diff options
Diffstat (limited to 'src')
-rw-r--r-- | src/btree.c | 28 | ||||
-rw-r--r-- | src/main.c | 44 | ||||
-rw-r--r-- | src/os_unix.c | 33 | ||||
-rw-r--r-- | src/pager.c | 41 | ||||
-rw-r--r-- | src/pager.h | 17 | ||||
-rw-r--r-- | src/sqlite.h.in | 1 | ||||
-rw-r--r-- | src/sqliteInt.h | 3 | ||||
-rw-r--r-- | src/wal.c | 333 | ||||
-rw-r--r-- | src/wal.h | 5 |
9 files changed, 315 insertions, 190 deletions
diff --git a/src/btree.c b/src/btree.c index 961de0ea4..3fe0e1c3c 100644 --- a/src/btree.c +++ b/src/btree.c @@ -2304,8 +2304,7 @@ static int btreeInvokeBusyHandler(void *pArg){ BtShared *pBt = (BtShared*)pArg; assert( pBt->db ); assert( sqlite3_mutex_held(pBt->db->mutex) ); - return sqlite3InvokeBusyHandler(&pBt->db->busyHandler, - sqlite3PagerFile(pBt->pPager)); + return sqlite3InvokeBusyHandler(&pBt->db->busyHandler); } /* @@ -3372,6 +3371,7 @@ int sqlite3BtreeNewDb(Btree *p){ */ int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ BtShared *pBt = p->pBt; + Pager *pPager = pBt->pPager; int rc = SQLITE_OK; sqlite3BtreeEnter(p); @@ -3387,7 +3387,7 @@ int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ assert( pBt->inTransaction==TRANS_WRITE || IfNotOmitAV(pBt->bDoTruncate)==0 ); if( (p->db->flags & SQLITE_ResetDatabase) - && sqlite3PagerIsreadonly(pBt->pPager)==0 + && sqlite3PagerIsreadonly(pPager)==0 ){ pBt->btsFlags &= ~BTS_READ_ONLY; } @@ -3434,7 +3434,19 @@ int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ pBt->btsFlags &= ~BTS_INITIALLY_EMPTY; if( pBt->nPage==0 ) pBt->btsFlags |= BTS_INITIALLY_EMPTY; + sqlite3PagerWalDb(pPager, p->db); do { + +#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 @@ -3448,7 +3460,7 @@ int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ if( (pBt->btsFlags & BTS_READ_ONLY)!=0 ){ rc = SQLITE_READONLY; }else{ - rc = sqlite3PagerBegin(pBt->pPager,wrflag>1,sqlite3TempInMemory(p->db)); + 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 ){ @@ -3461,11 +3473,15 @@ int sqlite3BtreeBeginTrans(Btree *p, int wrflag, int *pSchemaVersion){ } if( rc!=SQLITE_OK ){ + (void)sqlite3PagerWalWriteLock(pPager, 0); unlockBtreeIfUnused(pBt); } }while( (rc&0xFF)==SQLITE_BUSY && pBt->inTransaction==TRANS_NONE && btreeInvokeBusyHandler(pBt) ); - sqlite3PagerResetLockTimeout(pBt->pPager); + 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 ){ @@ -3517,7 +3533,7 @@ trans_begun: ** 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(pBt->pPager, p->db->nSavepoint); + rc = sqlite3PagerOpenSavepoint(pPager, p->db->nSavepoint); } } diff --git a/src/main.c b/src/main.c index 9e1a0e9da..9fc410fae 100644 --- a/src/main.c +++ b/src/main.c @@ -1614,8 +1614,7 @@ const char *sqlite3ErrStr(int rc){ */ static int sqliteDefaultBusyCallback( void *ptr, /* Database connection */ - int count, /* Number of times table has been busy */ - sqlite3_file *pFile /* The file on which the lock occurred */ + int count /* Number of times table has been busy */ ){ #if SQLITE_OS_WIN || HAVE_USLEEP /* This case is for systems that have support for sleeping for fractions of @@ -1629,31 +1628,6 @@ static int sqliteDefaultBusyCallback( int tmout = db->busyTimeout; int delay, prior; -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT - if( sqlite3OsFileControl(pFile,SQLITE_FCNTL_LOCK_TIMEOUT,&tmout)==SQLITE_OK ){ - if( count ){ - /* If this is the second or later invocation of the busy-handler, - ** but tmout==0, then code in wal.c must have disabled the blocking - ** lock before the SQLITE_BUSY error was hit. In this case, no delay - ** occurred while waiting for the lock, so fall through to the xSleep() - ** code below to delay a while before retrying the lock. - ** - ** Alternatively, if tmout!