *p){ + return p->nRef; +} + +/* +** Return the total number of pages in the cache. +*/ +int sqlite3PcachePagecount(PCache *pCache){ + assert( pCache->pCache!=0 ); + return sqlite3GlobalConfig.pcache2.xPagecount(pCache->pCache); +} + +#ifdef SQLITE_TEST +/* +** Get the suggested cache-size value. +*/ +int sqlite3PcacheGetCachesize(PCache *pCache){ + return numberOfCachePages(pCache); +} +#endif + +/* +** Set the suggested cache-size value. +*/ +void sqlite3PcacheSetCachesize(PCache *pCache, int mxPage){ + assert( pCache->pCache!=0 ); + pCache->szCache = mxPage; + sqlite3GlobalConfig.pcache2.xCachesize(pCache->pCache, + numberOfCachePages(pCache)); +} + +/* +** Set the suggested cache-spill value. Make no changes if if the +** argument is zero. Return the effective cache-spill size, which will +** be the larger of the szSpill and szCache. +*/ +int sqlite3PcacheSetSpillsize(PCache *p, int mxPage){ + int res; + assert( p->pCache!=0 ); + if( mxPage ){ + if( mxPage<0 ){ + mxPage = (int)((-1024*(i64)mxPage)/(p->szPage+p->szExtra)); + } + p->szSpill = mxPage; + } + res = numberOfCachePages(p); + if( resszSpill ) res = p->szSpill; + return res; +} + +/* +** Free up as much memory as possible from the page cache. +*/ +void sqlite3PcacheShrink(PCache *pCache){ + assert( pCache->pCache!=0 ); + sqlite3GlobalConfig.pcache2.xShrink(pCache->pCache); +} + +/* +** Return the size of the header added by this middleware layer +** in the page-cache hierarchy. +*/ +int sqlite3HeaderSizePcache(void){ return ROUND8(sizeof(PgHdr)); } + +/* +** Return the number of dirty pages currently in the cache, as a percentage +** of the configured cache size. +*/ +int sqlite3PCachePercentDirty(PCache *pCache){ + PgHdr *pDirty; + int nDirty = 0; + int nCache = numberOfCachePages(pCache); + for(pDirty=pCache->pDirty; pDirty; pDirty=pDirty->pDirtyNext) nDirty++; + return nCache ? (int)(((i64)nDirty * 100) / nCache) : 0; +} + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +/* +** Return true if there are one or more dirty pages in the cache. Else false. +*/ +int sqlite3PCacheIsDirty(PCache *pCache){ + return (pCache->pDirty!=0); +} +#endif + +#if defined(SQLITE_CHECK_PAGES) || defined(SQLITE_DEBUG) +/* +** For all dirty pages currently in the cache, invoke the specified +** callback. This is only used if the SQLITE_CHECK_PAGES macro is +** defined. +*/ +void sqlite3PcacheIterateDirty(PCache *pCache, void (*xIter)(PgHdr *)){ + PgHdr *pDirty; + for(pDirty=pCache->pDirty; pDirty; pDirty=pDirty->pDirtyNext){ + xIter(pDirty); + } +} +#endif diff --git a/source/libs/tdb/src/sqlite/pcache1.c b/source/libs/tdb/src/sqlite/pcache1.c new file mode 100644 index 0000000000..a93b146894 --- /dev/null +++ b/source/libs/tdb/src/sqlite/pcache1.c @@ -0,0 +1,1284 @@ +/* +** 2008 November 05 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file implements the default page cache implementation (the +** sqlite3_pcache interface). It also contains part of the implementation +** of the SQLITE_CONFIG_PAGECACHE and sqlite3_release_memory() features. +** If the default page cache implementation is overridden, then neither of +** these two features are available. +** +** A Page cache line looks like this: +** +** ------------------------------------------------------------- +** | database page content | PgHdr1 | MemPage | PgHdr | +** ------------------------------------------------------------- +** +** The database page content is up front (so that buffer overreads tend to +** flow harmlessly into the PgHdr1, MemPage, and PgHdr extensions). MemPage +** is the extension added by the btree.c module containing information such +** as the database page number and how that database page is used. PgHdr +** is added by the pcache.c layer and contains information used to keep track +** of which pages are "dirty". PgHdr1 is an extension added by this +** module (pcache1.c). The PgHdr1 header is a subclass of sqlite3_pcache_page. +** PgHdr1 contains information needed to look up a page by its page number. +** The superclass sqlite3_pcache_page.pBuf points to the start of the +** database page content and sqlite3_pcache_page.pExtra points to PgHdr. +** +** The size of the extension (MemPage+PgHdr+PgHdr1) can be determined at +** runtime using sqlite3_config(SQLITE_CONFIG_PCACHE_HDRSZ, &size). The +** sizes of the extensions sum to 272 bytes on x64 for 3.8.10, but this +** size can vary according to architecture, compile-time options, and +** SQLite library version number. +** +** If SQLITE_PCACHE_SEPARATE_HEADER is defined, then the extension is obtained +** using a separate memory allocation from the database page content. This +** seeks to overcome the "clownshoe" problem (also called "internal +** fragmentation" in academic literature) of allocating a few bytes more +** than a power of two with the memory allocator rounding up to the next +** power of two, and leaving the rounded-up space unused. +** +** This module tracks pointers to PgHdr1 objects. Only pcache.c communicates +** with this module. Information is passed back and forth as PgHdr1 pointers. +** +** The pcache.c and pager.c modules deal pointers to PgHdr objects. +** The btree.c module deals with pointers to MemPage objects. +** +** SOURCE OF PAGE CACHE MEMORY: +** +** Memory for a page might come from any of three sources: +** +** (1) The general-purpose memory allocator - sqlite3Malloc() +** (2) Global page-cache memory provided using sqlite3_config() with +** SQLITE_CONFIG_PAGECACHE. +** (3) PCache-local bulk allocation. +** +** The third case is a chunk of heap memory (defaulting to 100 pages worth) +** that is allocated when the page cache is created. The size of the local +** bulk allocation can be adjusted using +** +** sqlite3_config(SQLITE_CONFIG_PAGECACHE, (void*)0, 0, N). +** +** If N is positive, then N pages worth of memory are allocated using a single +** sqlite3Malloc() call and that memory is used for the first N pages allocated. +** Or if N is negative, then -1024*N bytes of memory are allocated and used +** for as many pages as can be accomodated. +** +** Only one of (2) or (3) can be used. Once the memory available to (2) or +** (3) is exhausted, subsequent allocations fail over to the general-purpose +** memory allocator (1). +** +** Earlier versions of SQLite used only methods (1) and (2). But experiments +** show that method (3) with N==100 provides about a 5% performance boost for +** common workloads. +*/ +#include "sqliteInt.h" + +typedef struct PCache1 PCache1; +typedef struct PgHdr1 PgHdr1; +typedef struct PgFreeslot PgFreeslot; +typedef struct PGroup PGroup; + +/* +** Each cache entry is represented by an instance of the following +** structure. Unless SQLITE_PCACHE_SEPARATE_HEADER is defined, a buffer of +** PgHdr1.pCache->szPage bytes is allocated directly before this structure +** in memory. +** +** Note: Variables isBulkLocal and isAnchor were once type "u8". That works, +** but causes a 2-byte gap in the structure for most architectures (since +** pointers must be either 4 or 8-byte aligned). As this structure is located +** in memory directly after the associated page data, if the database is +** corrupt, code at the b-tree layer may overread the page buffer and +** read part of this structure before the corruption is detected. This +** can cause a valgrind error if the unitialized gap is accessed. Using u16 +** ensures there is no such gap, and therefore no bytes of unitialized memory +** in the structure. +*/ +struct PgHdr1 { + sqlite3_pcache_page page; /* Base class. Must be first. pBuf & pExtra */ + unsigned int iKey; /* Key value (page number) */ + u16 isBulkLocal; /* This page from bulk local storage */ + u16 isAnchor; /* This is the PGroup.lru element */ + PgHdr1 *pNext; /* Next in hash table chain */ + PCache1 *pCache; /* Cache that currently owns this page */ + PgHdr1 *pLruNext; /* Next in LRU list of unpinned pages */ + PgHdr1 *pLruPrev; /* Previous in LRU list of unpinned pages */ + /* NB: pLruPrev is only valid if pLruNext!=0 */ +}; + +/* +** A page is pinned if it is not on the LRU list. To be "pinned" means +** that the page is in active use and must not be deallocated. +*/ +#define PAGE_IS_PINNED(p) ((p)->pLruNext==0) +#define PAGE_IS_UNPINNED(p) ((p)->pLruNext!=0) + +/* Each page cache (or PCache) belongs to a PGroup. A PGroup is a set +** of one or more PCaches that are able to recycle each other's unpinned +** pages when they are under memory pressure. A PGroup is an instance of +** the following object. +** +** This page cache implementation works in one of two modes: +** +** (1) Every PCache is the sole member of its own PGroup. There is +** one PGroup per PCache. +** +** (2) There is a single global PGroup that all PCaches are a member +** of. +** +** Mode 1 uses more memory (since PCache instances are not able to rob +** unused pages from other PCaches) but it also operates without a mutex, +** and is therefore often faster. Mode 2 requires a mutex in order to be +** threadsafe, but recycles pages more efficiently. +** +** For mode (1), PGroup.mutex is NULL. For mode (2) there is only a single +** PGroup which is the pcache1.grp global variable and its mutex is +** SQLITE_MUTEX_STATIC_LRU. +*/ +struct PGroup { + sqlite3_mutex *mutex; /* MUTEX_STATIC_LRU or NULL */ + unsigned int nMaxPage; /* Sum of nMax for purgeable caches */ + unsigned int nMinPage; /* Sum of nMin for purgeable caches */ + unsigned int mxPinned; /* nMaxpage + 10 - nMinPage */ + unsigned int nPurgeable; /* Number of purgeable pages allocated */ + PgHdr1 lru; /* The beginning and end of the LRU list */ +}; + +/* Each page cache is an instance of the following object. Every +** open database file (including each in-memory database and each +** temporary or transient database) has a single page cache which +** is an instance of this object. +** +** Pointers to structures of this type are cast and returned as +** opaque sqlite3_pcache* handles. +*/ +struct PCache1 { + /* Cache configuration parameters. Page size (szPage) and the purgeable + ** flag (bPurgeable) and the pnPurgeable pointer are all set when the + ** cache is created and are never changed thereafter. nMax may be + ** modified at any time by a call to the pcache1Cachesize() method. + ** The PGroup mutex must be held when accessing nMax. + */ + PGroup *pGroup; /* PGroup this cache belongs to */ + unsigned int *pnPurgeable; /* Pointer to pGroup->nPurgeable */ + int szPage; /* Size of database content section */ + int szExtra; /* sizeof(MemPage)+sizeof(PgHdr) */ + int szAlloc; /* Total size of one pcache line */ + int bPurgeable; /* True if cache is purgeable */ + unsigned int nMin; /* Minimum number of pages reserved */ + unsigned int nMax; /* Configured "cache_size" value */ + unsigned int n90pct; /* nMax*9/10 */ + unsigned int iMaxKey; /* Largest key seen since xTruncate() */ + unsigned int nPurgeableDummy; /* pnPurgeable points here when not used*/ + + /* Hash table of all pages. The following variables may only be accessed + ** when the accessor is holding the PGroup mutex. + */ + unsigned int nRecyclable; /* Number of pages in the LRU list */ + unsigned int nPage; /* Total number of pages in apHash */ + unsigned int nHash; /* Number of slots in apHash[] */ + PgHdr1 **apHash; /* Hash table for fast lookup by key */ + PgHdr1 *pFree; /* List of unused pcache-local pages */ + void *pBulk; /* Bulk memory used by pcache-local */ +}; + +/* +** Free slots in the allocator used to divide up the global page cache +** buffer provided using the SQLITE_CONFIG_PAGECACHE mechanism. +*/ +struct PgFreeslot { + PgFreeslot *pNext; /* Next free slot */ +}; + +/* +** Global data used by this cache. +*/ +static SQLITE_WSD struct PCacheGlobal { + PGroup grp; /* The global PGroup for mode (2) */ + + /* Variables related to SQLITE_CONFIG_PAGECACHE settings. The + ** szSlot, nSlot, pStart, pEnd, nReserve, and isInit values are all + ** fixed at sqlite3_initialize() time and do not require mutex protection. + ** The nFreeSlot and pFree values do require mutex protection. + */ + int isInit; /* True if initialized */ + int separateCache; /* Use a new PGroup for each PCache */ + int nInitPage; /* Initial bulk allocation size */ + int szSlot; /* Size of each free slot */ + int nSlot; /* The number of pcache slots */ + int nReserve; /* Try to keep nFreeSlot above this */ + void *pStart, *pEnd; /* Bounds of global page cache memory */ + /* Above requires no mutex. Use mutex below for variable that follow. */ + sqlite3_mutex *mutex; /* Mutex for accessing the following: */ + PgFreeslot *pFree; /* Free page blocks */ + int nFreeSlot; /* Number of unused pcache slots */ + /* The following value requires a mutex to change. We skip the mutex on + ** reading because (1) most platforms read a 32-bit integer atomically and + ** (2) even if an incorrect value is read, no great harm is done since this + ** is really just an optimization. */ + int bUnderPressure; /* True if low on PAGECACHE memory */ +} pcache1_g; + +/* +** All code in this file should access the global structure above via the +** alias "pcache1". This ensures that the WSD emulation is used when +** compiling for systems that do not support real WSD. +*/ +#define pcache1 (GLOBAL(struct PCacheGlobal, pcache1_g)) + +/* +** Macros to enter and leave the PCache LRU mutex. +*/ +#if !defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) || SQLITE_THREADSAFE==0 +# define pcache1EnterMutex(X) assert((X)->mutex==0) +# define pcache1LeaveMutex(X) assert((X)->mutex==0) +# define PCACHE1_MIGHT_USE_GROUP_MUTEX 0 +#else +# define pcache1EnterMutex(X) sqlite3_mutex_enter((X)->mutex) +# define pcache1LeaveMutex(X) sqlite3_mutex_leave((X)->mutex) +# define PCACHE1_MIGHT_USE_GROUP_MUTEX 1 +#endif + +/******************************************************************************/ +/******** Page Allocation/SQLITE_CONFIG_PCACHE Related Functions **************/ + + +/* +** This function is called during initialization if a static buffer is +** supplied to use for the page-cache by passing the SQLITE_CONFIG_PAGECACHE +** verb to sqlite3_config(). Parameter pBuf points to an allocation large +** enough to contain 'n' buffers of 'sz' bytes each. +** +** This routine is called from sqlite3_initialize() and so it is guaranteed +** to be serialized already. There is no need for further mutexing. +*/ +void sqlite3PCacheBufferSetup(void *pBuf, int sz, int n){ + if( pcache1.isInit ){ + PgFreeslot *p; + if( pBuf==0 ) sz = n = 0; + if( n==0 ) sz = 0; + sz = ROUNDDOWN8(sz); + pcache1.szSlot = sz; + pcache1.nSlot = pcache1.nFreeSlot = n; + pcache1.nReserve = n>90 ? 10 : (n/10 + 1); + pcache1.pStart = pBuf; + pcache1.pFree = 0; + pcache1.bUnderPressure = 0; + while( n-- ){ + p = (PgFreeslot*)pBuf; + p->pNext = pcache1.pFree; + pcache1.pFree = p; + pBuf = (void*)&((char*)pBuf)[sz]; + } + pcache1.pEnd = pBuf; + } +} + +/* +** Try to initialize the pCache->pFree and pCache->pBulk fields. Return +** true if pCache->pFree ends up containing one or more free pages. +*/ +static int pcache1InitBulk(PCache1 *pCache){ + i64 szBulk; + char *zBulk; + if( pcache1.nInitPage==0 ) return 0; + /* Do not bother with a bulk allocation if the cache size very small */ + if( pCache->nMax<3 ) return 0; + sqlite3BeginBenignMalloc(); + if( pcache1.nInitPage>0 ){ + szBulk = pCache->szAlloc * (i64)pcache1.nInitPage; + }else{ + szBulk = -1024 * (i64)pcache1.nInitPage; + } + if( szBulk > pCache->szAlloc*(i64)pCache->nMax ){ + szBulk = pCache->szAlloc*(i64)pCache->nMax; + } + zBulk = pCache->pBulk = sqlite3Malloc( szBulk ); + sqlite3EndBenignMalloc(); + if( zBulk ){ + int nBulk = sqlite3MallocSize(zBulk)/pCache->szAlloc; + do{ + PgHdr1 *pX = (PgHdr1*)&zBulk[pCache->szPage]; + pX->page.pBuf = zBulk; + pX->page.pExtra = &pX[1]; + pX->isBulkLocal = 1; + pX->isAnchor = 0; + pX->pNext = pCache->pFree; + pX->pLruPrev = 0; /* Initializing this saves a valgrind error */ + pCache->pFree = pX; + zBulk += pCache->szAlloc; + }while( --nBulk ); + } + return pCache->pFree!=0; +} + +/* +** Malloc function used within this file to allocate space from the buffer +** configured using sqlite3_config(SQLITE_CONFIG_PAGECACHE) option. If no +** such buffer exists or there is no space left in it, this function falls +** back to sqlite3Malloc(). +** +** Multiple threads can run this routine at the same time. Global variables +** in pcache1 need to be protected via mutex. +*/ +static void *pcache1Alloc(int nByte){ + void *p = 0; + assert( sqlite3_mutex_notheld(pcache1.grp.mutex) ); + if( nByte<=pcache1.szSlot ){ + sqlite3_mutex_enter(pcache1.mutex); + p = (PgHdr1 *)pcache1.pFree; + if( p ){ + pcache1.pFree = pcache1.pFree->pNext; + pcache1.nFreeSlot--; + pcache1.bUnderPressure = pcache1.nFreeSlot=0 ); + sqlite3StatusHighwater(SQLITE_STATUS_PAGECACHE_SIZE, nByte); + sqlite3StatusUp(SQLITE_STATUS_PAGECACHE_USED, 1); + } + sqlite3_mutex_leave(pcache1.mutex); + } + if( p==0 ){ + /* Memory is not available in the SQLITE_CONFIG_PAGECACHE pool. Get + ** it from sqlite3Malloc instead. + */ + p = sqlite3Malloc(nByte); +#ifndef SQLITE_DISABLE_PAGECACHE_OVERFLOW_STATS + if( p ){ + int sz = sqlite3MallocSize(p); + sqlite3_mutex_enter(pcache1.mutex); + sqlite3StatusHighwater(SQLITE_STATUS_PAGECACHE_SIZE, nByte); + sqlite3StatusUp(SQLITE_STATUS_PAGECACHE_OVERFLOW, sz); + sqlite3_mutex_leave(pcache1.mutex); + } +#endif + sqlite3MemdebugSetType(p, MEMTYPE_PCACHE); + } + return p; +} + +/* +** Free an allocated buffer obtained from pcache1Alloc(). +*/ +static void pcache1Free(void *p){ + if( p==0 ) return; + if( SQLITE_WITHIN(p, pcache1.pStart, pcache1.pEnd) ){ + PgFreeslot *pSlot; + sqlite3_mutex_enter(pcache1.mutex); + sqlite3StatusDown(SQLITE_STATUS_PAGECACHE_USED, 1); + pSlot = (PgFreeslot*)p; + pSlot->pNext = pcache1.pFree; + pcache1.pFree = pSlot; + pcache1.nFreeSlot++; + pcache1.bUnderPressure = pcache1.nFreeSlot=pcache1.pStart && ppGroup->mutex) ); + if( pCache->pFree || (pCache->nPage==0 && pcache1InitBulk(pCache)) ){ + assert( pCache->pFree!=0 ); + p = pCache->pFree; + pCache->pFree = p->pNext; + p->pNext = 0; + }else{ +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT + /* The group mutex must be released before pcache1Alloc() is called. This + ** is because it might call sqlite3_release_memory(), which assumes that + ** this mutex is not held. */ + assert( pcache1.separateCache==0 ); + assert( pCache->pGroup==&pcache1.grp ); + pcache1LeaveMutex(pCache->pGroup); +#endif + if( benignMalloc ){ sqlite3BeginBenignMalloc(); } +#ifdef SQLITE_PCACHE_SEPARATE_HEADER + pPg = pcache1Alloc(pCache->szPage); + p = sqlite3Malloc(sizeof(PgHdr1) + pCache->szExtra); + if( !pPg || !p ){ + pcache1Free(pPg); + sqlite3_free(p); + pPg = 0; + } +#else + pPg = pcache1Alloc(pCache->szAlloc); +#endif + if( benignMalloc ){ sqlite3EndBenignMalloc(); } +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT + pcache1EnterMutex(pCache->pGroup); +#endif + if( pPg==0 ) return 0; +#ifndef SQLITE_PCACHE_SEPARATE_HEADER + p = (PgHdr1 *)&((u8 *)pPg)[pCache->szPage]; +#endif + p->page.pBuf = pPg; + p->page.pExtra = &p[1]; + p->isBulkLocal = 0; + p->isAnchor = 0; + p->pLruPrev = 0; /* Initializing this saves a valgrind error */ + } + (*pCache->pnPurgeable)++; + return p; +} + +/* +** Free a page object allocated by pcache1AllocPage(). +*/ +static void pcache1FreePage(PgHdr1 *p){ + PCache1 *pCache; + assert( p!=0 ); + pCache = p->pCache; + assert( sqlite3_mutex_held(p->pCache->pGroup->mutex) ); + if( p->isBulkLocal ){ + p->pNext = pCache->pFree; + pCache->pFree = p; + }else{ + pcache1Free(p->page.pBuf); +#ifdef SQLITE_PCACHE_SEPARATE_HEADER + sqlite3_free(p); +#endif + } + (*pCache->pnPurgeable)--; +} + +/* +** Malloc function used by SQLite to obtain space from the buffer configured +** using sqlite3_config(SQLITE_CONFIG_PAGECACHE) option. If no such buffer +** exists, this function falls back to sqlite3Malloc(). +*/ +void *sqlite3PageMalloc(int sz){ + assert( sz<=65536+8 ); /* These allocations are never very large */ + return pcache1Alloc(sz); +} + +/* +** Free an allocated buffer obtained from sqlite3PageMalloc(). +*/ +void sqlite3PageFree(void *p){ + pcache1Free(p); +} + + +/* +** Return true if it desirable to avoid allocating a new page cache +** entry. +** +** If memory was allocated specifically to the page cache using +** SQLITE_CONFIG_PAGECACHE but that memory has all been used, then +** it is desirable to avoid allocating a new page cache entry because +** presumably SQLITE_CONFIG_PAGECACHE was suppose to be sufficient +** for all page cache needs and we should not need to spill the +** allocation onto the heap. +** +** Or, the heap is used for all page cache memory but the heap is +** under memory pressure, then again it is desirable to avoid +** allocating a new page cache entry in order to avoid stressing +** the heap even further. +*/ +static int pcache1UnderMemoryPressure(PCache1 *pCache){ + if( pcache1.nSlot && (pCache->szPage+pCache->szExtra)<=pcache1.szSlot ){ + return pcache1.bUnderPressure; + }else{ + return sqlite3HeapNearlyFull(); + } +} + +/******************************************************************************/ +/******** General Implementation Functions ************************************/ + +/* +** This function is used to resize the hash table used by the cache passed +** as the first argument. +** +** The PCache mutex must be held when this function is called. +*/ +static void pcache1ResizeHash(PCache1 *p){ + PgHdr1 **apNew; + unsigned int nNew; + unsigned int i; + + assert( sqlite3_mutex_held(p->pGroup->mutex) ); + + nNew = p->nHash*2; + if( nNew<256 ){ + nNew = 256; + } + + pcache1LeaveMutex(p->pGroup); + if( p->nHash ){ sqlite3BeginBenignMalloc(); } + apNew = (PgHdr1 **)sqlite3MallocZero(sizeof(PgHdr1 *)*nNew); + if( p->nHash ){ sqlite3EndBenignMalloc(); } + pcache1EnterMutex(p->pGroup); + if( apNew ){ + for(i=0; inHash; i++){ + PgHdr1 *pPage; + PgHdr1 *pNext = p->apHash[i]; + while( (pPage = pNext)!=0 ){ + unsigned int h = pPage->iKey % nNew; + pNext = pPage->pNext; + pPage->pNext = apNew[h]; + apNew[h] = pPage; + } + } + sqlite3_free(p->apHash); + p->apHash = apNew; + p->nHash = nNew; + } +} + +/* +** This function is used internally to remove the page pPage from the +** PGroup LRU list, if is part of it. If pPage is not part of the PGroup +** LRU list, then this function is a no-op. +** +** The PGroup mutex must be held when this function is called. +*/ +static PgHdr1 *pcache1PinPage(PgHdr1 *pPage){ + assert( pPage!=0 ); + assert( PAGE_IS_UNPINNED(pPage) ); + assert( pPage->pLruNext ); + assert( pPage->pLruPrev ); + assert( sqlite3_mutex_held(pPage->pCache->pGroup->mutex) ); + pPage->pLruPrev->pLruNext = pPage->pLruNext; + pPage->pLruNext->pLruPrev = pPage->pLruPrev; + pPage->pLruNext = 0; + /* pPage->pLruPrev = 0; + ** No need to clear pLruPrev as it is never accessed if pLruNext is 0 */ + assert( pPage->isAnchor==0 ); + assert( pPage->pCache->pGroup->lru.isAnchor==1 ); + pPage->pCache->nRecyclable--; + return pPage; +} + + +/* +** Remove the page supplied as an argument from the hash table +** (PCache1.apHash structure) that it is currently stored in. +** Also free the page if freePage is true. +** +** The PGroup mutex must be held when this function is called. +*/ +static void pcache1RemoveFromHash(PgHdr1 *pPage, int freeFlag){ + unsigned int h; + PCache1 *pCache = pPage->pCache; + PgHdr1 **pp; + + assert( sqlite3_mutex_held(pCache->pGroup->mutex) ); + h = pPage->iKey % pCache->nHash; + for(pp=&pCache->apHash[h]; (*pp)!=pPage; pp=&(*pp)->pNext); + *pp = (*pp)->pNext; + + pCache->nPage--; + if( freeFlag ) pcache1FreePage(pPage); +} + +/* +** If there are currently more than nMaxPage pages allocated, try +** to recycle pages to reduce the number allocated to nMaxPage. +*/ +static void pcache1EnforceMaxPage(PCache1 *pCache){ + PGroup *pGroup = pCache->pGroup; + PgHdr1 *p; + assert( sqlite3_mutex_held(pGroup->mutex) ); + while( pGroup->nPurgeable>pGroup->nMaxPage + && (p=pGroup->lru.pLruPrev)->isAnchor==0 + ){ + assert( p->pCache->pGroup==pGroup ); + assert( PAGE_IS_UNPINNED(p) ); + pcache1PinPage(p); + pcache1RemoveFromHash(p, 1); + } + if( pCache->nPage==0 && pCache->pBulk ){ + sqlite3_free(pCache->pBulk); + pCache->pBulk = pCache->pFree = 0; + } +} + +/* +** Discard all pages from cache pCache with a page number (key value) +** greater than or equal to iLimit. Any pinned pages that meet this +** criteria are unpinned before they are discarded. +** +** The PCache mutex must be held when this function is called. +*/ +static void pcache1TruncateUnsafe( + PCache1 *pCache, /* The cache to truncate */ + unsigned int iLimit /* Drop pages with this pgno or larger */ +){ + TESTONLY( int nPage = 0; ) /* To assert pCache->nPage is correct */ + unsigned int h, iStop; + assert( sqlite3_mutex_held(pCache->pGroup->mutex) ); + assert( pCache->iMaxKey >= iLimit ); + assert( pCache->nHash > 0 ); + if( pCache->iMaxKey - iLimit < pCache->nHash ){ + /* If we are just shaving the last few pages off the end of the + ** cache, then there is no point in scanning the entire hash table. + ** Only scan those hash slots that might contain pages that need to + ** be removed. */ + h = iLimit % pCache->nHash; + iStop = pCache->iMaxKey % pCache->nHash; + TESTONLY( nPage = -10; ) /* Disable the pCache->nPage validity check */ + }else{ + /* This is the general case where many pages are being removed. + ** It is necessary to scan the entire hash table */ + h = pCache->nHash/2; + iStop = h - 1; + } + for(;;){ + PgHdr1 **pp; + PgHdr1 *pPage; + assert( hnHash ); + pp = &pCache->apHash[h]; + while( (pPage = *pp)!=0 ){ + if( pPage->iKey>=iLimit ){ + pCache->nPage--; + *pp = pPage->pNext; + if( PAGE_IS_UNPINNED(pPage) ) pcache1PinPage(pPage); + pcache1FreePage(pPage); + }else{ + pp = &pPage->pNext; + TESTONLY( if( nPage>=0 ) nPage++; ) + } + } + if( h==iStop ) break; + h = (h+1) % pCache->nHash; + } + assert( nPage<0 || pCache->nPage==(unsigned)nPage ); +} + +/******************************************************************************/ +/******** sqlite3_pcache Methods **********************************************/ + +/* +** Implementation of the sqlite3_pcache.xInit method. +*/ +static int pcache1Init(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + assert( pcache1.isInit==0 ); + memset(&pcache1, 0, sizeof(pcache1)); + + + /* + ** The pcache1.separateCache variable is true if each PCache has its own + ** private PGroup (mode-1). pcache1.separateCache is false if the single + ** PGroup in pcache1.grp is used for all page caches (mode-2). + ** + ** * Always use a unified cache (mode-2) if ENABLE_MEMORY_MANAGEMENT + ** + ** * Use a unified cache in single-threaded applications that have + ** configured a start-time buffer for use as page-cache memory using + ** sqlite3_config(SQLITE_CONFIG_PAGECACHE, pBuf, sz, N) with non-NULL + ** pBuf argument. + ** + ** * Otherwise use separate caches (mode-1) + */ +#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) + pcache1.separateCache = 0; +#elif SQLITE_THREADSAFE + pcache1.separateCache = sqlite3GlobalConfig.pPage==0 + || sqlite3GlobalConfig.bCoreMutex>0; +#else + pcache1.separateCache = sqlite3GlobalConfig.pPage==0; +#endif + +#if SQLITE_THREADSAFE + if( sqlite3GlobalConfig.bCoreMutex ){ + pcache1.grp.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_LRU); + pcache1.mutex = sqlite3MutexAlloc(SQLITE_MUTEX_STATIC_PMEM); + } +#endif + if( pcache1.separateCache + && sqlite3GlobalConfig.nPage!=0 + && sqlite3GlobalConfig.pPage==0 + ){ + pcache1.nInitPage = sqlite3GlobalConfig.nPage; + }else{ + pcache1.nInitPage = 0; + } + pcache1.grp.mxPinned = 10; + pcache1.isInit = 1; + return SQLITE_OK; +} + +/* +** Implementation of the sqlite3_pcache.xShutdown method. +** Note that the static mutex allocated in xInit does +** not need to be freed. +*/ +static void pcache1Shutdown(void *NotUsed){ + UNUSED_PARAMETER(NotUsed); + assert( pcache1.isInit!=0 ); + memset(&pcache1, 0, sizeof(pcache1)); +} + +/* forward declaration */ +static void pcache1Destroy(sqlite3_pcache *p); + +/* +** Implementation of the sqlite3_pcache.xCreate method. +** +** Allocate a new cache. +*/ +static sqlite3_pcache *pcache1Create(int szPage, int szExtra, int bPurgeable){ + PCache1 *pCache; /* The newly created page cache */ + PGroup *pGroup; /* The group the new page cache will belong to */ + int sz; /* Bytes of memory required to allocate the new cache */ + + assert( (szPage & (szPage-1))==0 && szPage>=512 && szPage<=65536 ); + assert( szExtra < 300 ); + + sz = sizeof(PCache1) + sizeof(PGroup)*pcache1.separateCache; + pCache = (PCache1 *)sqlite3MallocZero(sz); + if( pCache ){ + if( pcache1.separateCache ){ + pGroup = (PGroup*)&pCache[1]; + pGroup->mxPinned = 10; + }else{ + pGroup = &pcache1.grp; + } + pcache1EnterMutex(pGroup); + if( pGroup->lru.isAnchor==0 ){ + pGroup->lru.isAnchor = 1; + pGroup->lru.pLruPrev = pGroup->lru.pLruNext = &pGroup->lru; + } + pCache->pGroup = pGroup; + pCache->szPage = szPage; + pCache->szExtra = szExtra; + pCache->szAlloc = szPage + szExtra + ROUND8(sizeof(PgHdr1)); + pCache->bPurgeable = (bPurgeable ? 1 : 0); + pcache1ResizeHash(pCache); + if( bPurgeable ){ + pCache->nMin = 10; + pGroup->nMinPage += pCache->nMin; + pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage; + pCache->pnPurgeable = &pGroup->nPurgeable; + }else{ + pCache->pnPurgeable = &pCache->nPurgeableDummy; + } + pcache1LeaveMutex(pGroup); + if( pCache->nHash==0 ){ + pcache1Destroy((sqlite3_pcache*)pCache); + pCache = 0; + } + } + return (sqlite3_pcache *)pCache; +} + +/* +** Implementation of the sqlite3_pcache.xCachesize method. +** +** Configure the cache_size limit for a cache. +*/ +static void pcache1Cachesize(sqlite3_pcache *p, int nMax){ + PCache1 *pCache = (PCache1 *)p; + u32 n; + assert( nMax>=0 ); + if( pCache->bPurgeable ){ + PGroup *pGroup = pCache->pGroup; + pcache1EnterMutex(pGroup); + n = (u32)nMax; + if( n > 0x7fff0000 - pGroup->nMaxPage + pCache->nMax ){ + n = 0x7fff0000 - pGroup->nMaxPage + pCache->nMax; + } + pGroup->nMaxPage += (n - pCache->nMax); + pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage; + pCache->nMax = n; + pCache->n90pct = pCache->nMax*9/10; + pcache1EnforceMaxPage(pCache); + pcache1LeaveMutex(pGroup); + } +} + +/* +** Implementation of the sqlite3_pcache.xShrink method. +** +** Free up as much memory as possible. +*/ +static void pcache1Shrink(sqlite3_pcache *p){ + PCache1 *pCache = (PCache1*)p; + if( pCache->bPurgeable ){ + PGroup *pGroup = pCache->pGroup; + unsigned int savedMaxPage; + pcache1EnterMutex(pGroup); + savedMaxPage = pGroup->nMaxPage; + pGroup->nMaxPage = 0; + pcache1EnforceMaxPage(pCache); + pGroup->nMaxPage = savedMaxPage; + pcache1LeaveMutex(pGroup); + } +} + +/* +** Implementation of the sqlite3_pcache.xPagecount method. +*/ +static int pcache1Pagecount(sqlite3_pcache *p){ + int n; + PCache1 *pCache = (PCache1*)p; + pcache1EnterMutex(pCache->pGroup); + n = pCache->nPage; + pcache1LeaveMutex(pCache->pGroup); + return n; +} + + +/* +** Implement steps 3, 4, and 5 of the pcache1Fetch() algorithm described +** in the header of the pcache1Fetch() procedure. +** +** This steps are broken out into a separate procedure because they are +** usually not needed, and by avoiding the stack initialization required +** for these steps, the main pcache1Fetch() procedure can run faster. +*/ +static SQLITE_NOINLINE PgHdr1 *pcache1FetchStage2( + PCache1 *pCache, + unsigned int iKey, + int createFlag +){ + unsigned int nPinned; + PGroup *pGroup = pCache->pGroup; + PgHdr1 *pPage = 0; + + /* Step 3: Abort if createFlag is 1 but the cache is nearly full */ + assert( pCache->nPage >= pCache->nRecyclable ); + nPinned = pCache->nPage - pCache->nRecyclable; + assert( pGroup->mxPinned == pGroup->nMaxPage + 10 - pGroup->nMinPage ); + assert( pCache->n90pct == pCache->nMax*9/10 ); + if( createFlag==1 && ( + nPinned>=pGroup->mxPinned + || nPinned>=pCache->n90pct + || (pcache1UnderMemoryPressure(pCache) && pCache->nRecyclablenPage>=pCache->nHash ) pcache1ResizeHash(pCache); + assert( pCache->nHash>0 && pCache->apHash ); + + /* Step 4. Try to recycle a page. */ + if( pCache->bPurgeable + && !pGroup->lru.pLruPrev->isAnchor + && ((pCache->nPage+1>=pCache->nMax) || pcache1UnderMemoryPressure(pCache)) + ){ + PCache1 *pOther; + pPage = pGroup->lru.pLruPrev; + assert( PAGE_IS_UNPINNED(pPage) ); + pcache1RemoveFromHash(pPage, 0); + pcache1PinPage(pPage); + pOther = pPage->pCache; + if( pOther->szAlloc != pCache->szAlloc ){ + pcache1FreePage(pPage); + pPage = 0; + }else{ + pGroup->nPurgeable -= (pOther->bPurgeable - pCache->bPurgeable); + } + } + + /* Step 5. If a usable page buffer has still not been found, + ** attempt to allocate a new one. + */ + if( !pPage ){ + pPage = pcache1AllocPage(pCache, createFlag==1); + } + + if( pPage ){ + unsigned int h = iKey % pCache->nHash; + pCache->nPage++; + pPage->iKey = iKey; + pPage->pNext = pCache->apHash[h]; + pPage->pCache = pCache; + pPage->pLruNext = 0; + /* pPage->pLruPrev = 0; + ** No need to clear pLruPrev since it is not accessed when pLruNext==0 */ + *(void **)pPage->page.pExtra = 0; + pCache->apHash[h] = pPage; + if( iKey>pCache->iMaxKey ){ + pCache->iMaxKey = iKey; + } + } + return pPage; +} + +/* +** Implementation of the sqlite3_pcache.xFetch method. +** +** Fetch a page by key value. +** +** Whether or not a new page may be allocated by this function depends on +** the value of the createFlag argument. 