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 new_lck
= (struct tdb_lock_type
*)realloc(
302 sizeof(*tdb
->lockrecs
) * (tdb
->num_lockrecs
+1));
303 if (new_lck
== NULL
) {
307 tdb
->lockrecs
= new_lck
;
309 /* Since fcntl locks don't nest, we do a lock for the first one,
310 and simply bump the count for future ones */
311 if (tdb_brlock(tdb
, ltype
, offset
, 1, flags
)) {
315 tdb
->lockrecs
[tdb
->num_lockrecs
].off
= offset
;
316 tdb
->lockrecs
[tdb
->num_lockrecs
].count
= 1;
317 tdb
->lockrecs
[tdb
->num_lockrecs
].ltype
= ltype
;
323 static int tdb_lock_and_recover(struct tdb_context
*tdb
)
327 /* We need to match locking order in transaction commit. */
328 if (tdb_brlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0, TDB_LOCK_WAIT
)) {
332 if (tdb_brlock(tdb
, F_WRLCK
, OPEN_LOCK
, 1, TDB_LOCK_WAIT
)) {
333 tdb_brunlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0);
337 ret
= tdb_transaction_recover(tdb
);
339 tdb_brunlock(tdb
, F_WRLCK
, OPEN_LOCK
, 1);
340 tdb_brunlock(tdb
, F_WRLCK
, FREELIST_TOP
, 0);
345 static bool have_data_locks(const struct tdb_context
*tdb
)
349 for (i
= 0; i
< tdb
->num_lockrecs
; i
++) {
350 if (tdb
->lockrecs
[i
].off
>= lock_offset(-1))
357 * A allrecord lock allows us to avoid per chain locks. Check if the allrecord
358 * lock is strong enough.
360 static int tdb_lock_covered_by_allrecord_lock(struct tdb_context
*tdb
,
363 if (ltype
== F_RDLCK
) {
365 * The allrecord_lock is equal (F_RDLCK) or stronger
371 if (tdb
->allrecord_lock
.ltype
== F_RDLCK
) {
373 * We ask for ltype==F_WRLCK, but the allrecord_lock
374 * is too weak. We can't upgrade here, so fail.
376 tdb
->ecode
= TDB_ERR_LOCK
;
381 * Asking for F_WRLCK, allrecord is F_WRLCK as well. Pass.
386 static int tdb_lock_list(struct tdb_context
*tdb
, int list
, int ltype
,
387 enum tdb_lock_flags waitflag
)
392 if (tdb
->allrecord_lock
.count
) {
393 return tdb_lock_covered_by_allrecord_lock(tdb
, ltype
);
397 * Check for recoveries: Someone might have kill -9'ed a process
400 check
= !have_data_locks(tdb
);
401 ret
= tdb_nest_lock(tdb
, lock_offset(list
), ltype
, waitflag
);
403 if (ret
== 0 && check
&& tdb_needs_recovery(tdb
)) {
404 tdb_nest_unlock(tdb
, lock_offset(list
), ltype
, false);
406 if (tdb_lock_and_recover(tdb
) == -1) {
409 return tdb_lock_list(tdb
, list
, ltype
, waitflag
);
414 /* lock a list in the database. list -1 is the alloc list */
415 int tdb_lock(struct tdb_context
*tdb
, int list
, int ltype
)
419 ret
= tdb_lock_list(tdb
, list
, ltype
, TDB_LOCK_WAIT
);
421 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_lock failed on list %d "
422 "ltype=%d (%s)\n", list
, ltype
, strerror(errno
)));
427 /* lock a list in the database. list -1 is the alloc list. non-blocking lock */
428 _PUBLIC_
int tdb_lock_nonblock(struct tdb_context
*tdb
, int list
, int ltype
)
430 return tdb_lock_list(tdb
, list
, ltype
, TDB_LOCK_NOWAIT
);
434 int tdb_nest_unlock(struct tdb_context
*tdb
, uint32_t offset
, int ltype
,
438 struct tdb_lock_type
*lck
;
440 if (tdb
->flags
& TDB_NOLOCK
)
444 if (offset
>= lock_offset(tdb
->hash_size
)) {
445 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlock: offset %u invalid (%d)\n", offset
, tdb
->hash_size
));
449 lck
= find_nestlock(tdb
, offset
);
450 if ((lck
== NULL
) || (lck
->count
== 0)) {
451 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlock: count is 0\n"));
455 if (lck
->count
> 1) {
461 * This lock has count==1 left, so we need to unlock it in the
462 * kernel. We don't bother with decrementing the in-memory array
463 * element, we're about to overwrite it with the last array element
470 ret
= tdb_brunlock(tdb
, ltype
, offset
, 1);
474 * Shrink the array by overwriting the element just unlocked with the
475 * last array element.
