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;
44 return fcntl(tdb->fd, F_SETLKW, &fl);
46 return fcntl(tdb->fd, F_SETLK, &fl);
49 static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
52 #if 0 /* Check they matched up locks and unlocks correctly. */
57 locks = fopen("/proc/locks", "r");
59 while (fgets(line, 80, locks)) {
63 /* eg. 1: FLOCK ADVISORY WRITE 2440 08:01:2180826 0 EOF */
64 p = strchr(line, ':') + 1;
65 if (strncmp(p, " POSIX ADVISORY ", strlen(" POSIX ADVISORY ")))
67 p += strlen(" FLOCK ADVISORY ");
68 if (strncmp(p, "READ ", strlen("READ ")) == 0)
70 else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
75 if (atoi(p) != getpid())
77 p = strchr(strchr(p, ' ') + 1, ' ') + 1;
79 p = strchr(p, ' ') + 1;
80 if (strncmp(p, "EOF", 3) == 0)
83 l = atoi(p) - start + 1;
87 fprintf(stderr, "Len %u should be %u: %s",
92 fprintf(stderr, "Type %s wrong: %s",
93 rw == F_RDLCK ? "READ" : "WRITE", line);
102 fprintf(stderr, "Unlock on %u@%u not found!\n",
111 fl.l_whence = SEEK_SET;
116 return fcntl(tdb->fd, F_SETLKW, &fl);
119 /* a byte range locking function - return 0 on success
120 this functions locks/unlocks 1 byte at the specified offset.
122 note that a len of zero means lock to end of file
124 static int tdb_brlock(struct tdb_context *tdb,
125 int rw_type, tdb_off_t offset, tdb_off_t len,
126 enum tdb_lock_flags flags)
130 if (tdb->flags & TDB_NOLOCK) {
134 if (rw_type == F_WRLCK && tdb->read_only) {
135 tdb->ecode = TDB_ERR_RDONLY;
139 /* A 32 bit system cannot open a 64-bit file, but it could have
140 * expanded since then: check here. */
141 if ((size_t)(offset + len) != offset + len) {
142 tdb->ecode = TDB_ERR_IO;
143 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
144 "tdb_brlock: lock on giant offset %llu\n",
145 (long long)(offset + len));
150 ret = fcntl_lock(tdb, rw_type, offset, len,
151 flags & TDB_LOCK_WAIT);
152 } while (ret == -1 && errno == EINTR);
155 tdb->ecode = TDB_ERR_LOCK;
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->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
161 "tdb_brlock failed (fd=%d) at"
162 " offset %llu rw_type=%d flags=%d len=%llu\n",
163 tdb->fd, (long long)offset, rw_type,
164 flags, (long long)len);
171 static int tdb_brunlock(struct tdb_context *tdb,
172 int rw_type, tdb_off_t offset, size_t len)
176 if (tdb->flags & TDB_NOLOCK) {
181 ret = fcntl_unlock(tdb, rw_type, offset, len);
182 } while (ret == -1 && errno == EINTR);
185 tdb->log(tdb, TDB_DEBUG_TRACE, tdb->log_priv,
186 "tdb_brunlock failed (fd=%d) at offset %llu"
187 " rw_type=%d len=%llu\n",
188 tdb->fd, (long long)offset, rw_type, (long long)len);
194 upgrade a read lock to a write lock. This needs to be handled in a
195 special way as some OSes (such as solaris) have too conservative
196 deadlock detection and claim a deadlock when progress can be
197 made. For those OSes we may loop for a while.
