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 static int fcntl_lock(struct tdb_context *tdb,
33 int rw, off_t off, off_t len, bool waitflag)
38 fl.l_whence = SEEK_SET;
43 add_stat(tdb, lock_lowlevel, 1);
45 return fcntl(tdb->fd, F_SETLKW, &fl);
47 add_stat(tdb, lock_nonblock, 1);
48 return fcntl(tdb->fd, F_SETLK, &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!",
114 fl.l_whence = SEEK_SET;
119 return fcntl(tdb->fd, F_SETLKW, &fl);
122 /* a byte range locking function - return 0 on success
123 this functions locks/unlocks 1 byte at the specified offset.
125 note that a len of zero means lock to end of file
127 static enum TDB_ERROR tdb_brlock(struct tdb_context *tdb,
128 int rw_type, tdb_off_t offset, tdb_off_t len,
129 enum tdb_lock_flags flags)
133 if (tdb->flags & TDB_NOLOCK) {
137 if (rw_type == F_WRLCK && tdb->read_only) {
138 return tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_LOG_USE_ERROR,
139 "Write lock attempted on read-only database");
142 /* A 32 bit system cannot open a 64-bit file, but it could have
143 * expanded since then: check here. */
144 if ((size_t)(offset + len) != offset + len) {
145 return tdb_logerr(tdb, TDB_ERR_IO, TDB_LOG_ERROR,
146 "tdb_brlock: lock on giant offset %llu",
147 (long long)(offset + len));
151 ret = fcntl_lock(tdb, rw_type, offset, len,
152 flags & TDB_LOCK_WAIT);
153 } while (ret == -1 && errno == EINTR);
156 /* Generic lock error. errno set by fcntl.
157 * EAGAIN is an expected return from non-blocking
159 if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
160 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
161 "tdb_brlock failed (fd=%d) at"
162 " offset %zu rw_type=%d flags=%d len=%zu:"
164 tdb->fd, (size_t)offset, rw_type,
165 flags, (size_t)len, strerror(errno));
172 static enum TDB_ERROR tdb_brunlock(struct tdb_context *tdb,
173 int rw_type, tdb_off_t offset, size_t len)
177 if (tdb->flags & TDB_NOLOCK) {
182 ret = fcntl_unlock(tdb, rw_type, offset, len);
183 } while (ret == -1 && errno == EINTR);
186 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
187 "tdb_brunlock failed (fd=%d) at offset %zu"
188 " rw_type=%d len=%zu",
189 tdb->fd, (size_t)offset, rw_type,
196 upgrade a read lock to a write lock. This needs to be handled in a
197 special way as some OSes (such as solaris) have too conservative
198 deadlock detection and claim a deadlock when progress can be
199 made. For those OSes we may loop for a while.
201 enum TDB_ERROR tdb_allrecord_upgrade(struct tdb_context *tdb)
205 if (tdb->allrecord_lock.count != 1) {
206 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
207 "tdb_allrecord_upgrade failed:"
208 " count %u too high",
209 tdb->allrecord_lock.count);
212 if (tdb->allrecord_lock.off != 1) {
213 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
214 "tdb_allrecord_upgrade failed:"
215 " already upgraded?");
220 if (tdb_brlock(tdb, F_WRLCK,
221 TDB_HASH_LOCK_START, 0,
222 TDB_LOCK_WAIT|TDB_LOCK_PROBE) == TDB_SUCCESS) {
223 tdb->allrecord_lock.ltype = F_WRLCK;
224 tdb->allrecord_lock.off = 0;
227 if (errno != EDEADLK) {
230 /* sleep for as short a time as we can - more portable than usleep() */
233 select(0, NULL, NULL, NULL, &tv);
235 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
236 "tdb_allrecord_upgrade failed");
239 static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
244 for (i=0; i<tdb->num_lockrecs; i++) {
245 if (tdb->lockrecs[i].off == offset) {
246 return &tdb->lockrecs[i];
252 enum TDB_ERROR tdb_lock_and_recover(struct tdb_context *tdb)
254 enum TDB_ERROR ecode;
256 ecode = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
258 if (ecode != TDB_SUCCESS) {
262 ecode = tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK);
263 if (ecode != TDB_SUCCESS) {
264 tdb_allrecord_unlock(tdb, F_WRLCK);
267 ecode = tdb_transaction_recover(tdb);
268 tdb_unlock_open(tdb);
269 tdb_allrecord_unlock(tdb, F_WRLCK);
274 /* lock an offset in the database. */
275 static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
276 tdb_off_t offset, int ltype,
277 enum tdb_lock_flags flags)
279 struct tdb_lock_type *new_lck;
280 enum TDB_ERROR ecode;
282 if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
283 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
284 "tdb_nest_lock: invalid offset %zu ltype=%d",
285 (size_t)offset, ltype);
288 if (tdb->flags & TDB_NOLOCK)
291 add_stat(tdb, locks, 1);
293 new_lck = find_nestlock(tdb, offset);
295 if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
296 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
298 " offset %zu has read lock",
301 /* Just increment the struct, posix locks don't stack. */
306 if (tdb->num_lockrecs
307 && offset >= TDB_HASH_LOCK_START
308 && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
309 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
310 "tdb_nest_lock: already have a hash lock?");
313 new_lck = (struct tdb_lock_type *)realloc(
315 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