477 *lck
= tdb
->lockrecs
[--tdb
->num_lockrecs
];
480 * We don't bother with realloc when the array shrinks, but if we have
481 * a completely idle tdb we should get rid of the locked array.
484 if (tdb
->num_lockrecs
== 0) {
485 SAFE_FREE(tdb
->lockrecs
);
489 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlock: An error occurred unlocking!\n"));
493 _PUBLIC_
int tdb_unlock(struct tdb_context
*tdb
, int list
, int ltype
)
495 /* a global lock allows us to avoid per chain locks */
496 if (tdb
->allrecord_lock
.count
) {
497 return tdb_lock_covered_by_allrecord_lock(tdb
, ltype
);
500 return tdb_nest_unlock(tdb
, lock_offset(list
), ltype
, false);
504 get the transaction lock
506 int tdb_transaction_lock(struct tdb_context
*tdb
, int ltype
,
507 enum tdb_lock_flags lockflags
)
509 return tdb_nest_lock(tdb
, TRANSACTION_LOCK
, ltype
, lockflags
);
513 release the transaction lock
515 int tdb_transaction_unlock(struct tdb_context
*tdb
, int ltype
)
517 return tdb_nest_unlock(tdb
, TRANSACTION_LOCK
, ltype
, false);
520 /* Returns 0 if all done, -1 if error, 1 if ok. */
521 static int tdb_allrecord_check(struct tdb_context
*tdb
, int ltype
,
522 enum tdb_lock_flags flags
, bool upgradable
)
524 /* There are no locks on read-only dbs */
525 if (tdb
->read_only
|| tdb
->traverse_read
) {
526 tdb
->ecode
= TDB_ERR_LOCK
;
530 if (tdb
->allrecord_lock
.count
&& tdb
->allrecord_lock
.ltype
== ltype
) {
531 tdb
->allrecord_lock
.count
++;
535 if (tdb
->allrecord_lock
.count
) {
536 /* a global lock of a different type exists */
537 tdb
->ecode
= TDB_ERR_LOCK
;
541 if (tdb_have_extra_locks(tdb
)) {
542 /* can't combine global and chain locks */
543 tdb
->ecode
= TDB_ERR_LOCK
;
547 if (upgradable
&& ltype
!= F_RDLCK
) {
548 /* tdb error: you can't upgrade a write lock! */
549 tdb
->ecode
= TDB_ERR_LOCK
;
555 /* We only need to lock individual bytes, but Linux merges consecutive locks
556 * so we lock in contiguous ranges. */
557 static int tdb_chainlock_gradual(struct tdb_context
*tdb
,
558 int ltype
, enum tdb_lock_flags flags
,
559 size_t off
, size_t len
)
562 enum tdb_lock_flags nb_flags
= (flags
& ~TDB_LOCK_WAIT
);
565 /* Single record. Just do blocking lock. */
566 return tdb_brlock(tdb
, ltype
, off
, len
, flags
);
569 /* First we try non-blocking. */
570 ret
= tdb_brlock(tdb
, ltype
, off
, len
, nb_flags
);
575 /* Try locking first half, then second. */
576 ret
= tdb_chainlock_gradual(tdb
, ltype
, flags
, off
, len
/ 2);
580 ret
= tdb_chainlock_gradual(tdb
, ltype
, flags
,
581 off
+ len
/ 2, len
- len
/ 2);
583 tdb_brunlock(tdb
, ltype
, off
, len
/ 2);
589 /* lock/unlock entire database. It can only be upgradable if you have some
590 * other way of guaranteeing exclusivity (ie. transaction write lock).
