aboutsummaryrefslogtreecommitdiff
path: root/src/os_unix.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/os_unix.c')
-rw-r--r--src/os_unix.c112
1 files changed, 55 insertions, 57 deletions
diff --git a/src/os_unix.c b/src/os_unix.c
index c94c0c111..784bc3517 100644
--- a/src/os_unix.c
+++ b/src/os_unix.c
@@ -213,7 +213,7 @@
# endif
#else /* !SQLITE_WASI */
# ifndef HAVE_FCHMOD
-# define HAVE_FCHMOD
+# define HAVE_FCHMOD 1
# endif
#endif /* SQLITE_WASI */
@@ -284,6 +284,7 @@ struct unixFile {
#endif
#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
unsigned iBusyTimeout; /* Wait this many millisec on locks */
+ int bBlockOnConnect; /* True to block for SHARED locks */
#endif
#if OS_VXWORKS
struct vxworksFileId *pId; /* Unique file ID */
@@ -322,7 +323,7 @@ static pid_t randomnessPid = 0;
#define UNIXFILE_EXCL 0x01 /* Connections from one process only */
#define UNIXFILE_RDONLY 0x02 /* Connection is read only */
#define UNIXFILE_PERSIST_WAL 0x04 /* Persistent WAL mode */
-#ifndef SQLITE_DISABLE_DIRSYNC
+#if !defined(SQLITE_DISABLE_DIRSYNC) && !defined(_AIX)
# define UNIXFILE_DIRSYNC 0x08 /* Directory sync needed */
#else
# define UNIXFILE_DIRSYNC 0x00
@@ -1664,7 +1665,7 @@ static int unixFileLock(unixFile *pFile, struct flock *pLock){
if( (pFile->ctrlFlags & (UNIXFILE_EXCL|UNIXFILE_RDONLY))==UNIXFILE_EXCL ){
if( pInode->bProcessLock==0 ){
struct flock lock;
- assert( pInode->nLock==0 );
+ /* assert( pInode->nLock==0 ); <-- Not true if unix-excl READONLY used */
lock.l_whence = SEEK_SET;
lock.l_start = SHARED_FIRST;
lock.l_len = SHARED_SIZE;
@@ -1677,6 +1678,13 @@ static int unixFileLock(unixFile *pFile, struct flock *pLock){
rc = 0;
}
}else{
+#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+ if( pFile->bBlockOnConnect && pLock->l_type==F_RDLCK
+ && pLock->l_start==SHARED_FIRST && pLock->l_len==SHARED_SIZE
+ ){
+ rc = osFcntl(pFile->h, F_SETLKW, pLock);
+ }else
+#endif
rc = osSetPosixAdvisoryLock(pFile->h, pLock, pFile);
}
return rc;
@@ -2464,54 +2472,33 @@ static int robust_flock(int fd, int op){
** is set to SQLITE_OK unless an I/O error occurs during lock checking.
*/
static int flockCheckReservedLock(sqlite3_file *id, int *pResOut){
- int rc = SQLITE_OK;
- int reserved = 0;
+#ifdef SQLITE_DEBUG
unixFile *pFile = (unixFile*)id;
+#else
+ UNUSED_PARAMETER(id);
+#endif
SimulateIOError( return SQLITE_IOERR_CHECKRESERVEDLOCK; );
assert( pFile );
+ assert( pFile->eFileLock<=SHARED_LOCK );
- /* Check if a thread in this process holds such a lock */
- if( pFile->eFileLock>SHARED_LOCK ){
- reserved = 1;
- }
-
- /* Otherwise see if some other process holds it. */
- if( !reserved ){
- /* attempt to get the lock */
- int lrc = robust_flock(pFile->h, LOCK_EX | LOCK_NB);
- if( !lrc ){
- /* got the lock, unlock it */
- lrc = robust_flock(pFile->h, LOCK_UN);
- if ( lrc ) {
- int tErrno = errno;
- /* unlock failed with an error */
- lrc = SQLITE_IOERR_UNLOCK;
- storeLastErrno(pFile, tErrno);
- rc = lrc;
- }
- } else {
- int tErrno = errno;
- reserved = 1;
- /* someone else might have it reserved */
- lrc = sqliteErrorFromPosixError(tErrno, SQLITE_IOERR_LOCK);
- if( IS_LOCK_ERROR(lrc) ){
- storeLastErrno(pFile, tErrno);
- rc = lrc;
- }
- }
- }
- OSTRACE(("TEST WR-LOCK %d %d %d (flock)\n", pFile->h, rc, reserved));
+ /* The flock VFS only ever takes exclusive locks (see function flockLock).
+ ** Therefore, if this connection is holding any lock at all, no other
+ ** connection may be holding a RESERVED lock. So set *pResOut to 0
+ ** in this case.
