diff options
Diffstat (limited to 'src/os_unix.c')
-rw-r--r-- | src/os_unix.c | 256 |
1 files changed, 132 insertions, 124 deletions
diff --git a/src/os_unix.c b/src/os_unix.c index f4e689fa9..f04b6af27 100644 --- a/src/os_unix.c +++ b/src/os_unix.c @@ -371,7 +371,7 @@ static struct unix_syscall { #else { "fallocate", (sqlite3_syscall_ptr)0, 0 }, #endif -#define osFallocate ((int(*)(int,off_t,off_t)aSyscall[15].pCurrent) +#define osFallocate ((int(*)(int,off_t,off_t))aSyscall[15].pCurrent) }; /* End of the overrideable system calls */ @@ -444,18 +444,16 @@ static sqlite3_syscall_ptr unixGetSystemCall( ** system call. */ static const char *unixNextSystemCall(sqlite3_vfs *p, const char *zName){ - unsigned int i; + int i = -1; UNUSED_PARAMETER(p); - if( zName==0 ){ - i = -1; - }else{ - for(i=0; i<sizeof(aSyscall)/sizeof(aSyscall[0])-1; i++){ - if( strcmp(zName, aSyscall[0].zName)==0 ) break; + if( zName ){ + for(i=0; i<ArraySize(aSyscall)-1; i++){ + if( strcmp(zName, aSyscall[i].zName)==0 ) break; } } - for(i++; i<sizeof(aSyscall)/sizeof(aSyscall[0]); i++){ - if( aSyscall[0].pCurrent!=0 ) return aSyscall[0].zName; + for(i++; i<ArraySize(aSyscall); i++){ + if( aSyscall[i].pCurrent!=0 ) return aSyscall[i].zName; } return 0; } @@ -595,9 +593,22 @@ static int robust_ftruncate(int h, sqlite3_int64 sz){ */ static int sqliteErrorFromPosixError(int posixError, int sqliteIOErr) { switch (posixError) { +#if 0 + /* At one point this code was not commented out. In theory, this branch + ** should never be hit, as this function should only be called after + ** a locking-related function (i.e. fcntl()) has returned non-zero with + ** the value of errno as the first argument. Since a system call has failed, + ** errno should be non-zero. + ** + ** Despite this, if errno really is zero, we still don't want to return + ** SQLITE_OK. The system call failed, and *some* SQLite error should be + ** propagated back to the caller. Commenting this branch out means errno==0 + ** will be handled by the "default:" case below. + */ case 0: return SQLITE_OK; - +#endif + case EAGAIN: case ETIMEDOUT: case EBUSY: @@ -619,8 +630,15 @@ static int sqliteErrorFromPosixError(int posixError, int sqliteIOErr) { case EPERM: return SQLITE_PERM; + /* EDEADLK is only possible if a call to fcntl(F_SETLKW) is made. And + ** this module never makes such a call. And the code in SQLite itself + ** asserts that SQLITE_IOERR_BLOCKED is never returned. For these reasons + ** this case is also commented out. If the system does set errno to EDEADLK, + ** the default SQLITE_IOERR_XXX code will be returned. */ +#if 0 case EDEADLK: return SQLITE_IOERR_BLOCKED; +#endif #if EOPNOTSUPP!=ENOTSUP case EOPNOTSUPP: @@ -1039,7 +1057,7 @@ static void closePendingFds(unixFile *pFile){ static void releaseInodeInfo(unixFile *pFile){ unixInodeInfo *pInode = pFile->pInode; assert( unixMutexHeld() ); - if( pInode ){ + if( ALWAYS(pInode) ){ pInode->nRef--; if( pInode->nRef==0 ){ assert( pInode->pShmNode==0 ); @@ -1181,10 +1199,9 @@ static int unixCheckReservedLock(sqlite3_file *id, int *pResOut){ lock.l_start = RESERVED_BYTE; lock.l_len = 1; lock.l_type = F_WRLCK; - if (-1 == osFcntl(pFile->h, F_GETLK, &lock)) { - int tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_CHECKRESERVEDLOCK); - pFile->lastErrno = tErrno; + if( osFcntl(pFile->h, F_GETLK, &lock) ){ + rc = SQLITE_IOERR_CHECKRESERVEDLOCK; + pFile->lastErrno = errno; } else if( lock.l_type!