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/>.
30 #include <ccan/build_assert/build_assert.h>
32 /* If we were threaded, we could wait for unlock, but we're not, so fail. */
33 static enum TDB_ERROR
owner_conflict(struct tdb_context
*tdb
, const char *call
)
35 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
36 "%s: lock owned by another tdb in this process.",
40 /* If we fork, we no longer really own locks. */
41 static bool check_lock_pid(struct tdb_context
*tdb
,
42 const char *call
, bool log
)
44 /* No locks? No problem! */
45 if (tdb
->file
->allrecord_lock
.count
== 0
46 && tdb
->file
->num_lockrecs
== 0) {
50 /* No fork? No problem! */
51 if (tdb
->file
->locker
== getpid()) {
56 tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
57 "%s: fork() detected after lock acquisition!"
58 " (%u vs %u)", call
, tdb
->file
->locker
, getpid());
63 int tdb_fcntl_lock(int fd
, int rw
, off_t off
, off_t len
, bool waitflag
,
71 fl
.l_whence
= SEEK_SET
;
76 ret
= fcntl(fd
, F_SETLKW
, &fl
);
78 ret
= fcntl(fd
, F_SETLK
, &fl
);
79 } while (ret
!= 0 && errno
== EINTR
);
83 int tdb_fcntl_unlock(int fd
, int rw
, off_t off
, off_t len
, void *unused
)
90 fl
.l_whence
= SEEK_SET
;
94 ret
= fcntl(fd
, F_SETLKW
, &fl
);
95 } while (ret
!= 0 && errno
== EINTR
);
99 static int lock(struct tdb_context
*tdb
,
100 int rw
, off_t off
, off_t len
, bool waitflag
)
103 if (tdb
->file
->allrecord_lock
.count
== 0
104 && tdb
->file
->num_lockrecs
== 0) {
105 tdb
->file
->locker
= getpid();
108 tdb
->stats
.lock_lowlevel
++;
109 ret
= tdb
->lock_fn(tdb
->file
->fd
, rw
, off
, len
, waitflag
,
112 tdb
->stats
.lock_nonblock
++;
114 tdb
->stats
.lock_nonblock_fail
++;
119 static int unlock(struct tdb_context
*tdb
, int rw
, off_t off
, off_t len
)
121 #if 0 /* Check they matched up locks and unlocks correctly. */
126 locks
= fopen("/proc/locks", "r");
128 while (fgets(line
, 80, locks
)) {
132 /* eg. 1: FLOCK ADVISORY WRITE 2440 08:01:2180826 0 EOF */
133 p
= strchr(line
, ':') + 1;
134 if (strncmp(p
, " POSIX ADVISORY ", strlen(" POSIX ADVISORY ")))
136 p
+= strlen(" FLOCK ADVISORY ");
137 if (strncmp(p
, "READ ", strlen("READ ")) == 0)
139 else if (strncmp(p
, "WRITE ", strlen("WRITE ")) == 0)
144 if (atoi(p
) != getpid())
146 p
= strchr(strchr(p
, ' ') + 1, ' ') + 1;
148 p
= strchr(p
, ' ') + 1;
149 if (strncmp(p
, "EOF", 3) == 0)
152 l
= atoi(p
) - start
+ 1;
156 fprintf(stderr
, "Len %u should be %u: %s",
161 fprintf(stderr
, "Type %s wrong: %s",
162 rw
== F_RDLCK
? "READ" : "WRITE", line
);
171 fprintf(stderr
, "Unlock on %u@%u not found!",
179 return tdb
->unlock_fn(tdb
->file
->fd
, rw
, off
, len
, tdb
->lock_data
);
182 /* a byte range locking function - return 0 on success
183 this functions locks len bytes at the specified offset.