+ **
+ ** Or, this connection may be holding no lock. In that case, set *pResOut to
+ ** 0 as well. The caller will then attempt to take an EXCLUSIVE lock on the
+ ** db in order to roll the hot journal back. If there is another connection
+ ** holding a lock, that attempt will fail and an SQLITE_BUSY returned to
+ ** the user. With other VFS, we try to avoid this, in order to allow a reader
+ ** to proceed while a writer is preparing its transaction. But that won't
+ ** work with the flock VFS - as it always takes EXCLUSIVE locks - so it is
+ ** not a problem in this case. */
+ *pResOut = 0;
-#ifdef SQLITE_IGNORE_FLOCK_LOCK_ERRORS
- if( (rc & 0xff) == SQLITE_IOERR ){
- rc = SQLITE_OK;
- reserved=1;
- }
-#endif /* SQLITE_IGNORE_FLOCK_LOCK_ERRORS */
- *pResOut = reserved;
- return rc;
+ return SQLITE_OK;
}
/*
@@ -4008,6 +3995,11 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
}
#endif /* __linux__ && SQLITE_ENABLE_BATCH_ATOMIC_WRITE */
+ case SQLITE_FCNTL_NULL_IO: {
+ osClose(pFile->h);
+ pFile->h = -1;
+ return SQLITE_OK;
+ }
case SQLITE_FCNTL_LOCKSTATE: {
*(int*)pArg = pFile->eFileLock;
return SQLITE_OK;
@@ -4054,8 +4046,9 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
case SQLITE_FCNTL_LOCK_TIMEOUT: {
int iOld = pFile->iBusyTimeout;
+ int iNew = *(int*)pArg;
#if SQLITE_ENABLE_SETLK_TIMEOUT==1
- pFile->iBusyTimeout = *(int*)pArg;
+ pFile->iBusyTimeout = iNew<0 ? 0x7FFFFFFF : (unsigned)iNew;
#elif SQLITE_ENABLE_SETLK_TIMEOUT==2
pFile->iBusyTimeout = !!(*(int*)pArg);
#else
@@ -4064,7 +4057,12 @@ static int unixFileControl(sqlite3_file *id, int op, void *pArg){
*(int*)pArg = iOld;
return SQLITE_OK;
}
-#endif
+ case SQLITE_FCNTL_BLOCK_ON_CONNECT: {
+ int iNew = *(int*)pArg;
+ pFile->bBlockOnConnect = iNew;
+ return SQLITE_OK;
+ }
+#endif /* SQLITE_ENABLE_SETLK_TIMEOUT */
#if SQLITE_MAX_MMAP_SIZE>0
case SQLITE_FCNTL_MMAP_SIZE: {
i64 newLimit = *(i64*)pArg;
@@ -4149,6 +4147,7 @@ static void setDeviceCharacteristics(unixFile *pFd){
if( pFd->ctrlFlags & UNIXFILE_PSOW ){
pFd->deviceCharacteristics |= SQLITE_IOCAP_POWERSAFE_OVERWRITE;
}
+ pFd->deviceCharacteristics |= SQLITE_IOCAP_SUBPAGE_READ;
pFd->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE;
}
@@ -4159,7 +4158,7 @@ static void setDeviceCharacteristics(unixFile *pFd){
static void setDeviceCharacteristics(unixFile *pFile){
if( pFile->sectorSize == 0 ){
struct statvfs fsInfo;
-
+
/* Set defaults for non-supported filesystems */
pFile->sectorSize = SQLITE_DEFAULT_SECTOR_SIZE;
pFile->deviceCharacteristics = 0;
@@ -4199,7 +4198,7 @@ static void setDeviceCharacteristics(unixFile *pFile){
pFile->sectorSize = fsInfo.f_bsize;
pFile->deviceCharacteristics =
/* full bitset of atomics from max sector size and smaller */
- ((pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) << 1) - 2 |
+ (((pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) << 1) - 2) |
SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind
** so it is ordered */
0;
@@ -4207,7 +4206,7 @@ static void setDeviceCharacteristics(unixFile *pFile){
pFile->sectorSize = fsInfo.f_bsize;
pFile->deviceCharacteristics =
/* full bitset of atomics from max sector size and smaller */
- ((pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) << 1) - 2 |
+ (((pFile->sectorSize / 512 * SQLITE_IOCAP_ATOMIC512) << 1) - 2) |
SQLITE_IOCAP_SEQUENTIAL | /* The ram filesystem has no write behind
** so it is ordered */
0;
@@ -5036,21 +5035,20 @@ static int unixShmLock(
/* 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. Write lock (ofst==0).
- ** 3. Read locks (ofst>=3 && ofst<SQLITE_SHM_NLOCK).
+ ** 1. Recovery lock (ofst==2).
+ ** 2. Checkpointer lock (ofst==1).
+ ** 3. Write lock (ofst==0).
+ ** 4. Read locks (ofst>=3 && ofst<SQLITE_SHM_NLOCK).
**
** 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.
- **
- ** It is not permitted to block on the RECOVER lock.
*/
-#ifdef SQLITE_ENABLE_SETLK_TIMEOUT
+#if defined(SQLITE_ENABLE_SETLK_TIMEOUT) && defined(SQLITE_DEBUG)
{
u16 lockMask = (p->exclMask|p->sharedMask);
assert( (flags & SQLITE_SHM_UNLOCK) || pDbFd->iBusyTimeout==0 || (
- (ofst!=2) /* not RECOVER */
+ (ofst!=2 || lockMask==0)
&& (ofst!=1 || lockMask==0 || lockMask==2)
&& (ofst!=0 || lockMask<3)
&& (ofst<3 || lockMask<(1<<ofst))
@@ -6855,7 +6853,7 @@ static int unixSleep(sqlite3_vfs *NotUsed, int microseconds){
/* Almost all modern unix systems support nanosleep(). But if you are
** compiling for one of the rare exceptions, you can use
- ** -DHAVE_NANOSLEEP=0 (perhaps in conjuction with -DHAVE_USLEEP if
+ ** -DHAVE_NANOSLEEP=0 (perhaps in conjunction with -DHAVE_USLEEP if
** usleep() is available) in order to bypass the use of nanosleep() */
nanosleep(&sp, NULL);
generated by cgit v1.2.3 (git 2.25.1) at 2025-07-18 12:45:56 +0000