=F_UNLCK ){ reserved = 1; } @@ -1213,6 +1230,9 @@ static int unixCheckReservedLock(sqlite3_file *id, int *pResOut){ ** This function is a pass-through to fcntl(F_SETLK) if pFile is using ** any VFS other than "unix-excl" or if pFile is opened on "unix-excl" ** and is read-only. +** +** Zero is returned if the call completes successfully, or -1 if a call +** to fcntl() fails. In this case, errno is set appropriately (by fcntl()). */ static int unixFileLock(unixFile *pFile, struct flock *pLock){ int rc; @@ -1309,7 +1329,6 @@ static int unixLock(sqlite3_file *id, int eFileLock){ unixFile *pFile = (unixFile*)id; unixInodeInfo *pInode = pFile->pInode; struct flock lock; - int s = 0; int tErrno = 0; assert( pFile ); @@ -1378,11 +1397,10 @@ static int unixLock(sqlite3_file *id, int eFileLock){ ){ lock.l_type = (eFileLock==SHARED_LOCK?F_RDLCK:F_WRLCK); lock.l_start = PENDING_BYTE; - s = unixFileLock(pFile, &lock); - if( s==(-1) ){ + if( unixFileLock(pFile, &lock) ){ tErrno = errno; rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); - if( IS_LOCK_ERROR(rc) ){ + if( rc!=SQLITE_BUSY ){ pFile->lastErrno = tErrno; } goto end_lock; @@ -1396,33 +1414,31 @@ static int unixLock(sqlite3_file *id, int eFileLock){ if( eFileLock==SHARED_LOCK ){ assert( pInode->nShared==0 ); assert( pInode->eFileLock==0 ); + assert( rc==SQLITE_OK ); /* Now get the read-lock */ lock.l_start = SHARED_FIRST; lock.l_len = SHARED_SIZE; - if( (s = unixFileLock(pFile, &lock))==(-1) ){ + if( unixFileLock(pFile, &lock) ){ tErrno = errno; + rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); } + /* Drop the temporary PENDING lock */ lock.l_start = PENDING_BYTE; lock.l_len = 1L; lock.l_type = F_UNLCK; - if( unixFileLock(pFile, &lock)!=0 ){ - if( s != -1 ){ - /* This could happen with a network mount */ - tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); - if( IS_LOCK_ERROR(rc) ){ - pFile->lastErrno = tErrno; - } - goto end_lock; - } + if( unixFileLock(pFile, &lock) && rc==SQLITE_OK ){ + /* This could happen with a network mount */ + tErrno = errno; + rc = SQLITE_IOERR_UNLOCK; } - if( s==(-1) ){ - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); - if( IS_LOCK_ERROR(rc) ){ + + if( rc ){ + if( rc!=SQLITE_BUSY ){ pFile->lastErrno = tErrno; } + goto end_lock; }else{ pFile->eFileLock = SHARED_LOCK; pInode->nLock++; @@ -1439,22 +1455,20 @@ static int unixLock(sqlite3_file *id, int eFileLock){ */ assert( 0!=pFile->eFileLock ); lock.l_type = F_WRLCK; - switch( eFileLock ){ - case RESERVED_LOCK: - lock.l_start = RESERVED_BYTE; - break; - case EXCLUSIVE_LOCK: - lock.l_start = SHARED_FIRST; - lock.l_len = SHARED_SIZE; - break; - default: - assert(0); + + assert( eFileLock==RESERVED_LOCK || eFileLock==EXCLUSIVE_LOCK ); + if( eFileLock==RESERVED_LOCK ){ + lock.l_start = RESERVED_BYTE; + lock.l_len = 1L; + }else{ + lock.l_start = SHARED_FIRST; + lock.l_len = SHARED_SIZE; } - s = unixFileLock(pFile, &lock); - if( s==(-1) ){ + + if( unixFileLock(pFile, &lock) ){ tErrno = errno; rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK); - if( IS_LOCK_ERROR(rc) ){ + if( rc!=SQLITE_BUSY ){ pFile->lastErrno = tErrno; } } @@ -1525,7 +1539,6 @@ static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ struct flock lock; int rc = SQLITE_OK; int h; - int tErrno; /* Error code from system call errors */ assert( pFile ); OSTRACE(("UNLOCK %d %d was %d(%d,%d) pid=%d (unix)\n", pFile->h, eFileLock, @@ -1580,15 +1593,16 @@ static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ #endif #if defined(__APPLE__) && SQLITE_ENABLE_LOCKING_STYLE if( handleNFSUnlock ){ + int tErrno; /* Error code from system call errors */ off_t divSize = SHARED_SIZE - 1; lock.l_type = F_UNLCK; lock.l_whence = SEEK_SET; lock.l_start = SHARED_FIRST; lock.l_len = divSize; - if( unixFileLock(pFile,, &lock)==(-1) ){ + if( unixFileLock(pFile, &lock)==(-1) ){ tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); + rc = SQLITE_IOERR_UNLOCK; if( IS_LOCK_ERROR(rc) ){ pFile->lastErrno = tErrno; } @@ -1612,7 +1626,7 @@ static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ lock.l_len = SHARED_SIZE-divSize; if( unixFileLock(pFile, &lock)==(-1) ){ tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); + rc = SQLITE_IOERR_UNLOCK; if( IS_LOCK_ERROR(rc) ){ pFile->lastErrno = tErrno; } @@ -1625,12 +1639,15 @@ static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ lock.l_whence = SEEK_SET; lock.l_start = SHARED_FIRST; lock.l_len = SHARED_SIZE; - if( unixFileLock(pFile, &lock)==(-1) ){ - tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_RDLOCK); - if( IS_LOCK_ERROR(rc) ){ - pFile->lastErrno = tErrno; - } + if( unixFileLock(pFile, &lock) ){ + /* In theory, the call to unixFileLock() cannot fail because another + ** process is holding an incompatible lock. If it does, this + ** indicates that the other process is not following the locking + ** protocol. If this happens, return SQLITE_IOERR_RDLOCK. Returning + ** SQLITE_BUSY would confuse the upper layer (in practice it causes + ** an assert to fail). */ + rc = SQLITE_IOERR_RDLOCK; + pFile->lastErrno = errno; goto end_unlock; } } @@ -1639,14 +1656,11 @@ static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ lock.l_whence = SEEK_SET; lock.l_start = PENDING_BYTE; lock.l_len = 2L; assert( PENDING_BYTE+1==RESERVED_BYTE ); - if( unixFileLock(pFile, &lock)!=(-1) ){ + if( unixFileLock(pFile, &lock)==0 ){ pInode->eFileLock = SHARED_LOCK; }else{ - tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); - if( IS_LOCK_ERROR(rc) ){ - pFile->lastErrno = tErrno; - } + rc = SQLITE_IOERR_UNLOCK; + pFile->lastErrno = errno; goto end_unlock; } } @@ -1663,14 +1677,11 @@ static int posixUnlock(sqlite3_file *id, int eFileLock, int handleNFSUnlock){ SimulateIOErrorBenign(1); SimulateIOError( h=(-1) ) SimulateIOErrorBenign(0); - if( unixFileLock(pFile, &lock)!=(-1) ){ + if( unixFileLock(pFile, &lock)==0 ){ pInode->eFileLock = NO_LOCK; }else{ - tErrno = errno; - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); - if( IS_LOCK_ERROR(rc) ){ - pFile->lastErrno = tErrno; - } + rc = SQLITE_IOERR_UNLOCK; + pFile->lastErrno = errno; pInode->eFileLock = NO_LOCK; pFile->eFileLock = NO_LOCK; } @@ -1716,29 +1727,27 @@ static int unixUnlock(sqlite3_file *id, int eFileLock){ */ static int closeUnixFile(sqlite3_file *id){ unixFile *pFile = (unixFile*)id; - if( pFile ){ - if( pFile->dirfd>=0 ){ - robust_close(pFile, pFile->dirfd, __LINE__); - pFile->dirfd=-1; - } - if( pFile->h>=0 ){ - robust_close(pFile, pFile->h, __LINE__); - pFile->h = -1; - } + if( pFile->dirfd>=0 ){ + robust_close(pFile, pFile->dirfd, __LINE__); + pFile->dirfd=-1; + } + if( pFile->h>=0 ){ + robust_close(pFile, pFile->h, __LINE__); + pFile->h = -1; + } #if OS_VXWORKS - if( pFile->pId ){ - if( pFile->isDelete ){ - unlink(pFile->pId->zCanonicalName); - } - vxworksReleaseFileId(pFile->pId); - pFile->pId = 0; + if( pFile->pId ){ + if( pFile->isDelete ){ + unlink(pFile->pId->zCanonicalName); } -#endif - OSTRACE(("CLOSE %-3d\n", pFile->h)); - OpenCounter(-1); - sqlite3_free(pFile->pUnused); - memset(pFile, 0, sizeof(unixFile)); + vxworksReleaseFileId(pFile->pId); + pFile->pId = 0; } +#endif + OSTRACE(("CLOSE %-3d\n", pFile->h)); + OpenCounter(-1); + sqlite3_free(pFile->pUnused); + memset(pFile, 0, sizeof(unixFile)); return SQLITE_OK; } @@ -1747,24 +1756,25 @@ static int closeUnixFile(sqlite3_file *id){ */ static int unixClose(sqlite3_file *id){ int rc = SQLITE_OK; - if( id ){ - unixFile *pFile = (unixFile *)id; - unixUnlock(id, NO_LOCK); - unixEnterMutex(); - assert( pFile->pInode==0 || pFile->pInode->nLock>0 - || pFile->pInode->bProcessLock==0 ); - if( pFile->pInode && pFile->pInode->nLock ){ - /* If there are outstanding locks, do not actually close the file just - ** yet because that would clear those