185 note that a len of zero means lock to end of file
187 static enum TDB_ERROR
tdb_brlock(struct tdb_context
*tdb
,
188 int rw_type
, tdb_off_t offset
, tdb_off_t len
,
189 enum tdb_lock_flags flags
)
193 if (tdb
->flags
& TDB_NOLOCK
) {
197 if (rw_type
== F_WRLCK
&& tdb
->read_only
) {
198 return tdb_logerr(tdb
, TDB_ERR_RDONLY
, TDB_LOG_USE_ERROR
,
199 "Write lock attempted on read-only database");
202 /* A 32 bit system cannot open a 64-bit file, but it could have
203 * expanded since then: check here. */
204 if ((size_t)(offset
+ len
) != offset
+ len
) {
205 return tdb_logerr(tdb
, TDB_ERR_IO
, TDB_LOG_ERROR
,
206 "tdb_brlock: lock on giant offset %llu",
207 (long long)(offset
+ len
));
210 ret
= lock(tdb
, rw_type
, offset
, len
, flags
& TDB_LOCK_WAIT
);
212 /* Generic lock error. errno set by fcntl.
213 * EAGAIN is an expected return from non-blocking
215 if (!(flags
& TDB_LOCK_PROBE
)
216 && (errno
!= EAGAIN
&& errno
!= EINTR
)) {
217 tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
218 "tdb_brlock failed (fd=%d) at"
219 " offset %zu rw_type=%d flags=%d len=%zu:"
221 tdb
->file
->fd
, (size_t)offset
, rw_type
,
222 flags
, (size_t)len
, strerror(errno
));
229 static enum TDB_ERROR
tdb_brunlock(struct tdb_context
*tdb
,
230 int rw_type
, tdb_off_t offset
, size_t len
)
232 if (tdb
->flags
& TDB_NOLOCK
) {
236 if (!check_lock_pid(tdb
, "tdb_brunlock", true))
239 if (unlock(tdb
, rw_type
, offset
, len
) == -1) {
240 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
241 "tdb_brunlock failed (fd=%d) at offset %zu"
242 " rw_type=%d len=%zu: %s",
243 tdb
->file
->fd
, (size_t)offset
, rw_type
,
244 (size_t)len
, strerror(errno
));
250 upgrade a read lock to a write lock. This needs to be handled in a
251 special way as some OSes (such as solaris) have too conservative
252 deadlock detection and claim a deadlock when progress can be
253 made. For those OSes we may loop for a while.
255 enum TDB_ERROR
tdb_allrecord_upgrade(struct tdb_context
*tdb
)
259 if (!check_lock_pid(tdb
, "tdb_transaction_prepare_commit", true))
262 if (tdb
->file
->allrecord_lock
.count
!= 1) {
263 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
264 "tdb_allrecord_upgrade failed:"
265 " count %u too high",
266 tdb
->file
->allrecord_lock
.count
);
269 if (tdb
->file
->allrecord_lock
.off
!= 1) {
270 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
271 "tdb_allrecord_upgrade failed:"
272 " already upgraded?");
275 if (tdb
->file
->allrecord_lock
.owner
!= tdb
) {
276 return owner_conflict(tdb
, "tdb_allrecord_upgrade");
281 if (tdb_brlock(tdb
, F_WRLCK
,
282 TDB_HASH_LOCK_START
, 0,
283 TDB_LOCK_WAIT
|TDB_LOCK_PROBE
) == TDB_SUCCESS
) {
284 tdb
->file
->allrecord_lock
.ltype
= F_WRLCK
;
285 tdb
->file
->allrecord_lock
.off
= 0;
288 if (errno
!= EDEADLK
) {
291 /* sleep for as short a time as we can - more portable than usleep() */
294 select(0, NULL
, NULL
, NULL
, &tv
);
297 if (errno
!= EAGAIN
&& errno
!= EINTR
)