0 means do not allocate a new +** page. 1 means allocate a new page if space is easily available. 2 +** means to try really hard to allocate a new page. +** +** For a non-purgeable cache (a cache used as the storage for an in-memory +** database) there is really no difference between createFlag 1 and 2. So +** the calling function (pcache.c) will never have a createFlag of 1 on +** a non-purgeable cache. +** +** There are three different approaches to obtaining space for a page, +** depending on the value of parameter createFlag (which may be 0, 1 or 2). +** +** 1. Regardless of the value of createFlag, the cache is searched for a +** copy of the requested page. If one is found, it is returned. +** +** 2. If createFlag==0 and the page is not already in the cache, NULL is +** returned. +** +** 3. If createFlag is 1, and the page is not already in the cache, then +** return NULL (do not allocate a new page) if any of the following +** conditions are true: +** +** (a) the number of pages pinned by the cache is greater than +** PCache1.nMax, or +** +** (b) the number of pages pinned by the cache is greater than +** the sum of nMax for all purgeable caches, less the sum of +** nMin for all other purgeable caches, or +** +** 4. If none of the first three conditions apply and the cache is marked +** as purgeable, and if one of the following is true: +** +** (a) The number of pages allocated for the cache is already +** PCache1.nMax, or +** +** (b) The number of pages allocated for all purgeable caches is +** already equal to or greater than the sum of nMax for all +** purgeable caches, +** +** (c) The system is under memory pressure and wants to avoid +** unnecessary pages cache entry allocations +** +** then attempt to recycle a page from the LRU list. If it is the right +** size, return the recycled buffer. Otherwise, free the buffer and +** proceed to step 5. +** +** 5. Otherwise, allocate and return a new page buffer. +** +** There are two versions of this routine. pcache1FetchWithMutex() is +** the general case. pcache1FetchNoMutex() is a faster implementation for +** the common case where pGroup->mutex is NULL. The pcache1Fetch() wrapper +** invokes the appropriate routine. +*/ +static PgHdr1 *pcache1FetchNoMutex( + sqlite3_pcache *p, + unsigned int iKey, + int createFlag +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage = 0; + + /* Step 1: Search the hash table for an existing entry. */ + pPage = pCache->apHash[iKey % pCache->nHash]; + while( pPage && pPage->iKey!=iKey ){ pPage = pPage->pNext; } + + /* Step 2: If the page was found in the hash table, then return it. + ** If the page was not in the hash table and createFlag is 0, abort. + ** Otherwise (page not in hash and createFlag!=0) continue with + ** subsequent steps to try to create the page. */ + if( pPage ){ + if( PAGE_IS_UNPINNED(pPage) ){ + return pcache1PinPage(pPage); + }else{ + return pPage; + } + }else if( createFlag ){ + /* Steps 3, 4, and 5 implemented by this subroutine */ + return pcache1FetchStage2(pCache, iKey, createFlag); + }else{ + return 0; + } +} +#if PCACHE1_MIGHT_USE_GROUP_MUTEX +static PgHdr1 *pcache1FetchWithMutex( + sqlite3_pcache *p, + unsigned int iKey, + int createFlag +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage; + + pcache1EnterMutex(pCache->pGroup); + pPage = pcache1FetchNoMutex(p, iKey, createFlag); + assert( pPage==0 || pCache->iMaxKey>=iKey ); + pcache1LeaveMutex(pCache->pGroup); + return pPage; +} +#endif +static sqlite3_pcache_page *pcache1Fetch( + sqlite3_pcache *p, + unsigned int iKey, + int createFlag +){ +#if PCACHE1_MIGHT_USE_GROUP_MUTEX || defined(SQLITE_DEBUG) + PCache1 *pCache = (PCache1 *)p; +#endif + + assert( offsetof(PgHdr1,page)==0 ); + assert( pCache->bPurgeable || createFlag!=1 ); + assert( pCache->bPurgeable || pCache->nMin==0 ); + assert( pCache->bPurgeable==0 || pCache->nMin==10 ); + assert( pCache->nMin==0 || pCache->bPurgeable ); + assert( pCache->nHash>0 ); +#if PCACHE1_MIGHT_USE_GROUP_MUTEX + if( pCache->pGroup->mutex ){ + return (sqlite3_pcache_page*)pcache1FetchWithMutex(p, iKey, createFlag); + }else +#endif + { + return (sqlite3_pcache_page*)pcache1FetchNoMutex(p, iKey, createFlag); + } +} + + +/* +** Implementation of the sqlite3_pcache.xUnpin method. +** +** Mark a page as unpinned (eligible for asynchronous recycling). +*/ +static void pcache1Unpin( + sqlite3_pcache *p, + sqlite3_pcache_page *pPg, + int reuseUnlikely +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage = (PgHdr1 *)pPg; + PGroup *pGroup = pCache->pGroup; + + assert( pPage->pCache==pCache ); + pcache1EnterMutex(pGroup); + + /* It is an error to call this function if the page is already + ** part of the PGroup LRU list. + */ + assert( pPage->pLruNext==0 ); + assert( PAGE_IS_PINNED(pPage) ); + + if( reuseUnlikely || pGroup->nPurgeable>pGroup->nMaxPage ){ + pcache1RemoveFromHash(pPage, 1); + }else{ + /* Add the page to the PGroup LRU list. */ + PgHdr1 **ppFirst = &pGroup->lru.pLruNext; + pPage->pLruPrev = &pGroup->lru; + (pPage->pLruNext = *ppFirst)->pLruPrev = pPage; + *ppFirst = pPage; + pCache->nRecyclable++; + } + + pcache1LeaveMutex(pCache->pGroup); +} + +/* +** Implementation of the sqlite3_pcache.xRekey method. +*/ +static void pcache1Rekey( + sqlite3_pcache *p, + sqlite3_pcache_page *pPg, + unsigned int iOld, + unsigned int iNew +){ + PCache1 *pCache = (PCache1 *)p; + PgHdr1 *pPage = (PgHdr1 *)pPg; + PgHdr1 **pp; + unsigned int h; + assert( pPage->iKey==iOld ); + assert( pPage->pCache==pCache ); + + pcache1EnterMutex(pCache->pGroup); + + h = iOld%pCache->nHash; + pp = &pCache->apHash[h]; + while( (*pp)!=pPage ){ + pp = &(*pp)->pNext; + } + *pp = pPage->pNext; + + h = iNew%pCache->nHash; + pPage->iKey = iNew; + pPage->pNext = pCache->apHash[h]; + pCache->apHash[h] = pPage; + if( iNew>pCache->iMaxKey ){ + pCache->iMaxKey = iNew; + } + + pcache1LeaveMutex(pCache->pGroup); +} + +/* +** Implementation of the sqlite3_pcache.xTruncate method. +** +** Discard all unpinned pages in the cache with a page number equal to +** or greater than parameter iLimit. Any pinned pages with a page number +** equal to or greater than iLimit are implicitly unpinned. +*/ +static void pcache1Truncate(sqlite3_pcache *p, unsigned int iLimit){ + PCache1 *pCache = (PCache1 *)p; + pcache1EnterMutex(pCache->pGroup); + if( iLimit<=pCache->iMaxKey ){ + pcache1TruncateUnsafe(pCache, iLimit); + pCache->iMaxKey = iLimit-1; + } + pcache1LeaveMutex(pCache->pGroup); +} + +/* +** Implementation of the sqlite3_pcache.xDestroy method. +** +** Destroy a cache allocated using pcache1Create(). +*/ +static void pcache1Destroy(sqlite3_pcache *p){ + PCache1 *pCache = (PCache1 *)p; + PGroup *pGroup = pCache->pGroup; + assert( pCache->bPurgeable || (pCache->nMax==0 && pCache->nMin==0) ); + pcache1EnterMutex(pGroup); + if( pCache->nPage ) pcache1TruncateUnsafe(pCache, 0); + assert( pGroup->nMaxPage >= pCache->nMax ); + pGroup->nMaxPage -= pCache->nMax; + assert( pGroup->nMinPage >= pCache->nMin ); + pGroup->nMinPage -= pCache->nMin; + pGroup->mxPinned = pGroup->nMaxPage + 10 - pGroup->nMinPage; + pcache1EnforceMaxPage(pCache); + pcache1LeaveMutex(pGroup); + sqlite3_free(pCache->pBulk); + sqlite3_free(pCache->apHash); + sqlite3_free(pCache); +} + +/* +** This function is called during initialization (sqlite3_initialize()) to +** install the default pluggable cache module, assuming the user has not +** already provided an alternative. +*/ +void sqlite3PCacheSetDefault(void){ + static const sqlite3_pcache_methods2 defaultMethods = { + 1, /* iVersion */ + 0, /* pArg */ + pcache1Init, /* xInit */ + pcache1Shutdown, /* xShutdown */ + pcache1Create, /* xCreate */ + pcache1Cachesize, /* xCachesize */ + pcache1Pagecount, /* xPagecount */ + pcache1Fetch, /* xFetch */ + pcache1Unpin, /* xUnpin */ + pcache1Rekey, /* xRekey */ + pcache1Truncate, /* xTruncate */ + pcache1Destroy, /* xDestroy */ + pcache1Shrink /* xShrink */ + }; + sqlite3_config(SQLITE_CONFIG_PCACHE2, &defaultMethods); +} + +/* +** Return the size of the header on each page of this PCACHE implementation. +*/ +int sqlite3HeaderSizePcache1(void){ return ROUND8(sizeof(PgHdr1)); } + +/* +** Return the global mutex used by this PCACHE implementation. The +** sqlite3_status() routine needs access to this mutex. +*/ +sqlite3_mutex *sqlite3Pcache1Mutex(void){ + return pcache1.mutex; +} + +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT +/* +** This function is called to free superfluous dynamically allocated memory +** held by the pager system. Memory in use by any SQLite pager allocated +** by the current thread may be sqlite3_free()ed. +** +** nReq is the number of bytes of memory required. Once this much has +** been released, the function returns. The return value is the total number +** of bytes of memory released. +*/ +int sqlite3PcacheReleaseMemory(int nReq){ + int nFree = 0; + assert( sqlite3_mutex_notheld(pcache1.grp.mutex) ); + assert( sqlite3_mutex_notheld(pcache1.mutex) ); + if( sqlite3GlobalConfig.pPage==0 ){ + PgHdr1 *p; + pcache1EnterMutex(&pcache1.grp); + while( (nReq<0 || nFreeisAnchor==0 + ){ + nFree += pcache1MemSize(p->page.pBuf); +#ifdef SQLITE_PCACHE_SEPARATE_HEADER + nFree += sqlite3MemSize(p); +#endif + assert( PAGE_IS_UNPINNED(p) ); + pcache1PinPage(p); + pcache1RemoveFromHash(p, 1); + } + pcache1LeaveMutex(&pcache1.grp); + } + return nFree; +} +#endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT */ + +#ifdef SQLITE_TEST +/* +** This function is used by test procedures to inspect the internal state +** of the global cache. +*/ +void sqlite3PcacheStats( + int *pnCurrent, /* OUT: Total number of pages cached */ + int *pnMax, /* OUT: Global maximum cache size */ + int *pnMin, /* OUT: Sum of PCache1.nMin for purgeable caches */ + int *pnRecyclable /* OUT: Total number of pages available for recycling */ +){ + PgHdr1 *p; + int nRecyclable = 0; + for(p=pcache1.grp.lru.pLruNext; p && !p->isAnchor; p=p->pLruNext){ + assert( PAGE_IS_UNPINNED(p) ); + nRecyclable++; + } + *pnCurrent = pcache1.grp.nPurgeable; + *pnMax = (int)pcache1.grp.nMaxPage; + *pnMin = (int)pcache1.grp.nMinPage; + *pnRecyclable = nRecyclable; +} +#endif diff --git a/source/libs/tdb/src/sqlite/wal.c b/source/libs/tdb/src/sqlite/wal.c new file mode 100644 index 0000000000..a0ec91965c --- /dev/null +++ b/source/libs/tdb/src/sqlite/wal.c @@ -0,0 +1,4153 @@ +/* +** 2010 February 1 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** +** This file contains the implementation of a write-ahead log (WAL) used in +** "journal_mode=WAL" mode. +** +** WRITE-AHEAD LOG (WAL) FILE FORMAT +** +** A WAL file consists of a header followed by zero or more "frames". +** Each frame records the revised content of a single page from the +** database file. All changes to the database are recorded by writing +** frames into the WAL. Transactions commit when a frame is written that +** contains a commit marker. A single WAL can and usually does record +** multiple transactions. Periodically, the content of the WAL is +** transferred back into the database file in an operation called a +** "checkpoint". +** +** A single WAL file can be used multiple times. In other words, the +** WAL can fill up with frames and then be checkpointed and then new +** frames can overwrite the old ones. A WAL always grows from beginning +** toward the end. Checksums and counters attached to each frame are +** used to determine which frames within the WAL are valid and which +** are leftovers from prior checkpoints. +** +** The WAL header is 32 bytes in size and consists of the following eight +** big-endian 32-bit unsigned integer values: +** +** 0: Magic number. 0x377f0682 or 0x377f0683 +** 4: File format version. Currently 3007000 +** 8: Database page size. Example: 1024 +** 12: Checkpoint sequence number +** 16: Salt-1, random integer incremented with each checkpoint +** 20: Salt-2, a different random integer changing with each ckpt +** 24: Checksum-1 (first part of checksum for first 24 bytes of header). +** 28: Checksum-2 (second part of checksum for first 24 bytes of header). +** +** Immediately following the wal-header are zero or more frames. Each +** frame consists of a 24-byte frame-header followed by a bytes +** of page data. The frame-header is six big-endian 32-bit unsigned +** integer values, as follows: +** +** 0: Page number. +** 4: For commit records, the size of the database image in pages +** after the commit. For all other records, zero. +** 8: Salt-1 (copied from the header) +** 12: Salt-2 (copied from the header) +** 16: Checksum-1. +** 20: Checksum-2. +** +** A frame is considered valid if and only if the following conditions are +** true: +** +** (1) The salt-1 and salt-2 values in the frame-header match +** salt values in the wal-header +** +** (2) The checksum values in the final 8 bytes of the frame-header +** exactly match the checksum computed consecutively on the +** WAL header and the first 8 bytes and the content of all frames +** up to and including the current frame. +** +** The checksum is computed using 32-bit big-endian integers if the +** magic number in the first 4 bytes of the WAL is 0x377f0683 and it +** is computed using little-endian if the magic number is 0x377f0682. +** The checksum values are always stored in the frame header in a +** big-endian format regardless of which byte order is used to compute +** the checksum. The checksum is computed by interpreting the input as +** an even number of unsigned 32-bit integers: x[0] through x[N]. The +** algorithm used for the checksum is as follows: +** +** for i from 0 to n-1 step 2: +** s0 += x[i] + s1; +** s1 += x[i+1] + s0; +** endfor +** +** Note that s0 and s1 are both weighted checksums using fibonacci weights +** in reverse order (the largest fibonacci weight occurs on the first element +** of the sequence being summed.) The s1 value spans all 32-bit +** terms of the sequence whereas s0 omits the final term. +** +** On a checkpoint, the WAL is first VFS.xSync-ed, then valid content of the +** WAL is transferred into the database, then the database is VFS.xSync-ed. +** The VFS.xSync operations serve as write barriers - all writes launched +** before the xSync must complete before any write that launches after the +** xSync begins. +** +** After each checkpoint, the salt-1 value is incremented and the salt-2 +** value is randomized. This prevents old and new frames in the WAL from +** being considered valid at the same time and being checkpointing together +** following a crash. +** +** READER ALGORITHM +** +** To read a page from the database (call it page number P), a reader +** first checks the WAL to see if it contains page P. If so, then the +** last valid instance of page P that is a followed by a commit frame +** or is a commit frame itself becomes the value read. If the WAL +** contains no copies of page P that are valid and which are a commit +** frame or are followed by a commit frame, then page P is read from +** the database file. +** +** To start a read transaction, the reader records the index of the last +** valid frame in the WAL. The reader uses this recorded "mxFrame" value +** for all subsequent read operations. New transactions can be appended +** to the WAL, but as long as the reader uses its original mxFrame value +** and ignores the newly appended content, it will see a consistent snapshot +** of the database from a single point in time. This technique allows +** multiple concurrent readers to view different versions of the database +** content simultaneously. +** +** The reader algorithm in the previous paragraphs works correctly, but +** because frames for page P can appear anywhere within the WAL, the +** reader has to scan the entire WAL looking for page P frames. If the +** WAL is large (multiple megabytes is typical) that scan can be slow, +** and read performance suffers. To overcome this problem, a separate +** data structure called the wal-index is maintained to expedite the +** search for frames of a particular page. +** +** WAL-INDEX FORMAT +** +** Conceptually, the wal-index is shared memory, though VFS implementations +** might choose to implement the wal-index using a mmapped file. Because +** the wal-index is shared memory, SQLite does not support journal_mode=WAL +** on a network filesystem. All users of the database must be able to +** share memory. +** +** In the default unix and windows implementation, the wal-index is a mmapped +** file whose name is the database name with a "-shm" suffix added. For that +** reason, the wal-index is sometimes called the "shm" file. +** +** The wal-index is transient. After a crash, the wal-index can (and should +** be) reconstructed from the original WAL file. In fact, the VFS is required +** to either truncate or zero the header of the wal-index when the last +** connection to it closes. Because the wal-index is transient, it can +** use an architecture-specific format; it does not have to be cross-platform. +** Hence, unlike the database and WAL file formats which store all values +** as big endian, the wal-index can store multi-byte values in the native +** byte order of the host computer. +** +** The purpose of the wal-index is to answer this question quickly: Given +** a page number P and a maximum frame index M, return the index of the +** last frame in the wal before frame M for page P in the WAL, or return +** NULL if there are no frames for page P in the WAL prior to M. +** +** The wal-index consists of a header region, followed by an one or +** more index blocks. +** +** The wal-index header contains the total number of frames within the WAL +** in the mxFrame field. +** +** Each index block except for the first contains information on +** HASHTABLE_NPAGE frames. The first index block contains information on +** HASHTABLE_NPAGE_ONE frames. The values of HASHTABLE_NPAGE_ONE and +** HASHTABLE_NPAGE are selected so that together the wal-index header and +** first index block are the same size as all other index blocks in the +** wal-index. The values are: +** +** HASHTABLE_NPAGE 4096 +** HASHTABLE_NPAGE_ONE 4062 +** +** Each index block contains two sections, a page-mapping that contains the +** database page number associated with each wal frame, and a hash-table +** that allows readers to query an index block for a specific page number. +** The page-mapping is an array of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE +** for the first index block) 32-bit page numbers. The first entry in the +** first index-block contains the database page number corresponding to the +** first frame in the WAL file. The first entry in the second index block +** in the WAL file corresponds to the (HASHTABLE_NPAGE_ONE+1)th frame in +** the log, and so on. +** +** The last index block in a wal-index usually contains less than the full +** complement of HASHTABLE_NPAGE (or HASHTABLE_NPAGE_ONE) page-numbers, +** depending on the contents of the WAL file. This does not change the +** allocated size of the page-mapping array - the page-mapping array merely +** contains unused entries. +** +** Even without using the hash table, the last frame for page P +** can be found by scanning the page-mapping sections of each index block +** starting with the last index block and moving toward the first, and +** within each index block, starting at the end and moving toward the +** beginning. The first entry that equals P corresponds to the frame +** holding the content for that page. +** +** The hash table consists of HASHTABLE_NSLOT 16-bit unsigned integers. +** HASHTABLE_NSLOT = 2*HASHTABLE_NPAGE, and there is one entry in the +** hash table for each page number in the mapping section, so the hash +** table is never more than half full. The expected number of collisions +** prior to finding a match is 1. Each entry of the hash table is an +** 1-based index of an entry in the mapping section of the same +** index block. Let K be the 1-based index of the largest entry in +** the mapping section. (For index blocks other than the last, K will +** always be exactly HASHTABLE_NPAGE (4096) and for the last index block +** K will be (mxFrame%HASHTABLE_NPAGE).) Unused slots of the hash table +** contain a value of 0. +** +** To look for page P in the hash table, first compute a hash iKey on +** P as follows: +** +** iKey = (P * 383) % HASHTABLE_NSLOT +** +** Then start scanning entries of the hash table, starting with iKey +** (wrapping around to the beginning when the end of the hash table is +** reached) until an unused hash slot is found. Let the first unused slot +** be at index iUnused. (iUnused might be less than iKey if there was +** wrap-around.) Because the hash table is never more than half full, +** the search is guaranteed to eventually hit an unused entry. Let +** iMax be the value between iKey and iUnused, closest to iUnused, +** where aHash[iMax]==P. If there is no iMax entry (if there exists +** no hash slot such that aHash[i]==p) then page P is not in the +** current index block. Otherwise the iMax-th mapping entry of the +** current index block corresponds to the last entry that references +** page P. +** +** A hash search begins with the last index block and moves toward the +** first index block, looking for entries corresponding to page P. On +** average, only two or three slots in each index block need to be +** examined in order to either find the last entry for page P, or to +** establish that no such entry exists in the block. Each index block +** holds over 4000 entries. So two or three index blocks are sufficient +** to cover a typical 10 megabyte WAL file, assuming 1K pages. 8 or 10 +** comparisons (on average) suffice to either locate a frame in the +** WAL or to establish that the frame does not exist in the WAL. This +** is much faster than scanning the entire 10MB WAL. +** +** Note that entries are added in order of increasing K. Hence, one +** reader might be using some value K0 and a second reader that started +** at a later time (after additional transactions were added to the WAL +** and to the wal-index) might be using a different value K1, where K1>K0. +** Both readers can use the same hash table and mapping section to get +** the correct result. There may be entries in the hash table with +** K>K0 but to the first reader, those entries will appear to be unused +** slots in the hash table and so the first reader will get an answer as +** if no values greater than K0 had ever been inserted into the hash table +** in the first place - which is what reader one wants. Meanwhile, the +** second reader using K1 will see additional values that were inserted +** later, which is exactly what reader two wants. +** +** When a rollback occurs, the value of K is decreased. Hash table entries +** that correspond to frames greater than the new K value are removed +** from the hash table at this point. +*/ +#ifndef SQLITE_OMIT_WAL + +#include "wal.h" + +/* +** Trace output macros +*/ +#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG) +int sqlite3WalTrace = 0; +# define WALTRACE(X) if(sqlite3WalTrace) sqlite3DebugPrintf X +#else +# define WALTRACE(X) +#endif + +/* +** The maximum (and only) versions of the wal and wal-index formats +** that may be interpreted by this version of SQLite. +** +** If a client begins recovering a WAL file and finds that (a) the checksum +** values in the wal-header are correct and (b) the version field is not +** WAL_MAX_VERSION, recovery fails and SQLite returns SQLITE_CANTOPEN. +** +** Similarly, if a client successfully reads a wal-index header (i.e. the +** checksum test is successful) and finds that the version field is not +** WALINDEX_MAX_VERSION, then no read-transaction is opened and SQLite +** returns SQLITE_CANTOPEN. +*/ +#define WAL_MAX_VERSION 3007000 +#define WALINDEX_MAX_VERSION 3007000 + +/* +** Index numbers for various locking bytes. WAL_NREADER is the number +** of available reader locks and should be at least 3. The default +** is SQLITE_SHM_NLOCK==8 and WAL_NREADER==5. +** +** Technically, the various VFSes are free to implement these locks however +** they see fit. However, compatibility is encouraged so that VFSes can +** interoperate. The standard implemention used on both unix and windows +** is for the index number to indicate a byte offset into the +** WalCkptInfo.aLock[] array in the wal-index header. In other words, all +** locks are on the shm file. The WALINDEX_LOCK_OFFSET constant (which +** should be 120) is the location in the shm file for the first locking +** byte. +*/ +#define WAL_WRITE_LOCK 0 +#define WAL_ALL_BUT_WRITE 1 +#define WAL_CKPT_LOCK 1 +#define WAL_RECOVER_LOCK 2 +#define WAL_READ_LOCK(I) (3+(I)) +#define WAL_NREADER (SQLITE_SHM_NLOCK-3) + + +/* Object declarations */ +typedef struct WalIndexHdr WalIndexHdr; +typedef struct WalIterator WalIterator; +typedef struct WalCkptInfo WalCkptInfo; + + +/* +** The following object holds a copy of the wal-index header content. +** +** The actual header in the wal-index consists of two copies of this +** object followed by one instance of the WalCkptInfo object. +** For all versions of SQLite through 3.10.0 and probably beyond, +** the locking bytes (WalCkptInfo.aLock) start at offset 120 and +** the total header size is 136 bytes. +** +** The szPage value can be any power of 2 between 512 and 32768, inclusive. +** Or it can be 1 to represent a 65536-byte page. The latter case was +** added in 3.7.1 when support for 64K pages was added. +*/ +struct WalIndexHdr { + u32 iVersion; /* Wal-index version */ + u32 unused; /* Unused (padding) field */ + u32 iChange; /* Counter incremented each transaction */ + u8 isInit; /* 1 when initialized */ + u8 bigEndCksum; /* True if checksums in WAL are big-endian */ + u16 szPage; /* Database page size in bytes. 