316 if (new_lck == NULL) {
317 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
319 " unable to allocate %zu lock struct",
320 tdb->num_lockrecs + 1);
322 tdb->lockrecs = new_lck;
324 /* Since fcntl locks don't nest, we do a lock for the first one,
325 and simply bump the count for future ones */
326 ecode = tdb_brlock(tdb, ltype, offset, 1, flags);
327 if (ecode != TDB_SUCCESS) {
331 /* First time we grab a lock, perhaps someone died in commit? */
332 if (!(flags & TDB_LOCK_NOCHECK)
333 && tdb->num_lockrecs == 0) {
334 tdb_bool_err berr = tdb_needs_recovery(tdb);
336 tdb_brunlock(tdb, ltype, offset, 1);
340 ecode = tdb_lock_and_recover(tdb);
341 if (ecode == TDB_SUCCESS) {
342 ecode = tdb_brlock(tdb, ltype, offset, 1,
345 if (ecode != TDB_SUCCESS) {
351 tdb->lockrecs[tdb->num_lockrecs].off = offset;
352 tdb->lockrecs[tdb->num_lockrecs].count = 1;
353 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
359 static enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
360 tdb_off_t off, int ltype)
362 struct tdb_lock_type *lck;
363 enum TDB_ERROR ecode;
365 if (tdb->flags & TDB_NOLOCK)
368 lck = find_nestlock(tdb, off);
369 if ((lck == NULL) || (lck->count == 0)) {
370 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
371 "tdb_nest_unlock: no lock for %zu",
375 if (lck->count > 1) {
381 * This lock has count==1 left, so we need to unlock it in the
382 * kernel. We don't bother with decrementing the in-memory array
383 * element, we're about to overwrite it with the last array element
386 ecode = tdb_brunlock(tdb, ltype, off, 1);
389 * Shrink the array by overwriting the element just unlocked with the
390 * last array element.
392 *lck = tdb->lockrecs[--tdb->num_lockrecs];
398 get the transaction lock
400 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype)
402 return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
406 release the transaction lock
408 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
410 tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
413 /* We only need to lock individual bytes, but Linux merges consecutive locks
414 * so we lock in contiguous ranges. */
415 static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
416 int ltype, enum tdb_lock_flags flags,
417 tdb_off_t off, tdb_off_t len)
419 enum TDB_ERROR ecode;
420 enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
423 /* 0 would mean to end-of-file... */
425 /* Single hash. Just do blocking lock. */
426 return tdb_brlock(tdb, ltype, off, len, flags);
429 /* First we try non-blocking. */
430 if (tdb_brlock(tdb, ltype, off, len, nb_flags) == TDB_SUCCESS) {
434 /* Try locking first half, then second. */
435 ecode = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
436 if (ecode != TDB_SUCCESS)
439 ecode = tdb_lock_gradual(tdb, ltype, flags,
440 off + len / 2, len - len / 2);
441 if (ecode != TDB_SUCCESS) {
442 tdb_brunlock(tdb, ltype, off, len / 2);
447 /* lock/unlock entire database. It can only be upgradable if you have some
448 * other way of guaranteeing exclusivity (ie. transaction write lock). */
449 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
450 enum tdb_lock_flags flags, bool upgradable)
452 enum TDB_ERROR ecode;
455 if (tdb->allrecord_lock.count
456 && (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
457 tdb->allrecord_lock.count++;
461 if (tdb->allrecord_lock.count) {
462 /* a global lock of a different type exists */
463 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
464 "tdb_allrecord_lock: already have %s lock",
465 tdb->allrecord_lock.ltype == F_RDLCK
469 if (tdb_has_hash_locks(tdb)) {
470 /* can't combine global and chain locks */
471 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
472 "tdb_allrecord_lock:"
473 " already have chain lock");
476 if (upgradable && ltype != F_RDLCK) {
477 /* tdb error: you can't upgrade a write lock! */
478 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
479 "tdb_allrecord_lock:"
480 " can't upgrade a write lock");
483 add_stat(tdb, locks, 1);
485 /* Lock hashes, gradually. */
486 ecode = tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
487 TDB_HASH_LOCK_RANGE);
488 if (ecode != TDB_SUCCESS) {
489 if (!(flags & TDB_LOCK_PROBE)) {
490 tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
491 "tdb_allrecord_lock hashes failed");
496 /* Lock free tables: there to end of file. */
497 ecode = tdb_brlock(tdb, ltype,
498 TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
500 if (ecode != TDB_SUCCESS) {
501 if (!(flags & TDB_LOCK_PROBE)) {
502 tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
503 "tdb_allrecord_lock freetables failed");
505 tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
506 TDB_HASH_LOCK_RANGE);
510 tdb->allrecord_lock.count = 1;
511 /* If it's upgradable, it's actually exclusive so we can treat
512 * it as a write lock. */
513 tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
514 tdb->allrecord_lock.off = upgradable;
516 /* Now check for needing recovery. */