=0, then SQLite has already waited - ** sqlite3.busyTimeout ms for a lock. In this case, return 0 to - ** indicate that the lock should not be retried and the SQLITE_BUSY - ** error returned to the application. */ - if( tmout ){ - tmout = 0; - sqlite3OsFileControl(pFile, SQLITE_FCNTL_LOCK_TIMEOUT, &tmout); - return 0; - } - }else{ - return 1; - } - } -#else - UNUSED_PARAMETER(pFile); -#endif assert( count>=0 ); if( count < NDELAY ){ delay = delays[count]; @@ -1673,7 +1647,6 @@ static int sqliteDefaultBusyCallback( ** must be done in increments of whole seconds */ sqlite3 *db = (sqlite3 *)ptr; int tmout = ((sqlite3 *)ptr)->busyTimeout; - UNUSED_PARAMETER(pFile); if( (count+1)*1000 > tmout ){ return 0; } @@ -1691,19 +1664,10 @@ static int sqliteDefaultBusyCallback( ** If this routine returns non-zero, the lock is retried. If it ** returns 0, the operation aborts with an SQLITE_BUSY error. */ -int sqlite3InvokeBusyHandler(BusyHandler *p, sqlite3_file *pFile){ +int sqlite3InvokeBusyHandler(BusyHandler *p){ int rc; if( p->xBusyHandler==0 || p->nBusy<0 ) return 0; - if( p->bExtraFileArg ){ - /* Add an extra parameter with the pFile pointer to the end of the - ** callback argument list */ - int (*xTra)(void*,int,sqlite3_file*); - xTra = (int(*)(void*,int,sqlite3_file*))p->xBusyHandler; - rc = xTra(p->pBusyArg, p->nBusy, pFile); - }else{ - /* Legacy style busy handler callback */ - rc = p->xBusyHandler(p->pBusyArg, p->nBusy); - } + rc = p->xBusyHandler(p->pBusyArg, p->nBusy); if( rc==0 ){ p->nBusy = -1; }else{ @@ -1728,7 +1692,6 @@ int sqlite3_busy_handler( db->busyHandler.xBusyHandler = xBusy; db->busyHandler.pBusyArg = pArg; db->busyHandler.nBusy = 0; - db->busyHandler.bExtraFileArg = 0; db->busyTimeout = 0; sqlite3_mutex_leave(db->mutex); return SQLITE_OK; @@ -1779,7 +1742,6 @@ int sqlite3_busy_timeout(sqlite3 *db, int ms){ sqlite3_busy_handler(db, (int(*)(void*,int))sqliteDefaultBusyCallback, (void*)db); db->busyTimeout = ms; - db->busyHandler.bExtraFileArg = 1; }else{ sqlite3_busy_handler(db, 0, 0); } diff --git a/src/os_unix.c b/src/os_unix.c index 56e53929e..d51071108 100644 --- a/src/os_unix.c +++ b/src/os_unix.c @@ -1565,8 +1565,9 @@ static int osSetPosixAdvisoryLock( struct flock *pLock, /* The description of the lock */ unixFile *pFile /* Structure holding timeout value */ ){ + int tm = pFile->iBusyTimeout; int rc = osFcntl(h,F_SETLK,pLock); - while( rc<0 && pFile->iBusyTimeout>0 ){ + while( rc<0 && tm>0 ){ /* On systems that support some kind of blocking file lock with a timeout, ** make appropriate changes here to invoke that blocking file lock. On ** generic posix, however, there is no such API. So we simply try the @@ -1574,7 +1575,7 @@ static int osSetPosixAdvisoryLock( ** the lock is obtained. */ usleep(1000); rc = osFcntl(h,F_SETLK,pLock); - pFile->iBusyTimeout--; + tm--; } return rc; } @@ -4316,13 +4317,20 @@ static int unixShmSystemLock( assert( n>=1 && n<=SQLITE_SHM_NLOCK ); if( pShmNode->hShm>=0 ){ + int res; /* Initialize the locking parameters */ f.l_type = lockType; f.l_whence = SEEK_SET; f.l_start = ofst; f.l_len = n; - rc = osSetPosixAdvisoryLock(pShmNode->hShm, &f, pFile); - rc = (rc!