note that a len of zero means lock to end of file
*/
-int tdb_brlock_len(struct tdb_context *tdb, tdb_off_t offset,
- int rw_type, int lck_type, int probe, size_t len)
+int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset,
+ int rw_type, int lck_type, int probe, size_t len)
{
struct flock fl;
int ret;
int count = 1000;
while (count--) {
struct timeval tv;
- if (tdb_brlock_len(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
+ if (tdb_brlock(tdb, offset, F_WRLCK, F_SETLKW, 1, len) == 0) {
return 0;
}
if (errno != EDEADLK) {
}
-/* a byte range locking function - return 0 on success
- this functions locks/unlocks 1 byte at the specified offset.
-
- On error, errno is also set so that errors are passed back properly
- through tdb_open(). */
-int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset,
- int rw_type, int lck_type, int probe)
-{
- return tdb_brlock_len(tdb, offset, rw_type, lck_type, probe, 1);
-}
-
/* lock a list in the database. list -1 is the alloc list */
int tdb_lock(struct tdb_context *tdb, int list, int ltype)
{
+ /* a global lock allows us to avoid per chain locks */
+ if (tdb->global_lock.count &&
+ (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
+ return 0;
+ }
+
+ if (tdb->global_lock.count) {
+ return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+ }
+
if (list < -1 || list >= (int)tdb->header.hash_size) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid list %d for ltype=%d\n",
list, ltype));
/* Since fcntl locks don't nest, we do a lock for the first one,
and simply bump the count for future ones */
if (tdb->locked[list+1].count == 0) {
- if (tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0)) {
+ if (tdb->methods->tdb_brlock(tdb,FREELIST_TOP+4*list,ltype,F_SETLKW, 0, 1)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d ltype=%d (%s)\n",
list, ltype, strerror(errno)));
return -1;
{
int ret = -1;
+ /* a global lock allows us to avoid per chain locks */
+ if (tdb->global_lock.count &&
+ (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
+ return 0;
+ }
+
+ if (tdb->global_lock.count) {
+ return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+ }
+
if (tdb->flags & TDB_NOLOCK)
return 0;
if (tdb->locked[list+1].count == 1) {
/* Down to last nested lock: unlock underneath */
- ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0);
+ ret = tdb->methods->tdb_brlock(tdb, FREELIST_TOP+4*list, F_UNLCK, F_SETLKW, 0, 1);
tdb->num_locks--;
} else {
ret = 0;
/* lock/unlock entire database */
-int tdb_lockall(struct tdb_context *tdb)
+static int _tdb_lockall(struct tdb_context *tdb, int ltype)
{
- u32 i;
-
/* There are no locks on read-only dbs */
if (tdb->read_only || tdb->traverse_read)
return TDB_ERRCODE(TDB_ERR_LOCK, -1);
- for (i = 0; i < tdb->header.hash_size; i++)
- if (tdb_lock(tdb, i, F_WRLCK))
- break;
- /* If error, release locks we have... */
- if (i < tdb->header.hash_size) {
- u32 j;
+ if (tdb->global_lock.count && tdb->global_lock.ltype == ltype) {
+ tdb->global_lock.count++;
+ return 0;
+ }
+
+ if (tdb->global_lock.count) {
+ /* a global lock of a different type exists */
+ return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+ }
+
+ if (tdb->num_locks != 0) {
+ /* can't combine global and chain locks */
+ return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+ }
+
+ if (tdb->methods->tdb_brlock(tdb, FREELIST_TOP, ltype, F_SETLKW,
+ 0, 4*tdb->header.hash_size)) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
+ return -1;
+ }
+
+ tdb->global_lock.count = 1;
+ tdb->global_lock.ltype = ltype;
+
+ return 0;
+}
+
+/* unlock entire db */
+static int _tdb_unlockall(struct tdb_context *tdb, int ltype)
+{
+ /* There are no locks on read-only dbs */
+ if (tdb->read_only || tdb->traverse_read) {
+ return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+ }
- for ( j = 0; j < i; j++)
- tdb_unlock(tdb, j, F_WRLCK);
- return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
+ if (tdb->global_lock.ltype != ltype || tdb->global_lock.count == 0) {
+ return TDB_ERRCODE(TDB_ERR_LOCK, -1);
+ }
+
+ if (tdb->global_lock.count > 1) {
+ tdb->global_lock.count--;
+ return 0;
}
+ if (tdb->methods->tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW,
+ 0, 4*tdb->header.hash_size)) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
+ return -1;
+ }
+
+ tdb->global_lock.count = 0;
+ tdb->global_lock.ltype = 0;
+
return 0;
}
-void tdb_unlockall(struct tdb_context *tdb)
+
+/* lock entire database with write lock */
+int tdb_lockall(struct tdb_context *tdb)
+{
+ return _tdb_lockall(tdb, F_WRLCK);
+}
+
+/* unlock entire database with write lock */
+int tdb_unlockall(struct tdb_context *tdb)
+{
+ return _tdb_unlockall(tdb, F_WRLCK);
+}
+
+/* lock entire database with read lock */
+int tdb_lockall_read(struct tdb_context *tdb)
+{
+ return _tdb_lockall(tdb, F_RDLCK);
+}
+
+/* unlock entire database with read lock */
+int tdb_unlockall_read(struct tdb_context *tdb)
{
- u32 i;
- for (i=0; i < tdb->header.hash_size; i++)
