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 if (tdb_transaction_recover(tdb) == -1) {
271 tdb_unlock_open(tdb);
272 tdb_allrecord_unlock(tdb, F_WRLCK);
277 /* lock an offset in the database. */
278 static enum TDB_ERROR tdb_nest_lock(struct tdb_context *tdb,
279 tdb_off_t offset, int ltype,
280 enum tdb_lock_flags flags)
282 struct tdb_lock_type *new_lck;
283 enum TDB_ERROR ecode;
285 if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
286 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
287 "tdb_nest_lock: invalid offset %zu ltype=%d",
288 (size_t)offset, ltype);
291 if (tdb->flags & TDB_NOLOCK)
294 add_stat(tdb, locks, 1);
296 new_lck = find_nestlock(tdb, offset);
298 if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
299 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
301 " offset %zu has read lock",
304 /* Just increment the struct, posix locks don't stack. */
309 if (tdb->num_lockrecs
310 && offset >= TDB_HASH_LOCK_START
311 && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
312 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
313 "tdb_nest_lock: already have a hash lock?");
316 new_lck = (struct tdb_lock_type *)realloc(
318 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
319 if (new_lck == NULL) {
320 return tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
322 " unable to allocate %zu lock struct",
323 tdb->num_lockrecs + 1);
325 tdb->lockrecs = new_lck;
327 /* Since fcntl locks don't nest, we do a lock for the first one,
328 and simply bump the count for future ones */
329 ecode = tdb_brlock(tdb, ltype, offset, 1, flags);
330 if (ecode != TDB_SUCCESS) {
334 /* First time we grab a lock, perhaps someone died in commit? */
335 if (!(flags & TDB_LOCK_NOCHECK)
336 && tdb->num_lockrecs == 0
337 && unlikely(tdb_needs_recovery(tdb))) {
338 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, flags);
344 if (ecode != TDB_SUCCESS) {
349 tdb->lockrecs[tdb->num_lockrecs].off = offset;
350 tdb->lockrecs[tdb->num_lockrecs].count = 1;
351 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
357 static enum TDB_ERROR tdb_nest_unlock(struct tdb_context *tdb,
358 tdb_off_t off, int ltype)
360 struct tdb_lock_type *lck;
361 enum TDB_ERROR ecode;
363 if (tdb->flags & TDB_NOLOCK)
366 lck = find_nestlock(tdb, off);
367 if ((lck == NULL) || (lck->count == 0)) {
368 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
369 "tdb_nest_unlock: no lock for %zu",
373 if (lck->count > 1) {
379 * This lock has count==1 left, so we need to unlock it in the
380 * kernel. We don't bother with decrementing the in-memory array
381 * element, we're about to overwrite it with the last array element
384 ecode = tdb_brunlock(tdb, ltype, off, 1);
387 * Shrink the array by overwriting the element just unlocked with the
388 * last array element.
390 *lck = tdb->lockrecs[--tdb->num_lockrecs];
396 get the transaction lock
398 enum TDB_ERROR tdb_transaction_lock(struct tdb_context *tdb, int ltype)
400 return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
404 release the transaction lock
406 void tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
408 tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
411 /* We only need to lock individual bytes, but Linux merges consecutive locks
412 * so we lock in contiguous ranges. */
413 static enum TDB_ERROR tdb_lock_gradual(struct tdb_context *tdb,
414 int ltype, enum tdb_lock_flags flags,
415 tdb_off_t off, tdb_off_t len)
417 enum TDB_ERROR ecode;
418 enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
421 /* 0 would mean to end-of-file... */
423 /* Single hash. Just do blocking lock. */
424 return tdb_brlock(tdb, ltype, off, len, flags);
427 /* First we try non-blocking. */
428 if (tdb_brlock(tdb, ltype, off, len, nb_flags) == TDB_SUCCESS) {
432 /* Try locking first half, then second. */
433 ecode = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
434 if (ecode != TDB_SUCCESS)
437 ecode = tdb_lock_gradual(tdb, ltype, flags,
438 off + len / 2, len - len / 2);
439 if (ecode != TDB_SUCCESS) {
440 tdb_brunlock(tdb, ltype, off, len / 2);
445 /* lock/unlock entire database. It can only be upgradable if you have some
446 * other way of guaranteeing exclusivity (ie. transaction write lock). */
447 enum TDB_ERROR tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
448 enum tdb_lock_flags flags, bool upgradable)
450 enum TDB_ERROR ecode;
452 /* FIXME: There are no locks on read-only dbs */
453 if (tdb->read_only) {
454 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
455 "tdb_allrecord_lock: read-only");
458 if (tdb->allrecord_lock.count
459 && (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
460 tdb->allrecord_lock.count++;
464 if (tdb->allrecord_lock.count) {
465 /* a global lock of a different type exists */
466 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
467 "tdb_allrecord_lock: already have %s lock",
468 tdb->allrecord_lock.ltype == F_RDLCK
472 if (tdb_has_hash_locks(tdb)) {
473 /* can't combine global and chain locks */
474 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
475 "tdb_allrecord_lock:"
476 " already have chain lock");
479 if (upgradable && ltype != F_RDLCK) {
480 /* tdb error: you can't upgrade a write lock! */
481 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
482 "tdb_allrecord_lock:"
483 " can't upgrade a write lock");
486 add_stat(tdb, locks, 1);
488 /* Lock hashes, gradually. */
489 ecode = tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
490 TDB_HASH_LOCK_RANGE);
491 if (ecode != TDB_SUCCESS) {
492 if (!(flags & TDB_LOCK_PROBE)) {
493 tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
494 "tdb_allrecord_lock hashes failed");