591 * We do the locking gradually to avoid being starved by smaller locks. */
592 int tdb_allrecord_lock(struct tdb_context
*tdb
, int ltype
,
593 enum tdb_lock_flags flags
, bool upgradable
)
595 switch (tdb_allrecord_check(tdb
, ltype
, flags
, upgradable
)) {
602 /* We cover two kinds of locks:
603 * 1) Normal chain locks. Taken for almost all operations.
604 * 2) Individual records locks. Taken after normal or free
607 * It is (1) which cause the starvation problem, so we're only
608 * gradual for that. */
609 if (tdb_chainlock_gradual(tdb
, ltype
, flags
, FREELIST_TOP
,
610 tdb
->hash_size
* 4) == -1) {
614 /* Grab individual record locks. */
615 if (tdb_brlock(tdb
, ltype
, lock_offset(tdb
->hash_size
), 0,
617 tdb_brunlock(tdb
, ltype
, FREELIST_TOP
,
622 tdb
->allrecord_lock
.count
= 1;
623 /* If it's upgradable, it's actually exclusive so we can treat
624 * it as a write lock. */
625 tdb
->allrecord_lock
.ltype
= upgradable
? F_WRLCK
: ltype
;
626 tdb
->allrecord_lock
.off
= upgradable
;
628 if (tdb_needs_recovery(tdb
)) {
629 bool mark
= flags
& TDB_LOCK_MARK_ONLY
;
630 tdb_allrecord_unlock(tdb
, ltype
, mark
);
632 tdb
->ecode
= TDB_ERR_LOCK
;
633 TDB_LOG((tdb
, TDB_DEBUG_ERROR
,
634 "tdb_lockall_mark cannot do recovery\n"));
637 if (tdb_lock_and_recover(tdb
) == -1) {
640 return tdb_allrecord_lock(tdb
, ltype
, flags
, upgradable
);
648 /* unlock entire db */
649 int tdb_allrecord_unlock(struct tdb_context
*tdb
, int ltype
, bool mark_lock
)
651 /* There are no locks on read-only dbs */
652 if (tdb
->read_only
|| tdb
->traverse_read
) {
653 tdb
->ecode
= TDB_ERR_LOCK
;
657 if (tdb
->allrecord_lock
.count
== 0) {
658 tdb
->ecode
= TDB_ERR_LOCK
;
662 /* Upgradable locks are marked as write locks. */
663 if (tdb
->allrecord_lock
.ltype
!= ltype
664 && (!tdb
->allrecord_lock
.off
|| ltype
!= F_RDLCK
)) {
665 tdb
->ecode
= TDB_ERR_LOCK
;
669 if (tdb
->allrecord_lock
.count
> 1) {
670 tdb
->allrecord_lock
.count
--;
674 if (!mark_lock
&& tdb_brunlock(tdb
, ltype
, FREELIST_TOP
, 0)) {
675 TDB_LOG((tdb
, TDB_DEBUG_ERROR
, "tdb_unlockall failed (%s)\n", strerror(errno
)));
679 tdb
->allrecord_lock
.count
= 0;
680 tdb
->allrecord_lock
.ltype
= 0;
685 /* lock entire database with write lock */
686 _PUBLIC_
int tdb_lockall(struct tdb_context
*tdb
)
688 tdb_trace(tdb
, "tdb_lockall");
689 return tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_WAIT
, false);
692 /* lock entire database with write lock - mark only */
693 _PUBLIC_
int tdb_lockall_mark(struct tdb_context
*tdb
)
695 tdb_trace(tdb
, "tdb_lockall_mark");
696 return tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_MARK_ONLY
, false);
699 /* unlock entire database with write lock - unmark only */
700 _PUBLIC_
int tdb_lockall_unmark(struct tdb_context
*tdb
)
702 tdb_trace(tdb
, "tdb_lockall_unmark");
703 return tdb_allrecord_unlock(tdb
, F_WRLCK
, true);
706 /* lock entire database with write lock - nonblocking varient */
707 _PUBLIC_
int tdb_lockall_nonblock(struct tdb_context
*tdb
)
709 int ret
= tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_NOWAIT
, false);
710 tdb_trace_ret(tdb
, "tdb_lockall_nonblock", ret
);
714 /* unlock entire database with write lock */
715 _PUBLIC_
