2 Unix SMB/CIFS implementation.
4 trivial database library
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
14 This library is free software; you can redistribute it and/or
15 modify it under the terms of the GNU Lesser General Public
16 License as published by the Free Software Foundation; either
17 version 3 of the License, or (at your option) any later version.
19 This library is distributed in the hope that it will be useful,
20 but WITHOUT ANY WARRANTY; without even the implied warranty of
21 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
22 Lesser General Public License for more details.
24 You should have received a copy of the GNU Lesser General Public
25 License along with this library; if not, see <http://www.gnu.org/licenses/>.
28 #include "tdb_private.h"
30 _PUBLIC_
void tdb_setalarm_sigptr(struct tdb_context
*tdb
, volatile sig_atomic_t *ptr
)
32 tdb
->interrupt_sig_ptr
= ptr
;
35 static int fcntl_lock(struct tdb_context
*tdb
,
36 int rw
, off_t off
, off_t len
, bool waitflag
)
42 fl
.l_whence
= SEEK_SET
;
47 cmd
= waitflag
? F_SETLKW
: F_SETLK
;
49 return fcntl(tdb
->fd
, cmd
, &fl
);
52 static int fcntl_unlock(struct tdb_context
*tdb
, int rw
, off_t off
, off_t len
)
55 #if 0 /* Check they matched up locks and unlocks correctly. */
60 locks
= fopen("/proc/locks", "r");
62 while (fgets(line
, 80, locks
)) {
66 /* eg. 1: FLOCK ADVISORY WRITE 2440 08:01:2180826 0 EOF */
67 p
= strchr(line
, ':') + 1;
68 if (strncmp(p
, " POSIX ADVISORY ", strlen(" POSIX ADVISORY ")))
70 p
+= strlen(" FLOCK ADVISORY ");
71 if (strncmp(p
, "READ ", strlen("READ ")) == 0)
73 else if (strncmp(p
, "WRITE ", strlen("WRITE ")) == 0)
78 if (atoi(p
) != getpid())
80 p
= strchr(strchr(p
, ' ') + 1, ' ') + 1;
82 p
= strchr(p
, ' ') + 1;
83 if (strncmp(p
, "EOF", 3) == 0)
86 l
= atoi(p
) - start
+ 1;
90 fprintf(stderr
, "Len %u should be %u: %s",
95 fprintf(stderr
, "Type %s wrong: %s",
96 rw
== F_RDLCK
? "READ" : "WRITE", line
);
105 fprintf(stderr
, "Unlock on %u@%u not found!\n",
114 fl
.l_whence
= SEEK_SET
;
119 return fcntl(tdb
->fd
, F_SETLKW
, &fl
);
122 /* list -1 is the alloc list, otherwise a hash chain. */
123 static tdb_off_t
lock_offset(int list
)
125 return FREELIST_TOP
+ 4*list
;
128 /* a byte range locking function - return 0 on success
129 this functions locks/unlocks "len" byte at the specified offset.
131 On error, errno is also set so that errors are passed back properly
134 note that a len of zero means lock to end of file
136 int tdb_brlock(struct tdb_context
*tdb
,
137 int rw_type
, tdb_off_t offset
, size_t len
,
138 enum tdb_lock_flags flags
)
142 if (tdb
->flags
& TDB_NOLOCK
) {
146 if (flags
& TDB_LOCK_MARK_ONLY
) {
150 if ((rw_type
== F_WRLCK
) && (tdb
->read_only
|| tdb
->traverse_read
)) {
151 tdb
->ecode
= TDB_ERR_RDONLY
;
156 ret
= fcntl_lock(tdb
, rw_type
, offset
, len
,
157 flags
& TDB_LOCK_WAIT
);
158 /* Check for a sigalarm break. */
159 if (ret
== -1 && errno
== EINTR
&&
160 tdb
->interrupt_sig_ptr
&&
161 *tdb
->interrupt_sig_ptr
) {
164 } while (ret
== -1 && errno
== EINTR
);
167 tdb
->ecode
= TDB_ERR_LOCK
;
168 /* Generic lock error. errno set by fcntl.