1==64K */ + u32 mxFrame; /* Index of last valid frame in the WAL */ + u32 nPage; /* Size of database in pages */ + u32 aFrameCksum[2]; /* Checksum of last frame in log */ + u32 aSalt[2]; /* Two salt values copied from WAL header */ + u32 aCksum[2]; /* Checksum over all prior fields */ +}; + +/* +** A copy of the following object occurs in the wal-index immediately +** following the second copy of the WalIndexHdr. This object stores +** information used by checkpoint. +** +** nBackfill is the number of frames in the WAL that have been written +** back into the database. (We call the act of moving content from WAL to +** database "backfilling".) The nBackfill number is never greater than +** WalIndexHdr.mxFrame. nBackfill can only be increased by threads +** holding the WAL_CKPT_LOCK lock (which includes a recovery thread). +** However, a WAL_WRITE_LOCK thread can move the value of nBackfill from +** mxFrame back to zero when the WAL is reset. +** +** nBackfillAttempted is the largest value of nBackfill that a checkpoint +** has attempted to achieve. Normally nBackfill==nBackfillAtempted, however +** the nBackfillAttempted is set before any backfilling is done and the +** nBackfill is only set after all backfilling completes. So if a checkpoint +** crashes, nBackfillAttempted might be larger than nBackfill. The +** WalIndexHdr.mxFrame must never be less than nBackfillAttempted. +** +** The aLock[] field is a set of bytes used for locking. These bytes should +** never be read or written. +** +** There is one entry in aReadMark[] for each reader lock. If a reader +** holds read-lock K, then the value in aReadMark[K] is no greater than +** the mxFrame for that reader. The value READMARK_NOT_USED (0xffffffff) +** for any aReadMark[] means that entry is unused. aReadMark[0] is +** a special case; its value is never used and it exists as a place-holder +** to avoid having to offset aReadMark[] indexs by one. Readers holding +** WAL_READ_LOCK(0) always ignore the entire WAL and read all content +** directly from the database. +** +** The value of aReadMark[K] may only be changed by a thread that +** is holding an exclusive lock on WAL_READ_LOCK(K). Thus, the value of +** aReadMark[K] cannot changed while there is a reader is using that mark +** since the reader will be holding a shared lock on WAL_READ_LOCK(K). +** +** The checkpointer may only transfer frames from WAL to database where +** the frame numbers are less than or equal to every aReadMark[] that is +** in use (that is, every aReadMark[j] for which there is a corresponding +** WAL_READ_LOCK(j)). New readers (usually) pick the aReadMark[] with the +** largest value and will increase an unused aReadMark[] to mxFrame if there +** is not already an aReadMark[] equal to mxFrame. The exception to the +** previous sentence is when nBackfill equals mxFrame (meaning that everything +** in the WAL has been backfilled into the database) then new readers +** will choose aReadMark[0] which has value 0 and hence such reader will +** get all their all content directly from the database file and ignore +** the WAL. +** +** Writers normally append new frames to the end of the WAL. However, +** if nBackfill equals mxFrame (meaning that all WAL content has been +** written back into the database) and if no readers are using the WAL +** (in other words, if there are no WAL_READ_LOCK(i) where i>0) then +** the writer will first "reset" the WAL back to the beginning and start +** writing new content beginning at frame 1. +** +** We assume that 32-bit loads are atomic and so no locks are needed in +** order to read from any aReadMark[] entries. +*/ +struct WalCkptInfo { + u32 nBackfill; /* Number of WAL frames backfilled into DB */ + u32 aReadMark[WAL_NREADER]; /* Reader marks */ + u8 aLock[SQLITE_SHM_NLOCK]; /* Reserved space for locks */ + u32 nBackfillAttempted; /* WAL frames perhaps written, or maybe not */ + u32 notUsed0; /* Available for future enhancements */ +}; +#define READMARK_NOT_USED 0xffffffff + +/* +** This is a schematic view of the complete 136-byte header of the +** wal-index file (also known as the -shm file): +** +** +-----------------------------+ +** 0: | iVersion | \ +** +-----------------------------+ | +** 4: | (unused padding) | | +** +-----------------------------+ | +** 8: | iChange | | +** +-------+-------+-------------+ | +** 12: | bInit | bBig | szPage | | +** +-------+-------+-------------+ | +** 16: | mxFrame | | First copy of the +** +-----------------------------+ | WalIndexHdr object +** 20: | nPage | | +** +-----------------------------+ | +** 24: | aFrameCksum | | +** | | | +** +-----------------------------+ | +** 32: | aSalt | | +** | | | +** +-----------------------------+ | +** 40: | aCksum | | +** | | / +** +-----------------------------+ +** 48: | iVersion | \ +** +-----------------------------+ | +** 52: | (unused padding) | | +** +-----------------------------+ | +** 56: | iChange | | +** +-------+-------+-------------+ | +** 60: | bInit | bBig | szPage | | +** +-------+-------+-------------+ | Second copy of the +** 64: | mxFrame | | WalIndexHdr +** +-----------------------------+ | +** 68: | nPage | | +** +-----------------------------+ | +** 72: | aFrameCksum | | +** | | | +** +-----------------------------+ | +** 80: | aSalt | | +** | | | +** +-----------------------------+ | +** 88: | aCksum | | +** | | / +** +-----------------------------+ +** 96: | nBackfill | +** +-----------------------------+ +** 100: | 5 read marks | +** | | +** | | +** | | +** | | +** +-------+-------+------+------+ +** 120: | Write | Ckpt | Rcvr | Rd0 | \ +** +-------+-------+------+------+ ) 8 lock bytes +** | Read1 | Read2 | Rd3 | Rd4 | / +** +-------+-------+------+------+ +** 128: | nBackfillAttempted | +** +-----------------------------+ +** 132: | (unused padding) | +** +-----------------------------+ +*/ + +/* A block of WALINDEX_LOCK_RESERVED bytes beginning at +** WALINDEX_LOCK_OFFSET is reserved for locks. Since some systems +** only support mandatory file-locks, we do not read or write data +** from the region of the file on which locks are applied. +*/ +#define WALINDEX_LOCK_OFFSET (sizeof(WalIndexHdr)*2+offsetof(WalCkptInfo,aLock)) +#define WALINDEX_HDR_SIZE (sizeof(WalIndexHdr)*2+sizeof(WalCkptInfo)) + +/* Size of header before each frame in wal */ +#define WAL_FRAME_HDRSIZE 24 + +/* Size of write ahead log header, including checksum. */ +#define WAL_HDRSIZE 32 + +/* WAL magic value. Either this value, or the same value with the least +** significant bit also set (WAL_MAGIC | 0x00000001) is stored in 32-bit +** big-endian format in the first 4 bytes of a WAL file. +** +** If the LSB is set, then the checksums for each frame within the WAL +** file are calculated by treating all data as an array of 32-bit +** big-endian words. Otherwise, they are calculated by interpreting +** all data as 32-bit little-endian words. +*/ +#define WAL_MAGIC 0x377f0682 + +/* +** Return the offset of frame iFrame in the write-ahead log file, +** assuming a database page size of szPage bytes. The offset returned +** is to the start of the write-ahead log frame-header. +*/ +#define walFrameOffset(iFrame, szPage) ( \ + WAL_HDRSIZE + ((iFrame)-1)*(i64)((szPage)+WAL_FRAME_HDRSIZE) \ +) + +/* +** An open write-ahead log file is represented by an instance of the +** following object. +*/ +struct Wal { + sqlite3_vfs *pVfs; /* The VFS used to create pDbFd */ + sqlite3_file *pDbFd; /* File handle for the database file */ + sqlite3_file *pWalFd; /* File handle for WAL file */ + u32 iCallback; /* Value to pass to log callback (or 0) */ + i64 mxWalSize; /* Truncate WAL to this size upon reset */ + int nWiData; /* Size of array apWiData */ + int szFirstBlock; /* Size of first block written to WAL file */ + volatile u32 **apWiData; /* Pointer to wal-index content in memory */ + u32 szPage; /* Database page size */ + i16 readLock; /* Which read lock is being held. -1 for none */ + u8 syncFlags; /* Flags to use to sync header writes */ + u8 exclusiveMode; /* Non-zero if connection is in exclusive mode */ + u8 writeLock; /* True if in a write transaction */ + u8 ckptLock; /* True if holding a checkpoint lock */ + u8 readOnly; /* WAL_RDWR, WAL_RDONLY, or WAL_SHM_RDONLY */ + u8 truncateOnCommit; /* True to truncate WAL file on commit */ + u8 syncHeader; /* Fsync the WAL header if true */ + u8 padToSectorBoundary; /* Pad transactions out to the next sector */ + u8 bShmUnreliable; /* SHM content is read-only and unreliable */ + WalIndexHdr hdr; /* Wal-index header for current transaction */ + u32 minFrame; /* Ignore wal frames before this one */ + u32 iReCksum; /* On commit, recalculate checksums from here */ + const char *zWalName; /* Name of WAL file */ + u32 nCkpt; /* Checkpoint sequence counter in the wal-header */ +#ifdef SQLITE_DEBUG + u8 lockError; /* True if a locking error has occurred */ +#endif +#ifdef SQLITE_ENABLE_SNAPSHOT + WalIndexHdr *pSnapshot; /* Start transaction here if not NULL */ +#endif +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + sqlite3 *db; +#endif +}; + +/* +** Candidate values for Wal.exclusiveMode. +*/ +#define WAL_NORMAL_MODE 0 +#define WAL_EXCLUSIVE_MODE 1 +#define WAL_HEAPMEMORY_MODE 2 + +/* +** Possible values for WAL.readOnly +*/ +#define WAL_RDWR 0 /* Normal read/write connection */ +#define WAL_RDONLY 1 /* The WAL file is readonly */ +#define WAL_SHM_RDONLY 2 /* The SHM file is readonly */ + +/* +** Each page of the wal-index mapping contains a hash-table made up of +** an array of HASHTABLE_NSLOT elements of the following type. +*/ +typedef u16 ht_slot; + +/* +** This structure is used to implement an iterator that loops through +** all frames in the WAL in database page order. Where two or more frames +** correspond to the same database page, the iterator visits only the +** frame most recently written to the WAL (in other words, the frame with +** the largest index). +** +** The internals of this structure are only accessed by: +** +** walIteratorInit() - Create a new iterator, +** walIteratorNext() - Step an iterator, +** walIteratorFree() - Free an iterator. +** +** This functionality is used by the checkpoint code (see walCheckpoint()). +*/ +struct WalIterator { + u32 iPrior; /* Last result returned from the iterator */ + int nSegment; /* Number of entries in aSegment[] */ + struct WalSegment { + int iNext; /* Next slot in aIndex[] not yet returned */ + ht_slot *aIndex; /* i0, i1, i2... such that aPgno[iN] ascend */ + u32 *aPgno; /* Array of page numbers. */ + int nEntry; /* Nr. of entries in aPgno[] and aIndex[] */ + int iZero; /* Frame number associated with aPgno[0] */ + } aSegment[1]; /* One for every 32KB page in the wal-index */ +}; + +/* +** Define the parameters of the hash tables in the wal-index file. There +** is a hash-table following every HASHTABLE_NPAGE page numbers in the +** wal-index. +** +** Changing any of these constants will alter the wal-index format and +** create incompatibilities. +*/ +#define HASHTABLE_NPAGE 4096 /* Must be power of 2 */ +#define HASHTABLE_HASH_1 383 /* Should be prime */ +#define HASHTABLE_NSLOT (HASHTABLE_NPAGE*2) /* Must be a power of 2 */ + +/* +** The block of page numbers associated with the first hash-table in a +** wal-index is smaller than usual. This is so that there is a complete +** hash-table on each aligned 32KB page of the wal-index. +*/ +#define HASHTABLE_NPAGE_ONE (HASHTABLE_NPAGE - (WALINDEX_HDR_SIZE/sizeof(u32))) + +/* The wal-index is divided into pages of WALINDEX_PGSZ bytes each. */ +#define WALINDEX_PGSZ ( \ + sizeof(ht_slot)*HASHTABLE_NSLOT + HASHTABLE_NPAGE*sizeof(u32) \ +) + +/* +** Obtain a pointer to the iPage'th page of the wal-index. The wal-index +** is broken into pages of WALINDEX_PGSZ bytes. Wal-index pages are +** numbered from zero. +** +** If the wal-index is currently smaller the iPage pages then the size +** of the wal-index might be increased, but only if it is safe to do +** so. It is safe to enlarge the wal-index if pWal->writeLock is true +** or pWal->exclusiveMode==WAL_HEAPMEMORY_MODE. +** +** Three possible result scenarios: +** +** (1) rc==SQLITE_OK and *ppPage==Requested-Wal-Index-Page +** (2) rc>=SQLITE_ERROR and *ppPage==NULL +** (3) rc==SQLITE_OK and *ppPage==NULL // only if iPage==0 +** +** Scenario (3) can only occur when pWal->writeLock is false and iPage==0 +*/ +static SQLITE_NOINLINE int walIndexPageRealloc( + Wal *pWal, /* The WAL context */ + int iPage, /* The page we seek */ + volatile u32 **ppPage /* Write the page pointer here */ +){ + int rc = SQLITE_OK; + + /* Enlarge the pWal->apWiData[] array if required */ + if( pWal->nWiData<=iPage ){ + sqlite3_int64 nByte = sizeof(u32*)*(iPage+1); + volatile u32 **apNew; + apNew = (volatile u32 **)sqlite3Realloc((void *)pWal->apWiData, nByte); + if( !apNew ){ + *ppPage = 0; + return SQLITE_NOMEM_BKPT; + } + memset((void*)&apNew[pWal->nWiData], 0, + sizeof(u32*)*(iPage+1-pWal->nWiData)); + pWal->apWiData = apNew; + pWal->nWiData = iPage+1; + } + + /* Request a pointer to the required page from the VFS */ + assert( pWal->apWiData[iPage]==0 ); + if( pWal->exclusiveMode==WAL_HEAPMEMORY_MODE ){ + pWal->apWiData[iPage] = (u32 volatile *)sqlite3MallocZero(WALINDEX_PGSZ); + if( !pWal->apWiData[iPage] ) rc = SQLITE_NOMEM_BKPT; + }else{ + rc = sqlite3OsShmMap(pWal->pDbFd, iPage, WALINDEX_PGSZ, + pWal->writeLock, (void volatile **)&pWal->apWiData[iPage] + ); + assert( pWal->apWiData[iPage]!=0 + || rc!=SQLITE_OK + || (pWal->writeLock==0 && iPage==0) ); + testcase( pWal->apWiData[iPage]==0 && rc==SQLITE_OK ); + if( rc==SQLITE_OK ){ + if( iPage>0 && sqlite3FaultSim(600) ) rc = SQLITE_NOMEM; + }else if( (rc&0xff)==SQLITE_READONLY ){ + pWal->readOnly |= WAL_SHM_RDONLY; + if( rc==SQLITE_READONLY ){ + rc = SQLITE_OK; + } + } + } + + *ppPage = pWal->apWiData[iPage]; + assert( iPage==0 || *ppPage || rc!=SQLITE_OK ); + return rc; +} +static int walIndexPage( + Wal *pWal, /* The WAL context */ + int iPage, /* The page we seek */ + volatile u32 **ppPage /* Write the page pointer here */ +){ + if( pWal->nWiData<=iPage || (*ppPage = pWal->apWiData[iPage])==0 ){ + return walIndexPageRealloc(pWal, iPage, ppPage); + } + return SQLITE_OK; +} + +/* +** Return a pointer to the WalCkptInfo structure in the wal-index. +*/ +static volatile WalCkptInfo *walCkptInfo(Wal *pWal){ + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + return (volatile WalCkptInfo*)&(pWal->apWiData[0][sizeof(WalIndexHdr)/2]); +} + +/* +** Return a pointer to the WalIndexHdr structure in the wal-index. +*/ +static volatile WalIndexHdr *walIndexHdr(Wal *pWal){ + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + return (volatile WalIndexHdr*)pWal->apWiData[0]; +} + +/* +** The argument to this macro must be of type u32. On a little-endian +** architecture, it returns the u32 value that results from interpreting +** the 4 bytes as a big-endian value. On a big-endian architecture, it +** returns the value that would be produced by interpreting the 4 bytes +** of the input value as a little-endian integer. +*/ +#define BYTESWAP32(x) ( \ + (((x)&0x000000FF)<<24) + (((x)&0x0000FF00)<<8) \ + + (((x)&0x00FF0000)>>8) + (((x)&0xFF000000)>>24) \ +) + +/* +** Generate or extend an 8 byte checksum based on the data in +** array aByte[] and the initial values of aIn[0] and aIn[1] (or +** initial values of 0 and 0 if aIn==NULL). +** +** The checksum is written back into aOut[] before returning. +** +** nByte must be a positive multiple of 8. +*/ +static void walChecksumBytes( + int nativeCksum, /* True for native byte-order, false for non-native */ + u8 *a, /* Content to be checksummed */ + int nByte, /* Bytes of content in a[]. Must be a multiple of 8. */ + const u32 *aIn, /* Initial checksum value input */ + u32 *aOut /* OUT: Final checksum value output */ +){ + u32 s1, s2; + u32 *aData = (u32 *)a; + u32 *aEnd = (u32 *)&a[nByte]; + + if( aIn ){ + s1 = aIn[0]; + s2 = aIn[1]; + }else{ + s1 = s2 = 0; + } + + assert( nByte>=8 ); + assert( (nByte&0x00000007)==0 ); + assert( nByte<=65536 ); + + if( nativeCksum ){ + do { + s1 += *aData++ + s2; + s2 += *aData++ + s1; + }while( aDataexclusiveMode!=WAL_HEAPMEMORY_MODE ){ + sqlite3OsShmBarrier(pWal->pDbFd); + } +} + +/* +** Add the SQLITE_NO_TSAN as part of the return-type of a function +** definition as a hint that the function contains constructs that +** might give false-positive TSAN warnings. +** +** See tag-20200519-1. +*/ +#if defined(__clang__) && !defined(SQLITE_NO_TSAN) +# define SQLITE_NO_TSAN __attribute__((no_sanitize_thread)) +#else +# define SQLITE_NO_TSAN +#endif + +/* +** Write the header information in pWal->hdr into the wal-index. +** +** The checksum on pWal->hdr is updated before it is written. +*/ +static SQLITE_NO_TSAN void walIndexWriteHdr(Wal *pWal){ + volatile WalIndexHdr *aHdr = walIndexHdr(pWal); + const int nCksum = offsetof(WalIndexHdr, aCksum); + + assert( pWal->writeLock ); + pWal->hdr.isInit = 1; + pWal->hdr.iVersion = WALINDEX_MAX_VERSION; + walChecksumBytes(1, (u8*)&pWal->hdr, nCksum, 0, pWal->hdr.aCksum); + /* Possible TSAN false-positive. See tag-20200519-1 */ + memcpy((void*)&aHdr[1], (const void*)&pWal->hdr, sizeof(WalIndexHdr)); + walShmBarrier(pWal); + memcpy((void*)&aHdr[0], (const void*)&pWal->hdr, sizeof(WalIndexHdr)); +} + +/* +** This function encodes a single frame header and writes it to a buffer +** supplied by the caller. A frame-header is made up of a series of +** 4-byte big-endian integers, as follows: +** +** 0: Page number. +** 4: For commit records, the size of the database image in pages +** after the commit. For all other records, zero. +** 8: Salt-1 (copied from the wal-header) +** 12: Salt-2 (copied from the wal-header) +** 16: Checksum-1. +** 20: Checksum-2. +*/ +static void walEncodeFrame( + Wal *pWal, /* The write-ahead log */ + u32 iPage, /* Database page number for frame */ + u32 nTruncate, /* New db size (or 0 for non-commit frames) */ + u8 *aData, /* Pointer to page data */ + u8 *aFrame /* OUT: Write encoded frame here */ +){ + int nativeCksum; /* True for native byte-order checksums */ + u32 *aCksum = pWal->hdr.aFrameCksum; + assert( WAL_FRAME_HDRSIZE==24 ); + sqlite3Put4byte(&aFrame[0], iPage); + sqlite3Put4byte(&aFrame[4], nTruncate); + if( pWal->iReCksum==0 ){ + memcpy(&aFrame[8], pWal->hdr.aSalt, 8); + + nativeCksum = (pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN); + walChecksumBytes(nativeCksum, aFrame, 8, aCksum, aCksum); + walChecksumBytes(nativeCksum, aData, pWal->szPage, aCksum, aCksum); + + sqlite3Put4byte(&aFrame[16], aCksum[0]); + sqlite3Put4byte(&aFrame[20], aCksum[1]); + }else{ + memset(&aFrame[8], 0, 16); + } +} + +/* +** Check to see if the frame with header in aFrame[] and content +** in aData[] is valid. If it is a valid frame, fill *piPage and +** *pnTruncate and return true. Return if the frame is not valid. +*/ +static int walDecodeFrame( + Wal *pWal, /* The write-ahead log */ + u32 *piPage, /* OUT: Database page number for frame */ + u32 *pnTruncate, /* OUT: New db size (or 0 if not commit) */ + u8 *aData, /* Pointer to page data (for checksum) */ + u8 *aFrame /* Frame data */ +){ + int nativeCksum; /* True for native byte-order checksums */ + u32 *aCksum = pWal->hdr.aFrameCksum; + u32 pgno; /* Page number of the frame */ + assert( WAL_FRAME_HDRSIZE==24 ); + + /* A frame is only valid if the salt values in the frame-header + ** match the salt values in the wal-header. + */ + if( memcmp(&pWal->hdr.aSalt, &aFrame[8], 8)!=0 ){ + return 0; + } + + /* A frame is only valid if the page number is creater than zero. + */ + pgno = sqlite3Get4byte(&aFrame[0]); + if( pgno==0 ){ + return 0; + } + + /* A frame is only valid if a checksum of the WAL header, + ** all prior frams, the first 16 bytes of this frame-header, + ** and the frame-data matches the checksum in the last 8 + ** bytes of this frame-header. + */ + nativeCksum = (pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN); + walChecksumBytes(nativeCksum, aFrame, 8, aCksum, aCksum); + walChecksumBytes(nativeCksum, aData, pWal->szPage, aCksum, aCksum); + if( aCksum[0]!=sqlite3Get4byte(&aFrame[16]) + || aCksum[1]!=sqlite3Get4byte(&aFrame[20]) + ){ + /* Checksum failed. */ + return 0; + } + + /* If we reach this point, the frame is valid. Return the page number + ** and the new database size. + */ + *piPage = pgno; + *pnTruncate = sqlite3Get4byte(&aFrame[4]); + return 1; +} + + +#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG) +/* +** Names of locks. This routine is used to provide debugging output and is not +** a part of an ordinary build. +*/ +static const char *walLockName(int lockIdx){ + if( lockIdx==WAL_WRITE_LOCK ){ + return "WRITE-LOCK"; + }else if( lockIdx==WAL_CKPT_LOCK ){ + return "CKPT-LOCK"; + }else if( lockIdx==WAL_RECOVER_LOCK ){ + return "RECOVER-LOCK"; + }else{ + static char zName[15]; + sqlite3_snprintf(sizeof(zName), zName, "READ-LOCK[%d]", + lockIdx-WAL_READ_LOCK(0)); + return zName; + } +} +#endif /*defined(SQLITE_TEST) || defined(SQLITE_DEBUG) */ + + +/* +** Set or release locks on the WAL. Locks are either shared or exclusive. +** A lock cannot be moved directly between shared and exclusive - it must go +** through the unlocked state first. +** +** In locking_mode=EXCLUSIVE, all of these routines become no-ops. +*/ +static int walLockShared(Wal *pWal, int lockIdx){ + int rc; + if( pWal->exclusiveMode ) return SQLITE_OK; + rc = sqlite3OsShmLock(pWal->pDbFd, lockIdx, 1, + SQLITE_SHM_LOCK | SQLITE_SHM_SHARED); + WALTRACE(("WAL%p: acquire SHARED-%s %s\n", pWal, + walLockName(lockIdx), rc ? "failed" : "ok")); + VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && (rc&0xFF)!=SQLITE_BUSY); ) + return rc; +} +static void walUnlockShared(Wal *pWal, int lockIdx){ + if( pWal->exclusiveMode ) return; + (void)sqlite3OsShmLock(pWal->pDbFd, lockIdx, 1, + SQLITE_SHM_UNLOCK | SQLITE_SHM_SHARED); + WALTRACE(("WAL%p: release SHARED-%s\n", pWal, walLockName(lockIdx))); +} +static int walLockExclusive(Wal *pWal, int lockIdx, int n){ + int rc; + if( pWal->exclusiveMode ) return SQLITE_OK; + rc = sqlite3OsShmLock(pWal->pDbFd, lockIdx, n, + SQLITE_SHM_LOCK | SQLITE_SHM_EXCLUSIVE); + WALTRACE(("WAL%p: acquire EXCLUSIVE-%s cnt=%d %s\n", pWal, + walLockName(lockIdx), n, rc ? "failed" : "ok")); + VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && (rc&0xFF)!=SQLITE_BUSY); ) + return rc; +} +static void walUnlockExclusive(Wal *pWal, int lockIdx, int n){ + if( pWal->exclusiveMode ) return; + (void)sqlite3OsShmLock(pWal->pDbFd, lockIdx, n, + SQLITE_SHM_UNLOCK | SQLITE_SHM_EXCLUSIVE); + WALTRACE(("WAL%p: release EXCLUSIVE-%s cnt=%d\n", pWal, + walLockName(lockIdx), n)); +} + +/* +** Compute a hash on a page number. The resulting hash value must land +** between 0 and (HASHTABLE_NSLOT-1). The walHashNext() function advances +** the hash to the next value in the event of a collision. +*/ +static int walHash(u32 iPage){ + assert( iPage>0 ); + assert( (HASHTABLE_NSLOT & (HASHTABLE_NSLOT-1))==0 ); + return (iPage*HASHTABLE_HASH_1) & (HASHTABLE_NSLOT-1); +} +static int walNextHash(int iPriorHash){ + return (iPriorHash+1)&(HASHTABLE_NSLOT-1); +} + +/* +** An instance of the WalHashLoc object is used to describe the location +** of a page hash table in the wal-index. This becomes the return value +** from walHashGet(). +*/ +typedef struct WalHashLoc WalHashLoc; +struct WalHashLoc { + volatile ht_slot *aHash; /* Start of the wal-index hash table */ + volatile u32 *aPgno; /* aPgno[1] is the page of first frame indexed */ + u32 iZero; /* One less than the frame number of first indexed*/ +}; + +/* +** Return pointers to the hash table and page number array stored on +** page iHash of the wal-index. The wal-index is broken into 32KB pages +** numbered starting from 0. +** +** Set output variable pLoc->aHash to point to the start of the hash table +** in the wal-index file. Set pLoc->iZero to one less than the frame +** number of the first frame indexed by this hash table. If a +** slot in the hash table is set to N, it refers to frame number +** (pLoc->iZero+N) in the log. +** +** Finally, set pLoc->aPgno so that pLoc->aPgno[0] is the page number of the +** first frame indexed by the hash table, frame (pLoc->iZero). +*/ +static int walHashGet( + Wal *pWal, /* WAL handle */ + int iHash, /* Find the iHash'th table */ + WalHashLoc *pLoc /* OUT: Hash table location */ +){ + int rc; /* Return code */ + + rc = walIndexPage(pWal, iHash, &pLoc->aPgno); + assert( rc==SQLITE_OK || iHash>0 ); + + if( pLoc->aPgno ){ + pLoc->aHash = (volatile ht_slot *)&pLoc->aPgno[HASHTABLE_NPAGE]; + if( iHash==0 ){ + pLoc->aPgno = &pLoc->aPgno[WALINDEX_HDR_SIZE/sizeof(u32)]; + pLoc->iZero = 0; + }else{ + pLoc->iZero = HASHTABLE_NPAGE_ONE + (iHash-1)*HASHTABLE_NPAGE; + } + }else if( NEVER(rc==SQLITE_OK) ){ + rc = SQLITE_ERROR; + } + return rc; +} + +/* +** Return the number of the wal-index page that contains the hash-table +** and page-number array that contain entries corresponding to WAL frame +** iFrame. The wal-index is broken up into 32KB pages. Wal-index pages +** are numbered starting from 0. +*/ +static int walFramePage(u32 iFrame){ + int iHash = (iFrame+HASHTABLE_NPAGE-HASHTABLE_NPAGE_ONE-1) / HASHTABLE_NPAGE; + assert( (iHash==0 || iFrame>HASHTABLE_NPAGE_ONE) + && (iHash>=1 || iFrame<=HASHTABLE_NPAGE_ONE) + && (iHash<=1 || iFrame>(HASHTABLE_NPAGE_ONE+HASHTABLE_NPAGE)) + && (iHash>=2 || iFrame<=HASHTABLE_NPAGE_ONE+HASHTABLE_NPAGE) + && (iHash<=2 || iFrame>(HASHTABLE_NPAGE_ONE+2*HASHTABLE_NPAGE)) + ); + assert( iHash>=0 ); + return iHash; +} + +/* +** Return the page number associated with frame iFrame in this WAL. +*/ +static u32 walFramePgno(Wal *pWal, u32 iFrame){ + int iHash = walFramePage(iFrame); + if( iHash==0 ){ + return pWal->apWiData[0][WALINDEX_HDR_SIZE/sizeof(u32) + iFrame - 1]; + } + return pWal->apWiData[iHash][(iFrame-1-HASHTABLE_NPAGE_ONE)%HASHTABLE_NPAGE]; +} + +/* +** Remove entries from the hash table that point to WAL slots greater +** than pWal->hdr.mxFrame. +** +** This function is called whenever pWal->hdr.mxFrame is decreased due +** to a rollback or savepoint. +** +** At most only the hash table containing pWal->hdr.mxFrame needs to be +** updated. Any later hash tables will be automatically cleared when +** pWal->hdr.mxFrame advances to the point where those hash tables are +** actually needed. +*/ +static void walCleanupHash(Wal *pWal){ + WalHashLoc sLoc; /* Hash table location */ + int iLimit = 0; /* Zero values greater than this */ + int nByte; /* Number of bytes to zero in aPgno[] */ + int i; /* Used to iterate through aHash[] */ + + assert( pWal->writeLock ); + testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE-1 ); + testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE ); + testcase( pWal->hdr.mxFrame==HASHTABLE_NPAGE_ONE+1 ); + + if( pWal->hdr.mxFrame==0 ) return; + + /* Obtain pointers to the hash-table and page-number array containing + ** the entry that corresponds to frame pWal->hdr.mxFrame. It is guaranteed + ** that the page said hash-table and array reside on is already mapped.(1) + */ + assert( pWal->nWiData>walFramePage(pWal->hdr.mxFrame) ); + assert( pWal->apWiData[walFramePage(pWal->hdr.mxFrame)] ); + i = walHashGet(pWal, walFramePage(pWal->hdr.mxFrame), &sLoc); + if( NEVER(i) ) return; /* Defense-in-depth, in case (1) above is wrong */ + + /* Zero all hash-table entries that correspond to frame numbers greater + ** than pWal->hdr.mxFrame. + */ + iLimit = pWal->hdr.mxFrame - sLoc.iZero; + assert( iLimit>0 ); + for(i=0; iiLimit ){ + sLoc.aHash[i] = 0; + } + } + + /* Zero the entries in the aPgno array that correspond to frames with + ** frame numbers greater than pWal->hdr.mxFrame. + */ + nByte = (int)((char *)sLoc.aHash - (char *)&sLoc.aPgno[iLimit]); + assert( nByte>=0 ); + memset((void *)&sLoc.aPgno[iLimit], 0, nByte); + +#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT + /* Verify that the every entry in the mapping region is still reachable + ** via the hash table even after the cleanup. + */ + if( iLimit ){ + int j; /* Loop counter */ + int iKey; /* Hash key */ + for(j=0; j=0 ); + memset((void*)sLoc.aPgno, 0, nByte); + } + + /* If the entry in aPgno[] is already set, then the previous writer + ** must have exited unexpectedly in the middle of a transaction (after + ** writing one or more dirty pages to the WAL to free up memory). + ** Remove the remnants of that writers uncommitted transaction from + ** the hash-table before writing any new entries. + */ + if( sLoc.aPgno[idx-1] ){ + walCleanupHash(pWal); + assert( !sLoc.aPgno[idx-1] ); + } + + /* Write the aPgno[] array entry and the hash-table slot. */ + nCollide = idx; + for(iKey=walHash(iPage); sLoc.aHash[iKey]; iKey=walNextHash(iKey)){ + if( (nCollide--)==0 ) return SQLITE_CORRUPT_BKPT; + } + sLoc.aPgno[idx-1] = iPage; + AtomicStore(&sLoc.aHash[iKey], (ht_slot)idx); + +#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT + /* Verify that the number of entries in the hash table exactly equals + ** the number of entries in the mapping region. + */ + { + int i; /* Loop counter */ + int nEntry = 0; /* Number of entries in the hash table */ + for(i=0; ickptLock==1 || pWal->ckptLock==0 ); + assert( WAL_ALL_BUT_WRITE==WAL_WRITE_LOCK+1 ); + assert( WAL_CKPT_LOCK==WAL_ALL_BUT_WRITE ); + assert( pWal->writeLock ); + iLock = WAL_ALL_BUT_WRITE + pWal->ckptLock; + rc = walLockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock); + if( rc ){ + return rc; + } + + WALTRACE(("WAL%p: recovery begin...\n", pWal)); + + memset(&pWal->hdr, 0, sizeof(WalIndexHdr)); + + rc = sqlite3OsFileSize(pWal->pWalFd, &nSize); + if( rc!=SQLITE_OK ){ + goto recovery_error; + } + + if( nSize>WAL_HDRSIZE ){ + u8 aBuf[WAL_HDRSIZE]; /* Buffer to load WAL header into */ + u32 *aPrivate = 0; /* Heap copy of *-shm hash being populated */ + u8 *aFrame = 0; /* Malloc'd buffer to load entire frame */ + int szFrame; /* Number of bytes in buffer aFrame[] */ + u8 *aData; /* Pointer to data part of aFrame buffer */ + int szPage; /* Page size according to the log */ + u32 magic; /* Magic value read from WAL header */ + u32 version; /* Magic value read from WAL header */ + int isValid; /* True if this frame is valid */ + u32 iPg; /* Current 32KB wal-index page */ + u32 iLastFrame; /* Last frame in wal, based on nSize alone */ + + /* Read in the WAL header. */ + rc = sqlite3OsRead(pWal->pWalFd, aBuf, WAL_HDRSIZE, 0); + if( rc!=SQLITE_OK ){ + goto recovery_error; + } + + /* If the database page size is not a power of two, or is greater than + ** SQLITE_MAX_PAGE_SIZE, conclude that the WAL file contains no valid + ** data. Similarly, if the 'magic' value is invalid, ignore the whole + ** WAL file. + */ + magic = sqlite3Get4byte(&aBuf[0]); + szPage = sqlite3Get4byte(&aBuf[8]); + if( (magic&0xFFFFFFFE)!=WAL_MAGIC + || szPage&(szPage-1) + || szPage>SQLITE_MAX_PAGE_SIZE + || szPage<512 + ){ + goto finished; + } + pWal->hdr.bigEndCksum = (u8)(magic&0x00000001); + pWal->szPage = szPage; + pWal->nCkpt = sqlite3Get4byte(&aBuf[12]); + memcpy(&pWal->hdr.aSalt, &aBuf[16], 8); + + /* Verify that the WAL header checksum is correct */ + walChecksumBytes(pWal->hdr.bigEndCksum==SQLITE_BIGENDIAN, + aBuf, WAL_HDRSIZE-2*4, 0, pWal->hdr.aFrameCksum + ); + if( pWal->hdr.aFrameCksum[0]!=sqlite3Get4byte(&aBuf[24]) + || pWal->hdr.aFrameCksum[1]!=sqlite3Get4byte(&aBuf[28]) + ){ + goto finished; + } + + /* Verify that the version number on the WAL format is one that + ** are able to understand */ + version = sqlite3Get4byte(&aBuf[4]); + if( version!=WAL_MAX_VERSION ){ + rc = SQLITE_CANTOPEN_BKPT; + goto finished; + } + + /* Malloc a buffer to read frames into. */ + szFrame = szPage + WAL_FRAME_HDRSIZE; + aFrame = (u8 *)sqlite3_malloc64(szFrame + WALINDEX_PGSZ); + if( !aFrame ){ + rc = SQLITE_NOMEM_BKPT; + goto recovery_error; + } + aData = &aFrame[WAL_FRAME_HDRSIZE]; + aPrivate = (u32*)&aData[szPage]; + + /* Read all frames from the log file. */ + iLastFrame = (nSize - WAL_HDRSIZE) / szFrame; + for(iPg=0; iPg<=(u32)walFramePage(iLastFrame); iPg++){ + u32 *aShare; + u32 iFrame; /* Index of last frame read */ + u32 iLast = MIN(iLastFrame, HASHTABLE_NPAGE_ONE+iPg*HASHTABLE_NPAGE); + u32 iFirst = 1 + (iPg==0?0:HASHTABLE_NPAGE_ONE+(iPg-1)*HASHTABLE_NPAGE); + u32 nHdr, nHdr32; + rc = walIndexPage(pWal, iPg, (volatile u32**)&aShare); + assert( aShare!=0 || rc!=SQLITE_OK ); + if( aShare==0 ) break; + pWal->apWiData[iPg] = aPrivate; + + for(iFrame=iFirst; iFrame<=iLast; iFrame++){ + i64 iOffset = walFrameOffset(iFrame, szPage); + u32 pgno; /* Database page number for frame */ + u32 nTruncate; /* dbsize field from frame header */ + + /* Read and decode the next log frame. */ + rc = sqlite3OsRead(pWal->pWalFd, aFrame, szFrame, iOffset); + if( rc!=SQLITE_OK ) break; + isValid = walDecodeFrame(pWal, &pgno, &nTruncate, aData, aFrame); + if( !isValid ) break; + rc = walIndexAppend(pWal, iFrame, pgno); + if( NEVER(rc!=SQLITE_OK) ) break; + + /* If nTruncate is non-zero, this is a commit record. */ + if( nTruncate ){ + pWal->hdr.mxFrame = iFrame; + pWal->hdr.nPage = nTruncate; + pWal->hdr.szPage = (u16)((szPage&0xff00) | (szPage>>16)); + testcase( szPage<=32768 ); + testcase( szPage>=65536 ); + aFrameCksum[0] = pWal->hdr.aFrameCksum[0]; + aFrameCksum[1] = pWal->hdr.aFrameCksum[1]; + } + } + pWal->apWiData[iPg] = aShare; + nHdr = (iPg==0 ? WALINDEX_HDR_SIZE : 0); + nHdr32 = nHdr / sizeof(u32); +#ifndef SQLITE_SAFER_WALINDEX_RECOVERY + /* Memcpy() should work fine here, on all reasonable implementations. + ** Technically, memcpy() might change the destination to some + ** intermediate value before setting to the final value, and that might + ** cause a concurrent reader to malfunction. Memcpy() is allowed to + ** do that, according to the spec, but no memcpy() implementation that + ** we know of actually does that, which is why we say that memcpy() + ** is safe for this. Memcpy() is certainly a lot faster. + */ + memcpy(&aShare[nHdr32], &aPrivate[nHdr32], WALINDEX_PGSZ-nHdr); +#else + /* In the event that some platform is found for which memcpy() + ** changes the destination to some intermediate value before + ** setting the final value, this alternative copy routine is + ** provided. + */ + { + int i; + for(i=nHdr32; ihdr.aFrameCksum[0] = aFrameCksum[0]; + pWal->hdr.aFrameCksum[1] = aFrameCksum[1]; + walIndexWriteHdr(pWal); + + /* Reset the checkpoint-header. This is safe because this thread is + ** currently holding locks that exclude all other writers and + ** checkpointers. Then set the values of read-mark slots 1 through N. + */ + pInfo = walCkptInfo(pWal); + pInfo->nBackfill = 0; + pInfo->nBackfillAttempted = pWal->hdr.mxFrame; + pInfo->aReadMark[0] = 0; + for(i=1; ihdr.mxFrame ){ + pInfo->aReadMark[i] = pWal->hdr.mxFrame; + }else{ + pInfo->aReadMark[i] = READMARK_NOT_USED; + } + walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1); + }else if( rc!