298 tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
299 "tdb_allrecord_upgrade failed");
303 static struct tdb_lock
*find_nestlock(struct tdb_context
*tdb
, tdb_off_t offset
,
304 const struct tdb_context
*owner
)
308 for (i
=0; i
<tdb
->file
->num_lockrecs
; i
++) {
309 if (tdb
->file
->lockrecs
[i
].off
== offset
) {
310 if (owner
&& tdb
->file
->lockrecs
[i
].owner
!= owner
)
312 return &tdb
->file
->lockrecs
[i
];
318 enum TDB_ERROR
tdb_lock_and_recover(struct tdb_context
*tdb
)
320 enum TDB_ERROR ecode
;
322 if (!check_lock_pid(tdb
, "tdb_transaction_prepare_commit", true))
325 ecode
= tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_WAIT
|TDB_LOCK_NOCHECK
,
327 if (ecode
!= TDB_SUCCESS
) {
331 ecode
= tdb_lock_open(tdb
, F_WRLCK
, TDB_LOCK_WAIT
|TDB_LOCK_NOCHECK
);
332 if (ecode
!= TDB_SUCCESS
) {
333 tdb_allrecord_unlock(tdb
, F_WRLCK
);
336 ecode
= tdb_transaction_recover(tdb
);
337 tdb_unlock_open(tdb
, F_WRLCK
);
338 tdb_allrecord_unlock(tdb
, F_WRLCK
);
343 /* lock an offset in the database. */
344 static enum TDB_ERROR
tdb_nest_lock(struct tdb_context
*tdb
,
345 tdb_off_t offset
, int ltype
,
346 enum tdb_lock_flags flags
)
348 struct tdb_lock
*new_lck
;
349 enum TDB_ERROR ecode
;
351 if (offset
> (TDB_HASH_LOCK_START
+ TDB_HASH_LOCK_RANGE
352 + tdb
->file
->map_size
/ 8)) {
353 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
354 "tdb_nest_lock: invalid offset %zu ltype=%d",
355 (size_t)offset
, ltype
);
358 if (tdb
->flags
& TDB_NOLOCK
)
361 if (!check_lock_pid(tdb
, "tdb_nest_lock", true)) {
367 new_lck
= find_nestlock(tdb
, offset
, NULL
);
369 if (new_lck
->owner
!= tdb
) {
370 return owner_conflict(tdb
, "tdb_nest_lock");
373 if (new_lck
->ltype
== F_RDLCK
&& ltype
== F_WRLCK
) {
374 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
376 " offset %zu has read lock",
379 /* Just increment the struct, posix locks don't stack. */
385 if (tdb
->file
->num_lockrecs
386 && offset
>= TDB_HASH_LOCK_START
387 && offset
< TDB_HASH_LOCK_START
+ TDB_HASH_LOCK_RANGE
) {
388 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
389 "tdb_nest_lock: already have a hash lock?");
393 new_lck
= (struct tdb_lock
*)realloc(
395 sizeof(*tdb
->file
->lockrecs
) * (tdb
->file
->num_lockrecs
+1));
396 if (new_lck
== NULL
) {
397 return tdb_logerr(tdb
, TDB_ERR_OOM
, TDB_LOG_ERROR
,
399 " unable to allocate %zu lock struct",
400 tdb
->file
->num_lockrecs
+ 1);
402 tdb
->file
->lockrecs
= new_lck
;
404 /* Since fcntl locks don't nest, we do a lock for the first one,
405 and simply bump the count for future ones */
406 ecode
= tdb_brlock(tdb
, ltype
, offset
, 1, flags
);
407 if (ecode
!= TDB_SUCCESS
) {
411 /* First time we grab a lock, perhaps someone died in commit? */
412 if (!(flags
& TDB_LOCK_NOCHECK
)
413 && tdb
->file
->num_lockrecs
== 0) {
414 tdb_bool_err berr
= tdb_needs_recovery(tdb
);
416 tdb_brunlock(tdb
, ltype
, offset
, 1);
420 ecode
= tdb_lock_and_recover(tdb
);
421 if (ecode
== TDB_SUCCESS
) {
422 ecode
= tdb_brlock(tdb
, ltype