169 * EAGAIN is an expected return from non-blocking
171 if (!(flags
& TDB_LOCK_PROBE
) && errno
!= EAGAIN
) {
172 TDB_LOG((tdb
, TDB_DEBUG_TRACE
,"tdb_brlock failed (fd=%d) at offset %u rw_type=%d flags=%d len=%zu\n",
173 tdb
->fd
, offset
, rw_type
, flags
, len
));
180 int tdb_brunlock(struct tdb_context
*tdb
,
181 int rw_type
, tdb_off_t offset
, size_t len
)
185 if (tdb
->flags
& TDB_NOLOCK
) {
190 ret
= fcntl_unlock(tdb
, rw_type
, offset
, len
);
191 } while (ret
== -1 && errno
== EINTR
);
194 TDB_LOG((tdb
, TDB_DEBUG_TRACE
,"tdb_brunlock failed (fd=%d) at offset %u rw_type=%u len=%zu\n",
195 tdb
->fd
, offset
, rw_type
, len
));
201 * Do a tdb_brlock in a loop. Some OSes (such as solaris) have too
202 * conservative deadlock detection and claim a deadlock when progress can be
203 * made. For those OSes we may loop for a while.
206 static int tdb_brlock_retry(struct tdb_context
*tdb
,
207 int rw_type
, tdb_off_t offset
, size_t len
,
208 enum tdb_lock_flags flags
)
216 ret
= tdb_brlock(tdb
, rw_type
, offset
, len
, flags
);
220 if (errno
!= EDEADLK
) {
223 /* sleep for as short a time as we can - more portable than usleep() */
226 select(0, NULL
, NULL
, NULL
, &tv
);
232 upgrade a read lock to a write lock.
234 int tdb_allrecord_upgrade(struct tdb_context
*tdb
)
238 if (tdb
->allrecord_lock
.count
!= 1) {
239 TDB_LOG((tdb
, TDB_DEBUG_ERROR
,
240 "tdb_allrecord_upgrade failed: count %u too high\n",
241 tdb
->allrecord_lock
.count
));
245 if (tdb
->allrecord_lock
.off
!= 1) {
246 TDB_LOG((tdb
, TDB_DEBUG_ERROR
,
247 "tdb_allrecord_upgrade failed: already upgraded?\n"));
251 ret
= tdb_brlock_retry(tdb
, F_WRLCK
, FREELIST_TOP
, 0,
252 TDB_LOCK_WAIT
|TDB_LOCK_PROBE
);
254 tdb
->allrecord_lock
.ltype
= F_WRLCK
;
255 tdb
->allrecord_lock
.off
= 0;
258 TDB_LOG((tdb
, TDB_DEBUG_TRACE
,"tdb_allrecord_upgrade failed\n"));
262 static struct tdb_lock_type
*find_nestlock(struct tdb_context
*tdb
,
267 for (i
=0; i
<tdb
->num_lockrecs
; i
++) {
268 if (tdb
->lockrecs
[i
].off
== offset
) {
269 return &tdb
->lockrecs
[i
];
275 /* lock an offset in the database. */
276 int tdb_nest_lock(struct tdb_context
*tdb
, uint32_t offset
, int ltype
,
277 enum tdb_lock_flags flags
)
279 struct tdb_lock_type
*new_lck
;
281 if (offset
>= lock_offset(tdb
->hash_size
)) {
282 tdb
->ecode
= TDB_ERR_LOCK
;
283 TDB_LOG((tdb
, TDB_DEBUG_ERROR
,"tdb_lock: invalid offset %u for ltype=%d\n",
287 if (tdb
->flags
& TDB_NOLOCK
)
290 new_lck
= find_nestlock(tdb
, offset
);
293 * Just increment the in-memory struct, posix locks
300 if (tdb
->num_lockrecs
== tdb
->lockrecs_array_length
) {
301 new_lck
= (struct tdb_lock_type
*)realloc(
303 sizeof(*tdb
->lockrecs
) * (tdb
->num_lockrecs
+1));
304 if (new_lck
== NULL
) {
308 tdb
->lockrecs_array_length
= tdb
->num_lockrecs
+1;
309 tdb
->lockrecs
= new_lck
;
312 /* Since fcntl locks don't nest, we do a lock for the first one,
313 and simply bump the count for future ones */
314 if (tdb_brlock(tdb
, ltype
, offset
, 1, flags
)) {
318 new_lck
= &tdb
->lockrecs
[tdb
->num_lockrecs
];
320 new_lck
->off
= offset
;
322 new_lck
->ltype
= ltype
;
328 static int tdb_lock_and_recover(struct tdb_context
*tdb
)