=SQLITE_BUSY ){ + goto recovery_error; + } + } + + /* If more than one frame was recovered from the log file, report an + ** event via sqlite3_log(). This is to help with identifying performance + ** problems caused by applications routinely shutting down without + ** checkpointing the log file. + */ + if( pWal->hdr.nPage ){ + sqlite3_log(SQLITE_NOTICE_RECOVER_WAL, + "recovered %d frames from WAL file %s", + pWal->hdr.mxFrame, pWal->zWalName + ); + } + } + +recovery_error: + WALTRACE(("WAL%p: recovery %s\n", pWal, rc ? "failed" : "ok")); + walUnlockExclusive(pWal, iLock, WAL_READ_LOCK(0)-iLock); + return rc; +} + +/* +** Close an open wal-index. +*/ +static void walIndexClose(Wal *pWal, int isDelete){ + if( pWal->exclusiveMode==WAL_HEAPMEMORY_MODE || pWal->bShmUnreliable ){ + int i; + for(i=0; inWiData; i++){ + sqlite3_free((void *)pWal->apWiData[i]); + pWal->apWiData[i] = 0; + } + } + if( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE ){ + sqlite3OsShmUnmap(pWal->pDbFd, isDelete); + } +} + +/* +** Open a connection to the WAL file zWalName. The database file must +** already be opened on connection pDbFd. The buffer that zWalName points +** to must remain valid for the lifetime of the returned Wal* handle. +** +** A SHARED lock should be held on the database file when this function +** is called. The purpose of this SHARED lock is to prevent any other +** client from unlinking the WAL or wal-index file. If another process +** were to do this just after this client opened one of these files, the +** system would be badly broken. +** +** If the log file is successfully opened, SQLITE_OK is returned and +** *ppWal is set to point to a new WAL handle. If an error occurs, +** an SQLite error code is returned and *ppWal is left unmodified. +*/ +int sqlite3WalOpen( + sqlite3_vfs *pVfs, /* vfs module to open wal and wal-index */ + sqlite3_file *pDbFd, /* The open database file */ + const char *zWalName, /* Name of the WAL file */ + int bNoShm, /* True to run in heap-memory mode */ + i64 mxWalSize, /* Truncate WAL to this size on reset */ + Wal **ppWal /* OUT: Allocated Wal handle */ +){ + int rc; /* Return Code */ + Wal *pRet; /* Object to allocate and return */ + int flags; /* Flags passed to OsOpen() */ + + assert( zWalName && zWalName[0] ); + assert( pDbFd ); + + /* Verify the values of various constants. Any changes to the values + ** of these constants would result in an incompatible on-disk format + ** for the -shm file. Any change that causes one of these asserts to + ** fail is a backward compatibility problem, even if the change otherwise + ** works. + ** + ** This table also serves as a helpful cross-reference when trying to + ** interpret hex dumps of the -shm file. + */ + assert( 48 == sizeof(WalIndexHdr) ); + assert( 40 == sizeof(WalCkptInfo) ); + assert( 120 == WALINDEX_LOCK_OFFSET ); + assert( 136 == WALINDEX_HDR_SIZE ); + assert( 4096 == HASHTABLE_NPAGE ); + assert( 4062 == HASHTABLE_NPAGE_ONE ); + assert( 8192 == HASHTABLE_NSLOT ); + assert( 383 == HASHTABLE_HASH_1 ); + assert( 32768 == WALINDEX_PGSZ ); + assert( 8 == SQLITE_SHM_NLOCK ); + assert( 5 == WAL_NREADER ); + assert( 24 == WAL_FRAME_HDRSIZE ); + assert( 32 == WAL_HDRSIZE ); + assert( 120 == WALINDEX_LOCK_OFFSET + WAL_WRITE_LOCK ); + assert( 121 == WALINDEX_LOCK_OFFSET + WAL_CKPT_LOCK ); + assert( 122 == WALINDEX_LOCK_OFFSET + WAL_RECOVER_LOCK ); + assert( 123 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(0) ); + assert( 124 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(1) ); + assert( 125 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(2) ); + assert( 126 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(3) ); + assert( 127 == WALINDEX_LOCK_OFFSET + WAL_READ_LOCK(4) ); + + /* In the amalgamation, the os_unix.c and os_win.c source files come before + ** this source file. Verify that the #defines of the locking byte offsets + ** in os_unix.c and os_win.c agree with the WALINDEX_LOCK_OFFSET value. + ** For that matter, if the lock offset ever changes from its initial design + ** value of 120, we need to know that so there is an assert() to check it. + */ +#ifdef WIN_SHM_BASE + assert( WIN_SHM_BASE==WALINDEX_LOCK_OFFSET ); +#endif +#ifdef UNIX_SHM_BASE + assert( UNIX_SHM_BASE==WALINDEX_LOCK_OFFSET ); +#endif + + + /* Allocate an instance of struct Wal to return. */ + *ppWal = 0; + pRet = (Wal*)sqlite3MallocZero(sizeof(Wal) + pVfs->szOsFile); + if( !pRet ){ + return SQLITE_NOMEM_BKPT; + } + + pRet->pVfs = pVfs; + pRet->pWalFd = (sqlite3_file *)&pRet[1]; + pRet->pDbFd = pDbFd; + pRet->readLock = -1; + pRet->mxWalSize = mxWalSize; + pRet->zWalName = zWalName; + pRet->syncHeader = 1; + pRet->padToSectorBoundary = 1; + pRet->exclusiveMode = (bNoShm ? WAL_HEAPMEMORY_MODE: WAL_NORMAL_MODE); + + /* Open file handle on the write-ahead log file. */ + flags = (SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE|SQLITE_OPEN_WAL); + rc = sqlite3OsOpen(pVfs, zWalName, pRet->pWalFd, flags, &flags); + if( rc==SQLITE_OK && flags&SQLITE_OPEN_READONLY ){ + pRet->readOnly = WAL_RDONLY; + } + + if( rc!=SQLITE_OK ){ + walIndexClose(pRet, 0); + sqlite3OsClose(pRet->pWalFd); + sqlite3_free(pRet); + }else{ + int iDC = sqlite3OsDeviceCharacteristics(pDbFd); + if( iDC & SQLITE_IOCAP_SEQUENTIAL ){ pRet->syncHeader = 0; } + if( iDC & SQLITE_IOCAP_POWERSAFE_OVERWRITE ){ + pRet->padToSectorBoundary = 0; + } + *ppWal = pRet; + WALTRACE(("WAL%d: opened\n", pRet)); + } + return rc; +} + +/* +** Change the size to which the WAL file is trucated on each reset. +*/ +void sqlite3WalLimit(Wal *pWal, i64 iLimit){ + if( pWal ) pWal->mxWalSize = iLimit; +} + +/* +** Find the smallest page number out of all pages held in the WAL that +** has not been returned by any prior invocation of this method on the +** same WalIterator object. Write into *piFrame the frame index where +** that page was last written into the WAL. Write into *piPage the page +** number. +** +** Return 0 on success. If there are no pages in the WAL with a page +** number larger than *piPage, then return 1. +*/ +static int walIteratorNext( + WalIterator *p, /* Iterator */ + u32 *piPage, /* OUT: The page number of the next page */ + u32 *piFrame /* OUT: Wal frame index of next page */ +){ + u32 iMin; /* Result pgno must be greater than iMin */ + u32 iRet = 0xFFFFFFFF; /* 0xffffffff is never a valid page number */ + int i; /* For looping through segments */ + + iMin = p->iPrior; + assert( iMin<0xffffffff ); + for(i=p->nSegment-1; i>=0; i--){ + struct WalSegment *pSegment = &p->aSegment[i]; + while( pSegment->iNextnEntry ){ + u32 iPg = pSegment->aPgno[pSegment->aIndex[pSegment->iNext]]; + if( iPg>iMin ){ + if( iPgiZero + pSegment->aIndex[pSegment->iNext]; + } + break; + } + pSegment->iNext++; + } + } + + *piPage = p->iPrior = iRet; + return (iRet==0xFFFFFFFF); +} + +/* +** This function merges two sorted lists into a single sorted list. +** +** aLeft[] and aRight[] are arrays of indices. The sort key is +** aContent[aLeft[]] and aContent[aRight[]]. Upon entry, the following +** is guaranteed for all J0 && nRight>0 ); + while( iRight=nRight || aContent[aLeft[iLeft]]=nLeft || aContent[aLeft[iLeft]]>dbpage ); + assert( iRight>=nRight || aContent[aRight[iRight]]>dbpage ); + } + + *paRight = aLeft; + *pnRight = iOut; + memcpy(aLeft, aTmp, sizeof(aTmp[0])*iOut); +} + +/* +** Sort the elements in list aList using aContent[] as the sort key. +** Remove elements with duplicate keys, preferring to keep the +** larger aList[] values. +** +** The aList[] entries are indices into aContent[]. The values in +** aList[] are to be sorted so that for all J0 ); + assert( HASHTABLE_NPAGE==(1<<(ArraySize(aSub)-1)) ); + + for(iList=0; iListaList && p->nList<=(1<aList==&aList[iList&~((2<aList, p->nList, &aMerge, &nMerge, aBuffer); + } + aSub[iSub].aList = aMerge; + aSub[iSub].nList = nMerge; + } + + for(iSub++; iSubnList<=(1<aList==&aList[nList&~((2<aList, p->nList, &aMerge, &nMerge, aBuffer); + } + } + assert( aMerge==aList ); + *pnList = nMerge; + +#ifdef SQLITE_DEBUG + { + int i; + for(i=1; i<*pnList; i++){ + assert( aContent[aList[i]] > aContent[aList[i-1]] ); + } + } +#endif +} + +/* +** Free an iterator allocated by walIteratorInit(). +*/ +static void walIteratorFree(WalIterator *p){ + sqlite3_free(p); +} + +/* +** Construct a WalInterator object that can be used to loop over all +** pages in the WAL following frame nBackfill in ascending order. Frames +** nBackfill or earlier may be included - excluding them is an optimization +** only. The caller must hold the checkpoint lock. +** +** On success, make *pp point to the newly allocated WalInterator object +** return SQLITE_OK. Otherwise, return an error code. If this routine +** returns an error, the value of *pp is undefined. +** +** The calling routine should invoke walIteratorFree() to destroy the +** WalIterator object when it has finished with it. +*/ +static int walIteratorInit(Wal *pWal, u32 nBackfill, WalIterator **pp){ + WalIterator *p; /* Return value */ + int nSegment; /* Number of segments to merge */ + u32 iLast; /* Last frame in log */ + sqlite3_int64 nByte; /* Number of bytes to allocate */ + int i; /* Iterator variable */ + ht_slot *aTmp; /* Temp space used by merge-sort */ + int rc = SQLITE_OK; /* Return Code */ + + /* This routine only runs while holding the checkpoint lock. And + ** it only runs if there is actually content in the log (mxFrame>0). + */ + assert( pWal->ckptLock && pWal->hdr.mxFrame>0 ); + iLast = pWal->hdr.mxFrame; + + /* Allocate space for the WalIterator object. */ + nSegment = walFramePage(iLast) + 1; + nByte = sizeof(WalIterator) + + (nSegment-1)*sizeof(struct WalSegment) + + iLast*sizeof(ht_slot); + p = (WalIterator *)sqlite3_malloc64(nByte); + if( !p ){ + return SQLITE_NOMEM_BKPT; + } + memset(p, 0, nByte); + p->nSegment = nSegment; + + /* Allocate temporary space used by the merge-sort routine. This block + ** of memory will be freed before this function returns. + */ + aTmp = (ht_slot *)sqlite3_malloc64( + sizeof(ht_slot) * (iLast>HASHTABLE_NPAGE?HASHTABLE_NPAGE:iLast) + ); + if( !aTmp ){ + rc = SQLITE_NOMEM_BKPT; + } + + for(i=walFramePage(nBackfill+1); rc==SQLITE_OK && iaSegment[p->nSegment])[sLoc.iZero]; + sLoc.iZero++; + + for(j=0; jaSegment[i].iZero = sLoc.iZero; + p->aSegment[i].nEntry = nEntry; + p->aSegment[i].aIndex = aIndex; + p->aSegment[i].aPgno = (u32 *)sLoc.aPgno; + } + } + sqlite3_free(aTmp); + + if( rc!=SQLITE_OK ){ + walIteratorFree(p); + p = 0; + } + *pp = p; + return rc; +} + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT +/* +** Attempt to enable blocking locks. Blocking locks are enabled only if (a) +** they are supported by the VFS, and (b) the database handle is configured +** with a busy-timeout. Return 1 if blocking locks are successfully enabled, +** or 0 otherwise. +*/ +static int walEnableBlocking(Wal *pWal){ + int res = 0; + if( pWal->db ){ + int tmout = pWal->db->busyTimeout; + if( tmout ){ + int rc; + rc = sqlite3OsFileControl( + pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout + ); + res = (rc==SQLITE_OK); + } + } + return res; +} + +/* +** Disable blocking locks. +*/ +static void walDisableBlocking(Wal *pWal){ + int tmout = 0; + sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout); +} + +/* +** If parameter bLock is true, attempt to enable blocking locks, take +** the WRITER lock, and then disable blocking locks. If blocking locks +** cannot be enabled, no attempt to obtain the WRITER lock is made. Return +** an SQLite error code if an error occurs, or SQLITE_OK otherwise. It is not +** an error if blocking locks can not be enabled. +** +** If the bLock parameter is false and the WRITER lock is held, release it. +*/ +int sqlite3WalWriteLock(Wal *pWal, int bLock){ + int rc = SQLITE_OK; + assert( pWal->readLock<0 || bLock==0 ); + if( bLock ){ + assert( pWal->db ); + if( walEnableBlocking(pWal) ){ + rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1); + if( rc==SQLITE_OK ){ + pWal->writeLock = 1; + } + walDisableBlocking(pWal); + } + }else if( pWal->writeLock ){ + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + pWal->writeLock = 0; + } + return rc; +} + +/* +** Set the database handle used to determine if blocking locks are required. +*/ +void sqlite3WalDb(Wal *pWal, sqlite3 *db){ + pWal->db = db; +} + +/* +** Take an exclusive WRITE lock. Blocking if so configured. +*/ +static int walLockWriter(Wal *pWal){ + int rc; + walEnableBlocking(pWal); + rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1); + walDisableBlocking(pWal); + return rc; +} +#else +# define walEnableBlocking(x) 0 +# define walDisableBlocking(x) +# define walLockWriter(pWal) walLockExclusive((pWal), WAL_WRITE_LOCK, 1) +# define sqlite3WalDb(pWal, db) +#endif /* ifdef SQLITE_ENABLE_SETLK_TIMEOUT */ + + +/* +** Attempt to obtain the exclusive WAL lock defined by parameters lockIdx and +** n. If the attempt fails and parameter xBusy is not NULL, then it is a +** busy-handler function. Invoke it and retry the lock until either the +** lock is successfully obtained or the busy-handler returns 0. +*/ +static int walBusyLock( + Wal *pWal, /* WAL connection */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int lockIdx, /* Offset of first byte to lock */ + int n /* Number of bytes to lock */ +){ + int rc; + do { + rc = walLockExclusive(pWal, lockIdx, n); + }while( xBusy && rc==SQLITE_BUSY && xBusy(pBusyArg) ); +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + if( rc==SQLITE_BUSY_TIMEOUT ){ + walDisableBlocking(pWal); + rc = SQLITE_BUSY; + } +#endif + return rc; +} + +/* +** The cache of the wal-index header must be valid to call this function. +** Return the page-size in bytes used by the database. +*/ +static int walPagesize(Wal *pWal){ + return (pWal->hdr.szPage&0xfe00) + ((pWal->hdr.szPage&0x0001)<<16); +} + +/* +** The following is guaranteed when this function is called: +** +** a) the WRITER lock is held, +** b) the entire log file has been checkpointed, and +** c) any existing readers are reading exclusively from the database +** file - there are no readers that may attempt to read a frame from +** the log file. +** +** This function updates the shared-memory structures so that the next +** client to write to the database (which may be this one) does so by +** writing frames into the start of the log file. +** +** The value of parameter salt1 is used as the aSalt[1] value in the +** new wal-index header. It should be passed a pseudo-random value (i.e. +** one obtained from sqlite3_randomness()). +*/ +static void walRestartHdr(Wal *pWal, u32 salt1){ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + int i; /* Loop counter */ + u32 *aSalt = pWal->hdr.aSalt; /* Big-endian salt values */ + pWal->nCkpt++; + pWal->hdr.mxFrame = 0; + sqlite3Put4byte((u8*)&aSalt[0], 1 + sqlite3Get4byte((u8*)&aSalt[0])); + memcpy(&pWal->hdr.aSalt[1], &salt1, 4); + walIndexWriteHdr(pWal); + AtomicStore(&pInfo->nBackfill, 0); + pInfo->nBackfillAttempted = 0; + pInfo->aReadMark[1] = 0; + for(i=2; iaReadMark[i] = READMARK_NOT_USED; + assert( pInfo->aReadMark[0]==0 ); +} + +/* +** Copy as much content as we can from the WAL back into the database file +** in response to an sqlite3_wal_checkpoint() request or the equivalent. +** +** The amount of information copies from WAL to database might be limited +** by active readers. This routine will never overwrite a database page +** that a concurrent reader might be using. +** +** All I/O barrier operations (a.k.a fsyncs) occur in this routine when +** SQLite is in WAL-mode in synchronous=NORMAL. That means that if +** checkpoints are always run by a background thread or background +** process, foreground threads will never block on a lengthy fsync call. +** +** Fsync is called on the WAL before writing content out of the WAL and +** into the database. This ensures that if the new content is persistent +** in the WAL and can be recovered following a power-loss or hard reset. +** +** Fsync is also called on the database file if (and only if) the entire +** WAL content is copied into the database file. This second fsync makes +** it safe to delete the WAL since the new content will persist in the +** database file. +** +** This routine uses and updates the nBackfill field of the wal-index header. +** This is the only routine that will increase the value of nBackfill. +** (A WAL reset or recovery will revert nBackfill to zero, but not increase +** its value.) +** +** The caller must be holding sufficient locks to ensure that no other +** checkpoint is running (in any other thread or process) at the same +** time. +*/ +static int walCheckpoint( + Wal *pWal, /* Wal connection */ + sqlite3 *db, /* Check for interrupts on this handle */ + int eMode, /* One of PASSIVE, FULL or RESTART */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int sync_flags, /* Flags for OsSync() (or 0) */ + u8 *zBuf /* Temporary buffer to use */ +){ + int rc = SQLITE_OK; /* Return code */ + int szPage; /* Database page-size */ + WalIterator *pIter = 0; /* Wal iterator context */ + u32 iDbpage = 0; /* Next database page to write */ + u32 iFrame = 0; /* Wal frame containing data for iDbpage */ + u32 mxSafeFrame; /* Max frame that can be backfilled */ + u32 mxPage; /* Max database page to write */ + int i; /* Loop counter */ + volatile WalCkptInfo *pInfo; /* The checkpoint status information */ + + szPage = walPagesize(pWal); + testcase( szPage<=32768 ); + testcase( szPage>=65536 ); + pInfo = walCkptInfo(pWal); + if( pInfo->nBackfillhdr.mxFrame ){ + + /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked + ** in the SQLITE_CHECKPOINT_PASSIVE mode. */ + assert( eMode!=SQLITE_CHECKPOINT_PASSIVE || xBusy==0 ); + + /* Compute in mxSafeFrame the index of the last frame of the WAL that is + ** safe to write into the database. Frames beyond mxSafeFrame might + ** overwrite database pages that are in use by active readers and thus + ** cannot be backfilled from the WAL. + */ + mxSafeFrame = pWal->hdr.mxFrame; + mxPage = pWal->hdr.nPage; + for(i=1; iaReadMark+i); + if( mxSafeFrame>y ){ + assert( y<=pWal->hdr.mxFrame ); + rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(i), 1); + if( rc==SQLITE_OK ){ + u32 iMark = (i==1 ? mxSafeFrame : READMARK_NOT_USED); + AtomicStore(pInfo->aReadMark+i, iMark); + walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1); + }else if( rc==SQLITE_BUSY ){ + mxSafeFrame = y; + xBusy = 0; + }else{ + goto walcheckpoint_out; + } + } + } + + /* Allocate the iterator */ + if( pInfo->nBackfillnBackfill, &pIter); + assert( rc==SQLITE_OK || pIter==0 ); + } + + if( pIter + && (rc = walBusyLock(pWal,xBusy,pBusyArg,WAL_READ_LOCK(0),1))==SQLITE_OK + ){ + u32 nBackfill = pInfo->nBackfill; + + pInfo->nBackfillAttempted = mxSafeFrame; + + /* Sync the WAL to disk */ + rc = sqlite3OsSync(pWal->pWalFd, CKPT_SYNC_FLAGS(sync_flags)); + + /* If the database may grow as a result of this checkpoint, hint + ** about the eventual size of the db file to the VFS layer. + */ + if( rc==SQLITE_OK ){ + i64 nReq = ((i64)mxPage * szPage); + i64 nSize; /* Current size of database file */ + sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_CKPT_START, 0); + rc = sqlite3OsFileSize(pWal->pDbFd, &nSize); + if( rc==SQLITE_OK && nSizehdr.mxFrame*szPage)pDbFd, SQLITE_FCNTL_SIZE_HINT,&nReq); + } + } + + } + + /* Iterate through the contents of the WAL, copying data to the db file */ + while( rc==SQLITE_OK && 0==walIteratorNext(pIter, &iDbpage, &iFrame) ){ + i64 iOffset; + assert( walFramePgno(pWal, iFrame)==iDbpage ); + if( AtomicLoad(&db->u1.isInterrupted) ){ + rc = db->mallocFailed ? SQLITE_NOMEM_BKPT : SQLITE_INTERRUPT; + break; + } + if( iFrame<=nBackfill || iFrame>mxSafeFrame || iDbpage>mxPage ){ + continue; + } + iOffset = walFrameOffset(iFrame, szPage) + WAL_FRAME_HDRSIZE; + /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL file */ + rc = sqlite3OsRead(pWal->pWalFd, zBuf, szPage, iOffset); + if( rc!=SQLITE_OK ) break; + iOffset = (iDbpage-1)*(i64)szPage; + testcase( IS_BIG_INT(iOffset) ); + rc = sqlite3OsWrite(pWal->pDbFd, zBuf, szPage, iOffset); + if( rc!=SQLITE_OK ) break; + } + sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_CKPT_DONE, 0); + + /* If work was actually accomplished... */ + if( rc==SQLITE_OK ){ + if( mxSafeFrame==walIndexHdr(pWal)->mxFrame ){ + i64 szDb = pWal->hdr.nPage*(i64)szPage; + testcase( IS_BIG_INT(szDb) ); + rc = sqlite3OsTruncate(pWal->pDbFd, szDb); + if( rc==SQLITE_OK ){ + rc = sqlite3OsSync(pWal->pDbFd, CKPT_SYNC_FLAGS(sync_flags)); + } + } + if( rc==SQLITE_OK ){ + AtomicStore(&pInfo->nBackfill, mxSafeFrame); + } + } + + /* Release the reader lock held while backfilling */ + walUnlockExclusive(pWal, WAL_READ_LOCK(0), 1); + } + + if( rc==SQLITE_BUSY ){ + /* Reset the return code so as not to report a checkpoint failure + ** just because there are active readers. */ + rc = SQLITE_OK; + } + } + + /* If this is an SQLITE_CHECKPOINT_RESTART or TRUNCATE operation, and the + ** entire wal file has been copied into the database file, then block + ** until all readers have finished using the wal file. This ensures that + ** the next process to write to the database restarts the wal file. + */ + if( rc==SQLITE_OK && eMode!=SQLITE_CHECKPOINT_PASSIVE ){ + assert( pWal->writeLock ); + if( pInfo->nBackfillhdr.mxFrame ){ + rc = SQLITE_BUSY; + }else if( eMode>=SQLITE_CHECKPOINT_RESTART ){ + u32 salt1; + sqlite3_randomness(4, &salt1); + assert( pInfo->nBackfill==pWal->hdr.mxFrame ); + rc = walBusyLock(pWal, xBusy, pBusyArg, WAL_READ_LOCK(1), WAL_NREADER-1); + if( rc==SQLITE_OK ){ + if( eMode==SQLITE_CHECKPOINT_TRUNCATE ){ + /* IMPLEMENTATION-OF: R-44699-57140 This mode works the same way as + ** SQLITE_CHECKPOINT_RESTART with the addition that it also + ** truncates the log file to zero bytes just prior to a + ** successful return. + ** + ** In theory, it might be safe to do this without updating the + ** wal-index header in shared memory, as all subsequent reader or + ** writer clients should see that the entire log file has been + ** checkpointed and behave accordingly. This seems unsafe though, + ** as it would leave the system in a state where the contents of + ** the wal-index header do not match the contents of the + ** file-system. To avoid this, update the wal-index header to + ** indicate that the log file contains zero valid frames. */ + walRestartHdr(pWal, salt1); + rc = sqlite3OsTruncate(pWal->pWalFd, 0); + } + walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + } + } + } + + walcheckpoint_out: + walIteratorFree(pIter); + return rc; +} + +/* +** If the WAL file is currently larger than nMax bytes in size, truncate +** it to exactly nMax bytes. If an error occurs while doing so, ignore it. +*/ +static void walLimitSize(Wal *pWal, i64 nMax){ + i64 sz; + int rx; + sqlite3BeginBenignMalloc(); + rx = sqlite3OsFileSize(pWal->pWalFd, &sz); + if( rx==SQLITE_OK && (sz > nMax ) ){ + rx = sqlite3OsTruncate(pWal->pWalFd, nMax); + } + sqlite3EndBenignMalloc(); + if( rx ){ + sqlite3_log(rx, "cannot limit WAL size: %s", pWal->zWalName); + } +} + +/* +** Close a connection to a log file. +*/ +int sqlite3WalClose( + Wal *pWal, /* Wal to close */ + sqlite3 *db, /* For interrupt flag */ + int sync_flags, /* Flags to pass to OsSync() (or 0) */ + int nBuf, + u8 *zBuf /* Buffer of at least nBuf bytes */ +){ + int rc = SQLITE_OK; + if( pWal ){ + int isDelete = 0; /* True to unlink wal and wal-index files */ + + /* If an EXCLUSIVE lock can be obtained on the database file (using the + ** ordinary, rollback-mode locking methods, this guarantees that the + ** connection associated with this log file is the only connection to + ** the database. In this case checkpoint the database and unlink both + ** the wal and wal-index files. + ** + ** The EXCLUSIVE lock is not released before returning. + */ + if( zBuf!=0 + && SQLITE_OK==(rc = sqlite3OsLock(pWal->pDbFd, SQLITE_LOCK_EXCLUSIVE)) + ){ + if( pWal->exclusiveMode==WAL_NORMAL_MODE ){ + pWal->exclusiveMode = WAL_EXCLUSIVE_MODE; + } + rc = sqlite3WalCheckpoint(pWal, db, + SQLITE_CHECKPOINT_PASSIVE, 0, 0, sync_flags, nBuf, zBuf, 0, 0 + ); + if( rc==SQLITE_OK ){ + int bPersist = -1; + sqlite3OsFileControlHint( + pWal->pDbFd, SQLITE_FCNTL_PERSIST_WAL, &bPersist + ); + if( bPersist!=1 ){ + /* Try to delete the WAL file if the checkpoint completed and + ** fsyned (rc==SQLITE_OK) and if we are not in persistent-wal + ** mode (!bPersist) */ + isDelete = 1; + }else if( pWal->mxWalSize>=0 ){ + /* Try to truncate the WAL file to zero bytes if the checkpoint + ** completed and fsynced (rc==SQLITE_OK) and we are in persistent + ** WAL mode (bPersist) and if the PRAGMA journal_size_limit is a + ** non-negative value (pWal->mxWalSize>=0). Note that we truncate + ** to zero bytes as truncating to the journal_size_limit might + ** leave a corrupt WAL file on disk. */ + walLimitSize(pWal, 0); + } + } + } + + walIndexClose(pWal, isDelete); + sqlite3OsClose(pWal->pWalFd); + if( isDelete ){ + sqlite3BeginBenignMalloc(); + sqlite3OsDelete(pWal->pVfs, pWal->zWalName, 0); + sqlite3EndBenignMalloc(); + } + WALTRACE(("WAL%p: closed\n", pWal)); + sqlite3_free((void *)pWal->apWiData); + sqlite3_free(pWal); + } + return rc; +} + +/* +** Try to read the wal-index header. Return 0 on success and 1 if +** there is a problem. +** +** The wal-index is in shared memory. Another thread or process might +** be writing the header at the same time this procedure is trying to +** read it, which might result in inconsistency. A dirty read is detected +** by verifying that both copies of the header are the same and also by +** a checksum on the header. +** +** If and only if the read is consistent and the header is different from +** pWal->hdr, then pWal->hdr is updated to the content of the new header +** and *pChanged is set to 1. +** +** If the checksum cannot be verified return non-zero. If the header +** is read successfully and the checksum verified, return zero. +*/ +static SQLITE_NO_TSAN int walIndexTryHdr(Wal *pWal, int *pChanged){ + u32 aCksum[2]; /* Checksum on the header content */ + WalIndexHdr h1, h2; /* Two copies of the header content */ + WalIndexHdr volatile *aHdr; /* Header in shared memory */ + + /* The first page of the wal-index must be mapped at this point. */ + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + + /* Read the header. This might happen concurrently with a write to the + ** same area of shared memory on a different CPU in a SMP, + ** meaning it is possible that an inconsistent snapshot is read + ** from the file. If this happens, return non-zero. + ** + ** tag-20200519-1: + ** There are two copies of the header at the beginning of the wal-index. + ** When reading, read [0] first then [1]. Writes are in the reverse order. + ** Memory barriers are used to prevent the compiler or the hardware from + ** reordering the reads and writes. TSAN and similar tools can sometimes + ** give false-positive warnings about these accesses because the tools do not + ** account for the double-read and the memory barrier. The use of mutexes + ** here would be problematic as the memory being accessed is potentially + ** shared among multiple processes and not all mutex implementions work + ** reliably in that environment. + */ + aHdr = walIndexHdr(pWal); + memcpy(&h1, (void *)&aHdr[0], sizeof(h1)); /* Possible TSAN false-positive */ + walShmBarrier(pWal); + memcpy(&h2, (void *)&aHdr[1], sizeof(h2)); + + if( memcmp(&h1, &h2, sizeof(h1))!=0 ){ + return 1; /* Dirty read */ + } + if( h1.isInit==0 ){ + return 1; /* Malformed header - probably all zeros */ + } + walChecksumBytes(1, (u8*)&h1, sizeof(h1)-sizeof(h1.aCksum), 0, aCksum); + if( aCksum[0]!=h1.aCksum[0] || aCksum[1]!=h1.aCksum[1] ){ + return 1; /* Checksum does not match */ + } + + if( memcmp(&pWal->hdr, &h1, sizeof(WalIndexHdr)) ){ + *pChanged = 1; + memcpy(&pWal->hdr, &h1, sizeof(WalIndexHdr)); + pWal->szPage = (pWal->hdr.szPage&0xfe00) + ((pWal->hdr.szPage&0x0001)<<16); + testcase( pWal->szPage<=32768 ); + testcase( pWal->szPage>=65536 ); + } + + /* The header was successfully read. Return zero. */ + return 0; +} + +/* +** This is the value that walTryBeginRead returns when it needs to +** be retried. +*/ +#define WAL_RETRY (-1) + +/* +** Read the wal-index header from the wal-index and into pWal->hdr. +** If the wal-header appears to be corrupt, try to reconstruct the +** wal-index from the WAL before returning. +** +** Set *pChanged to 1 if the wal-index header value in pWal->hdr is +** changed by this operation. If pWal->hdr is unchanged, set *pChanged +** to 0. +** +** If the wal-index header is successfully read, return SQLITE_OK. +** Otherwise an SQLite error code. +*/ +static int walIndexReadHdr(Wal *pWal, int *pChanged){ + int rc; /* Return code */ + int badHdr; /* True if a header read failed */ + volatile u32 *page0; /* Chunk of wal-index containing header */ + + /* Ensure that page 0 of the wal-index (the page that contains the + ** wal-index header) is mapped. Return early if an error occurs here. + */ + assert( pChanged ); + rc = walIndexPage(pWal, 0, &page0); + if( rc!=SQLITE_OK ){ + assert( rc!=SQLITE_READONLY ); /* READONLY changed to OK in walIndexPage */ + if( rc==SQLITE_READONLY_CANTINIT ){ + /* The SQLITE_READONLY_CANTINIT return means that the shared-memory + ** was openable but is not writable, and this thread is unable to + ** confirm that another write-capable connection has the shared-memory + ** open, and hence the content of the shared-memory is unreliable, + ** since the shared-memory might be inconsistent with the WAL file + ** and there is no writer on hand to fix it. */ + assert( page0==0 ); + assert( pWal->writeLock==0 ); + assert( pWal->readOnly & WAL_SHM_RDONLY ); + pWal->bShmUnreliable = 1; + pWal->exclusiveMode = WAL_HEAPMEMORY_MODE; + *pChanged = 1; + }else{ + return rc; /* Any other non-OK return is just an error */ + } + }else{ + /* page0 can be NULL if the SHM is zero bytes in size and pWal->writeLock + ** is zero, which prevents the SHM from growing */ + testcase( page0!=0 ); + } + assert( page0!=0 || pWal->writeLock==0 ); + + /* If the first page of the wal-index has been mapped, try to read the + ** wal-index header immediately, without holding any lock. This usually + ** works, but may fail if the wal-index header is corrupt or currently + ** being modified by another thread or process. + */ + badHdr = (page0 ? walIndexTryHdr(pWal, pChanged) : 1); + + /* If the first attempt failed, it might have been due to a race + ** with a writer. So get a WRITE lock and try again. + */ + if( badHdr ){ + if( pWal->bShmUnreliable==0 && (pWal->readOnly & WAL_SHM_RDONLY) ){ + if( SQLITE_OK==(rc = walLockShared(pWal, WAL_WRITE_LOCK)) ){ + walUnlockShared(pWal, WAL_WRITE_LOCK); + rc = SQLITE_READONLY_RECOVERY; + } + }else{ + int bWriteLock = pWal->writeLock; + if( bWriteLock || SQLITE_OK==(rc = walLockWriter(pWal)) ){ + pWal->writeLock = 1; + if( SQLITE_OK==(rc = walIndexPage(pWal, 0, &page0)) ){ + badHdr = walIndexTryHdr(pWal, pChanged); + if( badHdr ){ + /* If the wal-index header is still malformed even while holding + ** a WRITE lock, it can only mean that the header is corrupted and + ** needs to be reconstructed. So run recovery to do exactly that. + */ + rc = walIndexRecover(pWal); + *pChanged = 1; + } + } + if( bWriteLock==0 ){ + pWal->writeLock = 0; + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + } + } + } + } + + /* If the header is read successfully, check the version number to make + ** sure the wal-index was not constructed with some future format that + ** this version of SQLite cannot understand. + */ + if( badHdr==0 && pWal->hdr.iVersion!=WALINDEX_MAX_VERSION ){ + rc = SQLITE_CANTOPEN_BKPT; + } + if( pWal->bShmUnreliable ){ + if( rc!