, offset
, 1,
425 if (ecode
!= TDB_SUCCESS
) {
431 tdb
->file
->lockrecs
[tdb
->file
->num_lockrecs
].owner
= tdb
;
432 tdb
->file
->lockrecs
[tdb
->file
->num_lockrecs
].off
= offset
;
433 tdb
->file
->lockrecs
[tdb
->file
->num_lockrecs
].count
= 1;
434 tdb
->file
->lockrecs
[tdb
->file
->num_lockrecs
].ltype
= ltype
;
435 tdb
->file
->num_lockrecs
++;
440 static enum TDB_ERROR
tdb_nest_unlock(struct tdb_context
*tdb
,
441 tdb_off_t off
, int ltype
)
443 struct tdb_lock
*lck
;
444 enum TDB_ERROR ecode
;
446 if (tdb
->flags
& TDB_NOLOCK
)
449 lck
= find_nestlock(tdb
, off
, tdb
);
450 if ((lck
== NULL
) || (lck
->count
== 0)) {
451 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
452 "tdb_nest_unlock: no lock for %zu",
456 if (lck
->count
> 1) {
462 * This lock has count==1 left, so we need to unlock it in the
463 * kernel. We don't bother with decrementing the in-memory array
464 * element, we're about to overwrite it with the last array element
467 ecode
= tdb_brunlock(tdb
, ltype
, off
, 1);
470 * Shrink the array by overwriting the element just unlocked with the
471 * last array element.
473 *lck
= tdb
->file
->lockrecs
[--tdb
->file
->num_lockrecs
];
479 get the transaction lock
481 enum TDB_ERROR
tdb_transaction_lock(struct tdb_context
*tdb
, int ltype
)
483 return tdb_nest_lock(tdb
, TDB_TRANSACTION_LOCK
, ltype
, TDB_LOCK_WAIT
);
487 release the transaction lock
489 void tdb_transaction_unlock(struct tdb_context
*tdb
, int ltype
)
491 tdb_nest_unlock(tdb
, TDB_TRANSACTION_LOCK
, ltype
);
494 /* We only need to lock individual bytes, but Linux merges consecutive locks
495 * so we lock in contiguous ranges. */
496 static enum TDB_ERROR
tdb_lock_gradual(struct tdb_context
*tdb
,
497 int ltype
, enum tdb_lock_flags flags
,
498 tdb_off_t off
, tdb_off_t len
)
500 enum TDB_ERROR ecode
;
501 enum tdb_lock_flags nb_flags
= (flags
& ~TDB_LOCK_WAIT
);
504 /* 0 would mean to end-of-file... */
506 /* Single hash. Just do blocking lock. */
507 return tdb_brlock(tdb
, ltype
, off
, len
, flags
);
510 /* First we try non-blocking. */
511 if (tdb_brlock(tdb
, ltype
, off
, len
, nb_flags
) == TDB_SUCCESS
) {
515 /* Try locking first half, then second. */
516 ecode
= tdb_lock_gradual(tdb
, ltype
, flags
, off
, len
/ 2);
517 if (ecode
!= TDB_SUCCESS
)
520 ecode
= tdb_lock_gradual(tdb
, ltype
, flags
,
521 off
+ len
/ 2, len
- len
/ 2);
522 if (ecode
!= TDB_SUCCESS
) {
523 tdb_brunlock(tdb
, ltype
, off
, len
/ 2);
528 /* lock/unlock entire database. It can only be upgradable if you have some
529 * other way of guaranteeing exclusivity (ie. transaction write lock). */
530 enum TDB_ERROR
tdb_allrecord_lock(struct tdb_context
*tdb
, int ltype
,
531 enum tdb_lock_flags flags
, bool upgradable
)
533 enum TDB_ERROR ecode
;
536 if (tdb
->flags
& TDB_NOLOCK
)
539 if (!check_lock_pid(tdb
, "tdb_allrecord_lock", true)) {
543 if (tdb
->file
->allrecord_lock
.count
) {
544 if (tdb
->file
->allrecord_lock
.owner
!= tdb
) {
545 return owner_conflict(tdb