517 if (flags & TDB_LOCK_NOCHECK)
520 berr = tdb_needs_recovery(tdb);
521 if (likely(berr == false))
524 tdb_allrecord_unlock(tdb, ltype);
527 ecode = tdb_lock_and_recover(tdb);
528 if (ecode != TDB_SUCCESS) {
534 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
536 return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
539 void tdb_unlock_open(struct tdb_context *tdb)
541 tdb_nest_unlock(tdb, TDB_OPEN_LOCK, F_WRLCK);
544 bool tdb_has_open_lock(struct tdb_context *tdb)
546 return find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
549 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype)
551 /* Lock doesn't protect data, so don't check (we recurse if we do!) */
552 return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype,
553 TDB_LOCK_WAIT | TDB_LOCK_NOCHECK);
556 void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
558 tdb_nest_unlock(tdb, TDB_EXPANSION_LOCK, ltype);
561 /* unlock entire db */
562 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
564 if (tdb->allrecord_lock.count == 0) {
565 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
566 "tdb_allrecord_unlock: not locked!");
570 /* Upgradable locks are marked as write locks. */
571 if (tdb->allrecord_lock.ltype != ltype
572 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
573 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
574 "tdb_allrecord_unlock: have %s lock",
575 tdb->allrecord_lock.ltype == F_RDLCK
580 if (tdb->allrecord_lock.count > 1) {
581 tdb->allrecord_lock.count--;
585 tdb->allrecord_lock.count = 0;
586 tdb->allrecord_lock.ltype = 0;
588 tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
591 bool tdb_has_expansion_lock(struct tdb_context *tdb)
593 return find_nestlock(tdb, TDB_EXPANSION_LOCK) != NULL;
596 bool tdb_has_hash_locks(struct tdb_context *tdb)
600 for (i=0; i<tdb->num_lockrecs; i++) {
601 if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
602 && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
603 + TDB_HASH_LOCK_RANGE))
609 static bool tdb_has_free_lock(struct tdb_context *tdb)
613 for (i=0; i<tdb->num_lockrecs; i++) {
614 if (tdb->lockrecs[i].off
615 > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE)
621 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
623 tdb_len_t hash_range,
624 int ltype, enum tdb_lock_flags waitflag)
626 /* FIXME: Do this properly, using hlock_range */
627 unsigned lock = TDB_HASH_LOCK_START
628 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
630 /* a allrecord lock allows us to avoid per chain locks */
631 if (tdb->allrecord_lock.count &&
632 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
636 if (tdb->allrecord_lock.count) {
637 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
639 " already have %s allrecordlock",
640 tdb->allrecord_lock.ltype == F_RDLCK
644 if (tdb_has_free_lock(tdb)) {
645 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
646 "tdb_lock_hashes: already have free lock");
649 if (tdb_has_expansion_lock(tdb)) {
650 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
652 " already have expansion lock");
655 return tdb_nest_lock(tdb, lock, ltype, waitflag);
658 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
660 tdb_len_t hash_range, int ltype)
662 unsigned lock = TDB_HASH_LOCK_START
663 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
665 /* a allrecord lock allows us to avoid per chain locks */
666 if (tdb->allrecord_lock.count) {
667 if (tdb->allrecord_lock.ltype == F_RDLCK
668 && ltype == F_WRLCK) {
669 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
670 "tdb_unlock_hashes RO allrecord!");
675 return tdb_nest_unlock(tdb, lock, ltype);
678 /* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
679 * Then we begin; bucket offsets are sizeof(tdb_len_t) apart, so we divide.
680 * The result is that on 32 bit systems we don't use lock values > 2^31 on
681 * files that are less than 4GB.
683 static tdb_off_t free_lock_off(tdb_off_t b_off)
685 return TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
686 + b_off / sizeof(tdb_off_t);
689 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
690 enum tdb_lock_flags waitflag)
692 assert(b_off >= sizeof(struct tdb_header));
694 /* a allrecord lock allows us to avoid per chain locks */
695 if (tdb->allrecord_lock.count) {
696 if (tdb->allrecord_lock.ltype == F_WRLCK)
698 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
699 "tdb_lock_free_bucket with"
700 " read-only allrecordlock!");
704 if (tdb_has_expansion_lock(tdb)) {
705 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
706 "tdb_lock_free_bucket:"
707 " already have expansion lock");
711 return tdb_nest_lock(tdb, free_lock_off(b_off), F_WRLCK, waitflag);
714 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
716 if (tdb->allrecord_lock.count)
719 tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
722 void tdb_lock_init(struct tdb_context *tdb)
724 tdb->num_lockrecs = 0;
725 tdb->lockrecs = NULL;
726 tdb->allrecord_lock.count = 0;