199 int tdb_allrecord_upgrade(struct tdb_context *tdb)
203 if (tdb->allrecord_lock.count != 1) {
204 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
205 "tdb_allrecord_upgrade failed: count %u too high\n",
206 tdb->allrecord_lock.count);
210 if (tdb->allrecord_lock.off != 1) {
211 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
212 "tdb_allrecord_upgrade failed: already upgraded?\n");
218 if (tdb_brlock(tdb, F_WRLCK,
219 TDB_HASH_LOCK_START, 0,
220 TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
221 tdb->allrecord_lock.ltype = F_WRLCK;
222 tdb->allrecord_lock.off = 0;
225 if (errno != EDEADLK) {
228 /* sleep for as short a time as we can - more portable than usleep() */
231 select(0, NULL, NULL, NULL, &tv);
233 tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
234 "tdb_allrecord_upgrade failed\n");
238 static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
243 for (i=0; i<tdb->num_lockrecs; i++) {
244 if (tdb->lockrecs[i].off == offset) {
245 return &tdb->lockrecs[i];
251 int tdb_lock_and_recover(struct tdb_context *tdb)
255 if (tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
260 if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
261 tdb_allrecord_unlock(tdb, F_WRLCK);
264 ret = tdb_transaction_recover(tdb);
266 tdb_unlock_open(tdb);
267 tdb_allrecord_unlock(tdb, F_WRLCK);
272 /* lock an offset in the database. */
273 static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
274 enum tdb_lock_flags flags)
276 struct tdb_lock_type *new_lck;
278 if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
279 tdb->ecode = TDB_ERR_LOCK;
280 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
281 "tdb_nest_lock: invalid offset %llu ltype=%d\n",
282 (long long)offset, ltype);
286 if (tdb->flags & TDB_NOLOCK)
289 new_lck = find_nestlock(tdb, offset);
291 if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
292 tdb->ecode = TDB_ERR_LOCK;
293 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
294 "tdb_nest_lock: offset %llu has read lock\n",
298 /* Just increment the struct, posix locks don't stack. */
303 if (tdb->num_lockrecs
304 && offset >= TDB_HASH_LOCK_START
305 && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
306 tdb->ecode = TDB_ERR_LOCK;
307 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
308 "tdb_nest_lock: already have a hash lock?\n");
312 new_lck = (struct tdb_lock_type *)realloc(
314 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
315 if (new_lck == NULL) {
316 tdb->ecode = TDB_ERR_OOM;
317 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
318 "tdb_nest_lock: unable to allocate %llu lock struct",
319 (long long)(tdb->num_lockrecs + 1));
323 tdb->lockrecs = new_lck;
325 /* Since fcntl locks don't nest, we do a lock for the first one,
326 and simply bump the count for future ones */
327 if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
331 /* First time we grab a lock, perhaps someone died in commit? */
332 if (!(flags & TDB_LOCK_NOCHECK)
333 && tdb->num_lockrecs == 0
334 && unlikely(tdb_needs_recovery(tdb))) {
335 tdb_brunlock(tdb, ltype, offset, 1);
337 if (tdb_lock_and_recover(tdb) == -1) {
341 if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
346 tdb->lockrecs[tdb->num_lockrecs].off = offset;
347 tdb->lockrecs[tdb->num_lockrecs].count = 1;
348 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
354 static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
357 struct tdb_lock_type *lck;
359 if (tdb->flags & TDB_NOLOCK)
362 lck = find_nestlock(tdb, off);
363 if ((lck == NULL) || (lck->count == 0)) {
364 tdb->ecode = TDB_ERR_LOCK;
365 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
366 "tdb_nest_unlock: no lock for %llu\n", (long long)off);
370 if (lck->count > 1) {
376 * This lock has count==1 left, so we need to unlock it in the
377 * kernel. We don't bother with decrementing the in-memory array
378 * element, we're about to overwrite it with the last array element
381 ret = tdb_brunlock(tdb, ltype, off, 1);
384 * Shrink the array by overwriting the element just unlocked with the
385 * last array element.