332 /* We need to match locking order in transaction commit. */
333 if (tdb_brlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0, TDB_LOCK_WAIT
)) {
337 if (tdb_brlock(tdb
, F_WRLCK
, OPEN_LOCK
, 1, TDB_LOCK_WAIT
)) {
338 tdb_brunlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0);
342 ret
= tdb_transaction_recover(tdb
);
344 tdb_brunlock(tdb
, F_WRLCK
, OPEN_LOCK
, 1);
345 tdb_brunlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0);
350 static bool have_data_locks(const struct tdb_context
*tdb
)
354 for (i
= 0; i
< tdb
->num_lockrecs
; i
++) {
355 if (tdb
->lockrecs
[i
].off
>= lock_offset(-1))
362 * A allrecord lock allows us to avoid per chain locks. Check if the allrecord
363 * lock is strong enough.
365 static int tdb_lock_covered_by_allrecord_lock(struct tdb_context
*tdb
,
368 if (ltype
== F_RDLCK
) {
370 * The allrecord_lock is equal (F_RDLCK) or stronger
376 if (tdb
->allrecord_lock
.ltype
== F_RDLCK
) {
378 * We ask for ltype==F_WRLCK, but the allrecord_lock
379 * is too weak. We can't upgrade here, so fail.
381 tdb
->ecode
= TDB_ERR_LOCK
;
386 * Asking for F_WRLCK, allrecord is F_WRLCK as well. Pass.
391 static int tdb_lock_list(struct tdb_context
*tdb
, int list
, int ltype
,
392 enum tdb_lock_flags waitflag
)
397 if (tdb
->allrecord_lock
.count
) {
398 return tdb_lock_covered_by_allrecord_lock(tdb
, ltype
);
402 * Check for recoveries: Someone might have kill -9'ed a process
405 check
= !have_data_locks(tdb
);
406 ret
= tdb_nest_lock(tdb
, lock_offset(list
), ltype
, waitflag
);
408 if (ret
== 0 && check
&& tdb_needs_recovery(tdb
)) {
409 tdb_nest_unlock(tdb
, lock_offset(list
), ltype
, false);
411 if (tdb_lock_and_recover(tdb
) == -1) {
414 return tdb_lock_list(tdb
, list
, ltype
, waitflag
);
419 /* lock a list in the database. list -1 is the alloc list */
420 int tdb_lock(struct tdb_context
*tdb
, int list
, int ltype
)
424 ret
= tdb_lock_list(tdb
, list
, ltype
, TDB_LOCK_WAIT
);
426 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_lock failed on list %d "
427 "ltype=%d (%s)\n", list
, ltype
, strerror(errno
)));
432 /* lock a list in the database. list -1 is the alloc list. non-blocking lock */
433 _PUBLIC_
int tdb_lock_nonblock(struct tdb_context
*tdb
, int list
, int ltype
)
435 return tdb_lock_list(tdb
, list
, ltype
, TDB_LOCK_NOWAIT
);
439 int tdb_nest_unlock(struct tdb_context
*tdb
, uint32_t offset
, int ltype
,
443 struct tdb_lock_type
*lck
;
445 if (tdb
->flags
& TDB_NOLOCK
)
449 if (offset
>= lock_offset(tdb
->hash_size
)) {
450 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlock: offset %u invalid (%d)\n", offset
, tdb
->hash_size
));
454 lck
= find_nestlock(tdb
, offset
);
455 if ((lck
== NULL
) || (lck
->count
== 0)) {
456 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlock: count is 0\n"));
460 if (lck
->count
> 1) {
466 * This lock has count==1 left, so we need to unlock it in the
467 * kernel. We don't bother with decrementing the in-memory array
468 * element, we're about to overwrite it with the last array element
475 ret
= tdb_brunlock(tdb
, ltype
, offset
, 1);
479 * Shrink the array by overwriting the element just unlocked with the
480 * last array element.