=SQLITE_OK ){ + walIndexClose(pWal, 0); + pWal->bShmUnreliable = 0; + assert( pWal->nWiData>0 && pWal->apWiData[0]==0 ); + /* walIndexRecover() might have returned SHORT_READ if a concurrent + ** writer truncated the WAL out from under it. If that happens, it + ** indicates that a writer has fixed the SHM file for us, so retry */ + if( rc==SQLITE_IOERR_SHORT_READ ) rc = WAL_RETRY; + } + pWal->exclusiveMode = WAL_NORMAL_MODE; + } + + return rc; +} + +/* +** Open a transaction in a connection where the shared-memory is read-only +** and where we cannot verify that there is a separate write-capable connection +** on hand to keep the shared-memory up-to-date with the WAL file. +** +** This can happen, for example, when the shared-memory is implemented by +** memory-mapping a *-shm file, where a prior writer has shut down and +** left the *-shm file on disk, and now the present connection is trying +** to use that database but lacks write permission on the *-shm file. +** Other scenarios are also possible, depending on the VFS implementation. +** +** Precondition: +** +** The *-wal file has been read and an appropriate wal-index has been +** constructed in pWal->apWiData[] using heap memory instead of shared +** memory. +** +** If this function returns SQLITE_OK, then the read transaction has +** been successfully opened. In this case output variable (*pChanged) +** is set to true before returning if the caller should discard the +** contents of the page cache before proceeding. Or, if it returns +** WAL_RETRY, then the heap memory wal-index has been discarded and +** the caller should retry opening the read transaction from the +** beginning (including attempting to map the *-shm file). +** +** If an error occurs, an SQLite error code is returned. +*/ +static int walBeginShmUnreliable(Wal *pWal, int *pChanged){ + i64 szWal; /* Size of wal file on disk in bytes */ + i64 iOffset; /* Current offset when reading wal file */ + u8 aBuf[WAL_HDRSIZE]; /* Buffer to load WAL header into */ + u8 *aFrame = 0; /* Malloc'd buffer to load entire frame */ + int szFrame; /* Number of bytes in buffer aFrame[] */ + u8 *aData; /* Pointer to data part of aFrame buffer */ + volatile void *pDummy; /* Dummy argument for xShmMap */ + int rc; /* Return code */ + u32 aSaveCksum[2]; /* Saved copy of pWal->hdr.aFrameCksum */ + + assert( pWal->bShmUnreliable ); + assert( pWal->readOnly & WAL_SHM_RDONLY ); + assert( pWal->nWiData>0 && pWal->apWiData[0] ); + + /* Take WAL_READ_LOCK(0). This has the effect of preventing any + ** writers from running a checkpoint, but does not stop them + ** from running recovery. */ + rc = walLockShared(pWal, WAL_READ_LOCK(0)); + if( rc!=SQLITE_OK ){ + if( rc==SQLITE_BUSY ) rc = WAL_RETRY; + goto begin_unreliable_shm_out; + } + pWal->readLock = 0; + + /* Check to see if a separate writer has attached to the shared-memory area, + ** thus making the shared-memory "reliable" again. Do this by invoking + ** the xShmMap() routine of the VFS and looking to see if the return + ** is SQLITE_READONLY instead of SQLITE_READONLY_CANTINIT. + ** + ** If the shared-memory is now "reliable" return WAL_RETRY, which will + ** cause the heap-memory WAL-index to be discarded and the actual + ** shared memory to be used in its place. + ** + ** This step is important because, even though this connection is holding + ** the WAL_READ_LOCK(0) which prevents a checkpoint, a writer might + ** have already checkpointed the WAL file and, while the current + ** is active, wrap the WAL and start overwriting frames that this + ** process wants to use. + ** + ** Once sqlite3OsShmMap() has been called for an sqlite3_file and has + ** returned any SQLITE_READONLY value, it must return only SQLITE_READONLY + ** or SQLITE_READONLY_CANTINIT or some error for all subsequent invocations, + ** even if some external agent does a "chmod" to make the shared-memory + ** writable by us, until sqlite3OsShmUnmap() has been called. + ** This is a requirement on the VFS implementation. + */ + rc = sqlite3OsShmMap(pWal->pDbFd, 0, WALINDEX_PGSZ, 0, &pDummy); + assert( rc!=SQLITE_OK ); /* SQLITE_OK not possible for read-only connection */ + if( rc!=SQLITE_READONLY_CANTINIT ){ + rc = (rc==SQLITE_READONLY ? WAL_RETRY : rc); + goto begin_unreliable_shm_out; + } + + /* We reach this point only if the real shared-memory is still unreliable. + ** Assume the in-memory WAL-index substitute is correct and load it + ** into pWal->hdr. + */ + memcpy(&pWal->hdr, (void*)walIndexHdr(pWal), sizeof(WalIndexHdr)); + + /* Make sure some writer hasn't come in and changed the WAL file out + ** from under us, then disconnected, while we were not looking. + */ + rc = sqlite3OsFileSize(pWal->pWalFd, &szWal); + if( rc!=SQLITE_OK ){ + goto begin_unreliable_shm_out; + } + if( szWalhdr.mxFrame==0 ? SQLITE_OK : WAL_RETRY); + goto begin_unreliable_shm_out; + } + + /* Check the salt keys at the start of the wal file still match. */ + rc = sqlite3OsRead(pWal->pWalFd, aBuf, WAL_HDRSIZE, 0); + if( rc!=SQLITE_OK ){ + goto begin_unreliable_shm_out; + } + if( memcmp(&pWal->hdr.aSalt, &aBuf[16], 8) ){ + /* Some writer has wrapped the WAL file while we were not looking. + ** Return WAL_RETRY which will cause the in-memory WAL-index to be + ** rebuilt. */ + rc = WAL_RETRY; + goto begin_unreliable_shm_out; + } + + /* Allocate a buffer to read frames into */ + szFrame = pWal->hdr.szPage + WAL_FRAME_HDRSIZE; + aFrame = (u8 *)sqlite3_malloc64(szFrame); + if( aFrame==0 ){ + rc = SQLITE_NOMEM_BKPT; + goto begin_unreliable_shm_out; + } + aData = &aFrame[WAL_FRAME_HDRSIZE]; + + /* Check to see if a complete transaction has been appended to the + ** wal file since the heap-memory wal-index was created. If so, the + ** heap-memory wal-index is discarded and WAL_RETRY returned to + ** the caller. */ + aSaveCksum[0] = pWal->hdr.aFrameCksum[0]; + aSaveCksum[1] = pWal->hdr.aFrameCksum[1]; + for(iOffset=walFrameOffset(pWal->hdr.mxFrame+1, pWal->hdr.szPage); + iOffset+szFrame<=szWal; + iOffset+=szFrame + ){ + u32 pgno; /* Database page number for frame */ + u32 nTruncate; /* dbsize field from frame header */ + + /* Read and decode the next log frame. */ + rc = sqlite3OsRead(pWal->pWalFd, aFrame, szFrame, iOffset); + if( rc!=SQLITE_OK ) break; + if( !walDecodeFrame(pWal, &pgno, &nTruncate, aData, aFrame) ) break; + + /* If nTruncate is non-zero, then a complete transaction has been + ** appended to this wal file. Set rc to WAL_RETRY and break out of + ** the loop. */ + if( nTruncate ){ + rc = WAL_RETRY; + break; + } + } + pWal->hdr.aFrameCksum[0] = aSaveCksum[0]; + pWal->hdr.aFrameCksum[1] = aSaveCksum[1]; + + begin_unreliable_shm_out: + sqlite3_free(aFrame); + if( rc!=SQLITE_OK ){ + int i; + for(i=0; inWiData; i++){ + sqlite3_free((void*)pWal->apWiData[i]); + pWal->apWiData[i] = 0; + } + pWal->bShmUnreliable = 0; + sqlite3WalEndReadTransaction(pWal); + *pChanged = 1; + } + return rc; +} + +/* +** Attempt to start a read transaction. This might fail due to a race or +** other transient condition. When that happens, it returns WAL_RETRY to +** indicate to the caller that it is safe to retry immediately. +** +** On success return SQLITE_OK. On a permanent failure (such an +** I/O error or an SQLITE_BUSY because another process is running +** recovery) return a positive error code. +** +** The useWal parameter is true to force the use of the WAL and disable +** the case where the WAL is bypassed because it has been completely +** checkpointed. If useWal==0 then this routine calls walIndexReadHdr() +** to make a copy of the wal-index header into pWal->hdr. If the +** wal-index header has changed, *pChanged is set to 1 (as an indication +** to the caller that the local page cache is obsolete and needs to be +** flushed.) When useWal==1, the wal-index header is assumed to already +** be loaded and the pChanged parameter is unused. +** +** The caller must set the cnt parameter to the number of prior calls to +** this routine during the current read attempt that returned WAL_RETRY. +** This routine will start taking more aggressive measures to clear the +** race conditions after multiple WAL_RETRY returns, and after an excessive +** number of errors will ultimately return SQLITE_PROTOCOL. The +** SQLITE_PROTOCOL return indicates that some other process has gone rogue +** and is not honoring the locking protocol. There is a vanishingly small +** chance that SQLITE_PROTOCOL could be returned because of a run of really +** bad luck when there is lots of contention for the wal-index, but that +** possibility is so small that it can be safely neglected, we believe. +** +** On success, this routine obtains a read lock on +** WAL_READ_LOCK(pWal->readLock). The pWal->readLock integer is +** in the range 0 <= pWal->readLock < WAL_NREADER. If pWal->readLock==(-1) +** that means the Wal does not hold any read lock. The reader must not +** access any database page that is modified by a WAL frame up to and +** including frame number aReadMark[pWal->readLock]. The reader will +** use WAL frames up to and including pWal->hdr.mxFrame if pWal->readLock>0 +** Or if pWal->readLock==0, then the reader will ignore the WAL +** completely and get all content directly from the database file. +** If the useWal parameter is 1 then the WAL will never be ignored and +** this routine will always set pWal->readLock>0 on success. +** When the read transaction is completed, the caller must release the +** lock on WAL_READ_LOCK(pWal->readLock) and set pWal->readLock to -1. +** +** This routine uses the nBackfill and aReadMark[] fields of the header +** to select a particular WAL_READ_LOCK() that strives to let the +** checkpoint process do as much work as possible. This routine might +** update values of the aReadMark[] array in the header, but if it does +** so it takes care to hold an exclusive lock on the corresponding +** WAL_READ_LOCK() while changing values. +*/ +static int walTryBeginRead(Wal *pWal, int *pChanged, int useWal, int cnt){ + volatile WalCkptInfo *pInfo; /* Checkpoint information in wal-index */ + u32 mxReadMark; /* Largest aReadMark[] value */ + int mxI; /* Index of largest aReadMark[] value */ + int i; /* Loop counter */ + int rc = SQLITE_OK; /* Return code */ + u32 mxFrame; /* Wal frame to lock to */ + + assert( pWal->readLock<0 ); /* Not currently locked */ + + /* useWal may only be set for read/write connections */ + assert( (pWal->readOnly & WAL_SHM_RDONLY)==0 || useWal==0 ); + + /* Take steps to avoid spinning forever if there is a protocol error. + ** + ** Circumstances that cause a RETRY should only last for the briefest + ** instances of time. No I/O or other system calls are done while the + ** locks are held, so the locks should not be held for very long. But + ** if we are unlucky, another process that is holding a lock might get + ** paged out or take a page-fault that is time-consuming to resolve, + ** during the few nanoseconds that it is holding the lock. In that case, + ** it might take longer than normal for the lock to free. + ** + ** After 5 RETRYs, we begin calling sqlite3OsSleep(). The first few + ** calls to sqlite3OsSleep() have a delay of 1 microsecond. Really this + ** is more of a scheduler yield than an actual delay. But on the 10th + ** an subsequent retries, the delays start becoming longer and longer, + ** so that on the 100th (and last) RETRY we delay for 323 milliseconds. + ** The total delay time before giving up is less than 10 seconds. + */ + if( cnt>5 ){ + int nDelay = 1; /* Pause time in microseconds */ + if( cnt>100 ){ + VVA_ONLY( pWal->lockError = 1; ) + return SQLITE_PROTOCOL; + } + if( cnt>=10 ) nDelay = (cnt-9)*(cnt-9)*39; + sqlite3OsSleep(pWal->pVfs, nDelay); + } + + if( !useWal ){ + assert( rc==SQLITE_OK ); + if( pWal->bShmUnreliable==0 ){ + rc = walIndexReadHdr(pWal, pChanged); + } + if( rc==SQLITE_BUSY ){ + /* If there is not a recovery running in another thread or process + ** then convert BUSY errors to WAL_RETRY. If recovery is known to + ** be running, convert BUSY to BUSY_RECOVERY. There is a race here + ** which might cause WAL_RETRY to be returned even if BUSY_RECOVERY + ** would be technically correct. But the race is benign since with + ** WAL_RETRY this routine will be called again and will probably be + ** right on the second iteration. + */ + if( pWal->apWiData[0]==0 ){ + /* This branch is taken when the xShmMap() method returns SQLITE_BUSY. + ** We assume this is a transient condition, so return WAL_RETRY. The + ** xShmMap() implementation used by the default unix and win32 VFS + ** modules may return SQLITE_BUSY due to a race condition in the + ** code that determines whether or not the shared-memory region + ** must be zeroed before the requested page is returned. + */ + rc = WAL_RETRY; + }else if( SQLITE_OK==(rc = walLockShared(pWal, WAL_RECOVER_LOCK)) ){ + walUnlockShared(pWal, WAL_RECOVER_LOCK); + rc = WAL_RETRY; + }else if( rc==SQLITE_BUSY ){ + rc = SQLITE_BUSY_RECOVERY; + } + } + if( rc!=SQLITE_OK ){ + return rc; + } + else if( pWal->bShmUnreliable ){ + return walBeginShmUnreliable(pWal, pChanged); + } + } + + assert( pWal->nWiData>0 ); + assert( pWal->apWiData[0]!=0 ); + pInfo = walCkptInfo(pWal); + if( !useWal && AtomicLoad(&pInfo->nBackfill)==pWal->hdr.mxFrame +#ifdef SQLITE_ENABLE_SNAPSHOT + && (pWal->pSnapshot==0 || pWal->hdr.mxFrame==0) +#endif + ){ + /* The WAL has been completely backfilled (or it is empty). + ** and can be safely ignored. + */ + rc = walLockShared(pWal, WAL_READ_LOCK(0)); + walShmBarrier(pWal); + if( rc==SQLITE_OK ){ + if( memcmp((void *)walIndexHdr(pWal), &pWal->hdr, sizeof(WalIndexHdr)) ){ + /* It is not safe to allow the reader to continue here if frames + ** may have been appended to the log before READ_LOCK(0) was obtained. + ** When holding READ_LOCK(0), the reader ignores the entire log file, + ** which implies that the database file contains a trustworthy + ** snapshot. Since holding READ_LOCK(0) prevents a checkpoint from + ** happening, this is usually correct. + ** + ** However, if frames have been appended to the log (or if the log + ** is wrapped and written for that matter) before the READ_LOCK(0) + ** is obtained, that is not necessarily true. A checkpointer may + ** have started to backfill the appended frames but crashed before + ** it finished. Leaving a corrupt image in the database file. + */ + walUnlockShared(pWal, WAL_READ_LOCK(0)); + return WAL_RETRY; + } + pWal->readLock = 0; + return SQLITE_OK; + }else if( rc!=SQLITE_BUSY ){ + return rc; + } + } + + /* If we get this far, it means that the reader will want to use + ** the WAL to get at content from recent commits. The job now is + ** to select one of the aReadMark[] entries that is closest to + ** but not exceeding pWal->hdr.mxFrame and lock that entry. + */ + mxReadMark = 0; + mxI = 0; + mxFrame = pWal->hdr.mxFrame; +#ifdef SQLITE_ENABLE_SNAPSHOT + if( pWal->pSnapshot && pWal->pSnapshot->mxFramepSnapshot->mxFrame; + } +#endif + for(i=1; iaReadMark+i); + if( mxReadMark<=thisMark && thisMark<=mxFrame ){ + assert( thisMark!=READMARK_NOT_USED ); + mxReadMark = thisMark; + mxI = i; + } + } + if( (pWal->readOnly & WAL_SHM_RDONLY)==0 + && (mxReadMarkaReadMark+i,mxFrame); + mxReadMark = mxFrame; + mxI = i; + walUnlockExclusive(pWal, WAL_READ_LOCK(i), 1); + break; + }else if( rc!=SQLITE_BUSY ){ + return rc; + } + } + } + if( mxI==0 ){ + assert( rc==SQLITE_BUSY || (pWal->readOnly & WAL_SHM_RDONLY)!=0 ); + return rc==SQLITE_BUSY ? WAL_RETRY : SQLITE_READONLY_CANTINIT; + } + + rc = walLockShared(pWal, WAL_READ_LOCK(mxI)); + if( rc ){ + return rc==SQLITE_BUSY ? WAL_RETRY : rc; + } + /* Now that the read-lock has been obtained, check that neither the + ** value in the aReadMark[] array or the contents of the wal-index + ** header have changed. + ** + ** It is necessary to check that the wal-index header did not change + ** between the time it was read and when the shared-lock was obtained + ** on WAL_READ_LOCK(mxI) was obtained to account for the possibility + ** that the log file may have been wrapped by a writer, or that frames + ** that occur later in the log than pWal->hdr.mxFrame may have been + ** copied into the database by a checkpointer. If either of these things + ** happened, then reading the database with the current value of + ** pWal->hdr.mxFrame risks reading a corrupted snapshot. So, retry + ** instead. + ** + ** Before checking that the live wal-index header has not changed + ** since it was read, set Wal.minFrame to the first frame in the wal + ** file that has not yet been checkpointed. This client will not need + ** to read any frames earlier than minFrame from the wal file - they + ** can be safely read directly from the database file. + ** + ** Because a ShmBarrier() call is made between taking the copy of + ** nBackfill and checking that the wal-header in shared-memory still + ** matches the one cached in pWal->hdr, it is guaranteed that the + ** checkpointer that set nBackfill was not working with a wal-index + ** header newer than that cached in pWal->hdr. If it were, that could + ** cause a problem. The checkpointer could omit to checkpoint + ** a version of page X that lies before pWal->minFrame (call that version + ** A) on the basis that there is a newer version (version B) of the same + ** page later in the wal file. But if version B happens to like past + ** frame pWal->hdr.mxFrame - then the client would incorrectly assume + ** that it can read version A from the database file. However, since + ** we can guarantee that the checkpointer that set nBackfill could not + ** see any pages past pWal->hdr.mxFrame, this problem does not come up. + */ + pWal->minFrame = AtomicLoad(&pInfo->nBackfill)+1; + walShmBarrier(pWal); + if( AtomicLoad(pInfo->aReadMark+mxI)!=mxReadMark + || memcmp((void *)walIndexHdr(pWal), &pWal->hdr, sizeof(WalIndexHdr)) + ){ + walUnlockShared(pWal, WAL_READ_LOCK(mxI)); + return WAL_RETRY; + }else{ + assert( mxReadMark<=pWal->hdr.mxFrame ); + pWal->readLock = (i16)mxI; + } + return rc; +} + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* +** Attempt to reduce the value of the WalCkptInfo.nBackfillAttempted +** variable so that older snapshots can be accessed. To do this, loop +** through all wal frames from nBackfillAttempted to (nBackfill+1), +** comparing their content to the corresponding page with the database +** file, if any. Set nBackfillAttempted to the frame number of the +** first frame for which the wal file content matches the db file. +** +** This is only really safe if the file-system is such that any page +** writes made by earlier checkpointers were atomic operations, which +** is not always true. It is also possible that nBackfillAttempted +** may be left set to a value larger than expected, if a wal frame +** contains content that duplicate of an earlier version of the same +** page. +** +** SQLITE_OK is returned if successful, or an SQLite error code if an +** error occurs. It is not an error if nBackfillAttempted cannot be +** decreased at all. +*/ +int sqlite3WalSnapshotRecover(Wal *pWal){ + int rc; + + assert( pWal->readLock>=0 ); + rc = walLockExclusive(pWal, WAL_CKPT_LOCK, 1); + if( rc==SQLITE_OK ){ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + int szPage = (int)pWal->szPage; + i64 szDb; /* Size of db file in bytes */ + + rc = sqlite3OsFileSize(pWal->pDbFd, &szDb); + if( rc==SQLITE_OK ){ + void *pBuf1 = sqlite3_malloc(szPage); + void *pBuf2 = sqlite3_malloc(szPage); + if( pBuf1==0 || pBuf2==0 ){ + rc = SQLITE_NOMEM; + }else{ + u32 i = pInfo->nBackfillAttempted; + for(i=pInfo->nBackfillAttempted; i>AtomicLoad(&pInfo->nBackfill); i--){ + WalHashLoc sLoc; /* Hash table location */ + u32 pgno; /* Page number in db file */ + i64 iDbOff; /* Offset of db file entry */ + i64 iWalOff; /* Offset of wal file entry */ + + rc = walHashGet(pWal, walFramePage(i), &sLoc); + if( rc!=SQLITE_OK ) break; + assert( i - sLoc.iZero - 1 >=0 ); + pgno = sLoc.aPgno[i-sLoc.iZero-1]; + iDbOff = (i64)(pgno-1) * szPage; + + if( iDbOff+szPage<=szDb ){ + iWalOff = walFrameOffset(i, szPage) + WAL_FRAME_HDRSIZE; + rc = sqlite3OsRead(pWal->pWalFd, pBuf1, szPage, iWalOff); + + if( rc==SQLITE_OK ){ + rc = sqlite3OsRead(pWal->pDbFd, pBuf2, szPage, iDbOff); + } + + if( rc!=SQLITE_OK || 0==memcmp(pBuf1, pBuf2, szPage) ){ + break; + } + } + + pInfo->nBackfillAttempted = i-1; + } + } + + sqlite3_free(pBuf1); + sqlite3_free(pBuf2); + } + walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1); + } + + return rc; +} +#endif /* SQLITE_ENABLE_SNAPSHOT */ + +/* +** Begin a read transaction on the database. +** +** This routine used to be called sqlite3OpenSnapshot() and with good reason: +** it takes a snapshot of the state of the WAL and wal-index for the current +** instant in time. The current thread will continue to use this snapshot. +** Other threads might append new content to the WAL and wal-index but +** that extra content is ignored by the current thread. +** +** If the database contents have changes since the previous read +** transaction, then *pChanged is set to 1 before returning. The +** Pager layer will use this to know that its cache is stale and +** needs to be flushed. +*/ +int sqlite3WalBeginReadTransaction(Wal *pWal, int *pChanged){ + int rc; /* Return code */ + int cnt = 0; /* Number of TryBeginRead attempts */ +#ifdef SQLITE_ENABLE_SNAPSHOT + int bChanged = 0; + WalIndexHdr *pSnapshot = pWal->pSnapshot; +#endif + + assert( pWal->ckptLock==0 ); + +#ifdef SQLITE_ENABLE_SNAPSHOT + if( pSnapshot ){ + if( memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){ + bChanged = 1; + } + + /* It is possible that there is a checkpointer thread running + ** concurrent with this code. If this is the case, it may be that the + ** checkpointer has already determined that it will checkpoint + ** snapshot X, where X is later in the wal file than pSnapshot, but + ** has not yet set the pInfo->nBackfillAttempted variable to indicate + ** its intent. To avoid the race condition this leads to, ensure that + ** there is no checkpointer process by taking a shared CKPT lock + ** before checking pInfo->nBackfillAttempted. */ + (void)walEnableBlocking(pWal); + rc = walLockShared(pWal, WAL_CKPT_LOCK); + walDisableBlocking(pWal); + + if( rc!=SQLITE_OK ){ + return rc; + } + pWal->ckptLock = 1; + } +#endif + + do{ + rc = walTryBeginRead(pWal, pChanged, 0, ++cnt); + }while( rc==WAL_RETRY ); + testcase( (rc&0xff)==SQLITE_BUSY ); + testcase( (rc&0xff)==SQLITE_IOERR ); + testcase( rc==SQLITE_PROTOCOL ); + testcase( rc==SQLITE_OK ); + +#ifdef SQLITE_ENABLE_SNAPSHOT + if( rc==SQLITE_OK ){ + if( pSnapshot && memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){ + /* At this point the client has a lock on an aReadMark[] slot holding + ** a value equal to or smaller than pSnapshot->mxFrame, but pWal->hdr + ** is populated with the wal-index header corresponding to the head + ** of the wal file. Verify that pSnapshot is still valid before + ** continuing. Reasons why pSnapshot might no longer be valid: + ** + ** (1) The WAL file has been reset since the snapshot was taken. + ** In this case, the salt will have changed. + ** + ** (2) A checkpoint as been attempted that wrote frames past + ** pSnapshot->mxFrame into the database file. Note that the + ** checkpoint need not have completed for this to cause problems. + */ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + + assert( pWal->readLock>0 || pWal->hdr.mxFrame==0 ); + assert( pInfo->aReadMark[pWal->readLock]<=pSnapshot->mxFrame ); + + /* Check that the wal file has not been wrapped. Assuming that it has + ** not, also check that no checkpointer has attempted to checkpoint any + ** frames beyond pSnapshot->mxFrame. If either of these conditions are + ** true, return SQLITE_ERROR_SNAPSHOT. Otherwise, overwrite pWal->hdr + ** with *pSnapshot and set *pChanged as appropriate for opening the + ** snapshot. */ + if( !memcmp(pSnapshot->aSalt, pWal->hdr.aSalt, sizeof(pWal->hdr.aSalt)) + && pSnapshot->mxFrame>=pInfo->nBackfillAttempted + ){ + assert( pWal->readLock>0 ); + memcpy(&pWal->hdr, pSnapshot, sizeof(WalIndexHdr)); + *pChanged = bChanged; + }else{ + rc = SQLITE_ERROR_SNAPSHOT; + } + + /* A client using a non-current snapshot may not ignore any frames + ** from the start of the wal file. This is because, for a system + ** where (minFrame < iSnapshot < maxFrame), a checkpointer may + ** have omitted to checkpoint a frame earlier than minFrame in + ** the file because there exists a frame after iSnapshot that + ** is the same database page. */ + pWal->minFrame = 1; + + if( rc!=SQLITE_OK ){ + sqlite3WalEndReadTransaction(pWal); + } + } + } + + /* Release the shared CKPT lock obtained above. */ + if( pWal->ckptLock ){ + assert( pSnapshot ); + walUnlockShared(pWal, WAL_CKPT_LOCK); + pWal->ckptLock = 0; + } +#endif + return rc; +} + +/* +** Finish with a read transaction. All this does is release the +** read-lock. +*/ +void sqlite3WalEndReadTransaction(Wal *pWal){ + sqlite3WalEndWriteTransaction(pWal); + if( pWal->readLock>=0 ){ + walUnlockShared(pWal, WAL_READ_LOCK(pWal->readLock)); + pWal->readLock = -1; + } +} + +/* +** Search the wal file for page pgno. If found, set *piRead to the frame that +** contains the page. Otherwise, if pgno is not in the wal file, set *piRead +** to zero. +** +** Return SQLITE_OK if successful, or an error code if an error occurs. If an +** error does occur, the final value of *piRead is undefined. +*/ +int sqlite3WalFindFrame( + Wal *pWal, /* WAL handle */ + Pgno pgno, /* Database page number to read data for */ + u32 *piRead /* OUT: Frame number (or zero) */ +){ + u32 iRead = 0; /* If !=0, WAL frame to return data from */ + u32 iLast = pWal->hdr.mxFrame; /* Last page in WAL for this reader */ + int iHash; /* Used to loop through N hash tables */ + int iMinHash; + + /* This routine is only be called from within a read transaction. */ + assert( pWal->readLock>=0 || pWal->lockError ); + + /* If the "last page" field of the wal-index header snapshot is 0, then + ** no data will be read from the wal under any circumstances. Return early + ** in this case as an optimization. Likewise, if pWal->readLock==0, + ** then the WAL is ignored by the reader so return early, as if the + ** WAL were empty. + */ + if( iLast==0 || (pWal->readLock==0 && pWal->bShmUnreliable==0) ){ + *piRead = 0; + return SQLITE_OK; + } + + /* Search the hash table or tables for an entry matching page number + ** pgno. Each iteration of the following for() loop searches one + ** hash table (each hash table indexes up to HASHTABLE_NPAGE frames). + ** + ** This code might run concurrently to the code in walIndexAppend() + ** that adds entries to the wal-index (and possibly to this hash + ** table). This means the value just read from the hash + ** slot (aHash[iKey]) may have been added before or after the + ** current read transaction was opened. Values added after the + ** read transaction was opened may have been written incorrectly - + ** i.e. these slots may contain garbage data. However, we assume + ** that any slots written before the current read transaction was + ** opened remain unmodified. + ** + ** For the reasons above, the if(...) condition featured in the inner + ** loop of the following block is more stringent that would be required + ** if we had exclusive access to the hash-table: + ** + ** (aPgno[iFrame]==pgno): + ** This condition filters out normal hash-table collisions. + ** + ** (iFrame<=iLast): + ** This condition filters out entries that were added to the hash + ** table after the current read-transaction had started. + */ + iMinHash = walFramePage(pWal->minFrame); + for(iHash=walFramePage(iLast); iHash>=iMinHash; iHash--){ + WalHashLoc sLoc; /* Hash table location */ + int iKey; /* Hash slot index */ + int nCollide; /* Number of hash collisions remaining */ + int rc; /* Error code */ + u32 iH; + + rc = walHashGet(pWal, iHash, &sLoc); + if( rc!=SQLITE_OK ){ + return rc; + } + nCollide = HASHTABLE_NSLOT; + iKey = walHash(pgno); + while( (iH = AtomicLoad(&sLoc.aHash[iKey]))!=0 ){ + u32 iFrame = iH + sLoc.iZero; + if( iFrame<=iLast && iFrame>=pWal->minFrame && sLoc.aPgno[iH-1]==pgno ){ + assert( iFrame>iRead || CORRUPT_DB ); + iRead = iFrame; + } + if( (nCollide--)==0 ){ + return SQLITE_CORRUPT_BKPT; + } + iKey = walNextHash(iKey); + } + if( iRead ) break; + } + +#ifdef SQLITE_ENABLE_EXPENSIVE_ASSERT + /* If expensive assert() statements are available, do a linear search + ** of the wal-index file content. Make sure the results agree with the + ** result obtained using the hash indexes above. */ + { + u32 iRead2 = 0; + u32 iTest; + assert( pWal->bShmUnreliable || pWal->minFrame>0 ); + for(iTest=iLast; iTest>=pWal->minFrame && iTest>0; iTest--){ + if( walFramePgno(pWal, iTest)==pgno ){ + iRead2 = iTest; + break; + } + } + assert( iRead==iRead2 ); + } +#endif + + *piRead = iRead; + return SQLITE_OK; +} + +/* +** Read the contents of frame iRead from the wal file into buffer pOut +** (which is nOut bytes in size). Return SQLITE_OK if successful, or an +** error code otherwise. +*/ +int sqlite3WalReadFrame( + Wal *pWal, /* WAL handle */ + u32 iRead, /* Frame to read */ + int nOut, /* Size of buffer pOut in bytes */ + u8 *pOut /* Buffer to write page data to */ +){ + int sz; + i64 iOffset; + sz = pWal->hdr.szPage; + sz = (sz&0xfe00) + ((sz&0x0001)<<16); + testcase( sz<=32768 ); + testcase( sz>=65536 ); + iOffset = walFrameOffset(iRead, sz) + WAL_FRAME_HDRSIZE; + /* testcase( IS_BIG_INT(iOffset) ); // requires a 4GiB WAL */ + return sqlite3OsRead(pWal->pWalFd, pOut, (nOut>sz ? sz : nOut), iOffset); +} + +/* +** Return the size of the database in pages (or zero, if unknown). +*/ +Pgno sqlite3WalDbsize(Wal *pWal){ + if( pWal && ALWAYS(pWal->readLock>=0) ){ + return pWal->hdr.nPage; + } + return 0; +} + + +/* +** This function starts a write transaction on the WAL. +** +** A read transaction must have already been started by a prior call +** to sqlite3WalBeginReadTransaction(). +** +** If another thread or process has written into the database since +** the read transaction was started, then it is not possible for this +** thread to write as doing so would cause a fork. So this routine +** returns SQLITE_BUSY in that case and no write transaction is started. +** +** There can only be a single writer active at a time. +*/ +int sqlite3WalBeginWriteTransaction(Wal *pWal){ + int rc; + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + /* If the write-lock is already held, then it was obtained before the + ** read-transaction was even opened, making this call a no-op. + ** Return early. */ + if( pWal->writeLock ){ + assert( !memcmp(&pWal->hdr,(void *)walIndexHdr(pWal),sizeof(WalIndexHdr)) ); + return SQLITE_OK; + } +#endif + + /* Cannot start a write transaction without first holding a read + ** transaction. */ + assert( pWal->readLock>=0 ); + assert( pWal->writeLock==0 && pWal->iReCksum==0 ); + + if( pWal->readOnly ){ + return SQLITE_READONLY; + } + + /* Only one writer allowed at a time. Get the write lock. Return + ** SQLITE_BUSY if unable. + */ + rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1); + if( rc ){ + return rc; + } + pWal->writeLock = 1; + + /* If another connection has written to the database file since the + ** time the read transaction on this connection was started, then + ** the write is disallowed. + */ + if( memcmp(&pWal->hdr, (void *)walIndexHdr(pWal), sizeof(WalIndexHdr))!=0 ){ + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + pWal->writeLock = 0; + rc = SQLITE_BUSY_SNAPSHOT; + } + + return rc; +} + +/* +** End a write transaction. The commit has already been done. This +** routine merely releases the lock. +*/ +int sqlite3WalEndWriteTransaction(Wal *pWal){ + if( pWal->writeLock ){ + walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); + pWal->writeLock = 0; + pWal->iReCksum = 0; + pWal->truncateOnCommit = 0; + } + return SQLITE_OK; +} + +/* +** If any data has been written (but not committed) to the log file, this +** function moves the write-pointer back to the start of the transaction. +** +** Additionally, the callback function is invoked for each frame written +** to the WAL since the start of the transaction. If the callback returns +** other than SQLITE_OK, it is not invoked again and the error code is +** returned to the caller. +** +** Otherwise, if the callback function does not return an error, this +** function returns SQLITE_OK. +*/ +int sqlite3WalUndo(Wal *pWal, int (*xUndo)(void *, Pgno), void *pUndoCtx){ + int rc = SQLITE_OK; + if( ALWAYS(pWal->writeLock) ){ + Pgno iMax = pWal->hdr.mxFrame; + Pgno iFrame; + + /* Restore the clients cache of the wal-index header to the state it + ** was in before the client began writing to the database. + */ + memcpy(&pWal->hdr, (void *)walIndexHdr(pWal), sizeof(WalIndexHdr)); + + for(iFrame=pWal->hdr.mxFrame+1; + ALWAYS(rc==SQLITE_OK) && iFrame<=iMax; + iFrame++ + ){ + /* This call cannot fail. Unless the page for which the page number + ** is passed as the second argument is (a) in the cache and + ** (b) has an outstanding reference, then xUndo is either a no-op + ** (if (a) is false) or simply expels the page from the cache (if (b) + ** is false). + ** + ** If the upper layer is doing a rollback, it is guaranteed that there + ** are no outstanding references to any page other than page 1. And + ** page 1 is never written to the log until the transaction is + ** committed. As a result, the call to xUndo may not fail. + */ + assert( walFramePgno(pWal, iFrame)!