387 *lck = tdb->lockrecs[--tdb->num_lockrecs];
393 get the transaction lock
395 int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
397 return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
401 release the transaction lock
403 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
405 return tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
408 /* We only need to lock individual bytes, but Linux merges consecutive locks
409 * so we lock in contiguous ranges. */
410 static int tdb_lock_gradual(struct tdb_context *tdb,
411 int ltype, enum tdb_lock_flags flags,
412 tdb_off_t off, tdb_off_t len)
415 enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
418 /* 0 would mean to end-of-file... */
420 /* Single hash. Just do blocking lock. */
421 return tdb_brlock(tdb, ltype, off, len, flags);
424 /* First we try non-blocking. */
425 ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
430 /* Try locking first half, then second. */
431 ret = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
435 ret = tdb_lock_gradual(tdb, ltype, flags,
436 off + len / 2, len - len / 2);
438 tdb_brunlock(tdb, ltype, off, len / 2);
444 /* lock/unlock entire database. It can only be upgradable if you have some
445 * other way of guaranteeing exclusivity (ie. transaction write lock). */
446 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
447 enum tdb_lock_flags flags, bool upgradable)
449 /* FIXME: There are no locks on read-only dbs */
450 if (tdb->read_only) {
451 tdb->ecode = TDB_ERR_LOCK;
452 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
453 "tdb_allrecord_lock: read-only\n");
457 if (tdb->allrecord_lock.count
458 && (ltype == F_RDLCK || tdb->allrecord_lock.ltype == F_WRLCK)) {
459 tdb->allrecord_lock.count++;
463 if (tdb->allrecord_lock.count) {
464 /* a global lock of a different type exists */
465 tdb->ecode = TDB_ERR_LOCK;
466 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
467 "tdb_allrecord_lock: already have %s lock\n",
468 tdb->allrecord_lock.ltype == F_RDLCK
473 if (tdb_has_hash_locks(tdb)) {
474 /* can't combine global and chain locks */
475 tdb->ecode = TDB_ERR_LOCK;
476 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
477 "tdb_allrecord_lock: already have chain lock\n");
481 if (upgradable && ltype != F_RDLCK) {
482 /* tdb error: you can't upgrade a write lock! */
483 tdb->ecode = TDB_ERR_LOCK;
484 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
485 "tdb_allrecord_lock: can't upgrade a write lock\n");
490 /* Lock hashes, gradually. */
491 if (tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
492 TDB_HASH_LOCK_RANGE)) {
493 if (!(flags & TDB_LOCK_PROBE)) {
494 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
495 "tdb_allrecord_lock hashes failed (%s)\n",
501 /* Lock free lists: there to end of file. */
502 if (tdb_brlock(tdb, ltype, TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
504 if (!(flags & TDB_LOCK_PROBE)) {
505 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
506 "tdb_allrecord_lock freelist failed (%s)\n",
509 tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START,
510 TDB_HASH_LOCK_RANGE);
514 tdb->allrecord_lock.count = 1;
515 /* If it's upgradable, it's actually exclusive so we can treat
516 * it as a write lock. */
517 tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
518 tdb->allrecord_lock.off = upgradable;
520 /* Now check for needing recovery. */
521 if (!(flags & TDB_LOCK_NOCHECK) && unlikely(tdb_needs_recovery(tdb))) {
522 tdb_allrecord_unlock(tdb, ltype);
523 if (tdb_lock_and_recover(tdb) == -1) {
532 int 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 int 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 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
562 /* FIXME: There are no locks on read-only dbs */
563 if (tdb->read_only) {
564 tdb->ecode = TDB_ERR_LOCK;
565 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
566 "tdb_allrecord_unlock: read-only\n");
570 if (tdb->allrecord_lock.count == 0) {
571 tdb->ecode = TDB_ERR_LOCK;
572 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