482 *lck
= tdb
->lockrecs
[--tdb
->num_lockrecs
];
485 * We don't bother with realloc when the array shrinks, but if we have
486 * a completely idle tdb we should get rid of the locked array.
490 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlock: An error occurred unlocking!\n"));
494 _PUBLIC_
int tdb_unlock(struct tdb_context
*tdb
, int list
, int ltype
)
496 /* a global lock allows us to avoid per chain locks */
497 if (tdb
->allrecord_lock
.count
) {
498 return tdb_lock_covered_by_allrecord_lock(tdb
, ltype
);
501 return tdb_nest_unlock(tdb
, lock_offset(list
), ltype
, false);
505 get the transaction lock
507 int tdb_transaction_lock(struct tdb_context
*tdb
, int ltype
,
508 enum tdb_lock_flags lockflags
)
510 return tdb_nest_lock(tdb
, TRANSACTION_LOCK
, ltype
, lockflags
);
514 release the transaction lock
516 int tdb_transaction_unlock(struct tdb_context
*tdb
, int ltype
)
518 return tdb_nest_unlock(tdb
, TRANSACTION_LOCK
, ltype
, false);
521 /* Returns 0 if all done, -1 if error, 1 if ok. */
522 static int tdb_allrecord_check(struct tdb_context
*tdb
, int ltype
,
523 enum tdb_lock_flags flags
, bool upgradable
)
525 /* There are no locks on read-only dbs */
526 if (tdb
->read_only
|| tdb
->traverse_read
) {
527 tdb
->ecode
= TDB_ERR_LOCK
;
531 if (tdb
->allrecord_lock
.count
&& tdb
->allrecord_lock
.ltype
== ltype
) {
532 tdb
->allrecord_lock
.count
++;
536 if (tdb
->allrecord_lock
.count
) {
537 /* a global lock of a different type exists */
538 tdb
->ecode
= TDB_ERR_LOCK
;
542 if (tdb_have_extra_locks(tdb
)) {
543 /* can't combine global and chain locks */
544 tdb
->ecode
= TDB_ERR_LOCK
;
548 if (upgradable
&& ltype
!= F_RDLCK
) {
549 /* tdb error: you can't upgrade a write lock! */
550 tdb
->ecode
= TDB_ERR_LOCK
;
556 /* We only need to lock individual bytes, but Linux merges consecutive locks
557 * so we lock in contiguous ranges. */
558 static int tdb_chainlock_gradual(struct tdb_context
*tdb
,
559 int ltype
, enum tdb_lock_flags flags
,
560 size_t off
, size_t len
)
563 enum tdb_lock_flags nb_flags
= (flags
& ~TDB_LOCK_WAIT
);
566 /* Single record. Just do blocking lock. */
567 return tdb_brlock(tdb
, ltype
, off
, len
, flags
);
570 /* First we try non-blocking. */
571 ret
= tdb_brlock(tdb
, ltype
, off
, len
, nb_flags
);
576 /* Try locking first half, then second. */
577 ret
= tdb_chainlock_gradual(tdb
, ltype
, flags
, off
, len
/ 2);
581 ret
= tdb_chainlock_gradual(tdb
, ltype
, flags
,
582 off
+ len
/ 2, len
- len
/ 2);
584 tdb_brunlock(tdb
, ltype
, off
, len
/ 2);
590 /* lock/unlock entire database. It can only be upgradable if you have some
591 * other way of guaranteeing exclusivity (ie. transaction write lock).