- tdb_unlock(tdb, i, F_WRLCK);
+ return _tdb_unlockall(tdb, F_RDLCK);
}
/* lock/unlock one hash chain. This is meant to be used to reduce
/* record lock stops delete underneath */
int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
{
- return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0) : 0;
+ return off ? tdb->methods->tdb_brlock(tdb, off, F_RDLCK, F_SETLKW, 0, 1) : 0;
}
/*
for (i = &tdb->travlocks; i; i = i->next)
if (i->off == off)
return -1;
- return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1);
+ return tdb->methods->tdb_brlock(tdb, off, F_WRLCK, F_SETLK, 1, 1);
}
/*
*/
int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
{
- return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0);
+ return tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLK, 0, 1);
}
/* fcntl locks don't stack: avoid unlocking someone else's */
for (i = &tdb->travlocks; i; i = i->next)
if (i->off == off)
count++;
- return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0) : 0);
+ return (count == 1 ? tdb->methods->tdb_brlock(tdb, off, F_UNLCK, F_SETLKW, 0, 1) : 0);
}
}
/* ensure there is only one process initialising at once */
- if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
+ if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get global lock on %s: %s\n",
name, strerror(errno)));
goto fail; /* errno set by tdb_brlock */
/* we need to zero database if we are the only one with it open */
if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
- (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))) {
+ (locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
open_flags |= O_CREAT;
if (ftruncate(tdb->fd, 0) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
}
tdb_mmap(tdb);
if (locked) {
- if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
+ if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
"failed to take ACTIVE_LOCK on %s: %s\n",
name, strerror(errno)));
if (tdb_flags & TDB_CLEAR_IF_FIRST) {
/* leave this lock in place to indicate it's in use */
- if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
+ if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
goto fail;
}
/* Internal (memory-only) databases skip all the code above to
* do with disk files, and resume here by releasing their
* global lock and hooking into the active list. */
- if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
+ if (tdb->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -1)
goto fail;
tdb->next = tdbs;
tdbs = tdb;
return 0; /* Nothing to do. */
}
- if (tdb->num_locks != 0) {
+ if (tdb->num_locks != 0 || tdb->global_lock.count) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
goto fail;
}
goto fail;
}
if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
- (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)) {
+ (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
goto fail;
}
return;
}
- if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1) != 0) {
+ if (tdb_brlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, F_SETLKW, 1, 1) != 0) {
return;
}
seqnum++;
tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
- tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1);
+ tdb_brlock(tdb, TDB_SEQNUM_OFS, F_UNLCK, F_SETLKW, 1, 1);
}
void (*next_hash_chain)(struct tdb_context *, u32 *);
int (*tdb_oob)(struct tdb_context *, tdb_off_t , int );
int (*tdb_expand_file)(struct tdb_context *, tdb_off_t , tdb_off_t );
- int (*tdb_brlock)(struct tdb_context *, tdb_off_t , int, int, int);
+ int (*tdb_brlock)(struct tdb_context *, tdb_off_t , int, int, int, size_t);
};
struct tdb_context {
tdb_len_t map_size; /* how much space has been mapped */
int read_only; /* opened read-only */
int traverse_read; /* read-only traversal */
+ struct tdb_lock_type global_lock;
struct tdb_lock_type *locked; /* array of chain locks */
enum TDB_ERROR ecode; /* error code for last tdb error */
struct tdb_header header; /* a cached copy of the header */
void tdb_mmap(struct tdb_context *tdb);
int tdb_lock(struct tdb_context *tdb, int list, int ltype);
int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
-int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset, int rw_type, int lck_type, int probe);
+int tdb_brlock(struct tdb_context *tdb, tdb_off_t offset, int rw_type, int lck_type, int probe, size_t len);
int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len);
-int tdb_brlock_len(struct tdb_context *tdb, tdb_off_t offset,
- int rw_type, int lck_type, int probe, size_t len);
int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
brlock during a transaction - ignore them
*/
int transaction_brlock(struct tdb_context *tdb, tdb_off_t offset,
- int rw_type, int lck_type, int probe)
+ int rw_type, int lck_type, int probe, size_t len)
{
return 0;
}
return 0;
}
- if (tdb->num_locks != 0) {
+ if (tdb->num_locks != 0 || tdb->global_lock.count) {
/* the caller must not have any locks when starting a
transaction as otherwise we'll be screwed by lack