, "tdb_allrecord_lock");
549 || tdb
->file
->allrecord_lock
.ltype
== F_WRLCK
) {
550 tdb
->file
->allrecord_lock
.count
++;
554 /* a global lock of a different type exists */
555 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
556 "tdb_allrecord_lock: already have %s lock",
557 tdb
->file
->allrecord_lock
.ltype
== F_RDLCK
561 if (tdb_has_hash_locks(tdb
)) {
562 /* can't combine global and chain locks */
563 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
564 "tdb_allrecord_lock:"
565 " already have chain lock");
568 if (upgradable
&& ltype
!= F_RDLCK
) {
569 /* tdb error: you can't upgrade a write lock! */
570 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
571 "tdb_allrecord_lock:"
572 " can't upgrade a write lock");
577 /* Lock hashes, gradually. */
578 ecode
= tdb_lock_gradual(tdb
, ltype
, flags
, TDB_HASH_LOCK_START
,
579 TDB_HASH_LOCK_RANGE
);
580 if (ecode
!= TDB_SUCCESS
)
583 /* Lock free tables: there to end of file. */
584 ecode
= tdb_brlock(tdb
, ltype
,
585 TDB_HASH_LOCK_START
+ TDB_HASH_LOCK_RANGE
,
587 if (ecode
!= TDB_SUCCESS
) {
588 tdb_brunlock(tdb
, ltype
, TDB_HASH_LOCK_START
,
589 TDB_HASH_LOCK_RANGE
);
593 tdb
->file
->allrecord_lock
.owner
= tdb
;
594 tdb
->file
->allrecord_lock
.count
= 1;
595 /* If it's upgradable, it's actually exclusive so we can treat
596 * it as a write lock. */
597 tdb
->file
->allrecord_lock
.ltype
= upgradable
? F_WRLCK
: ltype
;
598 tdb
->file
->allrecord_lock
.off
= upgradable
;
600 /* Now check for needing recovery. */
601 if (flags
& TDB_LOCK_NOCHECK
)
604 berr
= tdb_needs_recovery(tdb
);
605 if (likely(berr
== false))
608 tdb_allrecord_unlock(tdb
, ltype
);
611 ecode
= tdb_lock_and_recover(tdb
);
612 if (ecode
!= TDB_SUCCESS
) {
618 enum TDB_ERROR
tdb_lock_open(struct tdb_context
*tdb
,
619 int ltype
, enum tdb_lock_flags flags
)
621 return tdb_nest_lock(tdb
, TDB_OPEN_LOCK
, ltype
, flags
);
624 void tdb_unlock_open(struct tdb_context
*tdb
, int ltype
)
626 tdb_nest_unlock(tdb
, TDB_OPEN_LOCK
, ltype
);
629 bool tdb_has_open_lock(struct tdb_context
*tdb
)
631 return !(tdb
->flags
& TDB_NOLOCK
)
632 && find_nestlock(tdb
, TDB_OPEN_LOCK
, tdb
) != NULL
;
635 enum TDB_ERROR
tdb_lock_expand(struct tdb_context
*tdb
, int ltype
)
637 /* Lock doesn't protect data, so don't check (we recurse if we do!) */
638 return tdb_nest_lock(tdb
, TDB_EXPANSION_LOCK
, ltype
,
639 TDB_LOCK_WAIT
| TDB_LOCK_NOCHECK
);
642 void tdb_unlock_expand(struct tdb_context
*tdb
, int ltype
)
644 tdb_nest_unlock(tdb
, TDB_EXPANSION_LOCK
, ltype
);
647 /* unlock entire db */
648 void tdb_allrecord_unlock(struct tdb_context
*tdb
, int ltype
)
650 if (tdb
->flags
& TDB_NOLOCK
)
653 if (tdb
->file
->allrecord_lock
.count
== 0) {
654 tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
655 "tdb_allrecord_unlock: not locked!");
659 if (tdb
->file
->allrecord_lock
.owner
!= tdb
) {
660 tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
661 "tdb_allrecord_unlock: not locked by us!");