592 * We do the locking gradually to avoid being starved by smaller locks. */
593 int tdb_allrecord_lock(struct tdb_context
*tdb
, int ltype
,
594 enum tdb_lock_flags flags
, bool upgradable
)
596 switch (tdb_allrecord_check(tdb
, ltype
, flags
, upgradable
)) {
603 /* We cover two kinds of locks:
604 * 1) Normal chain locks. Taken for almost all operations.
605 * 2) Individual records locks. Taken after normal or free
608 * It is (1) which cause the starvation problem, so we're only
609 * gradual for that. */
610 if (tdb_chainlock_gradual(tdb
, ltype
, flags
, FREELIST_TOP
,
611 tdb
->hash_size
* 4) == -1) {
615 /* Grab individual record locks. */
616 if (tdb_brlock(tdb
, ltype
, lock_offset(tdb
->hash_size
), 0,
618 tdb_brunlock(tdb
, ltype
, FREELIST_TOP
,
623 tdb
->allrecord_lock
.count
= 1;
624 /* If it's upgradable, it's actually exclusive so we can treat
625 * it as a write lock. */
626 tdb
->allrecord_lock
.ltype
= upgradable
? F_WRLCK
: ltype
;
627 tdb
->allrecord_lock
.off
= upgradable
;
629 if (tdb_needs_recovery(tdb
)) {
630 bool mark
= flags
& TDB_LOCK_MARK_ONLY
;
631 tdb_allrecord_unlock(tdb
, ltype
, mark
);
633 tdb
->ecode
= TDB_ERR_LOCK
;
634 TDB_LOG((tdb
, TDB_DEBUG_ERROR
,
635 "tdb_lockall_mark cannot do recovery\n"));
638 if (tdb_lock_and_recover(tdb
) == -1) {
641 return tdb_allrecord_lock(tdb
, ltype
, flags
, upgradable
);
649 /* unlock entire db */
650 int tdb_allrecord_unlock(struct tdb_context
*tdb
, int ltype
, bool mark_lock
)
652 /* There are no locks on read-only dbs */
653 if (tdb
->read_only
|| tdb
->traverse_read
) {
654 tdb
->ecode
= TDB_ERR_LOCK
;
658 if (tdb
->allrecord_lock
.count
== 0) {
659 tdb
->ecode
= TDB_ERR_LOCK
;
663 /* Upgradable locks are marked as write locks. */
664 if (tdb
->allrecord_lock
.ltype
!= ltype
665 && (!tdb
->allrecord_lock
.off
|| ltype
!= F_RDLCK
)) {
666 tdb
->ecode
= TDB_ERR_LOCK
;
670 if (tdb
->allrecord_lock
.count
> 1) {
671 tdb
->allrecord_lock
.count
--;
675 if (!mark_lock
&& tdb_brunlock(tdb
, ltype
, FREELIST_TOP
, 0)) {
676 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlockall failed (%s)\n", strerror(errno
)));
680 tdb
->allrecord_lock
.count
= 0;
681 tdb
->allrecord_lock
.ltype
= 0;
686 /* lock entire database with write lock */
687 _PUBLIC_
int tdb_lockall(struct tdb_context
*tdb
)
689 tdb_trace(tdb
, "tdb_lockall");
690 return tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_WAIT
, false);
693 /* lock entire database with write lock - mark only */
694 _PUBLIC_
int tdb_lockall_mark(struct tdb_context
*tdb
)
696 tdb_trace(tdb
, "tdb_lockall_mark");
697 return tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_MARK_ONLY
, false);
700 /* unlock entire database with write lock - unmark only */
701 _PUBLIC_
int tdb_lockall_unmark(struct tdb_context
*tdb
)
703 tdb_trace(tdb
, "tdb_lockall_unmark");
704 return tdb_allrecord_unlock(tdb
, F_WRLCK
, true);
707 /* lock entire database with write lock - nonblocking varient */
708 _PUBLIC_
int tdb_lockall_nonblock(struct tdb_context
*tdb
)
710 int ret
= tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_NOWAIT
, false);
711 tdb_trace_ret(tdb
, "tdb_lockall_nonblock", ret
);
715 /* unlock entire database with write lock */
716 _PUBLIC_