=1 ); + rc = xUndo(pUndoCtx, walFramePgno(pWal, iFrame)); + } + if( iMax!=pWal->hdr.mxFrame ) walCleanupHash(pWal); + } + return rc; +} + +/* +** Argument aWalData must point to an array of WAL_SAVEPOINT_NDATA u32 +** values. This function populates the array with values required to +** "rollback" the write position of the WAL handle back to the current +** point in the event of a savepoint rollback (via WalSavepointUndo()). +*/ +void sqlite3WalSavepoint(Wal *pWal, u32 *aWalData){ + assert( pWal->writeLock ); + aWalData[0] = pWal->hdr.mxFrame; + aWalData[1] = pWal->hdr.aFrameCksum[0]; + aWalData[2] = pWal->hdr.aFrameCksum[1]; + aWalData[3] = pWal->nCkpt; +} + +/* +** Move the write position of the WAL back to the point identified by +** the values in the aWalData[] array. aWalData must point to an array +** of WAL_SAVEPOINT_NDATA u32 values that has been previously populated +** by a call to WalSavepoint(). +*/ +int sqlite3WalSavepointUndo(Wal *pWal, u32 *aWalData){ + int rc = SQLITE_OK; + + assert( pWal->writeLock ); + assert( aWalData[3]!=pWal->nCkpt || aWalData[0]<=pWal->hdr.mxFrame ); + + if( aWalData[3]!=pWal->nCkpt ){ + /* This savepoint was opened immediately after the write-transaction + ** was started. Right after that, the writer decided to wrap around + ** to the start of the log. Update the savepoint values to match. + */ + aWalData[0] = 0; + aWalData[3] = pWal->nCkpt; + } + + if( aWalData[0]hdr.mxFrame ){ + pWal->hdr.mxFrame = aWalData[0]; + pWal->hdr.aFrameCksum[0] = aWalData[1]; + pWal->hdr.aFrameCksum[1] = aWalData[2]; + walCleanupHash(pWal); + } + + return rc; +} + +/* +** This function is called just before writing a set of frames to the log +** file (see sqlite3WalFrames()). It checks to see if, instead of appending +** to the current log file, it is possible to overwrite the start of the +** existing log file with the new frames (i.e. "reset" the log). If so, +** it sets pWal->hdr.mxFrame to 0. Otherwise, pWal->hdr.mxFrame is left +** unchanged. +** +** SQLITE_OK is returned if no error is encountered (regardless of whether +** or not pWal->hdr.mxFrame is modified). An SQLite error code is returned +** if an error occurs. +*/ +static int walRestartLog(Wal *pWal){ + int rc = SQLITE_OK; + int cnt; + + if( pWal->readLock==0 ){ + volatile WalCkptInfo *pInfo = walCkptInfo(pWal); + assert( pInfo->nBackfill==pWal->hdr.mxFrame ); + if( pInfo->nBackfill>0 ){ + u32 salt1; + sqlite3_randomness(4, &salt1); + rc = walLockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + if( rc==SQLITE_OK ){ + /* If all readers are using WAL_READ_LOCK(0) (in other words if no + ** readers are currently using the WAL), then the transactions + ** frames will overwrite the start of the existing log. Update the + ** wal-index header to reflect this. + ** + ** In theory it would be Ok to update the cache of the header only + ** at this point. But updating the actual wal-index header is also + ** safe and means there is no special case for sqlite3WalUndo() + ** to handle if this transaction is rolled back. */ + walRestartHdr(pWal, salt1); + walUnlockExclusive(pWal, WAL_READ_LOCK(1), WAL_NREADER-1); + }else if( rc!=SQLITE_BUSY ){ + return rc; + } + } + walUnlockShared(pWal, WAL_READ_LOCK(0)); + pWal->readLock = -1; + cnt = 0; + do{ + int notUsed; + rc = walTryBeginRead(pWal, ¬Used, 1, ++cnt); + }while( rc==WAL_RETRY ); + assert( (rc&0xff)!=SQLITE_BUSY ); /* BUSY not possible when useWal==1 */ + testcase( (rc&0xff)==SQLITE_IOERR ); + testcase( rc==SQLITE_PROTOCOL ); + testcase( rc==SQLITE_OK ); + } + return rc; +} + +/* +** Information about the current state of the WAL file and where +** the next fsync should occur - passed from sqlite3WalFrames() into +** walWriteToLog(). +*/ +typedef struct WalWriter { + Wal *pWal; /* The complete WAL information */ + sqlite3_file *pFd; /* The WAL file to which we write */ + sqlite3_int64 iSyncPoint; /* Fsync at this offset */ + int syncFlags; /* Flags for the fsync */ + int szPage; /* Size of one page */ +} WalWriter; + +/* +** Write iAmt bytes of content into the WAL file beginning at iOffset. +** Do a sync when crossing the p->iSyncPoint boundary. +** +** In other words, if iSyncPoint is in between iOffset and iOffset+iAmt, +** first write the part before iSyncPoint, then sync, then write the +** rest. +*/ +static int walWriteToLog( + WalWriter *p, /* WAL to write to */ + void *pContent, /* Content to be written */ + int iAmt, /* Number of bytes to write */ + sqlite3_int64 iOffset /* Start writing at this offset */ +){ + int rc; + if( iOffsetiSyncPoint && iOffset+iAmt>=p->iSyncPoint ){ + int iFirstAmt = (int)(p->iSyncPoint - iOffset); + rc = sqlite3OsWrite(p->pFd, pContent, iFirstAmt, iOffset); + if( rc ) return rc; + iOffset += iFirstAmt; + iAmt -= iFirstAmt; + pContent = (void*)(iFirstAmt + (char*)pContent); + assert( WAL_SYNC_FLAGS(p->syncFlags)!=0 ); + rc = sqlite3OsSync(p->pFd, WAL_SYNC_FLAGS(p->syncFlags)); + if( iAmt==0 || rc ) return rc; + } + rc = sqlite3OsWrite(p->pFd, pContent, iAmt, iOffset); + return rc; +} + +/* +** Write out a single frame of the WAL +*/ +static int walWriteOneFrame( + WalWriter *p, /* Where to write the frame */ + PgHdr *pPage, /* The page of the frame to be written */ + int nTruncate, /* The commit flag. Usually 0. >0 for commit */ + sqlite3_int64 iOffset /* Byte offset at which to write */ +){ + int rc; /* Result code from subfunctions */ + void *pData; /* Data actually written */ + u8 aFrame[WAL_FRAME_HDRSIZE]; /* Buffer to assemble frame-header in */ + pData = pPage->pData; + walEncodeFrame(p->pWal, pPage->pgno, nTruncate, pData, aFrame); + rc = walWriteToLog(p, aFrame, sizeof(aFrame), iOffset); + if( rc ) return rc; + /* Write the page data */ + rc = walWriteToLog(p, pData, p->szPage, iOffset+sizeof(aFrame)); + return rc; +} + +/* +** This function is called as part of committing a transaction within which +** one or more frames have been overwritten. It updates the checksums for +** all frames written to the wal file by the current transaction starting +** with the earliest to have been overwritten. +** +** SQLITE_OK is returned if successful, or an SQLite error code otherwise. +*/ +static int walRewriteChecksums(Wal *pWal, u32 iLast){ + const int szPage = pWal->szPage;/* Database page size */ + int rc = SQLITE_OK; /* Return code */ + u8 *aBuf; /* Buffer to load data from wal file into */ + u8 aFrame[WAL_FRAME_HDRSIZE]; /* Buffer to assemble frame-headers in */ + u32 iRead; /* Next frame to read from wal file */ + i64 iCksumOff; + + aBuf = sqlite3_malloc(szPage + WAL_FRAME_HDRSIZE); + if( aBuf==0 ) return SQLITE_NOMEM_BKPT; + + /* Find the checksum values to use as input for the recalculating the + ** first checksum. If the first frame is frame 1 (implying that the current + ** transaction restarted the wal file), these values must be read from the + ** wal-file header. Otherwise, read them from the frame header of the + ** previous frame. */ + assert( pWal->iReCksum>0 ); + if( pWal->iReCksum==1 ){ + iCksumOff = 24; + }else{ + iCksumOff = walFrameOffset(pWal->iReCksum-1, szPage) + 16; + } + rc = sqlite3OsRead(pWal->pWalFd, aBuf, sizeof(u32)*2, iCksumOff); + pWal->hdr.aFrameCksum[0] = sqlite3Get4byte(aBuf); + pWal->hdr.aFrameCksum[1] = sqlite3Get4byte(&aBuf[sizeof(u32)]); + + iRead = pWal->iReCksum; + pWal->iReCksum = 0; + for(; rc==SQLITE_OK && iRead<=iLast; iRead++){ + i64 iOff = walFrameOffset(iRead, szPage); + rc = sqlite3OsRead(pWal->pWalFd, aBuf, szPage+WAL_FRAME_HDRSIZE, iOff); + if( rc==SQLITE_OK ){ + u32 iPgno, nDbSize; + iPgno = sqlite3Get4byte(aBuf); + nDbSize = sqlite3Get4byte(&aBuf[4]); + + walEncodeFrame(pWal, iPgno, nDbSize, &aBuf[WAL_FRAME_HDRSIZE], aFrame); + rc = sqlite3OsWrite(pWal->pWalFd, aFrame, sizeof(aFrame), iOff); + } + } + + sqlite3_free(aBuf); + return rc; +} + +/* +** Write a set of frames to the log. The caller must hold the write-lock +** on the log file (obtained using sqlite3WalBeginWriteTransaction()). +*/ +int sqlite3WalFrames( + Wal *pWal, /* Wal handle to write to */ + int szPage, /* Database page-size in bytes */ + PgHdr *pList, /* List of dirty pages to write */ + Pgno nTruncate, /* Database size after this commit */ + int isCommit, /* True if this is a commit */ + int sync_flags /* Flags to pass to OsSync() (or 0) */ +){ + int rc; /* Used to catch return codes */ + u32 iFrame; /* Next frame address */ + PgHdr *p; /* Iterator to run through pList with. */ + PgHdr *pLast = 0; /* Last frame in list */ + int nExtra = 0; /* Number of extra copies of last page */ + int szFrame; /* The size of a single frame */ + i64 iOffset; /* Next byte to write in WAL file */ + WalWriter w; /* The writer */ + u32 iFirst = 0; /* First frame that may be overwritten */ + WalIndexHdr *pLive; /* Pointer to shared header */ + + assert( pList ); + assert( pWal->writeLock ); + + /* If this frame set completes a transaction, then nTruncate>0. If + ** nTruncate==0 then this frame set does not complete the transaction. */ + assert( (isCommit!=0)==(nTruncate!=0) ); + +#if defined(SQLITE_TEST) && defined(SQLITE_DEBUG) + { int cnt; for(cnt=0, p=pList; p; p=p->pDirty, cnt++){} + WALTRACE(("WAL%p: frame write begin. %d frames. mxFrame=%d. %s\n", + pWal, cnt, pWal->hdr.mxFrame, isCommit ? "Commit" : "Spill")); + } +#endif + + pLive = (WalIndexHdr*)walIndexHdr(pWal); + if( memcmp(&pWal->hdr, (void *)pLive, sizeof(WalIndexHdr))!=0 ){ + iFirst = pLive->mxFrame+1; + } + + /* See if it is possible to write these frames into the start of the + ** log file, instead of appending to it at pWal->hdr.mxFrame. + */ + if( SQLITE_OK!=(rc = walRestartLog(pWal)) ){ + return rc; + } + + /* If this is the first frame written into the log, write the WAL + ** header to the start of the WAL file. See comments at the top of + ** this source file for a description of the WAL header format. + */ + iFrame = pWal->hdr.mxFrame; + if( iFrame==0 ){ + u8 aWalHdr[WAL_HDRSIZE]; /* Buffer to assemble wal-header in */ + u32 aCksum[2]; /* Checksum for wal-header */ + + sqlite3Put4byte(&aWalHdr[0], (WAL_MAGIC | SQLITE_BIGENDIAN)); + sqlite3Put4byte(&aWalHdr[4], WAL_MAX_VERSION); + sqlite3Put4byte(&aWalHdr[8], szPage); + sqlite3Put4byte(&aWalHdr[12], pWal->nCkpt); + if( pWal->nCkpt==0 ) sqlite3_randomness(8, pWal->hdr.aSalt); + memcpy(&aWalHdr[16], pWal->hdr.aSalt, 8); + walChecksumBytes(1, aWalHdr, WAL_HDRSIZE-2*4, 0, aCksum); + sqlite3Put4byte(&aWalHdr[24], aCksum[0]); + sqlite3Put4byte(&aWalHdr[28], aCksum[1]); + + pWal->szPage = szPage; + pWal->hdr.bigEndCksum = SQLITE_BIGENDIAN; + pWal->hdr.aFrameCksum[0] = aCksum[0]; + pWal->hdr.aFrameCksum[1] = aCksum[1]; + pWal->truncateOnCommit = 1; + + rc = sqlite3OsWrite(pWal->pWalFd, aWalHdr, sizeof(aWalHdr), 0); + WALTRACE(("WAL%p: wal-header write %s\n", pWal, rc ? "failed" : "ok")); + if( rc!=SQLITE_OK ){ + return rc; + } + + /* Sync the header (unless SQLITE_IOCAP_SEQUENTIAL is true or unless + ** all syncing is turned off by PRAGMA synchronous=OFF). Otherwise + ** an out-of-order write following a WAL restart could result in + ** database corruption. See the ticket: + ** + ** https://sqlite.org/src/info/ff5be73dee + */ + if( pWal->syncHeader ){ + rc = sqlite3OsSync(pWal->pWalFd, CKPT_SYNC_FLAGS(sync_flags)); + if( rc ) return rc; + } + } + assert( (int)pWal->szPage==szPage ); + + /* Setup information needed to write frames into the WAL */ + w.pWal = pWal; + w.pFd = pWal->pWalFd; + w.iSyncPoint = 0; + w.syncFlags = sync_flags; + w.szPage = szPage; + iOffset = walFrameOffset(iFrame+1, szPage); + szFrame = szPage + WAL_FRAME_HDRSIZE; + + /* Write all frames into the log file exactly once */ + for(p=pList; p; p=p->pDirty){ + int nDbSize; /* 0 normally. Positive == commit flag */ + + /* Check if this page has already been written into the wal file by + ** the current transaction. If so, overwrite the existing frame and + ** set Wal.writeLock to WAL_WRITELOCK_RECKSUM - indicating that + ** checksums must be recomputed when the transaction is committed. */ + if( iFirst && (p->pDirty || isCommit==0) ){ + u32 iWrite = 0; + VVA_ONLY(rc =) sqlite3WalFindFrame(pWal, p->pgno, &iWrite); + assert( rc==SQLITE_OK || iWrite==0 ); + if( iWrite>=iFirst ){ + i64 iOff = walFrameOffset(iWrite, szPage) + WAL_FRAME_HDRSIZE; + void *pData; + if( pWal->iReCksum==0 || iWriteiReCksum ){ + pWal->iReCksum = iWrite; + } + pData = p->pData; + rc = sqlite3OsWrite(pWal->pWalFd, pData, szPage, iOff); + if( rc ) return rc; + p->flags &= ~PGHDR_WAL_APPEND; + continue; + } + } + + iFrame++; + assert( iOffset==walFrameOffset(iFrame, szPage) ); + nDbSize = (isCommit && p->pDirty==0) ? nTruncate : 0; + rc = walWriteOneFrame(&w, p, nDbSize, iOffset); + if( rc ) return rc; + pLast = p; + iOffset += szFrame; + p->flags |= PGHDR_WAL_APPEND; + } + + /* Recalculate checksums within the wal file if required. */ + if( isCommit && pWal->iReCksum ){ + rc = walRewriteChecksums(pWal, iFrame); + if( rc ) return rc; + } + + /* If this is the end of a transaction, then we might need to pad + ** the transaction and/or sync the WAL file. + ** + ** Padding and syncing only occur if this set of frames complete a + ** transaction and if PRAGMA synchronous=FULL. If synchronous==NORMAL + ** or synchronous==OFF, then no padding or syncing are needed. + ** + ** If SQLITE_IOCAP_POWERSAFE_OVERWRITE is defined, then padding is not + ** needed and only the sync is done. If padding is needed, then the + ** final frame is repeated (with its commit mark) until the next sector + ** boundary is crossed. Only the part of the WAL prior to the last + ** sector boundary is synced; the part of the last frame that extends + ** past the sector boundary is written after the sync. + */ + if( isCommit && WAL_SYNC_FLAGS(sync_flags)!=0 ){ + int bSync = 1; + if( pWal->padToSectorBoundary ){ + int sectorSize = sqlite3SectorSize(pWal->pWalFd); + w.iSyncPoint = ((iOffset+sectorSize-1)/sectorSize)*sectorSize; + bSync = (w.iSyncPoint==iOffset); + testcase( bSync ); + while( iOffsettruncateOnCommit && pWal->mxWalSize>=0 ){ + i64 sz = pWal->mxWalSize; + if( walFrameOffset(iFrame+nExtra+1, szPage)>pWal->mxWalSize ){ + sz = walFrameOffset(iFrame+nExtra+1, szPage); + } + walLimitSize(pWal, sz); + pWal->truncateOnCommit = 0; + } + + /* Append data to the wal-index. It is not necessary to lock the + ** wal-index to do this as the SQLITE_SHM_WRITE lock held on the wal-index + ** guarantees that there are no other writers, and no data that may + ** be in use by existing readers is being overwritten. + */ + iFrame = pWal->hdr.mxFrame; + for(p=pList; p && rc==SQLITE_OK; p=p->pDirty){ + if( (p->flags & PGHDR_WAL_APPEND)==0 ) continue; + iFrame++; + rc = walIndexAppend(pWal, iFrame, p->pgno); + } + assert( pLast!=0 || nExtra==0 ); + while( rc==SQLITE_OK && nExtra>0 ){ + iFrame++; + nExtra--; + rc = walIndexAppend(pWal, iFrame, pLast->pgno); + } + + if( rc==SQLITE_OK ){ + /* Update the private copy of the header. */ + pWal->hdr.szPage = (u16)((szPage&0xff00) | (szPage>>16)); + testcase( szPage<=32768 ); + testcase( szPage>=65536 ); + pWal->hdr.mxFrame = iFrame; + if( isCommit ){ + pWal->hdr.iChange++; + pWal->hdr.nPage = nTruncate; + } + /* If this is a commit, update the wal-index header too. */ + if( isCommit ){ + walIndexWriteHdr(pWal); + pWal->iCallback = iFrame; + } + } + + WALTRACE(("WAL%p: frame write %s\n", pWal, rc ? "failed" : "ok")); + return rc; +} + +/* +** This routine is called to implement sqlite3_wal_checkpoint() and +** related interfaces. +** +** Obtain a CHECKPOINT lock and then backfill as much information as +** we can from WAL into the database. +** +** If parameter xBusy is not NULL, it is a pointer to a busy-handler +** callback. In this case this function runs a blocking checkpoint. +*/ +int sqlite3WalCheckpoint( + Wal *pWal, /* Wal connection */ + sqlite3 *db, /* Check this handle's interrupt flag */ + int eMode, /* PASSIVE, FULL, RESTART, or TRUNCATE */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int sync_flags, /* Flags to sync db file with (or 0) */ + int nBuf, /* Size of temporary buffer */ + u8 *zBuf, /* Temporary buffer to use */ + int *pnLog, /* OUT: Number of frames in WAL */ + int *pnCkpt /* OUT: Number of backfilled frames in WAL */ +){ + int rc; /* Return code */ + int isChanged = 0; /* True if a new wal-index header is loaded */ + int eMode2 = eMode; /* Mode to pass to walCheckpoint() */ + int (*xBusy2)(void*) = xBusy; /* Busy handler for eMode2 */ + + assert( pWal->ckptLock==0 ); + assert( pWal->writeLock==0 ); + + /* EVIDENCE-OF: R-62920-47450 The busy-handler callback is never invoked + ** in the SQLITE_CHECKPOINT_PASSIVE mode. */ + assert( eMode!=SQLITE_CHECKPOINT_PASSIVE || xBusy==0 ); + + if( pWal->readOnly ) return SQLITE_READONLY; + WALTRACE(("WAL%p: checkpoint begins\n", pWal)); + + /* Enable blocking locks, if possible. If blocking locks are successfully + ** enabled, set xBusy2=0 so that the busy-handler is never invoked. */ + sqlite3WalDb(pWal, db); + (void)walEnableBlocking(pWal); + + /* IMPLEMENTATION-OF: R-62028-47212 All calls obtain an exclusive + ** "checkpoint" lock on the database file. + ** EVIDENCE-OF: R-10421-19736 If any other process is running a + ** checkpoint operation at the same time, the lock cannot be obtained and + ** SQLITE_BUSY is returned. + ** EVIDENCE-OF: R-53820-33897 Even if there is a busy-handler configured, + ** it will not be invoked in this case. + */ + rc = walLockExclusive(pWal, WAL_CKPT_LOCK, 1); + testcase( rc==SQLITE_BUSY ); + testcase( rc!=SQLITE_OK && xBusy2!=0 ); + if( rc==SQLITE_OK ){ + pWal->ckptLock = 1; + + /* IMPLEMENTATION-OF: R-59782-36818 The SQLITE_CHECKPOINT_FULL, RESTART and + ** TRUNCATE modes also obtain the exclusive "writer" lock on the database + ** file. + ** + ** EVIDENCE-OF: R-60642-04082 If the writer lock cannot be obtained + ** immediately, and a busy-handler is configured, it is invoked and the + ** writer lock retried until either the busy-handler returns 0 or the + ** lock is successfully obtained. + */ + if( eMode!=SQLITE_CHECKPOINT_PASSIVE ){ + rc = walBusyLock(pWal, xBusy2, pBusyArg, WAL_WRITE_LOCK, 1); + if( rc==SQLITE_OK ){ + pWal->writeLock = 1; + }else if( rc==SQLITE_BUSY ){ + eMode2 = SQLITE_CHECKPOINT_PASSIVE; + xBusy2 = 0; + rc = SQLITE_OK; + } + } + } + + + /* Read the wal-index header. */ + if( rc==SQLITE_OK ){ + walDisableBlocking(pWal); + rc = walIndexReadHdr(pWal, &isChanged); + (void)walEnableBlocking(pWal); + if( isChanged && pWal->pDbFd->pMethods->iVersion>=3 ){ + sqlite3OsUnfetch(pWal->pDbFd, 0, 0); + } + } + + /* Copy data from the log to the database file. */ + if( rc==SQLITE_OK ){ + + if( pWal->hdr.mxFrame && walPagesize(pWal)!=nBuf ){ + rc = SQLITE_CORRUPT_BKPT; + }else{ + rc = walCheckpoint(pWal, db, eMode2, xBusy2, pBusyArg, sync_flags, zBuf); + } + + /* If no error occurred, set the output variables. */ + if( rc==SQLITE_OK || rc==SQLITE_BUSY ){ + if( pnLog ) *pnLog = (int)pWal->hdr.mxFrame; + if( pnCkpt ) *pnCkpt = (int)(walCkptInfo(pWal)->nBackfill); + } + } + + if( isChanged ){ + /* If a new wal-index header was loaded before the checkpoint was + ** performed, then the pager-cache associated with pWal is now + ** out of date. So zero the cached wal-index header to ensure that + ** next time the pager opens a snapshot on this database it knows that + ** the cache needs to be reset. + */ + memset(&pWal->hdr, 0, sizeof(WalIndexHdr)); + } + + walDisableBlocking(pWal); + sqlite3WalDb(pWal, 0); + + /* Release the locks. */ + sqlite3WalEndWriteTransaction(pWal); + if( pWal->ckptLock ){ + walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1); + pWal->ckptLock = 0; + } + WALTRACE(("WAL%p: checkpoint %s\n", pWal, rc ? "failed" : "ok")); +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + if( rc==SQLITE_BUSY_TIMEOUT ) rc = SQLITE_BUSY; +#endif + return (rc==SQLITE_OK && eMode!=eMode2 ? SQLITE_BUSY : rc); +} + +/* Return the value to pass to a sqlite3_wal_hook callback, the +** number of frames in the WAL at the point of the last commit since +** sqlite3WalCallback() was called. If no commits have occurred since +** the last call, then return 0. +*/ +int sqlite3WalCallback(Wal *pWal){ + u32 ret = 0; + if( pWal ){ + ret = pWal->iCallback; + pWal->iCallback = 0; + } + return (int)ret; +} + +/* +** This function is called to change the WAL subsystem into or out +** of locking_mode=EXCLUSIVE. +** +** If op is zero, then attempt to change from locking_mode=EXCLUSIVE +** into locking_mode=NORMAL. This means that we must acquire a lock +** on the pWal->readLock byte. If the WAL is already in locking_mode=NORMAL +** or if the acquisition of the lock fails, then return 0. If the +** transition out of exclusive-mode is successful, return 1. This +** operation must occur while the pager is still holding the exclusive +** lock on the main database file. +** +** If op is one, then change from locking_mode=NORMAL into +** locking_mode=EXCLUSIVE. This means that the pWal->readLock must +** be released. Return 1 if the transition is made and 0 if the +** WAL is already in exclusive-locking mode - meaning that this +** routine is a no-op. The pager must already hold the exclusive lock +** on the main database file before invoking this operation. +** +** If op is negative, then do a dry-run of the op==1 case but do +** not actually change anything. The pager uses this to see if it +** should acquire the database exclusive lock prior to invoking +** the op==1 case. +*/ +int sqlite3WalExclusiveMode(Wal *pWal, int op){ + int rc; + assert( pWal->writeLock==0 ); + assert( pWal->exclusiveMode!=WAL_HEAPMEMORY_MODE || op==-1 ); + + /* pWal->readLock is usually set, but might be -1 if there was a + ** prior error while attempting to acquire are read-lock. This cannot + ** happen if the connection is actually in exclusive mode (as no xShmLock + ** locks are taken in this case). Nor should the pager attempt to + ** upgrade to exclusive-mode following such an error. + */ + assert( pWal->readLock>=0 || pWal->lockError ); + assert( pWal->readLock>=0 || (op<=0 && pWal->exclusiveMode==0) ); + + if( op==0 ){ + if( pWal->exclusiveMode!=WAL_NORMAL_MODE ){ + pWal->exclusiveMode = WAL_NORMAL_MODE; + if( walLockShared(pWal, WAL_READ_LOCK(pWal->readLock))!=SQLITE_OK ){ + pWal->exclusiveMode = WAL_EXCLUSIVE_MODE; + } + rc = pWal->exclusiveMode==WAL_NORMAL_MODE; + }else{ + /* Already in locking_mode=NORMAL */ + rc = 0; + } + }else if( op>0 ){ + assert( pWal->exclusiveMode==WAL_NORMAL_MODE ); + assert( pWal->readLock>=0 ); + walUnlockShared(pWal, WAL_READ_LOCK(pWal->readLock)); + pWal->exclusiveMode = WAL_EXCLUSIVE_MODE; + rc = 1; + }else{ + rc = pWal->exclusiveMode==WAL_NORMAL_MODE; + } + return rc; +} + +/* +** Return true if the argument is non-NULL and the WAL module is using +** heap-memory for the wal-index. Otherwise, if the argument is NULL or the +** WAL module is using shared-memory, return false. +*/ +int sqlite3WalHeapMemory(Wal *pWal){ + return (pWal && pWal->exclusiveMode==WAL_HEAPMEMORY_MODE ); +} + +#ifdef SQLITE_ENABLE_SNAPSHOT +/* Create a snapshot object. The content of a snapshot is opaque to +** every other subsystem, so the WAL module can put whatever it needs +** in the object. +*/ +int sqlite3WalSnapshotGet(Wal *pWal, sqlite3_snapshot **ppSnapshot){ + int rc = SQLITE_OK; + WalIndexHdr *pRet; + static const u32 aZero[4] = { 0, 0, 0, 0 }; + + assert( pWal->readLock>=0 && pWal->writeLock==0 ); + + if( memcmp(&pWal->hdr.aFrameCksum[0],aZero,16)==0 ){ + *ppSnapshot = 0; + return SQLITE_ERROR; + } + pRet = (WalIndexHdr*)sqlite3_malloc(sizeof(WalIndexHdr)); + if( pRet==0 ){ + rc = SQLITE_NOMEM_BKPT; + }else{ + memcpy(pRet, &pWal->hdr, sizeof(WalIndexHdr)); + *ppSnapshot = (sqlite3_snapshot*)pRet; + } + + return rc; +} + +/* Try to open on pSnapshot when the next read-transaction starts +*/ +void sqlite3WalSnapshotOpen( + Wal *pWal, + sqlite3_snapshot *pSnapshot +){ + pWal->pSnapshot = (WalIndexHdr*)pSnapshot; +} + +/* +** Return a +ve value if snapshot p1 is newer than p2. A -ve value if +** p1 is older than p2 and zero if p1 and p2 are the same snapshot. +*/ +int sqlite3_snapshot_cmp(sqlite3_snapshot *p1, sqlite3_snapshot *p2){ + WalIndexHdr *pHdr1 = (WalIndexHdr*)p1; + WalIndexHdr *pHdr2 = (WalIndexHdr*)p2; + + /* aSalt[0] is a copy of the value stored in the wal file header. It + ** is incremented each time the wal file is restarted. */ + if( pHdr1->aSalt[0]aSalt[0] ) return -1; + if( pHdr1->aSalt[0]>pHdr2->aSalt[0] ) return +1; + if( pHdr1->mxFramemxFrame ) return -1; + if( pHdr1->mxFrame>pHdr2->mxFrame ) return +1; + return 0; +} + +/* +** The caller currently has a read transaction open on the database. +** This function takes a SHARED lock on the CHECKPOINTER slot and then +** checks if the snapshot passed as the second argument is still +** available. If so, SQLITE_OK is returned. +** +** If the snapshot is not available, SQLITE_ERROR is returned. Or, if +** the CHECKPOINTER lock cannot be obtained, SQLITE_BUSY. If any error +** occurs (any value other than SQLITE_OK is returned), the CHECKPOINTER +** lock is released before returning. +*/ +int sqlite3WalSnapshotCheck(Wal *pWal, sqlite3_snapshot *pSnapshot){ + int rc; + rc = walLockShared(pWal, WAL_CKPT_LOCK); + if( rc==SQLITE_OK ){ + WalIndexHdr *pNew = (WalIndexHdr*)pSnapshot; + if( memcmp(pNew->aSalt, pWal->hdr.aSalt, sizeof(pWal->hdr.aSalt)) + || pNew->mxFramenBackfillAttempted + ){ + rc = SQLITE_ERROR_SNAPSHOT; + walUnlockShared(pWal, WAL_CKPT_LOCK); + } + } + return rc; +} + +/* +** Release a lock obtained by an earlier successful call to +** sqlite3WalSnapshotCheck(). +*/ +void sqlite3WalSnapshotUnlock(Wal *pWal){ + assert( pWal ); + walUnlockShared(pWal, WAL_CKPT_LOCK); +} + + +#endif /* SQLITE_ENABLE_SNAPSHOT */ + +#ifdef SQLITE_ENABLE_ZIPVFS +/* +** If the argument is not NULL, it points to a Wal object that holds a +** read-lock. This function returns the database page-size if it is known, +** or zero if it is not (or if pWal is NULL). +*/ +int sqlite3WalFramesize(Wal *pWal){ + assert( pWal==0 || pWal->readLock>=0 ); + return (pWal ? pWal->szPage : 0); +} +#endif + +/* Return the sqlite3_file object for the WAL file +*/ +sqlite3_file *sqlite3WalFile(Wal *pWal){ + return pWal->pWalFd; +} + +#endif /* #ifndef SQLITE_OMIT_WAL */ diff --git a/source/libs/tdb/src/sqliteinc/btree.h b/source/libs/tdb/src/sqliteinc/btree.h new file mode 100644 index 0000000000..f80ba4a97b --- /dev/null +++ b/source/libs/tdb/src/sqliteinc/btree.h @@ -0,0 +1,412 @@ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the sqlite B-Tree file +** subsystem. See comments in the source code for a detailed description +** of what each interface routine does. +*/ +#ifndef SQLITE_BTREE_H +#define SQLITE_BTREE_H + +/* TODO: This definition is just included so other modules compile. It +** needs to be revisited. +*/ +#define SQLITE_N_BTREE_META 16 + +/* +** If defined as non-zero, auto-vacuum is enabled by default. Otherwise +** it must be turned on for each database using "PRAGMA auto_vacuum = 1". +*/ +#ifndef SQLITE_DEFAULT_AUTOVACUUM + #define SQLITE_DEFAULT_AUTOVACUUM 0 +#endif + +#define BTREE_AUTOVACUUM_NONE 0 /* Do not do auto-vacuum */ +#define BTREE_AUTOVACUUM_FULL 1 /* Do full auto-vacuum */ +#define BTREE_AUTOVACUUM_INCR 2 /* Incremental vacuum */ + +/* +** Forward declarations of structure +*/ +typedef struct Btree Btree; +typedef struct BtCursor BtCursor; +typedef struct BtShared BtShared; +typedef struct BtreePayload BtreePayload; + + +int sqlite3BtreeOpen( + sqlite3_vfs *pVfs, /* VFS to use with this b-tree */ + const char *zFilename, /* Name of database file to open */ + sqlite3 *db, /* Associated database connection */ + Btree **ppBtree, /* Return open Btree* here */ + int flags, /* Flags */ + int vfsFlags /* Flags passed through to VFS open */ +); + +/* The flags parameter to sqlite3BtreeOpen can be the bitwise or of the +** following values. +** +** NOTE: These values must match the corresponding PAGER_ values in +** pager.h. +*/ +#define BTREE_OMIT_JOURNAL 1 /* Do not create or use a rollback journal */ +#define BTREE_MEMORY 2 /* This is an in-memory DB */ +#define BTREE_SINGLE 4 /* The file contains at most 1 b-tree */ +#define BTREE_UNORDERED 8 /* Use of a hash implementation is OK */ + +int sqlite3BtreeClose(Btree*); +int sqlite3BtreeSetCacheSize(Btree*,int); +int sqlite3BtreeSetSpillSize(Btree*,int); +#if SQLITE_MAX_MMAP_SIZE>0 + int sqlite3BtreeSetMmapLimit(Btree*,sqlite3_int64); +#endif +int sqlite3BtreeSetPagerFlags(Btree*,unsigned); +int sqlite3BtreeSetPageSize(Btree *p, int nPagesize, int nReserve, int eFix); +int sqlite3BtreeGetPageSize(Btree*); +Pgno sqlite3BtreeMaxPageCount(Btree*,Pgno); +Pgno sqlite3BtreeLastPage(Btree*); +int sqlite3BtreeSecureDelete(Btree*,int); +int sqlite3BtreeGetRequestedReserve(Btree*); +int sqlite3BtreeGetReserveNoMutex(Btree *p); +int sqlite3BtreeSetAutoVacuum(Btree *, int); +int sqlite3BtreeGetAutoVacuum(Btree *); +int sqlite3BtreeBeginTrans(Btree*,int,int*); +int sqlite3BtreeCommitPhaseOne(Btree*, const char*); +int sqlite3BtreeCommitPhaseTwo(Btree*, int); +int sqlite3BtreeCommit(Btree*); +int sqlite3BtreeRollback(Btree*,int,int); +int sqlite3BtreeBeginStmt(Btree*,int); +int sqlite3BtreeCreateTable(Btree*, Pgno*, int flags); +int sqlite3BtreeTxnState(Btree*); +int sqlite3BtreeIsInBackup(Btree*); + +void *sqlite3BtreeSchema(Btree *, int, void(*)(void *)); +int sqlite3BtreeSchemaLocked(Btree *pBtree); +#ifndef SQLITE_OMIT_SHARED_CACHE +int sqlite3BtreeLockTable(Btree *pBtree, int iTab, u8 isWriteLock); +#endif + +/* Savepoints are named, nestable SQL transactions mostly implemented */ +/* in vdbe.c and pager.c See https://sqlite.org/lang_savepoint.html */ +int sqlite3BtreeSavepoint(Btree *, int, int); + +/* "Checkpoint" only refers to WAL. See https://sqlite.org/wal.html#ckpt */ +#ifndef SQLITE_OMIT_WAL + int sqlite3BtreeCheckpoint(Btree*, int, int *, int *); +#endif + +const char *sqlite3BtreeGetFilename(Btree *); +const char *sqlite3BtreeGetJournalname(Btree *); +int sqlite3BtreeCopyFile(Btree *, Btree *); + +int sqlite3BtreeIncrVacuum(Btree *); + +/* The flags parameter to sqlite3BtreeCreateTable can be the bitwise OR +** of the flags shown below. +** +** Every SQLite table must have either BTREE_INTKEY or BTREE_BLOBKEY set. +** With BTREE_INTKEY, the table key is a 64-bit integer and arbitrary data +** is stored in the leaves. (BTREE_INTKEY is used for SQL tables.) With +** BTREE_BLOBKEY, the key is an arbitrary BLOB and no content is stored +** anywhere - the key is the content. (BTREE_BLOBKEY is used for SQL +** indices.) +*/ +#define BTREE_INTKEY 1 /* Table has only 64-bit signed integer keys */ +#define BTREE_BLOBKEY 2 /* Table has keys only - no data */ + +int sqlite3BtreeDropTable(Btree*, int, int*); +int sqlite3BtreeClearTable(Btree*, int, i64*); +int sqlite3BtreeClearTableOfCursor(BtCursor*); +int sqlite3BtreeTripAllCursors(Btree*, int, int); + +void sqlite3BtreeGetMeta(Btree *pBtree, int idx, u32 *pValue); +int sqlite3BtreeUpdateMeta(Btree*, int idx, u32 value); + +int sqlite3BtreeNewDb(Btree *p); + +/* +** The second parameter to sqlite3BtreeGetMeta or sqlite3BtreeUpdateMeta +** should be one of the following values. The integer values are assigned +** to constants so that the offset of the corresponding field in an +** SQLite database header may be found using the following formula: +** +** offset = 36 + (idx * 4) +** +** For example, the free-page-count field is located at byte offset 36 of +** the database file header. The incr-vacuum-flag field is located at +** byte offset 64 (== 36+4*7). +** +** The BTREE_DATA_VERSION value is not really a value stored in the header. +** It is a read-only number computed by the pager. But we merge it with +** the header value access routines since its access pattern is the same. +** Call it a "virtual meta value". +*/ +#define BTREE_FREE_PAGE_COUNT 0 +#define BTREE_SCHEMA_VERSION 1 +#define BTREE_FILE_FORMAT 2 +#define BTREE_DEFAULT_CACHE_SIZE 3 +#define BTREE_LARGEST_ROOT_PAGE 4 +#define BTREE_TEXT_ENCODING 5 +#define BTREE_USER_VERSION 6 +#define BTREE_INCR_VACUUM 7 +#define BTREE_APPLICATION_ID 8 +#define BTREE_DATA_VERSION 15 /* A virtual meta-value */ + +/* +** Kinds of hints that can be passed into the sqlite3BtreeCursorHint() +** interface. +** +** BTREE_HINT_RANGE (arguments: Expr*, Mem*) +** +** The first argument is an Expr* (which is guaranteed to be constant for +** the lifetime of the cursor) that defines constraints on which rows +** might be fetched with this cursor. The Expr* tree may contain +** TK_REGISTER nodes that refer to values stored in the array of registers +** passed as the second parameter. In other words, if Expr.op==TK_REGISTER +** then the value of the node is the value in Mem[pExpr.iTable]. Any +** TK_COLUMN node in the expression tree refers to the Expr.iColumn-th +** column of the b-tree of the cursor. The Expr tree will not contain +** any function calls nor subqueries nor references to b-trees other than +** the cursor being hinted. +** +** The design of the _RANGE hint is aid b-tree implementations that try +** to prefetch content from remote machines - to provide those +** implementations with limits on what needs to be prefetched and thereby +** reduce network bandwidth. +** +** Note that BTREE_HINT_FLAGS with BTREE_BULKLOAD is the only hint used by +** standard SQLite. The other hints are provided for extentions that use +** the SQLite parser and code generator but substitute their own storage +** engine. +*/ +#define BTREE_HINT_RANGE 0 /* Range constraints on queries */ + +/* +** Values that may be OR'd together to form the argument to the +** BTREE_HINT_FLAGS hint for sqlite3BtreeCursorHint(): +** +** The BTREE_BULKLOAD flag is set on index cursors when the index is going +** to be filled with content that is already in sorted order. +** +** The BTREE_SEEK_EQ flag is set on cursors that will get OP_SeekGE or +** OP_SeekLE opcodes for a range search, but where the range of entries +** selected will all have the same key. In other words, the cursor will +** be used only for equality key searches. +** +*/ +#define BTREE_BULKLOAD 0x00000001 /* Used to full index in sorted order */ +#define BTREE_SEEK_EQ 0x00000002 /* EQ seeks only - no range seeks */ + +/* +** Flags passed as the third argument to sqlite3BtreeCursor(). +** +** For read-only cursors the wrFlag argument is always zero. For read-write +** cursors it may be set to either (BTREE_WRCSR|BTREE_FORDELETE) or just +** (BTREE_WRCSR). If the BTREE_FORDELETE bit is set, then the cursor will +** only be used by SQLite for the following: +** +** * to seek to and then delete specific entries, and/or +** +** * to read values that will be used to create keys that other +** BTREE_FORDELETE cursors will seek to and delete. +** +** The BTREE_FORDELETE flag is an optimization hint. It is not used by +** by this, the native b-tree engine of SQLite, but it is available to +** alternative storage engines that might be substituted in place of this +** b-tree system. For alternative storage engines in which a delete of +** the main table row automatically deletes corresponding index rows, +** the FORDELETE flag hint allows those alternative storage engines to +** skip a lot of work. Namely: FORDELETE cursors may treat all SEEK +** and DELETE operations as no-ops, and any READ operation against a +** FORDELETE cursor may return a null row: 0x01 0x00. +*/ +#define BTREE_WRCSR 0x00000004 /* read-write cursor */ +#define BTREE_FORDELETE 0x00000008 /* Cursor is for seek/delete only */ + +int sqlite3BtreeCursor( + Btree*, /* BTree containing table to open */ + Pgno iTable, /* Index of root page */ + int wrFlag, /* 1 for writing. 