665 /* Upgradable locks are marked as write locks. */
666 if (tdb
->file
->allrecord_lock
.ltype
!= ltype
667 && (!tdb
->file
->allrecord_lock
.off
|| ltype
!= F_RDLCK
)) {
668 tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
669 "tdb_allrecord_unlock: have %s lock",
670 tdb
->file
->allrecord_lock
.ltype
== F_RDLCK
675 if (tdb
->file
->allrecord_lock
.count
> 1) {
676 tdb
->file
->allrecord_lock
.count
--;
680 tdb
->file
->allrecord_lock
.count
= 0;
681 tdb
->file
->allrecord_lock
.ltype
= 0;
683 tdb_brunlock(tdb
, ltype
, TDB_HASH_LOCK_START
, 0);
686 bool tdb_has_expansion_lock(struct tdb_context
*tdb
)
688 return find_nestlock(tdb
, TDB_EXPANSION_LOCK
, tdb
) != NULL
;
691 bool tdb_has_hash_locks(struct tdb_context
*tdb
)
695 for (i
=0; i
<tdb
->file
->num_lockrecs
; i
++) {
696 if (tdb
->file
->lockrecs
[i
].off
>= TDB_HASH_LOCK_START
697 && tdb
->file
->lockrecs
[i
].off
< (TDB_HASH_LOCK_START
698 + TDB_HASH_LOCK_RANGE
))
704 static bool tdb_has_free_lock(struct tdb_context
*tdb
)
708 if (tdb
->flags
& TDB_NOLOCK
)
711 for (i
=0; i
<tdb
->file
->num_lockrecs
; i
++) {
712 if (tdb
->file
->lockrecs
[i
].off
713 > TDB_HASH_LOCK_START
+ TDB_HASH_LOCK_RANGE
)
719 enum TDB_ERROR
tdb_lock_hashes(struct tdb_context
*tdb
,
721 tdb_len_t hash_range
,
722 int ltype
, enum tdb_lock_flags waitflag
)
724 /* FIXME: Do this properly, using hlock_range */
725 unsigned l
= TDB_HASH_LOCK_START
726 + (hash_lock
>> (64 - TDB_HASH_LOCK_RANGE_BITS
));
728 /* a allrecord lock allows us to avoid per chain locks */
729 if (tdb
->file
->allrecord_lock
.count
) {
730 if (!check_lock_pid(tdb
, "tdb_lock_hashes", true))
733 if (tdb
->file
->allrecord_lock
.owner
!= tdb
)
734 return owner_conflict(tdb
, "tdb_lock_hashes");
735 if (ltype
== tdb
->file
->allrecord_lock
.ltype
736 || ltype
== F_RDLCK
) {
740 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_USE_ERROR
,
742 " already have %s allrecordlock",
743 tdb
->file
->allrecord_lock
.ltype
== F_RDLCK
747 if (tdb_has_free_lock(tdb
)) {
748 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
749 "tdb_lock_hashes: already have free lock");
752 if (tdb_has_expansion_lock(tdb
)) {
753 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
755 " already have expansion lock");
758 return tdb_nest_lock(tdb
, l
, ltype
, waitflag
);
761 enum TDB_ERROR
tdb_unlock_hashes(struct tdb_context
*tdb
,
763 tdb_len_t hash_range
, int ltype
)
765 unsigned l
= TDB_HASH_LOCK_START
766 + (hash_lock
>> (64 - TDB_HASH_LOCK_RANGE_BITS
));
768 if (tdb
->flags
& TDB_NOLOCK
)
771 /* a allrecord lock allows us to avoid per chain locks */
772 if (tdb
->file
->allrecord_lock
.count
) {
773 if (tdb
->file
->allrecord_lock
.ltype
== F_RDLCK
774 && ltype
== F_WRLCK
) {
775 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
776 "tdb_unlock_hashes RO allrecord!");
781 return tdb_nest_unlock(tdb
, l
, ltype
);
784 /* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
785 * Then we begin; bucket offsets are sizeof(tdb_len_t) apart, so we divide.