int tdb_unlockall(struct tdb_context
*tdb
)
718 tdb_trace(tdb
, "tdb_unlockall");
719 return tdb_allrecord_unlock(tdb
, F_WRLCK
, false);
722 /* lock entire database with read lock */
723 _PUBLIC_
int tdb_lockall_read(struct tdb_context
*tdb
)
725 tdb_trace(tdb
, "tdb_lockall_read");
726 return tdb_allrecord_lock(tdb
, F_RDLCK
, TDB_LOCK_WAIT
, false);
729 /* lock entire database with read lock - nonblock varient */
730 _PUBLIC_
int tdb_lockall_read_nonblock(struct tdb_context
*tdb
)
732 int ret
= tdb_allrecord_lock(tdb
, F_RDLCK
, TDB_LOCK_NOWAIT
, false);
733 tdb_trace_ret(tdb
, "tdb_lockall_read_nonblock", ret
);
737 /* unlock entire database with read lock */
738 _PUBLIC_
int tdb_unlockall_read(struct tdb_context
*tdb
)
740 tdb_trace(tdb
, "tdb_unlockall_read");
741 return tdb_allrecord_unlock(tdb
, F_RDLCK
, false);
744 /* lock/unlock one hash chain. This is meant to be used to reduce
745 contention - it cannot guarantee how many records will be locked */
746 _PUBLIC_
int tdb_chainlock(struct tdb_context
*tdb
, TDB_DATA key
)
748 int ret
= tdb_lock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_WRLCK
);
749 tdb_trace_1rec(tdb
, "tdb_chainlock", key
);
753 /* lock/unlock one hash chain, non-blocking. This is meant to be used
754 to reduce contention - it cannot guarantee how many records will be
756 _PUBLIC_
int tdb_chainlock_nonblock(struct tdb_context
*tdb
, TDB_DATA key
)
758 int ret
= tdb_lock_nonblock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_WRLCK
);
759 tdb_trace_1rec_ret(tdb
, "tdb_chainlock_nonblock", key
, ret
);
763 /* mark a chain as locked without actually locking it. Warning! use with great caution! */
764 _PUBLIC_
int tdb_chainlock_mark(struct tdb_context
*tdb
, TDB_DATA key
)
766 int ret
= tdb_nest_lock(tdb
, lock_offset(BUCKET(tdb
->hash_fn(&key
))),
767 F_WRLCK
, TDB_LOCK_MARK_ONLY
);
768 tdb_trace_1rec(tdb
, "tdb_chainlock_mark", key
);
772 /* unmark a chain as locked without actually locking it. Warning! use with great caution! */
773 _PUBLIC_
int tdb_chainlock_unmark(struct tdb_context
*tdb
, TDB_DATA key
)
775 tdb_trace_1rec(tdb
, "tdb_chainlock_unmark", key
);
776 return tdb_nest_unlock(tdb
, lock_offset(BUCKET(tdb
->hash_fn(&key
))),
780 _PUBLIC_
int tdb_chainunlock(struct tdb_context
*tdb
, TDB_DATA key
)
782 tdb_trace_1rec(tdb
, "tdb_chainunlock", key
);
783 return tdb_unlock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_WRLCK
);
786 _PUBLIC_
int tdb_chainlock_read(struct tdb_context
*tdb
, TDB_DATA key
)
789 ret
= tdb_lock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_RDLCK
);
790 tdb_trace_1rec(tdb
, "tdb_chainlock_read", key
);
794 _PUBLIC_
int tdb_chainunlock_read(struct tdb_context
*tdb
, TDB_DATA key
)
796 tdb_trace_1rec(tdb
, "tdb_chainunlock_read", key
);
797 return tdb_unlock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_RDLCK
);
800 /* record lock stops delete underneath */
801 int tdb_lock_record(struct tdb_context
*tdb
, tdb_off_t off
)
803 if (tdb
->allrecord_lock
.count
) {
806 return off
? tdb_brlock(tdb
, F_RDLCK
, off
, 1, TDB_LOCK_WAIT
) : 0;
810 Write locks override our own fcntl readlocks, so check it here.
811 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
812 an error to fail to get the lock here.