int tdb_unlockall(struct tdb_context
*tdb
)
717 tdb_trace(tdb
, "tdb_unlockall");
718 return tdb_allrecord_unlock(tdb
, F_WRLCK
, false);
721 /* lock entire database with read lock */
722 _PUBLIC_
int tdb_lockall_read(struct tdb_context
*tdb
)
724 tdb_trace(tdb
, "tdb_lockall_read");
725 return tdb_allrecord_lock(tdb
, F_RDLCK
, TDB_LOCK_WAIT
, false);
728 /* lock entire database with read lock - nonblock varient */
729 _PUBLIC_
int tdb_lockall_read_nonblock(struct tdb_context
*tdb
)
731 int ret
= tdb_allrecord_lock(tdb
, F_RDLCK
, TDB_LOCK_NOWAIT
, false);
732 tdb_trace_ret(tdb
, "tdb_lockall_read_nonblock", ret
);
736 /* unlock entire database with read lock */
737 _PUBLIC_
int tdb_unlockall_read(struct tdb_context
*tdb
)
739 tdb_trace(tdb
, "tdb_unlockall_read");
740 return tdb_allrecord_unlock(tdb
, F_RDLCK
, false);
743 /* lock/unlock one hash chain. This is meant to be used to reduce
744 contention - it cannot guarantee how many records will be locked */
745 _PUBLIC_
int tdb_chainlock(struct tdb_context
*tdb
, TDB_DATA key
)
747 int ret
= tdb_lock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_WRLCK
);
748 tdb_trace_1rec(tdb
, "tdb_chainlock", key
);
752 /* lock/unlock one hash chain, non-blocking. This is meant to be used
753 to reduce contention - it cannot guarantee how many records will be
755 _PUBLIC_
int tdb_chainlock_nonblock(struct tdb_context
*tdb
, TDB_DATA key
)
757 int ret
= tdb_lock_nonblock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_WRLCK
);
758 tdb_trace_1rec_ret(tdb
, "tdb_chainlock_nonblock", key
, ret
);
762 /* mark a chain as locked without actually locking it. Warning! use with great caution! */
763 _PUBLIC_
int tdb_chainlock_mark(struct tdb_context
*tdb
, TDB_DATA key
)
765 int ret
= tdb_nest_lock(tdb
, lock_offset(BUCKET(tdb
->hash_fn(&key
))),
766 F_WRLCK
, TDB_LOCK_MARK_ONLY
);
767 tdb_trace_1rec(tdb
, "tdb_chainlock_mark", key
);
771 /* unmark a chain as locked without actually locking it. Warning! use with great caution! */
772 _PUBLIC_
int tdb_chainlock_unmark(struct tdb_context
*tdb
, TDB_DATA key
)
774 tdb_trace_1rec(tdb
, "tdb_chainlock_unmark", key
);
775 return tdb_nest_unlock(tdb
, lock_offset(BUCKET(tdb
->hash_fn(&key
))),
779 _PUBLIC_
int tdb_chainunlock(struct tdb_context
*tdb
, TDB_DATA key
)
781 tdb_trace_1rec(tdb
, "tdb_chainunlock", key
);
782 return tdb_unlock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_WRLCK
);
785 _PUBLIC_
int tdb_chainlock_read(struct tdb_context
*tdb
, TDB_DATA key
)
788 ret
= tdb_lock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_RDLCK
);
789 tdb_trace_1rec(tdb
, "tdb_chainlock_read", key
);
793 _PUBLIC_
int tdb_chainunlock_read(struct tdb_context
*tdb
, TDB_DATA key
)
795 tdb_trace_1rec(tdb
, "tdb_chainunlock_read", key
);
796 return tdb_unlock(tdb
, BUCKET(tdb
->hash_fn(&key
)), F_RDLCK
);
799 /* record lock stops delete underneath */
800 int tdb_lock_record(struct tdb_context
*tdb
, tdb_off_t off
)
802 if (tdb
->allrecord_lock
.count
) {
805 return off
? tdb_brlock(tdb
, F_RDLCK
, off
, 1, TDB_LOCK_WAIT
) : 0;
809 Write locks override our own fcntl readlocks, so check it here.
810 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
811 an error to fail to get the lock here.
