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 int 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 tdb_logerr(tdb, TDB_ERR_RDONLY, TDB_DEBUG_WARNING,
139 "Write lock attempted on read-only database");
143 /* A 32 bit system cannot open a 64-bit file, but it could have
144 * expanded since then: check here. */
145 if ((size_t)(offset + len) != offset + len) {
146 tdb_logerr(tdb, TDB_ERR_IO, TDB_DEBUG_ERROR,
147 "tdb_brlock: lock on giant offset %llu",
148 (long long)(offset + len));
153 ret = fcntl_lock(tdb, rw_type, offset, len,
154 flags & TDB_LOCK_WAIT);
155 } while (ret == -1 && errno == EINTR);
158 tdb->ecode = TDB_ERR_LOCK;
159 /* Generic lock error. errno set by fcntl.
160 * EAGAIN is an expected return from non-blocking
162 if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
163 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
164 "tdb_brlock failed (fd=%d) at"
165 " offset %zu rw_type=%d flags=%d len=%zu:"
167 tdb->fd, (size_t)offset, rw_type,
168 flags, (size_t)len, strerror(errno));
175 static int tdb_brunlock(struct tdb_context *tdb,
176 int rw_type, tdb_off_t offset, size_t len)
180 if (tdb->flags & TDB_NOLOCK) {
185 ret = fcntl_unlock(tdb, rw_type, offset, len);
186 } while (ret == -1 && errno == EINTR);
189 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_TRACE,
190 "tdb_brunlock failed (fd=%d) at offset %zu"
191 " rw_type=%d len=%zu",
192 tdb->fd, (size_t)offset, rw_type, (size_t)len);
198 upgrade a read lock to a write lock. This needs to be handled in a
199 special way as some OSes (such as solaris) have too conservative
200 deadlock detection and claim a deadlock when progress can be
201 made. For those OSes we may loop for a while.
203 int tdb_allrecord_upgrade(struct tdb_context *tdb)
207 if (tdb->allrecord_lock.count != 1) {
208 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
209 "tdb_allrecord_upgrade failed: count %u too high",
210 tdb->allrecord_lock.count);
214 if (tdb->allrecord_lock.off != 1) {
215 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
216 "tdb_allrecord_upgrade failed: already upgraded?");
222 if (tdb_brlock(tdb, F_WRLCK,
223 TDB_HASH_LOCK_START, 0,
224 TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
225 tdb->allrecord_lock.ltype = F_WRLCK;
226 tdb->allrecord_lock.off = 0;
229 if (errno != EDEADLK) {
232 /* sleep for as short a time as we can - more portable than usleep() */
235 select(0, NULL, NULL, NULL, &tv);
237 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_WARNING,
238 "tdb_allrecord_upgrade failed");
242 static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
247 for (i=0; i<tdb->num_lockrecs; i++) {
248 if (tdb->lockrecs[i].off == offset) {
249 return &tdb->lockrecs[i];
255 int tdb_lock_and_recover(struct tdb_context *tdb)
259 if (tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK,
264 if (tdb_lock_open(tdb, TDB_LOCK_WAIT|TDB_LOCK_NOCHECK) == -1) {
265 tdb_allrecord_unlock(tdb, F_WRLCK);
268 ret = tdb_transaction_recover(tdb);
270 tdb_unlock_open(tdb);
271 tdb_allrecord_unlock(tdb, F_WRLCK);
276 /* lock an offset in the database. */
277 static int tdb_nest_lock(struct tdb_context *tdb, tdb_off_t offset, int ltype,
278 enum tdb_lock_flags flags)
280 struct tdb_lock_type *new_lck;
282 if (offset > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE + tdb->map_size / 8) {
283 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_FATAL,
284 "tdb_nest_lock: invalid offset %zu ltype=%d",
285 (size_t)offset, ltype);
289 if (tdb->flags & TDB_NOLOCK)
292 add_stat(tdb, locks, 1);
294 new_lck = find_nestlock(tdb, offset);
296 if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
297 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_FATAL,
298 "tdb_nest_lock: offset %zu has read lock",
302 /* Just increment the struct, posix locks don't stack. */
307 if (tdb->num_lockrecs
308 && offset >= TDB_HASH_LOCK_START
309 && offset < TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE) {
310 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_FATAL,
311 "tdb_nest_lock: already have a hash lock?");
315 new_lck = (struct tdb_lock_type *)realloc(
317 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
318 if (new_lck == NULL) {
319 tdb_logerr(tdb, TDB_ERR_OOM, TDB_DEBUG_ERROR,
320 "tdb_nest_lock: unable to allocate %zu lock struct",
321 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 if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
333 /* First time we grab a lock, perhaps someone died in commit? */
334 if (!(flags & TDB_LOCK_NOCHECK)
335 && tdb->num_lockrecs == 0
336 && unlikely(tdb_needs_recovery(tdb))) {
337 tdb_brunlock(tdb, ltype, offset, 1);
339 if (tdb_lock_and_recover(tdb) == -1) {
343 if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
348 tdb->lockrecs[tdb->num_lockrecs].off = offset;
349 tdb->lockrecs[tdb->num_lockrecs].count = 1;
350 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
356 static int tdb_nest_unlock(struct tdb_context *tdb, tdb_off_t off, int ltype)
359 struct tdb_lock_type *lck;
361 if (tdb->flags & TDB_NOLOCK)
364 lck = find_nestlock(tdb, off);
365 if ((lck == NULL) || (lck->count == 0)) {
366 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
367 "tdb_nest_unlock: no lock for %zu", (size_t)off);
371 if (lck->count > 1) {
377 * This lock has count==1 left, so we need to unlock it in the
378 * kernel. We don't bother with decrementing the in-memory array
379 * element, we're about to overwrite it with the last array element
382 ret = tdb_brunlock(tdb, ltype, off, 1);
385 * Shrink the array by overwriting the element just unlocked with the
386 * last array element.