0 for read-only */ + struct KeyInfo*, /* First argument to compare function */ + BtCursor *pCursor /* Space to write cursor structure */ +); +BtCursor *sqlite3BtreeFakeValidCursor(void); +int sqlite3BtreeCursorSize(void); +void sqlite3BtreeCursorZero(BtCursor*); +void sqlite3BtreeCursorHintFlags(BtCursor*, unsigned); +#ifdef SQLITE_ENABLE_CURSOR_HINTS +void sqlite3BtreeCursorHint(BtCursor*, int, ...); +#endif + +int sqlite3BtreeCloseCursor(BtCursor*); +int sqlite3BtreeTableMoveto( + BtCursor*, + i64 intKey, + int bias, + int *pRes +); +int sqlite3BtreeIndexMoveto( + BtCursor*, + UnpackedRecord *pUnKey, + int *pRes +); +int sqlite3BtreeCursorHasMoved(BtCursor*); +int sqlite3BtreeCursorRestore(BtCursor*, int*); +int sqlite3BtreeDelete(BtCursor*, u8 flags); + +/* Allowed flags for sqlite3BtreeDelete() and sqlite3BtreeInsert() */ +#define BTREE_SAVEPOSITION 0x02 /* Leave cursor pointing at NEXT or PREV */ +#define BTREE_AUXDELETE 0x04 /* not the primary delete operation */ +#define BTREE_APPEND 0x08 /* Insert is likely an append */ +#define BTREE_PREFORMAT 0x80 /* Inserted data is a preformated cell */ + +/* An instance of the BtreePayload object describes the content of a single +** entry in either an index or table btree. +** +** Index btrees (used for indexes and also WITHOUT ROWID tables) contain +** an arbitrary key and no data. These btrees have pKey,nKey set to the +** key and the pData,nData,nZero fields are uninitialized. The aMem,nMem +** fields give an array of Mem objects that are a decomposition of the key. +** The nMem field might be zero, indicating that no decomposition is available. +** +** Table btrees (used for rowid tables) contain an integer rowid used as +** the key and passed in the nKey field. The pKey field is zero. +** pData,nData hold the content of the new entry. nZero extra zero bytes +** are appended to the end of the content when constructing the entry. +** The aMem,nMem fields are uninitialized for table btrees. +** +** Field usage summary: +** +** Table BTrees Index Btrees +** +** pKey always NULL encoded key +** nKey the ROWID length of pKey +** pData data not used +** aMem not used decomposed key value +** nMem not used entries in aMem +** nData length of pData not used +** nZero extra zeros after pData not used +** +** This object is used to pass information into sqlite3BtreeInsert(). The +** same information used to be passed as five separate parameters. But placing +** the information into this object helps to keep the interface more +** organized and understandable, and it also helps the resulting code to +** run a little faster by using fewer registers for parameter passing. +*/ +struct BtreePayload { + const void *pKey; /* Key content for indexes. NULL for tables */ + sqlite3_int64 nKey; /* Size of pKey for indexes. PRIMARY KEY for tabs */ + const void *pData; /* Data for tables. */ + sqlite3_value *aMem; /* First of nMem value in the unpacked pKey */ + u16 nMem; /* Number of aMem[] value. Might be zero */ + int nData; /* Size of pData. 0 if none. */ + int nZero; /* Extra zero data appended after pData,nData */ +}; + +int sqlite3BtreeInsert(BtCursor*, const BtreePayload *pPayload, + int flags, int seekResult); +int sqlite3BtreeFirst(BtCursor*, int *pRes); +int sqlite3BtreeLast(BtCursor*, int *pRes); +int sqlite3BtreeNext(BtCursor*, int flags); +int sqlite3BtreeEof(BtCursor*); +int sqlite3BtreePrevious(BtCursor*, int flags); +i64 sqlite3BtreeIntegerKey(BtCursor*); +void sqlite3BtreeCursorPin(BtCursor*); +void sqlite3BtreeCursorUnpin(BtCursor*); +#ifdef SQLITE_ENABLE_OFFSET_SQL_FUNC +i64 sqlite3BtreeOffset(BtCursor*); +#endif +int sqlite3BtreePayload(BtCursor*, u32 offset, u32 amt, void*); +const void *sqlite3BtreePayloadFetch(BtCursor*, u32 *pAmt); +u32 sqlite3BtreePayloadSize(BtCursor*); +sqlite3_int64 sqlite3BtreeMaxRecordSize(BtCursor*); + +char *sqlite3BtreeIntegrityCheck(sqlite3*,Btree*,Pgno*aRoot,int nRoot,int,int*); +struct Pager *sqlite3BtreePager(Btree*); +i64 sqlite3BtreeRowCountEst(BtCursor*); + +#ifndef SQLITE_OMIT_INCRBLOB +int sqlite3BtreePayloadChecked(BtCursor*, u32 offset, u32 amt, void*); +int sqlite3BtreePutData(BtCursor*, u32 offset, u32 amt, void*); +void sqlite3BtreeIncrblobCursor(BtCursor *); +#endif +void sqlite3BtreeClearCursor(BtCursor *); +int sqlite3BtreeSetVersion(Btree *pBt, int iVersion); +int sqlite3BtreeCursorHasHint(BtCursor*, unsigned int mask); +int sqlite3BtreeIsReadonly(Btree *pBt); +int sqlite3HeaderSizeBtree(void); + +#ifdef SQLITE_DEBUG +sqlite3_uint64 sqlite3BtreeSeekCount(Btree*); +#else +# define sqlite3BtreeSeekCount(X) 0 +#endif + +#ifndef NDEBUG +int sqlite3BtreeCursorIsValid(BtCursor*); +#endif +int sqlite3BtreeCursorIsValidNN(BtCursor*); + +int sqlite3BtreeCount(sqlite3*, BtCursor*, i64*); + +#ifdef SQLITE_TEST +int sqlite3BtreeCursorInfo(BtCursor*, int*, int); +void sqlite3BtreeCursorList(Btree*); +#endif + +#ifndef SQLITE_OMIT_WAL + int sqlite3BtreeCheckpoint(Btree*, int, int *, int *); +#endif + +int sqlite3BtreeTransferRow(BtCursor*, BtCursor*, i64); + +/* +** If we are not using shared cache, then there is no need to +** use mutexes to access the BtShared structures. So make the +** Enter and Leave procedures no-ops. +*/ +#ifndef SQLITE_OMIT_SHARED_CACHE + void sqlite3BtreeEnter(Btree*); + void sqlite3BtreeEnterAll(sqlite3*); + int sqlite3BtreeSharable(Btree*); + void sqlite3BtreeEnterCursor(BtCursor*); + int sqlite3BtreeConnectionCount(Btree*); +#else +# define sqlite3BtreeEnter(X) +# define sqlite3BtreeEnterAll(X) +# define sqlite3BtreeSharable(X) 0 +# define sqlite3BtreeEnterCursor(X) +# define sqlite3BtreeConnectionCount(X) 1 +#endif + +#if !defined(SQLITE_OMIT_SHARED_CACHE) && SQLITE_THREADSAFE + void sqlite3BtreeLeave(Btree*); + void sqlite3BtreeLeaveCursor(BtCursor*); + void sqlite3BtreeLeaveAll(sqlite3*); +#ifndef NDEBUG + /* These routines are used inside assert() statements only. */ + int sqlite3BtreeHoldsMutex(Btree*); + int sqlite3BtreeHoldsAllMutexes(sqlite3*); + int sqlite3SchemaMutexHeld(sqlite3*,int,Schema*); +#endif +#else + +# define sqlite3BtreeLeave(X) +# define sqlite3BtreeLeaveCursor(X) +# define sqlite3BtreeLeaveAll(X) + +# define sqlite3BtreeHoldsMutex(X) 1 +# define sqlite3BtreeHoldsAllMutexes(X) 1 +# define sqlite3SchemaMutexHeld(X,Y,Z) 1 +#endif + + +#endif /* SQLITE_BTREE_H */ diff --git a/source/libs/tdb/src/sqliteinc/btreeInt.h b/source/libs/tdb/src/sqliteinc/btreeInt.h new file mode 100644 index 0000000000..1076fd8f2c --- /dev/null +++ b/source/libs/tdb/src/sqliteinc/btreeInt.h @@ -0,0 +1,729 @@ +/* +** 2004 April 6 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This file implements an external (disk-based) database using BTrees. +** For a detailed discussion of BTrees, refer to +** +** Donald E. Knuth, THE ART OF COMPUTER PROGRAMMING, Volume 3: +** "Sorting And Searching", pages 473-480. Addison-Wesley +** Publishing Company, Reading, Massachusetts. +** +** The basic idea is that each page of the file contains N database +** entries and N+1 pointers to subpages. +** +** ---------------------------------------------------------------- +** | Ptr(0) | Key(0) | Ptr(1) | Key(1) | ... | Key(N-1) | Ptr(N) | +** ---------------------------------------------------------------- +** +** All of the keys on the page that Ptr(0) points to have values less +** than Key(0). All of the keys on page Ptr(1) and its subpages have +** values greater than Key(0) and less than Key(1). All of the keys +** on Ptr(N) and its subpages have values greater than Key(N-1). And +** so forth. +** +** Finding a particular key requires reading O(log(M)) pages from the +** disk where M is the number of entries in the tree. +** +** In this implementation, a single file can hold one or more separate +** BTrees. Each BTree is identified by the index of its root page. The +** key and data for any entry are combined to form the "payload". A +** fixed amount of payload can be carried directly on the database +** page. If the payload is larger than the preset amount then surplus +** bytes are stored on overflow pages. The payload for an entry +** and the preceding pointer are combined to form a "Cell". Each +** page has a small header which contains the Ptr(N) pointer and other +** information such as the size of key and data. +** +** FORMAT DETAILS +** +** The file is divided into pages. The first page is called page 1, +** the second is page 2, and so forth. A page number of zero indicates +** "no such page". The page size can be any power of 2 between 512 and 65536. +** Each page can be either a btree page, a freelist page, an overflow +** page, or a pointer-map page. +** +** The first page is always a btree page. The first 100 bytes of the first +** page contain a special header (the "file header") that describes the file. +** The format of the file header is as follows: +** +** OFFSET SIZE DESCRIPTION +** 0 16 Header string: "SQLite format 3\000" +** 16 2 Page size in bytes. (1 means 65536) +** 18 1 File format write version +** 19 1 File format read version +** 20 1 Bytes of unused space at the end of each page +** 21 1 Max embedded payload fraction (must be 64) +** 22 1 Min embedded payload fraction (must be 32) +** 23 1 Min leaf payload fraction (must be 32) +** 24 4 File change counter +** 28 4 Reserved for future use +** 32 4 First freelist page +** 36 4 Number of freelist pages in the file +** 40 60 15 4-byte meta values passed to higher layers +** +** 40 4 Schema cookie +** 44 4 File format of schema layer +** 48 4 Size of page cache +** 52 4 Largest root-page (auto/incr_vacuum) +** 56 4 1=UTF-8 2=UTF16le 3=UTF16be +** 60 4 User version +** 64 4 Incremental vacuum mode +** 68 4 Application-ID +** 72 20 unused +** 92 4 The version-valid-for number +** 96 4 SQLITE_VERSION_NUMBER +** +** All of the integer values are big-endian (most significant byte first). +** +** The file change counter is incremented when the database is changed +** This counter allows other processes to know when the file has changed +** and thus when they need to flush their cache. +** +** The max embedded payload fraction is the amount of the total usable +** space in a page that can be consumed by a single cell for standard +** B-tree (non-LEAFDATA) tables. A value of 255 means 100%. The default +** is to limit the maximum cell size so that at least 4 cells will fit +** on one page. Thus the default max embedded payload fraction is 64. +** +** If the payload for a cell is larger than the max payload, then extra +** payload is spilled to overflow pages. Once an overflow page is allocated, +** as many bytes as possible are moved into the overflow pages without letting +** the cell size drop below the min embedded payload fraction. +** +** The min leaf payload fraction is like the min embedded payload fraction +** except that it applies to leaf nodes in a LEAFDATA tree. The maximum +** payload fraction for a LEAFDATA tree is always 100% (or 255) and it +** not specified in the header. +** +** Each btree pages is divided into three sections: The header, the +** cell pointer array, and the cell content area. Page 1 also has a 100-byte +** file header that occurs before the page header. +** +** |----------------| +** | file header | 100 bytes. Page 1 only. +** |----------------| +** | page header | 8 bytes for leaves. 12 bytes for interior nodes +** |----------------| +** | cell pointer | | 2 bytes per cell. Sorted order. +** | array | | Grows downward +** | | v +** |----------------| +** | unallocated | +** | space | +** |----------------| ^ Grows upwards +** | cell content | | Arbitrary order interspersed with freeblocks. +** | area | | and free space fragments. +** |----------------| +** +** The page headers looks like this: +** +** OFFSET SIZE DESCRIPTION +** 0 1 Flags. 1: intkey, 2: zerodata, 4: leafdata, 8: leaf +** 1 2 byte offset to the first freeblock +** 3 2 number of cells on this page +** 5 2 first byte of the cell content area +** 7 1 number of fragmented free bytes +** 8 4 Right child (the Ptr(N) value). Omitted on leaves. +** +** The flags define the format of this btree page. The leaf flag means that +** this page has no children. The zerodata flag means that this page carries +** only keys and no data. The intkey flag means that the key is an integer +** which is stored in the key size entry of the cell header rather than in +** the payload area. +** +** The cell pointer array begins on the first byte after the page header. +** The cell pointer array contains zero or more 2-byte numbers which are +** offsets from the beginning of the page to the cell content in the cell +** content area. The cell pointers occur in sorted order. The system strives +** to keep free space after the last cell pointer so that new cells can +** be easily added without having to defragment the page. +** +** Cell content is stored at the very end of the page and grows toward the +** beginning of the page. +** +** Unused space within the cell content area is collected into a linked list of +** freeblocks. Each freeblock is at least 4 bytes in size. The byte offset +** to the first freeblock is given in the header. Freeblocks occur in +** increasing order. Because a freeblock must be at least 4 bytes in size, +** any group of 3 or fewer unused bytes in the cell content area cannot +** exist on the freeblock chain. A group of 3 or fewer free bytes is called +** a fragment. The total number of bytes in all fragments is recorded. +** in the page header at offset 7. +** +** SIZE DESCRIPTION +** 2 Byte offset of the next freeblock +** 2 Bytes in this freeblock +** +** Cells are of variable length. Cells are stored in the cell content area at +** the end of the page. Pointers to the cells are in the cell pointer array +** that immediately follows the page header. Cells is not necessarily +** contiguous or in order, but cell pointers are contiguous and in order. +** +** Cell content makes use of variable length integers. A variable +** length integer is 1 to 9 bytes where the lower 7 bits of each +** byte are used. The integer consists of all bytes that have bit 8 set and +** the first byte with bit 8 clear. The most significant byte of the integer +** appears first. A variable-length integer may not be more than 9 bytes long. +** As a special case, all 8 bytes of the 9th byte are used as data. This +** allows a 64-bit integer to be encoded in 9 bytes. +** +** 0x00 becomes 0x00000000 +** 0x7f becomes 0x0000007f +** 0x81 0x00 becomes 0x00000080 +** 0x82 0x00 becomes 0x00000100 +** 0x80 0x7f becomes 0x0000007f +** 0x8a 0x91 0xd1 0xac 0x78 becomes 0x12345678 +** 0x81 0x81 0x81 0x81 0x01 becomes 0x10204081 +** +** Variable length integers are used for rowids and to hold the number of +** bytes of key and data in a btree cell. +** +** The content of a cell looks like this: +** +** SIZE DESCRIPTION +** 4 Page number of the left child. Omitted if leaf flag is set. +** var Number of bytes of data. Omitted if the zerodata flag is set. +** var Number of bytes of key. Or the key itself if intkey flag is set. +** * Payload +** 4 First page of the overflow chain. Omitted if no overflow +** +** Overflow pages form a linked list. Each page except the last is completely +** filled with data (pagesize - 4 bytes). The last page can have as little +** as 1 byte of data. +** +** SIZE DESCRIPTION +** 4 Page number of next overflow page +** * Data +** +** Freelist pages come in two subtypes: trunk pages and leaf pages. The +** file header points to the first in a linked list of trunk page. Each trunk +** page points to multiple leaf pages. The content of a leaf page is +** unspecified. A trunk page looks like this: +** +** SIZE DESCRIPTION +** 4 Page number of next trunk page +** 4 Number of leaf pointers on this page +** * zero or more pages numbers of leaves +*/ +#include "sqliteInt.h" + + +/* The following value is the maximum cell size assuming a maximum page +** size give above. +*/ +#define MX_CELL_SIZE(pBt) ((int)(pBt->pageSize-8)) + +/* The maximum number of cells on a single page of the database. This +** assumes a minimum cell size of 6 bytes (4 bytes for the cell itself +** plus 2 bytes for the index to the cell in the page header). Such +** small cells will be rare, but they are possible. +*/ +#define MX_CELL(pBt) ((pBt->pageSize-8)/6) + +/* Forward declarations */ +typedef struct MemPage MemPage; +typedef struct BtLock BtLock; +typedef struct CellInfo CellInfo; + +/* +** This is a magic string that appears at the beginning of every +** SQLite database in order to identify the file as a real database. +** +** You can change this value at compile-time by specifying a +** -DSQLITE_FILE_HEADER="..." on the compiler command-line. The +** header must be exactly 16 bytes including the zero-terminator so +** the string itself should be 15 characters long. If you change +** the header, then your custom library will not be able to read +** databases generated by the standard tools and the standard tools +** will not be able to read databases created by your custom library. +*/ +#ifndef SQLITE_FILE_HEADER /* 123456789 123456 */ +# define SQLITE_FILE_HEADER "SQLite format 3" +#endif + +/* +** Page type flags. An ORed combination of these flags appear as the +** first byte of on-disk image of every BTree page. +*/ +#define PTF_INTKEY 0x01 +#define PTF_ZERODATA 0x02 +#define PTF_LEAFDATA 0x04 +#define PTF_LEAF 0x08 + +/* +** An instance of this object stores information about each a single database +** page that has been loaded into memory. The information in this object +** is derived from the raw on-disk page content. +** +** As each database page is loaded into memory, the pager allocats an +** instance of this object and zeros the first 8 bytes. (This is the +** "extra" information associated with each page of the pager.) +** +** Access to all fields of this structure is controlled by the mutex +** stored in MemPage.pBt->mutex. +*/ +struct MemPage { + u8 isInit; /* True if previously initialized. MUST BE FIRST! */ + u8 intKey; /* True if table b-trees. False for index b-trees */ + u8 intKeyLeaf; /* True if the leaf of an intKey table */ + Pgno pgno; /* Page number for this page */ + /* Only the first 8 bytes (above) are zeroed by pager.c when a new page + ** is allocated. All fields that follow must be initialized before use */ + u8 leaf; /* True if a leaf page */ + u8 hdrOffset; /* 100 for page 1. 0 otherwise */ + u8 childPtrSize; /* 0 if leaf==1. 4 if leaf==0 */ + u8 max1bytePayload; /* min(maxLocal,127) */ + u8 nOverflow; /* Number of overflow cell bodies in aCell[] */ + u16 maxLocal; /* Copy of BtShared.maxLocal or BtShared.maxLeaf */ + u16 minLocal; /* Copy of BtShared.minLocal or BtShared.minLeaf */ + u16 cellOffset; /* Index in aData of first cell pointer */ + int nFree; /* Number of free bytes on the page. -1 for unknown */ + u16 nCell; /* Number of cells on this page, local and ovfl */ + u16 maskPage; /* Mask for page offset */ + u16 aiOvfl[4]; /* Insert the i-th overflow cell before the aiOvfl-th + ** non-overflow cell */ + u8 *apOvfl[4]; /* Pointers to the body of overflow cells */ + BtShared *pBt; /* Pointer to BtShared that this page is part of */ + u8 *aData; /* Pointer to disk image of the page data */ + u8 *aDataEnd; /* One byte past the end of usable data */ + u8 *aCellIdx; /* The cell index area */ + u8 *aDataOfst; /* Same as aData for leaves. aData+4 for interior */ + DbPage *pDbPage; /* Pager page handle */ + u16 (*xCellSize)(MemPage*,u8*); /* cellSizePtr method */ + void (*xParseCell)(MemPage*,u8*,CellInfo*); /* btreeParseCell method */ +}; + +/* +** A linked list of the following structures is stored at BtShared.pLock. +** Locks are added (or upgraded from READ_LOCK to WRITE_LOCK) when a cursor +** is opened on the table with root page BtShared.iTable. Locks are removed +** from this list when a transaction is committed or rolled back, or when +** a btree handle is closed. +*/ +struct BtLock { + Btree *pBtree; /* Btree handle holding this lock */ + Pgno iTable; /* Root page of table */ + u8 eLock; /* READ_LOCK or WRITE_LOCK */ + BtLock *pNext; /* Next in BtShared.pLock list */ +}; + +/* Candidate values for BtLock.eLock */ +#define READ_LOCK 1 +#define WRITE_LOCK 2 + +/* A Btree handle +** +** A database connection contains a pointer to an instance of +** this object for every database file that it has open. This structure +** is opaque to the database connection. The database connection cannot +** see the internals of this structure and only deals with pointers to +** this structure. +** +** For some database files, the same underlying database cache might be +** shared between multiple connections. In that case, each connection +** has it own instance of this object. But each instance of this object +** points to the same BtShared object. The database cache and the +** schema associated with the database file are all contained within +** the BtShared object. +** +** All fields in this structure are accessed under sqlite3.mutex. +** The pBt pointer itself may not be changed while there exists cursors +** in the referenced BtShared that point back to this Btree since those +** cursors have to go through this Btree to find their BtShared and +** they often do so without holding sqlite3.mutex. +*/ +struct Btree { + sqlite3 *db; /* The database connection holding this btree */ + BtShared *pBt; /* Sharable content of this btree */ + u8 inTrans; /* TRANS_NONE, TRANS_READ or TRANS_WRITE */ + u8 sharable; /* True if we can share pBt with another db */ + u8 locked; /* True if db currently has pBt locked */ + u8 hasIncrblobCur; /* True if there are one or more Incrblob cursors */ + int wantToLock; /* Number of nested calls to sqlite3BtreeEnter() */ + int nBackup; /* Number of backup operations reading this btree */ + u32 iBDataVersion; /* Combines with pBt->pPager->iDataVersion */ + Btree *pNext; /* List of other sharable Btrees from the same db */ + Btree *pPrev; /* Back pointer of the same list */ +#ifdef SQLITE_DEBUG + u64 nSeek; /* Calls to sqlite3BtreeMovetoUnpacked() */ +#endif +#ifndef SQLITE_OMIT_SHARED_CACHE + BtLock lock; /* Object used to lock page 1 */ +#endif +}; + +/* +** Btree.inTrans may take one of the following values. +** +** If the shared-data extension is enabled, there may be multiple users +** of the Btree structure. At most one of these may open a write transaction, +** but any number may have active read transactions. +** +** These values must match SQLITE_TXN_NONE, SQLITE_TXN_READ, and +** SQLITE_TXN_WRITE +*/ +#define TRANS_NONE 0 +#define TRANS_READ 1 +#define TRANS_WRITE 2 + +#if TRANS_NONE!=SQLITE_TXN_NONE +# error wrong numeric code for no-transaction +#endif +#if TRANS_READ!=SQLITE_TXN_READ +# error wrong numeric code for read-transaction +#endif +#if TRANS_WRITE!=SQLITE_TXN_WRITE +# error wrong numeric code for write-transaction +#endif + + +/* +** An instance of this object represents a single database file. +** +** A single database file can be in use at the same time by two +** or more database connections. When two or more connections are +** sharing the same database file, each connection has it own +** private Btree object for the file and each of those Btrees points +** to this one BtShared object. BtShared.nRef is the number of +** connections currently sharing this database file. +** +** Fields in this structure are accessed under the BtShared.mutex +** mutex, except for nRef and pNext which are accessed under the +** global SQLITE_MUTEX_STATIC_MAIN mutex. The pPager field +** may not be modified once it is initially set as long as nRef>0. +** The pSchema field may be set once under BtShared.mutex and +** thereafter is unchanged as long as nRef>0. +** +** isPending: +** +** If a BtShared client fails to obtain a write-lock on a database +** table (because there exists one or more read-locks on the table), +** the shared-cache enters 'pending-lock' state and isPending is +** set to true. +** +** The shared-cache leaves the 'pending lock' state when either of +** the following occur: +** +** 1) The current writer (BtShared.pWriter) concludes its transaction, OR +** 2) The number of locks held by other connections drops to zero. +** +** while in the 'pending-lock' state, no connection may start a new +** transaction. +** +** This feature is included to help prevent writer-starvation. +*/ +struct BtShared { + Pager *pPager; /* The page cache */ + sqlite3 *db; /* Database connection currently using this Btree */ + BtCursor *pCursor; /* A list of all open cursors */ + MemPage *pPage1; /* First page of the database */ + u8 openFlags; /* Flags to sqlite3BtreeOpen() */ +#ifndef SQLITE_OMIT_AUTOVACUUM + u8 autoVacuum; /* True if auto-vacuum is enabled */ + u8 incrVacuum; /* True if incr-vacuum is enabled */ + u8 bDoTruncate; /* True to truncate db on commit */ +#endif + u8 inTransaction; /* Transaction state */ + u8 max1bytePayload; /* Maximum first byte of cell for a 1-byte payload */ + u8 nReserveWanted; /* Desired number of extra bytes per page */ + u16 btsFlags; /* Boolean parameters. See BTS_* macros below */ + u16 maxLocal; /* Maximum local payload in non-LEAFDATA tables */ + u16 minLocal; /* Minimum local payload in non-LEAFDATA tables */ + u16 maxLeaf; /* Maximum local payload in a LEAFDATA table */ + u16 minLeaf; /* Minimum local payload in a LEAFDATA table */ + u32 pageSize; /* Total number of bytes on a page */ + u32 usableSize; /* Number of usable bytes on each page */ + int nTransaction; /* Number of open transactions (read + write) */ + u32 nPage; /* Number of pages in the database */ + void *pSchema; /* Pointer to space allocated by sqlite3BtreeSchema() */ + void (*xFreeSchema)(void*); /* Destructor for BtShared.pSchema */ + sqlite3_mutex *mutex; /* Non-recursive mutex required to access this object */ + Bitvec *pHasContent; /* Set of pages moved to free-list this transaction */ +#ifndef SQLITE_OMIT_SHARED_CACHE + int nRef; /* Number of references to this structure */ + BtShared *pNext; /* Next on a list of sharable BtShared structs */ + BtLock *pLock; /* List of locks held on this shared-btree struct */ + Btree *pWriter; /* Btree with currently open write transaction */ +#endif + u8 *pTmpSpace; /* Temp space sufficient to hold a single cell */ + int nPreformatSize; /* Size of last cell written by TransferRow() */ +}; + +/* +** Allowed values for BtShared.btsFlags +*/ +#define BTS_READ_ONLY 0x0001 /* Underlying file is readonly */ +#define BTS_PAGESIZE_FIXED 0x0002 /* Page size can no longer be changed */ +#define BTS_SECURE_DELETE 0x0004 /* PRAGMA secure_delete is enabled */ +#define BTS_OVERWRITE 0x0008 /* Overwrite deleted content with zeros */ +#define BTS_FAST_SECURE 0x000c /* Combination of the previous two */ +#define BTS_INITIALLY_EMPTY 0x0010 /* Database was empty at trans start */ +#define BTS_NO_WAL 0x0020 /* Do not open write-ahead-log files */ +#define BTS_EXCLUSIVE 0x0040 /* pWriter has an exclusive lock */ +#define BTS_PENDING 0x0080 /* Waiting for read-locks to clear */ + +/* +** An instance of the following structure is used to hold information +** about a cell. The parseCellPtr() function fills in this structure +** based on information extract from the raw disk page. +*/ +struct CellInfo { + i64 nKey; /* The key for INTKEY tables, or nPayload otherwise */ + u8 *pPayload; /* Pointer to the start of payload */ + u32 nPayload; /* Bytes of payload */ + u16 nLocal; /* Amount of payload held locally, not on overflow */ + u16 nSize; /* Size of the cell content on the main b-tree page */ +}; + +/* +** Maximum depth of an SQLite B-Tree structure. Any B-Tree deeper than +** this will be declared corrupt. This value is calculated based on a +** maximum database size of 2^31 pages a minimum fanout of 2 for a +** root-node and 3 for all other internal nodes. +** +** If a tree that appears to be taller than this is encountered, it is +** assumed that the database is corrupt. +*/ +#define BTCURSOR_MAX_DEPTH 20 + +/* +** A cursor is a pointer to a particular entry within a particular +** b-tree within a database file. +** +** The entry is identified by its MemPage and the index in +** MemPage.aCell[] of the entry. +** +** A single database file can be shared by two more database connections, +** but cursors cannot be shared. Each cursor is associated with a +** particular database connection identified BtCursor.pBtree.db. +** +** Fields in this structure are accessed under the BtShared.mutex +** found at self->pBt->mutex. +** +** skipNext meaning: +** The meaning of skipNext depends on the value of eState: +** +** eState Meaning of skipNext +** VALID skipNext is meaningless and is ignored +** INVALID skipNext is meaningless and is ignored +** SKIPNEXT sqlite3BtreeNext() is a no-op if skipNext>0 and +** sqlite3BtreePrevious() is no-op if skipNext<0. +** REQUIRESEEK restoreCursorPosition() restores the cursor to +** eState=SKIPNEXT if skipNext!