=(-1)) ? SQLITE_OK : SQLITE_BUSY; + res = osSetPosixAdvisoryLock(pShmNode->hShm, &f, pFile); + if( res==-1 ){ +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + rc = (pFile->iBusyTimeout ? SQLITE_BUSY_TIMEOUT : SQLITE_BUSY); +#else + rc = SQLITE_BUSY; +#endif + } } /* Update the global lock state and do debug tracing */ @@ -4819,22 +4827,23 @@ static int unixShmLock( assert( pShmNode->hShm>=0 || pDbFd->pInode->bProcessLock==1 ); assert( pShmNode->hShm<0 || pDbFd->pInode->bProcessLock==0 ); - /* Check that, if this to be a blocking lock, that locks have been - ** obtained in the following order. + /* Check that, if this to be a blocking lock, no locks that occur later + ** in the following list than the lock being obtained are already held: ** ** 1. Checkpointer lock (ofst==1). - ** 2. Recover lock (ofst==2). + ** 2. Write lock (ofst==0). ** 3. Read locks (ofst>=3 && ofst<SQLITE_SHM_NLOCK). - ** 4. Write lock (ofst==0). ** ** In other words, if this is a blocking lock, none of the locks that ** occur later in the above list than the lock being obtained may be ** held. */ #ifdef SQLITE_ENABLE_SETLK_TIMEOUT - assert( pDbFd->iBusyTimeout==0 - || (flags & SQLITE_SHM_UNLOCK) || ofst==0 - || ((p->exclMask|p->sharedMask)&~((1<<ofst)-2))==0 - ); + assert( (flags & SQLITE_SHM_UNLOCK) || pDbFd->iBusyTimeout==0 || ( + (ofst!=2) /* not RECOVER */ + && (ofst!=1 || (p->exclMask|p->sharedMask)==0) + && (ofst!=0 || (p->exclMask|p->sharedMask)<3) + && (ofst<3 || (p->exclMask|p->sharedMask)<(1<<ofst)) + )); #endif mask = (1<<(ofst+n)) - (1<<ofst); diff --git a/src/pager.c b/src/pager.c index 0d08a2dfa..a826f256b 100644 --- a/src/pager.c +++ b/src/pager.c @@ -5705,7 +5705,6 @@ void sqlite3PagerUnrefPageOne(DbPage *pPg){ assert( pPg->pgno==1 ); assert( (pPg->flags & PGHDR_MMAP)==0 ); /* Page1 is never memory mapped */ pPager = pPg->pPager; - sqlite3PagerResetLockTimeout(pPager); sqlite3PcacheRelease(pPg); pagerUnlockIfUnused(pPager); } @@ -6998,16 +6997,6 @@ sqlite3_file *sqlite3PagerFile(Pager *pPager){ return pPager->fd; } -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT -/* -** Reset the lock timeout for pager. -*/ -void sqlite3PagerResetLockTimeout(Pager *pPager){ - int x = 0; - sqlite3OsFileControl(pPager->fd, SQLITE_FCNTL_LOCK_TIMEOUT, &x); -} -#endif - /* ** Return the file handle for the journal file (if it exists). ** This will be either the rollback journal or the WAL file. @@ -7421,7 +7410,6 @@ int sqlite3PagerCheckpoint( pPager->walSyncFlags, pPager->pageSize, (u8 *)pPager->pTmpSpace, pnLog, pnCkpt ); - sqlite3PagerResetLockTimeout(pPager); } return rc; } @@ -7586,7 +7574,31 @@ int sqlite3PagerCloseWal(Pager *pPager, sqlite3 *db){ 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 /* @@ -7606,7 +7618,10 @@ int sqlite3PagerSnapshotGet(Pager *pPager, sqlite3_snapshot **ppSnapshot){ ** 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 sqlite3PagerSnapshotOpen( + Pager *pPager, + sqlite3_snapshot *pSnapshot +){ int rc = SQLITE_OK; if( pPager->pWal ){ sqlite3WalSnapshotOpen(pPager->pWal, pSnapshot); diff --git a/src/pager.h b/src/pager.h index 2c99d67a9..62a371ef8 100644 --- a/src/pager.h +++ b/src/pager.h @@ -177,14 +177,22 @@ int sqlite3PagerSharedLock(Pager *pPager); int sqlite3PagerOpenWal(Pager *pPager, int *pisOpen); int sqlite3PagerCloseWal(Pager *pPager, sqlite3*); # ifdef SQLITE_ENABLE_SNAPSHOT - int sqlite3PagerSnapshotGet(Pager *pPager, sqlite3_snapshot **ppSnapshot); - int sqlite3PagerSnapshotOpen(Pager *pPager, sqlite3_snapshot *pSnapshot); + 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 @@ -210,11 +218,6 @@ int sqlite3PagerIsMemdb(Pager*); void sqlite3PagerCacheStat(Pager *, int, int, int *); void sqlite3PagerClearCache(Pager*); int sqlite3SectorSize(sqlite3_file *); -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT -void sqlite3PagerResetLockTimeout(Pager *pPager); -#else -# define sqlite3PagerResetLockTimeout(X) -#endif /* Functions used to truncate the database file. */ void