813 int tdb_write_lock_record(struct tdb_context
*tdb
, tdb_off_t off
)
815 struct tdb_traverse_lock
*i
;
816 for (i
= &tdb
->travlocks
; i
; i
= i
->next
)
819 if (tdb
->allrecord_lock
.count
) {
820 if (tdb
->allrecord_lock
.ltype
== F_WRLCK
) {
825 return tdb_brlock(tdb
, F_WRLCK
, off
, 1, TDB_LOCK_NOWAIT
|TDB_LOCK_PROBE
);
828 int tdb_write_unlock_record(struct tdb_context
*tdb
, tdb_off_t off
)
830 if (tdb
->allrecord_lock
.count
) {
833 return tdb_brunlock(tdb
, F_WRLCK
, off
, 1);
836 /* fcntl locks don't stack: avoid unlocking someone else's */
837 int tdb_unlock_record(struct tdb_context
*tdb
, tdb_off_t off
)
839 struct tdb_traverse_lock
*i
;
842 if (tdb
->allrecord_lock
.count
) {
848 for (i
= &tdb
->travlocks
; i
; i
= i
->next
)
851 return (count
== 1 ? tdb_brunlock(tdb
, F_RDLCK
, off
, 1) : 0);
854 bool tdb_have_extra_locks(struct tdb_context
*tdb
)
856 unsigned int extra
= tdb
->num_lockrecs
;
858 /* A transaction holds the lock for all records. */
859 if (!tdb
->transaction
&& tdb
->allrecord_lock
.count
) {
863 /* We always hold the active lock if CLEAR_IF_FIRST. */
864 if (find_nestlock(tdb
, ACTIVE_LOCK
)) {
868 /* In a transaction, we expect to hold the transaction lock */
869 if (tdb
->transaction
&& find_nestlock(tdb
, TRANSACTION_LOCK
)) {
876 /* The transaction code uses this to remove all locks. */
877 void tdb_release_transaction_locks(struct tdb_context
*tdb
)
879 unsigned int i
, active
= 0;
881 if (tdb
->allrecord_lock
.count
!= 0) {
882 tdb_allrecord_unlock(tdb
, tdb
->allrecord_lock
.ltype
, false);
883 tdb
->allrecord_lock
.count
= 0;
886 for (i
=0;i
<tdb
->num_lockrecs
;i
++) {
887 struct tdb_lock_type
*lck
= &tdb
->lockrecs
[i
];
889 /* Don't release the active lock! Copy it to first entry. */
890 if (lck
->off
== ACTIVE_LOCK
) {
891 tdb
->lockrecs
[active
++] = *lck
;
893 tdb_brunlock(tdb
, lck
->ltype
, lck
->off
, 1);
896 tdb
->num_lockrecs
= active
;
897 if (tdb
->num_lockrecs
== 0) {
898 SAFE_FREE(tdb
->lockrecs
);
902 /* Following functions are added specifically to support CTDB. */
904 /* Don't do actual fcntl locking, just mark tdb locked */
905 int tdb_transaction_write_lock_mark(struct tdb_context
*tdb
);
906 _PUBLIC_
int tdb_transaction_write_lock_mark(struct tdb_context
*tdb
)
908 return tdb_transaction_lock(tdb
, F_WRLCK
, TDB_LOCK_MARK_ONLY
);
911 /* Don't do actual fcntl unlocking, just mark tdb unlocked */
912 int tdb_transaction_write_lock_unmark(struct tdb_context
*tdb
);
913 _PUBLIC_
int tdb_transaction_write_lock_unmark(struct tdb_context
*tdb
)
915 return tdb_nest_unlock(tdb
, TRANSACTION_LOCK
, F_WRLCK
, true);