of nested locks in posix */
/* get the transaction write lock. This is a blocking lock. As
discussed with Volker, there are a number of ways we could
make this async, which we will probably do in the future */
- if (tdb_brlock_len(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+ if (tdb_brlock(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get transaction lock\n"));
tdb->ecode = TDB_ERR_LOCK;
SAFE_FREE(tdb->transaction);
/* get a read lock from the freelist to the end of file. This
is upgraded to a write lock during the commit */
- if (tdb_brlock_len(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
+ if (tdb_brlock(tdb, FREELIST_TOP, F_RDLCK, F_SETLKW, 0, 0) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
tdb->ecode = TDB_ERR_LOCK;
goto fail;
return 0;
fail:
- tdb_brlock_len(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
- tdb_brlock_len(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+ tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
SAFE_FREE(tdb->transaction->hash_heads);
SAFE_FREE(tdb->transaction);
return -1;
free(el);
}
+ /* remove any global lock created during the transaction */
+ if (tdb->global_lock.count != 0) {
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 4*tdb->header.hash_size);
+ tdb->global_lock.count = 0;
+ }
+
/* remove any locks created during the transaction */
if (tdb->num_locks != 0) {
int h;
for (h=0;h<tdb->header.hash_size+1;h++) {
if (tdb->locked[h].count != 0) {
- tdb_brlock_len(tdb,FREELIST_TOP+4*h,F_UNLCK,F_SETLKW, 0, 1);
+ tdb_brlock(tdb,FREELIST_TOP+4*h,F_UNLCK,F_SETLKW, 0, 1);
tdb->locked[h].count = 0;
}
}
/* restore the normal io methods */
tdb->methods = tdb->transaction->io_methods;
- tdb_brlock_len(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
- tdb_brlock_len(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, FREELIST_TOP, F_UNLCK, F_SETLKW, 0, 0);
+ tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
SAFE_FREE(tdb->transaction->hash_heads);
SAFE_FREE(tdb->transaction);
/* if there are any locks pending then the caller has not
nested their locks properly, so fail the transaction */
- if (tdb->num_locks) {
+ if (tdb->num_locks || tdb->global_lock.count) {
tdb->ecode = TDB_ERR_LOCK;
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: locks pending on commit\n"));
tdb_transaction_cancel(tdb);
/* get the global lock - this prevents new users attaching to the database
during the commit */
- if (tdb_brlock_len(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
+ if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: failed to get global lock\n"));
tdb->ecode = TDB_ERR_LOCK;
tdb_transaction_cancel(tdb);
/* write the recovery data to the end of the file */
if (transaction_setup_recovery(tdb, &magic_offset) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: failed to setup recovery data\n"));
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
tdb_transaction_cancel(tdb);
return -1;
}
tdb->transaction->old_map_size) == -1) {
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: expansion failed\n"));
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
tdb_transaction_cancel(tdb);
return -1;
}
tdb_transaction_recover(tdb);
tdb_transaction_cancel(tdb);
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
return -1;
}
}
- tdb_brlock_len(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
+ tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1);
/*
TODO: maybe write to some dummy hdr field, or write to magic
/* we need to get a read lock on the transaction lock here to
cope with the lock ordering semantics of solaris10 */
- if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_RDLCK, F_SETLKW, 0) == -1) {
+ if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_traverse_read: failed to get transaction lock\n"));
tdb->ecode = TDB_ERR_LOCK;
return -1;
ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
tdb->traverse_read--;
- tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0);
+ tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
return ret;
}
return tdb_traverse_read(tdb, fn, private_data);
}
- if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
+ if (tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_traverse: failed to get transaction lock\n"));
tdb->ecode = TDB_ERR_LOCK;
return -1;
ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
- tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0);
+ tdb->methods->tdb_brlock(tdb, TRANSACTION_LOCK, F_UNLCK, F_SETLKW, 0, 1);
return ret;
}
int tdb_traverse_read(struct tdb_context *tdb, tdb_traverse_func fn, void *);
int tdb_exists(struct tdb_context *tdb, TDB_DATA key);
int tdb_lockall(struct tdb_context *tdb);
-void tdb_unlockall(struct tdb_context *tdb);
+int tdb_unlockall(struct tdb_context *tdb);
+int tdb_lockall_read(struct tdb_context *tdb);
+int tdb_unlockall_read(struct tdb_context *tdb);
const char *tdb_name(struct tdb_context *tdb);
int tdb_fd(struct tdb_context *tdb);
tdb_log_func tdb_log_fn(struct tdb_context *tdb);