sqlite3PagerTruncateImage(Pager*,Pgno); diff --git a/src/sqlite.h.in b/src/sqlite.h.in index aa0ac9b0f..90544db29 100644 --- a/src/sqlite.h.in +++ b/src/sqlite.h.in @@ -508,6 +508,7 @@ int sqlite3_exec( #define SQLITE_LOCKED_VTAB (SQLITE_LOCKED | (2<<8)) #define SQLITE_BUSY_RECOVERY (SQLITE_BUSY | (1<<8)) #define SQLITE_BUSY_SNAPSHOT (SQLITE_BUSY | (2<<8)) +#define SQLITE_BUSY_TIMEOUT (SQLITE_BUSY | (3<<8)) #define SQLITE_CANTOPEN_NOTEMPDIR (SQLITE_CANTOPEN | (1<<8)) #define SQLITE_CANTOPEN_ISDIR (SQLITE_CANTOPEN | (2<<8)) #define SQLITE_CANTOPEN_FULLPATH (SQLITE_CANTOPEN | (3<<8)) diff --git a/src/sqliteInt.h b/src/sqliteInt.h index 4a78d0cfd..5ae69e873 100644 --- a/src/sqliteInt.h +++ b/src/sqliteInt.h @@ -994,7 +994,6 @@ struct BusyHandler { int (*xBusyHandler)(void *,int); /* The busy callback */ void *pBusyArg; /* First arg to busy callback */ int nBusy; /* Incremented with each busy call */ - u8 bExtraFileArg; /* Include sqlite3_file as callback arg */ }; /* @@ -4585,7 +4584,7 @@ void sqlite3RenameExprlistUnmap(Parse*, ExprList*); CollSeq *sqlite3GetCollSeq(Parse*, u8, CollSeq *, const char*); char sqlite3AffinityType(const char*, Column*); void sqlite3Analyze(Parse*, Token*, Token*); -int sqlite3InvokeBusyHandler(BusyHandler*, sqlite3_file*); +int sqlite3InvokeBusyHandler(BusyHandler*); int sqlite3FindDb(sqlite3*, Token*); int sqlite3FindDbName(sqlite3 *, const char *); int sqlite3AnalysisLoad(sqlite3*,int iDB); @@ -467,6 +467,9 @@ struct Wal { #ifdef SQLITE_ENABLE_SNAPSHOT WalIndexHdr *pSnapshot; /* Start transaction here if not NULL */ #endif +#ifdef SQLITE_ENABLE_SETLK_TIMEOUT + sqlite3 *db; +#endif }; /* @@ -840,7 +843,7 @@ static int walLockShared(Wal *pWal, int lockIdx){ 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!=SQLITE_BUSY); ) + VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && (rc&0xFF)!=SQLITE_BUSY); ) return rc; } static void walUnlockShared(Wal *pWal, int lockIdx){ @@ -856,7 +859,7 @@ static int walLockExclusive(Wal *pWal, int lockIdx, int 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!=SQLITE_BUSY); ) + VVA_ONLY( pWal->lockError = (u8)(rc!=SQLITE_OK && (rc&0xFF)!=SQLITE_BUSY); ) return rc; } static void walUnlockExclusive(Wal *pWal, int lockIdx, int n){ @@ -1128,11 +1131,6 @@ static int walIndexRecover(Wal *pWal){ u32 aFrameCksum[2] = {0, 0}; int iLock; /* Lock offset to lock for checkpoint */ -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT - int tmout = 0; - sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout); -#endif - /* Obtain an exclusive lock on all byte in the locking range not already ** locked by the caller. The caller is guaranteed to have locked the ** WAL_WRITE_LOCK byte, and may have also locked the WAL_CKPT_LOCK byte. @@ -1681,6 +1679,89 @@ static int walIteratorInit(Wal *pWal, u32 nBackfill, WalIterator **pp){ 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 @@ -1698,6 +1779,12 @@ static int walBusyLock( 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; } @@ -2177,28 +2264,32 @@ static int walIndexReadHdr(Wal *pWal, int *pChanged){ /* If the first attempt failed, it might have been due to a race ** with a writer. So get a WRITE lock and try again. */ - assert( badHdr==0 || pWal->writeLock==0 ); 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 if( SQLITE_OK==(rc = walLockExclusive(pWal, WAL_WRITE_LOCK, 1)) ){ - 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; + }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); } } - pWal->writeLock = 0; - walUnlockExclusive(pWal, WAL_WRITE_LOCK, 1); } } @@ -2750,22 +2841,34 @@ int sqlite3WalSnapshotRecover(Wal *pWal){ int