573 "tdb_allrecord_unlock: not locked!\n");
577 /* Upgradable locks are marked as write locks. */
578 if (tdb->allrecord_lock.ltype != ltype
579 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
580 tdb->ecode = TDB_ERR_LOCK;
581 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
582 "tdb_allrecord_unlock: have %s lock\n",
583 tdb->allrecord_lock.ltype == F_RDLCK
588 if (tdb->allrecord_lock.count > 1) {
589 tdb->allrecord_lock.count--;
593 tdb->allrecord_lock.count = 0;
594 tdb->allrecord_lock.ltype = 0;
596 return tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
599 bool tdb_has_expansion_lock(struct tdb_context *tdb)
601 return find_nestlock(tdb, TDB_EXPANSION_LOCK) != NULL;
604 bool tdb_has_hash_locks(struct tdb_context *tdb)
608 for (i=0; i<tdb->num_lockrecs; i++) {
609 if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
610 && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
611 + TDB_HASH_LOCK_RANGE))
617 static bool tdb_has_free_lock(struct tdb_context *tdb)
621 for (i=0; i<tdb->num_lockrecs; i++) {
622 if (tdb->lockrecs[i].off
623 > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE)
629 int tdb_lock_hashes(struct tdb_context *tdb,
631 tdb_len_t hash_range,
632 int ltype, enum tdb_lock_flags waitflag)
634 /* FIXME: Do this properly, using hlock_range */
635 unsigned lock = TDB_HASH_LOCK_START
636 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
638 /* a allrecord lock allows us to avoid per chain locks */
639 if (tdb->allrecord_lock.count &&
640 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
644 if (tdb->allrecord_lock.count) {
645 tdb->ecode = TDB_ERR_LOCK;
646 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
647 "tdb_lock_hashes: have %s allrecordlock\n",
648 tdb->allrecord_lock.ltype == F_RDLCK
653 if (tdb_has_free_lock(tdb)) {
654 tdb->ecode = TDB_ERR_LOCK;
655 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
656 "tdb_lock_hashes: have free lock already\n");
660 if (tdb_has_expansion_lock(tdb)) {
661 tdb->ecode = TDB_ERR_LOCK;
662 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
663 "tdb_lock_hashes: have expansion lock already\n");
667 return tdb_nest_lock(tdb, lock, ltype, waitflag);
670 int tdb_unlock_hashes(struct tdb_context *tdb,
672 tdb_len_t hash_range, int ltype)
674 unsigned lock = TDB_HASH_LOCK_START
675 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
677 /* a allrecord lock allows us to avoid per chain locks */
678 if (tdb->allrecord_lock.count) {
679 if (tdb->allrecord_lock.ltype == F_RDLCK
680 && ltype == F_WRLCK) {
681 tdb->ecode = TDB_ERR_LOCK;
682 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
683 "tdb_unlock_hashes RO allrecord!\n");
689 return tdb_nest_unlock(tdb, lock, ltype);
692 /* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
693 * Then we begin; bucket offsets are sizeof(tdb_len_t) apart, so we divide.
694 * The result is that on 32 bit systems we don't use lock values > 2^31 on
695 * files that are less than 4GB.
697 static tdb_off_t free_lock_off(tdb_off_t b_off)
699 return TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
700 + b_off / sizeof(tdb_off_t);
703 int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
704 enum tdb_lock_flags waitflag)
706 assert(b_off >= sizeof(struct tdb_header));
708 /* a allrecord lock allows us to avoid per chain locks */
709 if (tdb->allrecord_lock.count) {
710 if (tdb->allrecord_lock.ltype == F_WRLCK)
712 tdb->ecode = TDB_ERR_LOCK;
713 tdb->log(tdb, TDB_DEBUG_FATAL, tdb->log_priv,
714 "tdb_lock_free_bucket with RO allrecordlock!\n");
719 if (tdb_has_expansion_lock(tdb)) {
720 tdb->ecode = TDB_ERR_LOCK;
721 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
722 "tdb_lock_free_bucket: have expansion lock already\n");
727 return tdb_nest_lock(tdb, free_lock_off(b_off), F_WRLCK, waitflag);
730 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
732 if (tdb->allrecord_lock.count)
735 tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
738 void tdb_lock_init(struct tdb_context *tdb)
740 tdb->num_lockrecs = 0;
741 tdb->lockrecs = NULL;
742 tdb->allrecord_lock.count = 0;