499 /* Lock free tables: there to end of file. */
500 ecode = tdb_brlock(tdb, ltype,
501 TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
503 if (ecode != TDB_SUCCESS) {
504 if (!(flags & TDB_LOCK_PROBE)) {
505 tdb_logerr(tdb, ecode, TDB_LOG_ERROR,
506 "tdb_allrecord_lock freetables failed");
508 tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
509 TDB_HASH_LOCK_RANGE);
513 tdb->allrecord_lock.count = 1;
514 /* If it's upgradable, it's actually exclusive so we can treat
515 * it as a write lock. */
516 tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
517 tdb->allrecord_lock.off = upgradable;
519 /* Now check for needing recovery. */
520 if (!(flags & TDB_LOCK_NOCHECK) && unlikely(tdb_needs_recovery(tdb))) {
521 tdb_allrecord_unlock(tdb, ltype);
522 ecode = tdb_lock_and_recover(tdb);
523 if (ecode != TDB_SUCCESS) {
532 enum TDB_ERROR tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
534 return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
537 void tdb_unlock_open(struct tdb_context *tdb)
539 tdb_nest_unlock(tdb, TDB_OPEN_LOCK, F_WRLCK);
542 bool tdb_has_open_lock(struct tdb_context *tdb)
544 return find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
547 enum TDB_ERROR tdb_lock_expand(struct tdb_context *tdb, int ltype)
549 /* Lock doesn't protect data, so don't check (we recurse if we do!) */
550 return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype,
551 TDB_LOCK_WAIT | TDB_LOCK_NOCHECK);
554 void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
556 tdb_nest_unlock(tdb, TDB_EXPANSION_LOCK, ltype);
559 /* unlock entire db */
560 void tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
562 if (tdb->allrecord_lock.count == 0) {
563 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
564 "tdb_allrecord_unlock: not locked!");
568 /* Upgradable locks are marked as write locks. */
569 if (tdb->allrecord_lock.ltype != ltype
570 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
571 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
572 "tdb_allrecord_unlock: have %s lock",
573 tdb->allrecord_lock.ltype == F_RDLCK
578 if (tdb->allrecord_lock.count > 1) {
579 tdb->allrecord_lock.count--;
583 tdb->allrecord_lock.count = 0;
584 tdb->allrecord_lock.ltype = 0;
586 tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
589 bool tdb_has_expansion_lock(struct tdb_context *tdb)
591 return find_nestlock(tdb, TDB_EXPANSION_LOCK) != NULL;
594 bool tdb_has_hash_locks(struct tdb_context *tdb)
598 for (i=0; i<tdb->num_lockrecs; i++) {
599 if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
600 && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
601 + TDB_HASH_LOCK_RANGE))
607 static bool tdb_has_free_lock(struct tdb_context *tdb)
611 for (i=0; i<tdb->num_lockrecs; i++) {
612 if (tdb->lockrecs[i].off
613 > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE)
619 enum TDB_ERROR tdb_lock_hashes(struct tdb_context *tdb,
621 tdb_len_t hash_range,
622 int ltype, enum tdb_lock_flags waitflag)
624 /* FIXME: Do this properly, using hlock_range */
625 unsigned lock = TDB_HASH_LOCK_START
626 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
628 /* a allrecord lock allows us to avoid per chain locks */
629 if (tdb->allrecord_lock.count &&
630 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
634 if (tdb->allrecord_lock.count) {
635 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_USE_ERROR,
637 " already have %s allrecordlock",
638 tdb->allrecord_lock.ltype == F_RDLCK
642 if (tdb_has_free_lock(tdb)) {
643 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
644 "tdb_lock_hashes: already have free lock");
647 if (tdb_has_expansion_lock(tdb)) {
648 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
650 " already have expansion lock");
653 return tdb_nest_lock(tdb, lock, ltype, waitflag);
656 enum TDB_ERROR tdb_unlock_hashes(struct tdb_context *tdb,
658 tdb_len_t hash_range, int ltype)
660 unsigned lock = TDB_HASH_LOCK_START
661 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
663 /* a allrecord lock allows us to avoid per chain locks */
664 if (tdb->allrecord_lock.count) {
665 if (tdb->allrecord_lock.ltype == F_RDLCK
666 && ltype == F_WRLCK) {
667 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
668 "tdb_unlock_hashes RO allrecord!");
673 return tdb_nest_unlock(tdb, lock, ltype);
676 /* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
677 * Then we begin; bucket offsets are sizeof(tdb_len_t) apart, so we divide.
678 * The result is that on 32 bit systems we don't use lock values > 2^31 on
679 * files that are less than 4GB.
681 static tdb_off_t free_lock_off(tdb_off_t b_off)
683 return TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
684 + b_off / sizeof(tdb_off_t);
687 enum TDB_ERROR tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
688 enum tdb_lock_flags waitflag)
690 assert(b_off >= sizeof(struct tdb_header));
692 /* a allrecord lock allows us to avoid per chain locks */
693 if (tdb->allrecord_lock.count) {
694 if (tdb->allrecord_lock.ltype == F_WRLCK)
696 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
697 "tdb_lock_free_bucket with"
698 " read-only allrecordlock!");
702 if (tdb_has_expansion_lock(tdb)) {
703 return tdb_logerr(tdb, TDB_ERR_LOCK, TDB_LOG_ERROR,
704 "tdb_lock_free_bucket:"
705 " already have expansion lock");
709 return tdb_nest_lock(tdb, free_lock_off(b_off), F_WRLCK, waitflag);
712 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
714 if (tdb->allrecord_lock.count)
717 tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
720 void tdb_lock_init(struct tdb_context *tdb)
722 tdb->num_lockrecs = 0;
723 tdb->lockrecs = NULL;
724 tdb->allrecord_lock.count = 0;