=0 +** FAULT skipNext holds the cursor fault error code. +*/ +struct BtCursor { + u8 eState; /* One of the CURSOR_XXX constants (see below) */ + u8 curFlags; /* zero or more BTCF_* flags defined below */ + u8 curPagerFlags; /* Flags to send to sqlite3PagerGet() */ + u8 hints; /* As configured by CursorSetHints() */ + int skipNext; /* Prev() is noop if negative. Next() is noop if positive. + ** Error code if eState==CURSOR_FAULT */ + Btree *pBtree; /* The Btree to which this cursor belongs */ + Pgno *aOverflow; /* Cache of overflow page locations */ + void *pKey; /* Saved key that was cursor last known position */ + /* All fields above are zeroed when the cursor is allocated. See + ** sqlite3BtreeCursorZero(). Fields that follow must be manually + ** initialized. */ +#define BTCURSOR_FIRST_UNINIT pBt /* Name of first uninitialized field */ + BtShared *pBt; /* The BtShared this cursor points to */ + BtCursor *pNext; /* Forms a linked list of all cursors */ + CellInfo info; /* A parse of the cell we are pointing at */ + i64 nKey; /* Size of pKey, or last integer key */ + Pgno pgnoRoot; /* The root page of this tree */ + i8 iPage; /* Index of current page in apPage */ + u8 curIntKey; /* Value of apPage[0]->intKey */ + u16 ix; /* Current index for apPage[iPage] */ + u16 aiIdx[BTCURSOR_MAX_DEPTH-1]; /* Current index in apPage[i] */ + struct KeyInfo *pKeyInfo; /* Arg passed to comparison function */ + MemPage *pPage; /* Current page */ + MemPage *apPage[BTCURSOR_MAX_DEPTH-1]; /* Stack of parents of current page */ +}; + +/* +** Legal values for BtCursor.curFlags +*/ +#define BTCF_WriteFlag 0x01 /* True if a write cursor */ +#define BTCF_ValidNKey 0x02 /* True if info.nKey is valid */ +#define BTCF_ValidOvfl 0x04 /* True if aOverflow is valid */ +#define BTCF_AtLast 0x08 /* Cursor is pointing ot the last entry */ +#define BTCF_Incrblob 0x10 /* True if an incremental I/O handle */ +#define BTCF_Multiple 0x20 /* Maybe another cursor on the same btree */ +#define BTCF_Pinned 0x40 /* Cursor is busy and cannot be moved */ + +/* +** Potential values for BtCursor.eState. +** +** CURSOR_INVALID: +** Cursor does not point to a valid entry. This can happen (for example) +** because the table is empty or because BtreeCursorFirst() has not been +** called. +** +** CURSOR_VALID: +** Cursor points to a valid entry. getPayload() etc. may be called. +** +** CURSOR_SKIPNEXT: +** Cursor is valid except that the Cursor.skipNext field is non-zero +** indicating that the next sqlite3BtreeNext() or sqlite3BtreePrevious() +** operation should be a no-op. +** +** CURSOR_REQUIRESEEK: +** The table that this cursor was opened on still exists, but has been +** modified since the cursor was last used. The cursor position is saved +** in variables BtCursor.pKey and BtCursor.nKey. When a cursor is in +** this state, restoreCursorPosition() can be called to attempt to +** seek the cursor to the saved position. +** +** CURSOR_FAULT: +** An unrecoverable error (an I/O error or a malloc failure) has occurred +** on a different connection that shares the BtShared cache with this +** cursor. The error has left the cache in an inconsistent state. +** Do nothing else with this cursor. Any attempt to use the cursor +** should return the error code stored in BtCursor.skipNext +*/ +#define CURSOR_VALID 0 +#define CURSOR_INVALID 1 +#define CURSOR_SKIPNEXT 2 +#define CURSOR_REQUIRESEEK 3 +#define CURSOR_FAULT 4 + +/* +** The database page the PENDING_BYTE occupies. This page is never used. +*/ +# define PENDING_BYTE_PAGE(pBt) PAGER_MJ_PGNO(pBt) + +/* +** These macros define the location of the pointer-map entry for a +** database page. The first argument to each is the number of usable +** bytes on each page of the database (often 1024). The second is the +** page number to look up in the pointer map. +** +** PTRMAP_PAGENO returns the database page number of the pointer-map +** page that stores the required pointer. PTRMAP_PTROFFSET returns +** the offset of the requested map entry. +** +** If the pgno argument passed to PTRMAP_PAGENO is a pointer-map page, +** then pgno is returned. So (pgno==PTRMAP_PAGENO(pgsz, pgno)) can be +** used to test if pgno is a pointer-map page. PTRMAP_ISPAGE implements +** this test. +*/ +#define PTRMAP_PAGENO(pBt, pgno) ptrmapPageno(pBt, pgno) +#define PTRMAP_PTROFFSET(pgptrmap, pgno) (5*(pgno-pgptrmap-1)) +#define PTRMAP_ISPAGE(pBt, pgno) (PTRMAP_PAGENO((pBt),(pgno))==(pgno)) + +/* +** The pointer map is a lookup table that identifies the parent page for +** each child page in the database file. The parent page is the page that +** contains a pointer to the child. Every page in the database contains +** 0 or 1 parent pages. (In this context 'database page' refers +** to any page that is not part of the pointer map itself.) Each pointer map +** entry consists of a single byte 'type' and a 4 byte parent page number. +** The PTRMAP_XXX identifiers below are the valid types. +** +** The purpose of the pointer map is to facility moving pages from one +** position in the file to another as part of autovacuum. When a page +** is moved, the pointer in its parent must be updated to point to the +** new location. The pointer map is used to locate the parent page quickly. +** +** PTRMAP_ROOTPAGE: The database page is a root-page. The page-number is not +** used in this case. +** +** PTRMAP_FREEPAGE: The database page is an unused (free) page. The page-number +** is not used in this case. +** +** PTRMAP_OVERFLOW1: The database page is the first page in a list of +** overflow pages. The page number identifies the page that +** contains the cell with a pointer to this overflow page. +** +** PTRMAP_OVERFLOW2: The database page is the second or later page in a list of +** overflow pages. The page-number identifies the previous +** page in the overflow page list. +** +** PTRMAP_BTREE: The database page is a non-root btree page. The page number +** identifies the parent page in the btree. +*/ +#define PTRMAP_ROOTPAGE 1 +#define PTRMAP_FREEPAGE 2 +#define PTRMAP_OVERFLOW1 3 +#define PTRMAP_OVERFLOW2 4 +#define PTRMAP_BTREE 5 + +/* A bunch of assert() statements to check the transaction state variables +** of handle p (type Btree*) are internally consistent. +*/ +#define btreeIntegrity(p) \ + assert( p->pBt->inTransaction!=TRANS_NONE || p->pBt->nTransaction==0 ); \ + assert( p->pBt->inTransaction>=p->inTrans ); + + +/* +** The ISAUTOVACUUM macro is used within balance_nonroot() to determine +** if the database supports auto-vacuum or not. Because it is used +** within an expression that is an argument to another macro +** (sqliteMallocRaw), it is not possible to use conditional compilation. +** So, this macro is defined instead. +*/ +#ifndef SQLITE_OMIT_AUTOVACUUM +#define ISAUTOVACUUM (pBt->autoVacuum) +#else +#define ISAUTOVACUUM 0 +#endif + + +/* +** This structure is passed around through all the sanity checking routines +** in order to keep track of some global state information. +** +** The aRef[] array is allocated so that there is 1 bit for each page in +** the database. As the integrity-check proceeds, for each page used in +** the database the corresponding bit is set. This allows integrity-check to +** detect pages that are used twice and orphaned pages (both of which +** indicate corruption). +*/ +typedef struct IntegrityCk IntegrityCk; +struct IntegrityCk { + BtShared *pBt; /* The tree being checked out */ + Pager *pPager; /* The associated pager. Also accessible by pBt->pPager */ + u8 *aPgRef; /* 1 bit per page in the db (see above) */ + Pgno nPage; /* Number of pages in the database */ + int mxErr; /* Stop accumulating errors when this reaches zero */ + int nErr; /* Number of messages written to zErrMsg so far */ + int bOomFault; /* A memory allocation error has occurred */ + const char *zPfx; /* Error message prefix */ + Pgno v1; /* Value for first %u substitution in zPfx */ + int v2; /* Value for second %d substitution in zPfx */ + StrAccum errMsg; /* Accumulate the error message text here */ + u32 *heap; /* Min-heap used for analyzing cell coverage */ + sqlite3 *db; /* Database connection running the check */ +}; + +/* +** Routines to read or write a two- and four-byte big-endian integer values. +*/ +#define get2byte(x) ((x)[0]<<8 | (x)[1]) +#define put2byte(p,v) ((p)[0] = (u8)((v)>>8), (p)[1] = (u8)(v)) +#define get4byte sqlite3Get4byte +#define put4byte sqlite3Put4byte + +/* +** get2byteAligned(), unlike get2byte(), requires that its argument point to a +** two-byte aligned address. get2bytea() is only used for accessing the +** cell addresses in a btree header. +*/ +#if SQLITE_BYTEORDER==4321 +# define get2byteAligned(x) (*(u16*)(x)) +#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4008000 +# define get2byteAligned(x) __builtin_bswap16(*(u16*)(x)) +#elif SQLITE_BYTEORDER==1234 && MSVC_VERSION>=1300 +# define get2byteAligned(x) _byteswap_ushort(*(u16*)(x)) +#else +# define get2byteAligned(x) ((x)[0]<<8 | (x)[1]) +#endif diff --git a/source/libs/tdb/src/sqliteinc/pager.h b/source/libs/tdb/src/sqliteinc/pager.h new file mode 100644 index 0000000000..8d899bd1a6 --- /dev/null +++ b/source/libs/tdb/src/sqliteinc/pager.h @@ -0,0 +1,242 @@ +/* +** 2001 September 15 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the sqlite page cache +** subsystem. The page cache subsystem reads and writes a file a page +** at a time and provides a journal for rollback. +*/ + +#ifndef SQLITE_PAGER_H +#define SQLITE_PAGER_H + +/* +** Default maximum size for persistent journal files. A negative +** value means no limit. This value may be overridden using the +** sqlite3PagerJournalSizeLimit() API. See also "PRAGMA journal_size_limit". +*/ +#ifndef SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT + #define SQLITE_DEFAULT_JOURNAL_SIZE_LIMIT -1 +#endif + +/* +** The type used to represent a page number. The first page in a file +** is called page 1. 0 is used to represent "not a page". +*/ +typedef u32 Pgno; + +/* +** Each open file is managed by a separate instance of the "Pager" structure. +*/ +typedef struct Pager Pager; + +/* +** Handle type for pages. +*/ +typedef struct PgHdr DbPage; + +/* +** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is +** reserved for working around a windows/posix incompatibility). It is +** used in the journal to signify that the remainder of the journal file +** is devoted to storing a super-journal name - there are no more pages to +** roll back. See comments for function writeSuperJournal() in pager.c +** for details. +*/ +#define PAGER_MJ_PGNO(x) ((Pgno)((PENDING_BYTE/((x)->pageSize))+1)) + +/* +** Allowed values for the flags parameter to sqlite3PagerOpen(). +** +** NOTE: These values must match the corresponding BTREE_ values in btree.h. +*/ +#define PAGER_OMIT_JOURNAL 0x0001 /* Do not use a rollback journal */ +#define PAGER_MEMORY 0x0002 /* In-memory database */ + +/* +** Valid values for the second argument to sqlite3PagerLockingMode(). +*/ +#define PAGER_LOCKINGMODE_QUERY -1 +#define PAGER_LOCKINGMODE_NORMAL 0 +#define PAGER_LOCKINGMODE_EXCLUSIVE 1 + +/* +** Numeric constants that encode the journalmode. +** +** The numeric values encoded here (other than PAGER_JOURNALMODE_QUERY) +** are exposed in the API via the "PRAGMA journal_mode" command and +** therefore cannot be changed without a compatibility break. +*/ +#define PAGER_JOURNALMODE_QUERY (-1) /* Query the value of journalmode */ +#define PAGER_JOURNALMODE_DELETE 0 /* Commit by deleting journal file */ +#define PAGER_JOURNALMODE_PERSIST 1 /* Commit by zeroing journal header */ +#define PAGER_JOURNALMODE_OFF 2 /* Journal omitted. */ +#define PAGER_JOURNALMODE_TRUNCATE 3 /* Commit by truncating journal */ +#define PAGER_JOURNALMODE_MEMORY 4 /* In-memory journal file */ +#define PAGER_JOURNALMODE_WAL 5 /* Use write-ahead logging */ + +/* +** Flags that make up the mask passed to sqlite3PagerGet(). +*/ +#define PAGER_GET_NOCONTENT 0x01 /* Do not load data from disk */ +#define PAGER_GET_READONLY 0x02 /* Read-only page is acceptable */ + +/* +** Flags for sqlite3PagerSetFlags() +** +** Value constraints (enforced via assert()): +** PAGER_FULLFSYNC == SQLITE_FullFSync +** PAGER_CKPT_FULLFSYNC == SQLITE_CkptFullFSync +** PAGER_CACHE_SPILL == SQLITE_CacheSpill +*/ +#define PAGER_SYNCHRONOUS_OFF 0x01 /* PRAGMA synchronous=OFF */ +#define PAGER_SYNCHRONOUS_NORMAL 0x02 /* PRAGMA synchronous=NORMAL */ +#define PAGER_SYNCHRONOUS_FULL 0x03 /* PRAGMA synchronous=FULL */ +#define PAGER_SYNCHRONOUS_EXTRA 0x04 /* PRAGMA synchronous=EXTRA */ +#define PAGER_SYNCHRONOUS_MASK 0x07 /* Mask for four values above */ +#define PAGER_FULLFSYNC 0x08 /* PRAGMA fullfsync=ON */ +#define PAGER_CKPT_FULLFSYNC 0x10 /* PRAGMA checkpoint_fullfsync=ON */ +#define PAGER_CACHESPILL 0x20 /* PRAGMA cache_spill=ON */ +#define PAGER_FLAGS_MASK 0x38 /* All above except SYNCHRONOUS */ + +/* +** The remainder of this file contains the declarations of the functions +** that make up the Pager sub-system API. See source code comments for +** a detailed description of each routine. +*/ + +/* Open and close a Pager connection. */ +int sqlite3PagerOpen( + sqlite3_vfs*, + Pager **ppPager, + const char*, + int, + int, + int, + void(*)(DbPage*) +); +int sqlite3PagerClose(Pager *pPager, sqlite3*); +int sqlite3PagerReadFileheader(Pager*, int, unsigned char*); + +/* Functions used to configure a Pager object. */ +void sqlite3PagerSetBusyHandler(Pager*, int(*)(void *), void *); +int sqlite3PagerSetPagesize(Pager*, u32*, int); +Pgno sqlite3PagerMaxPageCount(Pager*, Pgno); +void sqlite3PagerSetCachesize(Pager*, int); +int sqlite3PagerSetSpillsize(Pager*, int); +void sqlite3PagerSetMmapLimit(Pager *, sqlite3_int64); +void sqlite3PagerShrink(Pager*); +void sqlite3PagerSetFlags(Pager*,unsigned); +int sqlite3PagerLockingMode(Pager *, int); +int sqlite3PagerSetJournalMode(Pager *, int); +int sqlite3PagerGetJournalMode(Pager*); +int sqlite3PagerOkToChangeJournalMode(Pager*); +i64 sqlite3PagerJournalSizeLimit(Pager *, i64); +sqlite3_backup **sqlite3PagerBackupPtr(Pager*); +int sqlite3PagerFlush(Pager*); + +/* Functions used to obtain and release page references. */ +int sqlite3PagerGet(Pager *pPager, Pgno pgno, DbPage **ppPage, int clrFlag); +DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno); +void sqlite3PagerRef(DbPage*); +void sqlite3PagerUnref(DbPage*); +void sqlite3PagerUnrefNotNull(DbPage*); +void sqlite3PagerUnrefPageOne(DbPage*); + +/* Operations on page references. */ +int sqlite3PagerWrite(DbPage*); +void sqlite3PagerDontWrite(DbPage*); +int sqlite3PagerMovepage(Pager*,DbPage*,Pgno,int); +int sqlite3PagerPageRefcount(DbPage*); +void *sqlite3PagerGetData(DbPage *); +void *sqlite3PagerGetExtra(DbPage *); + +/* Functions used to manage pager transactions and savepoints. */ +void sqlite3PagerPagecount(Pager*, int*); +int sqlite3PagerBegin(Pager*, int exFlag, int); +int sqlite3PagerCommitPhaseOne(Pager*,const char *zSuper, int); +int sqlite3PagerExclusiveLock(Pager*); +int sqlite3PagerSync(Pager *pPager, const char *zSuper); +int sqlite3PagerCommitPhaseTwo(Pager*); +int sqlite3PagerRollback(Pager*); +int sqlite3PagerOpenSavepoint(Pager *pPager, int n); +int sqlite3PagerSavepoint(Pager *pPager, int op, int iSavepoint); +int sqlite3PagerSharedLock(Pager *pPager); + +#ifndef SQLITE_OMIT_WAL + int sqlite3PagerCheckpoint(Pager *pPager, sqlite3*, int, int*, int*); + int sqlite3PagerWalSupported(Pager *pPager); + int sqlite3PagerWalCallback(Pager *pPager); + int sqlite3PagerOpenWal(Pager *pPager, int *pisOpen); + int sqlite3PagerCloseWal(Pager *pPager, sqlite3*); +# ifdef SQLITE_ENABLE_SNAPSHOT + int sqlite3PagerSnapshotGet(Pager*, sqlite3_snapshot **ppSnapshot); + int sqlite3PagerSnapshotOpen(Pager*, sqlite3_snapshot *pSnapshot); + int sqlite3PagerSnapshotRecover(Pager *pPager); + int sqlite3PagerSnapshotCheck(Pager *pPager, sqlite3_snapshot *pSnapshot); + void sqlite3PagerSnapshotUnlock(Pager *pPager); +# endif +#endif + +#if !defined(SQLITE_OMIT_WAL) && defined(SQLITE_ENABLE_SETLK_TIMEOUT) + int sqlite3PagerWalWriteLock(Pager*, int); + void sqlite3PagerWalDb(Pager*, sqlite3*); +#else +# define sqlite3PagerWalWriteLock(y,z) SQLITE_OK +# define sqlite3PagerWalDb(x,y) +#endif + +#ifdef SQLITE_DIRECT_OVERFLOW_READ + int sqlite3PagerDirectReadOk(Pager *pPager, Pgno pgno); +#endif + +#ifdef SQLITE_ENABLE_ZIPVFS + int sqlite3PagerWalFramesize(Pager *pPager); +#endif + +/* Functions used to query pager state and configuration. */ +u8 sqlite3PagerIsreadonly(Pager*); +u32 sqlite3PagerDataVersion(Pager*); +#ifdef SQLITE_DEBUG + int sqlite3PagerRefcount(Pager*); +#endif +int sqlite3PagerMemUsed(Pager*); +const char *sqlite3PagerFilename(const Pager*, int); +sqlite3_vfs *sqlite3PagerVfs(Pager*); +sqlite3_file *sqlite3PagerFile(Pager*); +sqlite3_file *sqlite3PagerJrnlFile(Pager*); +const char *sqlite3PagerJournalname(Pager*); +void *sqlite3PagerTempSpace(Pager*); +int sqlite3PagerIsMemdb(Pager*); +void sqlite3PagerCacheStat(Pager *, int, int, int *); +void sqlite3PagerClearCache(Pager*); +int sqlite3SectorSize(sqlite3_file *); + +/* Functions used to truncate the database file. */ +void sqlite3PagerTruncateImage(Pager*,Pgno); + +void sqlite3PagerRekey(DbPage*, Pgno, u16); + +/* Functions to support testing and debugging. */ +#if !defined(NDEBUG) || defined(SQLITE_TEST) + Pgno sqlite3PagerPagenumber(DbPage*); + int sqlite3PagerIswriteable(DbPage*); +#endif +#ifdef SQLITE_TEST + int *sqlite3PagerStats(Pager*); + void sqlite3PagerRefdump(Pager*); + void disable_simulated_io_errors(void); + void enable_simulated_io_errors(void); +#else +# define disable_simulated_io_errors() +# define enable_simulated_io_errors() +#endif + +#endif /* SQLITE_PAGER_H */ diff --git a/source/libs/tdb/src/sqliteinc/pcache.h b/source/libs/tdb/src/sqliteinc/pcache.h new file mode 100644 index 0000000000..eb55396afa --- /dev/null +++ b/source/libs/tdb/src/sqliteinc/pcache.h @@ -0,0 +1,190 @@ +/* +** 2008 August 05 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface that the sqlite page cache +** subsystem. +*/ + +#ifndef _PCACHE_H_ + +typedef struct PgHdr PgHdr; +typedef struct PCache PCache; + +/* +** Every page in the cache is controlled by an instance of the following +** structure. +*/ +struct PgHdr { + sqlite3_pcache_page *pPage; /* Pcache object page handle */ + void *pData; /* Page data */ + void *pExtra; /* Extra content */ + PCache *pCache; /* PRIVATE: Cache that owns this page */ + PgHdr *pDirty; /* Transient list of dirty sorted by pgno */ + Pager *pPager; /* The pager this page is part of */ + Pgno pgno; /* Page number for this page */ +#ifdef SQLITE_CHECK_PAGES + u32 pageHash; /* Hash of page content */ +#endif + u16 flags; /* PGHDR flags defined below */ + + /********************************************************************** + ** Elements above, except pCache, are public. All that follow are + ** private to pcache.c and should not be accessed by other modules. + ** pCache is grouped with the public elements for efficiency. + */ + i16 nRef; /* Number of users of this page */ + PgHdr *pDirtyNext; /* Next element in list of dirty pages */ + PgHdr *pDirtyPrev; /* Previous element in list of dirty pages */ + /* NB: pDirtyNext and pDirtyPrev are undefined if the + ** PgHdr object is not dirty */ +}; + +/* Bit values for PgHdr.flags */ +#define PGHDR_CLEAN 0x001 /* Page not on the PCache.pDirty list */ +#define PGHDR_DIRTY 0x002 /* Page is on the PCache.pDirty list */ +#define PGHDR_WRITEABLE 0x004 /* Journaled and ready to modify */ +#define PGHDR_NEED_SYNC 0x008 /* Fsync the rollback journal before + ** writing this page to the database */ +#define PGHDR_DONT_WRITE 0x010 /* Do not write content to disk */ +#define PGHDR_MMAP 0x020 /* This is an mmap page object */ + +#define PGHDR_WAL_APPEND 0x040 /* Appended to wal file */ + +/* Initialize and shutdown the page cache subsystem */ +int sqlite3PcacheInitialize(void); +void sqlite3PcacheShutdown(void); + +/* Page cache buffer management: +** These routines implement SQLITE_CONFIG_PAGECACHE. +*/ +void sqlite3PCacheBufferSetup(void *, int sz, int n); + +/* Create a new pager cache. +** Under memory stress, invoke xStress to try to make pages clean. +** Only clean and unpinned pages can be reclaimed. +*/ +int sqlite3PcacheOpen( + int szPage, /* Size of every page */ + int szExtra, /* Extra space associated with each page */ + int bPurgeable, /* True if pages are on backing store */ + int (*xStress)(void*, PgHdr*), /* Call to try to make pages clean */ + void *pStress, /* Argument to xStress */ + PCache *pToInit /* Preallocated space for the PCache */ +); + +/* Modify the page-size after the cache has been created. */ +int sqlite3PcacheSetPageSize(PCache *, int); + +/* Return the size in bytes of a PCache object. Used to preallocate +** storage space. +*/ +int sqlite3PcacheSize(void); + +/* One release per successful fetch. Page is pinned until released. +** Reference counted. +*/ +sqlite3_pcache_page *sqlite3PcacheFetch(PCache*, Pgno, int createFlag); +int sqlite3PcacheFetchStress(PCache*, Pgno, sqlite3_pcache_page**); +PgHdr *sqlite3PcacheFetchFinish(PCache*, Pgno, sqlite3_pcache_page *pPage); +void sqlite3PcacheRelease(PgHdr*); + +void sqlite3PcacheDrop(PgHdr*); /* Remove page from cache */ +void sqlite3PcacheMakeDirty(PgHdr*); /* Make sure page is marked dirty */ +void sqlite3PcacheMakeClean(PgHdr*); /* Mark a single page as clean */ +void sqlite3PcacheCleanAll(PCache*); /* Mark all dirty list pages as clean */ +void sqlite3PcacheClearWritable(PCache*); + +/* Change a page number. Used by incr-vacuum. */ +void sqlite3PcacheMove(PgHdr*, Pgno); + +/* Remove all pages with pgno>x. Reset the cache if x==0 */ +void sqlite3PcacheTruncate(PCache*, Pgno x); + +/* Get a list of all dirty pages in the cache, sorted by page number */ +PgHdr *sqlite3PcacheDirtyList(PCache*); + +/* Reset and close the cache object */ +void sqlite3PcacheClose(PCache*); + +/* Clear flags from pages of the page cache */ +void sqlite3PcacheClearSyncFlags(PCache *); + +/* Discard the contents of the cache */ +void sqlite3PcacheClear(PCache*); + +/* Return the total number of outstanding page references */ +int sqlite3PcacheRefCount(PCache*); + +/* Increment the reference count of an existing page */ +void sqlite3PcacheRef(PgHdr*); + +int sqlite3PcachePageRefcount(PgHdr*); + +/* Return the total number of pages stored in the cache */ +int sqlite3PcachePagecount(PCache*); + +#if defined(SQLITE_CHECK_PAGES) || defined(SQLITE_DEBUG) +/* Iterate through all dirty pages currently stored in the cache. This +** interface is only available if SQLITE_CHECK_PAGES is defined when the +** library is built. +*/ +void sqlite3PcacheIterateDirty(PCache *pCache, void (*xIter)(PgHdr *)); +#endif + +#if defined(SQLITE_DEBUG) +/* Check invariants on a PgHdr object */ +int sqlite3PcachePageSanity(PgHdr*); +#endif + +/* Set and get the suggested cache-size for the specified pager-cache. +** +** If no global maximum is configured, then the system attempts to limit +** the total number of pages cached by purgeable pager-caches to the sum +** of the suggested cache-sizes. +*/ +void sqlite3PcacheSetCachesize(PCache *, int); +#ifdef SQLITE_TEST +int sqlite3PcacheGetCachesize(PCache *); +#endif + +/* Set or get the suggested spill-size for the specified pager-cache. +** +** The spill-size is the minimum number of pages in cache before the cache +** will attempt to spill dirty pages by calling xStress. +*/ +int sqlite3PcacheSetSpillsize(PCache *, int); + +/* Free up as much memory as possible from the page cache */ +void sqlite3PcacheShrink(PCache*); + +#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT +/* Try to return memory used by the pcache module to the main memory heap */ +int sqlite3PcacheReleaseMemory(int); +#endif + +#ifdef SQLITE_TEST +void sqlite3PcacheStats(int*,int*,int*,int*); +#endif + +void sqlite3PCacheSetDefault(void); + +/* Return the header size */ +int sqlite3HeaderSizePcache(void); +int sqlite3HeaderSizePcache1(void); + +/* Number of dirty pages as a percentage of the configured cache size */ +int sqlite3PCachePercentDirty(PCache*); + +#ifdef SQLITE_DIRECT_OVERFLOW_READ +int sqlite3PCacheIsDirty(PCache *pCache); +#endif + +#endif /* _PCACHE_H_ */ diff --git a/source/libs/tdb/src/sqliteinc/sqliteInt.h b/source/libs/tdb/src/sqliteinc/sqliteInt.h new file mode 100644 index 0000000000..e69de29bb2 diff --git a/source/libs/tdb/src/sqliteinc/wal.h b/source/libs/tdb/src/sqliteinc/wal.h new file mode 100644 index 0000000000..02e2bab360 --- /dev/null +++ b/source/libs/tdb/src/sqliteinc/wal.h @@ -0,0 +1,155 @@ +/* +** 2010 February 1 +** +** The author disclaims copyright to this source code. In place of +** a legal notice, here is a blessing: +** +** May you do good and not evil. +** May you find forgiveness for yourself and forgive others. +** May you share freely, never taking more than you give. +** +************************************************************************* +** This header file defines the interface to the write-ahead logging +** system. Refer to the comments below and the header comment attached to +** the implementation of each function in log.c for further details. +*/ + +#ifndef SQLITE_WAL_H +#define SQLITE_WAL_H + +#include "sqliteInt.h" + +/* Macros for extracting appropriate sync flags for either transaction +** commits (WAL_SYNC_FLAGS(X)) or for checkpoint ops (CKPT_SYNC_FLAGS(X)): +*/ +#define WAL_SYNC_FLAGS(X) ((X)&0x03) +#define CKPT_SYNC_FLAGS(X) (((X)>>2)&0x03) + +#ifdef SQLITE_OMIT_WAL +# define sqlite3WalOpen(x,y,z) 0 +# define sqlite3WalLimit(x,y) +# define sqlite3WalClose(v,w,x,y,z) 0 +# define sqlite3WalBeginReadTransaction(y,z) 0 +# define sqlite3WalEndReadTransaction(z) +# define sqlite3WalDbsize(y) 0 +# define sqlite3WalBeginWriteTransaction(y) 0 +# define sqlite3WalEndWriteTransaction(x) 0 +# define sqlite3WalUndo(x,y,z) 0 +# define sqlite3WalSavepoint(y,z) +# define sqlite3WalSavepointUndo(y,z) 0 +# define sqlite3WalFrames(u,v,w,x,y,z) 0 +# define sqlite3WalCheckpoint(q,r,s,t,u,v,w,x,y,z) 0 +# define sqlite3WalCallback(z) 0 +# define sqlite3WalExclusiveMode(y,z) 0 +# define sqlite3WalHeapMemory(z) 0 +# define sqlite3WalFramesize(z) 0 +# define sqlite3WalFindFrame(x,y,z) 0 +# define sqlite3WalFile(x) 0 +#else + +#define WAL_SAVEPOINT_NDATA 4 + +/* Connection to a write-ahead log (WAL) file. +** There is one object of this type for each pager. +*/ +typedef struct Wal Wal; + +/* Open and close a connection to a write-ahead log. */ +int sqlite3WalOpen(sqlite3_vfs*, sqlite3_file*, const char *, int, i64, Wal**); +int sqlite3WalClose(Wal *pWal, sqlite3*, int sync_flags, int, u8 *); + +/* Set the limiting size of a WAL file. */ +void sqlite3WalLimit(Wal*, i64); + +/* Used by readers to open (lock) and close (unlock) a snapshot. A +** snapshot is like a read-transaction. It is the state of the database +** at an instant in time. sqlite3WalOpenSnapshot gets a read lock and +** preserves the current state even if the other threads or processes +** write to or checkpoint the WAL. sqlite3WalCloseSnapshot() closes the +** transaction and releases the lock. +*/ +int sqlite3WalBeginReadTransaction(Wal *pWal, int *); +void sqlite3WalEndReadTransaction(Wal *pWal); + +/* Read a page from the write-ahead log, if it is present. */ +int sqlite3WalFindFrame(Wal *, Pgno, u32 *); +int sqlite3WalReadFrame(Wal *, u32, int, u8 *); + +/* If the WAL is not empty, return the size of the database. */ +Pgno sqlite3WalDbsize(Wal *pWal); + +/* Obtain or release the WRITER lock. */ +int sqlite3WalBeginWriteTransaction(Wal *pWal); +int sqlite3WalEndWriteTransaction(Wal *pWal); + +/* Undo any frames written (but not committed) to the log */ +int sqlite3WalUndo(Wal *pWal, int (*xUndo)(void *, Pgno), void *pUndoCtx); + +/* Return an integer that records the current (uncommitted) write +** position in the WAL */ +void sqlite3WalSavepoint(Wal *pWal, u32 *aWalData); + +/* Move the write position of the WAL back to iFrame. Called in +** response to a ROLLBACK TO command. */ +int sqlite3WalSavepointUndo(Wal *pWal, u32 *aWalData); + +/* Write a frame or frames to the log. */ +int sqlite3WalFrames(Wal *pWal, int, PgHdr *, Pgno, int, int); + +/* Copy pages from the log to the database file */ +int sqlite3WalCheckpoint( + Wal *pWal, /* Write-ahead log connection */ + sqlite3 *db, /* Check this handle's interrupt flag */ + int eMode, /* One of PASSIVE, FULL and RESTART */ + int (*xBusy)(void*), /* Function to call when busy */ + void *pBusyArg, /* Context argument for xBusyHandler */ + int sync_flags, /* Flags to sync db file with (or 0) */ + int nBuf, /* Size of buffer nBuf */ + u8 *zBuf, /* Temporary buffer to use */ + int *pnLog, /* OUT: Number of frames in WAL */ + int *pnCkpt /* OUT: Number of backfilled frames in WAL */ +); + +/* Return the value to pass to a sqlite3_wal_hook callback, the +** number of frames in the WAL at the point of the last commit since +** sqlite3WalCallback() was called. If no commits have occurred since +** the last call, then return 0. +*/ +int sqlite3WalCallback(Wal *pWal); + +/* Tell the wal layer that an EXCLUSIVE lock has been obtained (or released) +** by the pager layer on the database file. +*/ +int sqlite3WalExclusiveMode(Wal *pWal, int op); + +/* Return true if the argument is non-NULL and the WAL module is using +** heap-memory for the wal-index. Otherwise, if the argument is NULL or the +** WAL module is using shared-memory, return false. +*/ +int sqlite3WalHeapMemory(Wal *pWal); + +#ifdef SQLITE_ENABLE_SNAPSHOT +int sqlite3WalSnapshotGet(Wal *pWal, sqlite3_snapshot **ppSnapshot); +void sqlite3WalSnapshotOpen(Wal *pWal, sqlite3_snapshot *pSnapshot); +int sqlite3WalSnapshotRecover(Wal *pWal); +int sqlite3WalSnapshotCheck(Wal *pWal, sqlite3_snapshot *pSnapshot); +void sqlite3WalSnapshotUnlock(Wal *pWal); +#endif + +#ifdef SQLITE_ENABLE_ZIPVFS +/* If the WAL file is not empty, return the number of bytes of content +** stored in each frame (i.e. the db page-size when the WAL was created). +*/ +int sqlite3WalFramesize(Wal *pWal); +#endif + +/* Return the sqlite3_file object for the WAL file */ +sqlite3_file *sqlite3WalFile(Wal *pWal); + +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT +int sqlite3WalWriteLock(Wal *pWal, int bLock); +void sqlite3WalDb(Wal *pWal, sqlite3 *db); +#endif + +#endif /* ifndef SQLITE_OMIT_WAL */ +#endif /* SQLITE_WAL_H */