sqlite3WalBeginReadTransaction(Wal *pWal, int *pChanged){ int rc; /* Return code */ int cnt = 0; /* Number of TryBeginRead attempts */ -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT - int tmout = 0; -#endif + + assert( pWal->ckptLock==0 ); #ifdef SQLITE_ENABLE_SNAPSHOT int bChanged = 0; WalIndexHdr *pSnapshot = pWal->pSnapshot; - if( pSnapshot && memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){ - bChanged = 1; - } -#endif + 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); -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT - /* Disable blocking locks. They are not useful when trying to open a - ** read-transaction, and blocking may cause deadlock anyway. */ - sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout); + if( rc!=SQLITE_OK ){ + return rc; + } + pWal->ckptLock = 1; + } #endif do{ @@ -2776,16 +2879,6 @@ int sqlite3WalBeginReadTransaction(Wal *pWal, int *pChanged){ testcase( rc==SQLITE_PROTOCOL ); testcase( rc==SQLITE_OK ); -#ifdef SQLITE_ENABLE_SETLK_TIMEOUT - /* If they were disabled earlier and the read-transaction has been - ** successfully opened, re-enable blocking locks. This is because the - ** connection may attempt to upgrade to a write-transaction, which does - ** benefit from using blocking locks. */ - if( rc==SQLITE_OK ){ - sqlite3OsFileControl(pWal->pDbFd, SQLITE_FCNTL_LOCK_TIMEOUT, (void*)&tmout); - } -#endif - #ifdef SQLITE_ENABLE_SNAPSHOT if( rc==SQLITE_OK ){ if( pSnapshot && memcmp(pSnapshot, &pWal->hdr, sizeof(WalIndexHdr))!=0 ){ @@ -2807,48 +2900,42 @@ int sqlite3WalBeginReadTransaction(Wal *pWal, int *pChanged){ assert( pWal->readLock>0 || pWal->hdr.mxFrame==0 ); assert( pInfo->aReadMark[pWal->readLock]<=pSnapshot->mxFrame ); - /* 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. - ** - ** TODO: Does the aReadMark[] lock prevent a checkpointer from doing - ** this already? - */ - rc = walLockShared(pWal, WAL_CKPT_LOCK); - - if( rc==SQLITE_OK ){ - /* 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; - } - - /* Release the shared CKPT lock obtained above. */ - walUnlockShared(pWal, WAL_CKPT_LOCK); - pWal->minFrame = 1; + /* 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; } @@ -3019,6 +3106,16 @@ Pgno sqlite3WalDbsize(Wal *pWal){ 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 ); @@ -3595,45 +3692,52 @@ int sqlite3WalCheckpoint( 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. */ + ** "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); - if( rc ){ - /* 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. - */ - testcase( rc==SQLITE_BUSY ); - testcase( xBusy!=0 ); - return rc; - } - pWal->ckptLock = 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, xBusy, 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; + /* 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); } @@ -3665,11 +3769,19 @@ int sqlite3WalCheckpoint( memset(&pWal->hdr, 0, sizeof(WalIndexHdr)); } + walDisableBlocking(pWal); + sqlite3WalDb(pWal, 0); + /* Release the locks. */ sqlite3WalEndWriteTransaction(pWal); - walUnlockExclusive(pWal, WAL_CKPT_LOCK, 1); - pWal->ckptLock = 0; + 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); } @@ -3786,7 +3898,10 @@ int sqlite3WalSnapshotGet(Wal *pWal, sqlite3_snapshot **ppSnapshot){ /* Try to open on pSnapshot when the next read-transaction starts */ -void sqlite3WalSnapshotOpen(Wal *pWal, sqlite3_snapshot *pSnapshot){ +void sqlite3WalSnapshotOpen( + Wal *pWal, + sqlite3_snapshot *pSnapshot +){ pWal->pSnapshot = (WalIndexHdr*)pSnapshot; } @@ -146,5 +146,10 @@ int sqlite3WalFramesize(Wal *pWal); /* 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 */ |