814 int tdb_write_lock_record(struct tdb_context
*tdb
, tdb_off_t off
)
816 struct tdb_traverse_lock
*i
;
817 for (i
= &tdb
->travlocks
; i
; i
= i
->next
)
820 if (tdb
->allrecord_lock
.count
) {
821 if (tdb
->allrecord_lock
.ltype
== F_WRLCK
) {
826 return tdb_brlock(tdb
, F_WRLCK
, off
, 1, TDB_LOCK_NOWAIT
|TDB_LOCK_PROBE
);
829 int tdb_write_unlock_record(struct tdb_context
*tdb
, tdb_off_t off
)
831 if (tdb
->allrecord_lock
.count
) {
834 return tdb_brunlock(tdb
, F_WRLCK
, off
, 1);
837 /* fcntl locks don't stack: avoid unlocking someone else's */
838 int tdb_unlock_record(struct tdb_context
*tdb
, tdb_off_t off
)
840 struct tdb_traverse_lock
*i
;
843 if (tdb
->allrecord_lock
.count
) {
849 for (i
= &tdb
->travlocks
; i
; i
= i
->next
)
852 return (count
== 1 ? tdb_brunlock(tdb
, F_RDLCK
, off
, 1) : 0);
855 bool tdb_have_extra_locks(struct tdb_context
*tdb
)
857 unsigned int extra
= tdb
->num_lockrecs
;
859 /* A transaction holds the lock for all records. */
860 if (!tdb
->transaction
&& tdb
->allrecord_lock
.count
) {
864 /* We always hold the active lock if CLEAR_IF_FIRST. */
865 if (find_nestlock(tdb
, ACTIVE_LOCK
)) {
869 /* In a transaction, we expect to hold the transaction lock */
870 if (tdb
->transaction
&& find_nestlock(tdb
, TRANSACTION_LOCK
)) {
877 /* The transaction code uses this to remove all locks. */
878 void tdb_release_transaction_locks(struct tdb_context
*tdb
)
880 unsigned int i
, active
= 0;
882 if (tdb
->allrecord_lock
.count
!= 0) {
883 tdb_allrecord_unlock(tdb
, tdb
->allrecord_lock
.ltype
, false);
884 tdb
->allrecord_lock
.count
= 0;
887 for (i
=0;i
<tdb
->num_lockrecs
;i
++) {
888 struct tdb_lock_type
*lck
= &tdb
->lockrecs
[i
];
890 /* Don't release the active lock! Copy it to first entry. */
891 if (lck
->off
== ACTIVE_LOCK
) {
892 tdb
->lockrecs
[active
++] = *lck
;
894 tdb_brunlock(tdb
, lck
->ltype
, lck
->off
, 1);
897 tdb
->num_lockrecs
= active
;
900 /* Following functions are added specifically to support CTDB. */
902 /* Don't do actual fcntl locking, just mark tdb locked */
903 int tdb_transaction_write_lock_mark(struct tdb_context
*tdb
);
904 _PUBLIC_
int tdb_transaction_write_lock_mark(struct tdb_context
*tdb
)
906 return tdb_transaction_lock(tdb
, F_WRLCK
, TDB_LOCK_MARK_ONLY
);
909 /* Don't do actual fcntl unlocking, just mark tdb unlocked */
910 int tdb_transaction_write_lock_unmark(struct tdb_context
*tdb
);
911 _PUBLIC_
int tdb_transaction_write_lock_unmark(struct tdb_context
*tdb
)
913 return tdb_nest_unlock(tdb
, TRANSACTION_LOCK
, F_WRLCK
, true);