786 * The result is that on 32 bit systems we don't use lock values > 2^31 on
787 * files that are less than 4GB.
789 static tdb_off_t
free_lock_off(tdb_off_t b_off
)
791 return TDB_HASH_LOCK_START
+ TDB_HASH_LOCK_RANGE
792 + b_off
/ sizeof(tdb_off_t
);
795 enum TDB_ERROR
tdb_lock_free_bucket(struct tdb_context
*tdb
, tdb_off_t b_off
,
796 enum tdb_lock_flags waitflag
)
798 assert(b_off
>= sizeof(struct tdb_header
));
800 if (tdb
->flags
& TDB_NOLOCK
)
803 /* a allrecord lock allows us to avoid per chain locks */
804 if (tdb
->file
->allrecord_lock
.count
) {
805 if (!check_lock_pid(tdb
, "tdb_lock_free_bucket", true))
808 if (tdb
->file
->allrecord_lock
.ltype
== F_WRLCK
)
810 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
811 "tdb_lock_free_bucket with"
812 " read-only allrecordlock!");
816 if (tdb_has_expansion_lock(tdb
)) {
817 return tdb_logerr(tdb
, TDB_ERR_LOCK
, TDB_LOG_ERROR
,
818 "tdb_lock_free_bucket:"
819 " already have expansion lock");
823 return tdb_nest_lock(tdb
, free_lock_off(b_off
), F_WRLCK
, waitflag
);
826 void tdb_unlock_free_bucket(struct tdb_context
*tdb
, tdb_off_t b_off
)
828 if (tdb
->file
->allrecord_lock
.count
)
831 tdb_nest_unlock(tdb
, free_lock_off(b_off
), F_WRLCK
);
834 enum TDB_ERROR
tdb_lockall(struct tdb_context
*tdb
)
836 return tdb_allrecord_lock(tdb
, F_WRLCK
, TDB_LOCK_WAIT
, false);
839 void tdb_unlockall(struct tdb_context
*tdb
)
841 tdb_allrecord_unlock(tdb
, F_WRLCK
);
844 enum TDB_ERROR
tdb_lockall_read(struct tdb_context
*tdb
)
846 return tdb_allrecord_lock(tdb
, F_RDLCK
, TDB_LOCK_WAIT
, false);
849 void tdb_unlockall_read(struct tdb_context
*tdb
)
851 tdb_allrecord_unlock(tdb
, F_RDLCK
);
854 void tdb_lock_cleanup(struct tdb_context
*tdb
)
858 /* We don't want to warn: they're allowed to close tdb after fork. */
859 if (!check_lock_pid(tdb
, "tdb_close", false))
862 while (tdb
->file
->allrecord_lock
.count
863 && tdb
->file
->allrecord_lock
.owner
== tdb
) {
864 tdb_allrecord_unlock(tdb
, tdb
->file
->allrecord_lock
.ltype
);
867 for (i
=0; i
<tdb
->file
->num_lockrecs
; i
++) {
868 if (tdb
->file
->lockrecs
[i
].owner
== tdb
) {
870 tdb
->file
->lockrecs
[i
].off
,
871 tdb
->file
->lockrecs
[i
].ltype
);