locks. Instead, add the file - ** descriptor to pInode->pUnused list. It will be automatically closed - ** when the last lock is cleared. - */ - setPendingFd(pFile); - } - releaseInodeInfo(pFile); - rc = closeUnixFile(id); - unixLeaveMutex(); + unixFile *pFile = (unixFile *)id; + unixUnlock(id, NO_LOCK); + unixEnterMutex(); + + /* unixFile.pInode is always valid here. Otherwise, a different close + ** routine (e.g. nolockClose()) would be called instead. + */ + assert( pFile->pInode->nLock>0 || pFile->pInode->bProcessLock==0 ); + if( ALWAYS(pFile->pInode) && pFile->pInode->nLock ){ + /* If there are outstanding locks, do not actually close the file just + ** yet because that would clear those locks. Instead, add the file + ** descriptor to pInode->pUnused list. It will be automatically closed + ** when the last lock is cleared. + */ + setPendingFd(pFile); } + releaseInodeInfo(pFile); + rc = closeUnixFile(id); + unixLeaveMutex(); return rc; } @@ -1979,7 +1989,7 @@ static int dotlockUnlock(sqlite3_file *id, int eFileLock) { int rc = 0; int tErrno = errno; if( ENOENT != tErrno ){ - rc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); + rc = SQLITE_IOERR_UNLOCK; } if( IS_LOCK_ERROR(rc) ){ pFile->lastErrno = tErrno; @@ -2067,7 +2077,7 @@ static int flockCheckReservedLock(sqlite3_file *id, int *pResOut){ if ( lrc ) { int tErrno = errno; /* unlock failed with an error */ - lrc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); + lrc = SQLITE_IOERR_UNLOCK; if( IS_LOCK_ERROR(lrc) ){ pFile->lastErrno = tErrno; rc = lrc; @@ -2189,21 +2199,12 @@ static int flockUnlock(sqlite3_file *id, int eFileLock) { } /* no, really, unlock. */ - int rc = robust_flock(pFile->h, LOCK_UN); - if (rc) { - int r, tErrno = errno; - r = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_UNLOCK); - if( IS_LOCK_ERROR(r) ){ - pFile->lastErrno = tErrno; - } + if( robust_flock(pFile->h, LOCK_UN) ){ #ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS - if( (r & SQLITE_IOERR) == SQLITE_IOERR ){ - r = SQLITE_BUSY; - } + return SQLITE_OK; #endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */ - - return r; - } else { + return SQLITE_IOERR_UNLOCK; + }else{ pFile->eFileLock = NO_LOCK; return SQLITE_OK; } @@ -3009,6 +3010,7 @@ static int seekAndWrite(unixFile *id, i64 offset, const void *pBuf, int cnt){ do{ got = osPwrite64(id->h, pBuf, cnt, offset);}while( got<0 && errno==EINTR); #else newOffset = lseek(id->h, offset, SEEK_SET); + SimulateIOError( newOffset-- ); if( newOffset!=offset ){ if( newOffset == -1 ){ ((unixFile*)id)->lastErrno = errno; @@ -3377,12 +3379,16 @@ static int fcntlSizeHint(unixFile *pFile, i64 nByte){ nSize = ((nByte+pFile->szChunk-1) / pFile->szChunk) * pFile->szChunk; if( nSize>(i64)buf.st_size ){ + #if defined(HAVE_POSIX_FALLOCATE) && HAVE_POSIX_FALLOCATE - int rc; + /* The code below is handling the return value of osFallocate() + ** correctly. posix_fallocate() is defined to "returns zero on success, + ** or an error number on failure". See the manpage for details. */ + int err; do{ - rc = osFallocate(pFile->.h, buf.st_size, nSize-buf.st_size; - }while( rc<0 && errno=EINTR ); - if( rc ) return SQLITE_IOERR_WRITE; + err = osFallocate(pFile->h, buf.st_size, nSize-buf.st_size); + }while( err==EINTR ); + if( err ) return SQLITE_IOERR_WRITE; #else /* If the OS does not have posix_fallocate(), fake it. First use ** ftruncate() to set the file size, then write a single byte to @@ -5735,7 +5741,9 @@ static int proxyCreateUnixFile( } memset(pNew, 0, sizeof(unixFile)); pNew->openFlags = openFlags; + memset(&dummyVfs, 0, sizeof(dummyVfs)); dummyVfs.pAppData = (void*)&autolockIoFinder; + dummyVfs.zName = "dummy"; pUnused->fd = fd; pUnused->flags = openFlags; pNew->pUnused = pUnused; |