388 *lck = tdb->lockrecs[--tdb->num_lockrecs];
394 get the transaction lock
396 int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
398 return tdb_nest_lock(tdb, TDB_TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
402 release the transaction lock
404 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
406 return tdb_nest_unlock(tdb, TDB_TRANSACTION_LOCK, ltype);
409 /* We only need to lock individual bytes, but Linux merges consecutive locks
410 * so we lock in contiguous ranges. */
411 static int tdb_lock_gradual(struct tdb_context *tdb,
412 int ltype, enum tdb_lock_flags flags,
413 tdb_off_t off, tdb_off_t len)
416 enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
419 /* 0 would mean to end-of-file... */
421 /* Single hash. Just do blocking lock. */
422 return tdb_brlock(tdb, ltype, off, len, flags);
425 /* First we try non-blocking. */
426 ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
431 /* Try locking first half, then second. */
432 ret = tdb_lock_gradual(tdb, ltype, flags, off, len / 2);
436 ret = tdb_lock_gradual(tdb, ltype, flags,
437 off + len / 2, len - len / 2);
439 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 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
448 enum tdb_lock_flags flags, bool upgradable)
450 /* FIXME: There are no locks on read-only dbs */
451 if (tdb->read_only) {
452 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
453 "tdb_allrecord_lock: read-only");
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_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
466 "tdb_allrecord_lock: already have %s lock",
467 tdb->allrecord_lock.ltype == F_RDLCK
472 if (tdb_has_hash_locks(tdb)) {
473 /* can't combine global and chain locks */
474 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
475 "tdb_allrecord_lock: already have chain lock");
479 if (upgradable && ltype != F_RDLCK) {
480 /* tdb error: you can't upgrade a write lock! */
481 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
482 "tdb_allrecord_lock: can't upgrade a write lock");
486 add_stat(tdb, locks, 1);
488 /* Lock hashes, gradually. */
489 if (tdb_lock_gradual(tdb, ltype, flags, TDB_HASH_LOCK_START,
490 TDB_HASH_LOCK_RANGE)) {
491 if (!(flags & TDB_LOCK_PROBE)) {
492 tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_ERROR,
493 "tdb_allrecord_lock hashes failed");
498 /* Lock free tables: there to end of file. */
499 if (tdb_brlock(tdb, ltype, TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE,
501 if (!(flags & TDB_LOCK_PROBE)) {
502 tdb_logerr(tdb, tdb->ecode, TDB_DEBUG_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) && unlikely(tdb_needs_recovery(tdb))) {
518 tdb_allrecord_unlock(tdb, ltype);
519 if (tdb_lock_and_recover(tdb) == -1) {
528 int tdb_lock_open(struct tdb_context *tdb, enum tdb_lock_flags flags)
530 return tdb_nest_lock(tdb, TDB_OPEN_LOCK, F_WRLCK, flags);
533 void tdb_unlock_open(struct tdb_context *tdb)
535 tdb_nest_unlock(tdb, TDB_OPEN_LOCK, F_WRLCK);
538 bool tdb_has_open_lock(struct tdb_context *tdb)
540 return find_nestlock(tdb, TDB_OPEN_LOCK) != NULL;
543 int tdb_lock_expand(struct tdb_context *tdb, int ltype)
545 /* Lock doesn't protect data, so don't check (we recurse if we do!) */
546 return tdb_nest_lock(tdb, TDB_EXPANSION_LOCK, ltype,
547 TDB_LOCK_WAIT | TDB_LOCK_NOCHECK);
550 void tdb_unlock_expand(struct tdb_context *tdb, int ltype)
552 tdb_nest_unlock(tdb, TDB_EXPANSION_LOCK, ltype);
555 /* unlock entire db */
556 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype)
558 if (tdb->allrecord_lock.count == 0) {
559 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
560 "tdb_allrecord_unlock: not locked!");
564 /* Upgradable locks are marked as write locks. */
565 if (tdb->allrecord_lock.ltype != ltype
566 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
567 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
568 "tdb_allrecord_unlock: have %s lock",
569 tdb->allrecord_lock.ltype == F_RDLCK
574 if (tdb->allrecord_lock.count > 1) {
575 tdb->allrecord_lock.count--;
579 tdb->allrecord_lock.count = 0;
580 tdb->allrecord_lock.ltype = 0;
582 return tdb_brunlock(tdb, ltype, TDB_HASH_LOCK_START, 0);
585 bool tdb_has_expansion_lock(struct tdb_context *tdb)
587 return find_nestlock(tdb, TDB_EXPANSION_LOCK) != NULL;
590 bool tdb_has_hash_locks(struct tdb_context *tdb)
594 for (i=0; i<tdb->num_lockrecs; i++) {
595 if (tdb->lockrecs[i].off >= TDB_HASH_LOCK_START
596 && tdb->lockrecs[i].off < (TDB_HASH_LOCK_START
597 + TDB_HASH_LOCK_RANGE))
603 static bool tdb_has_free_lock(struct tdb_context *tdb)
607 for (i=0; i<tdb->num_lockrecs; i++) {
608 if (tdb->lockrecs[i].off
609 > TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE)
615 int tdb_lock_hashes(struct tdb_context *tdb,
617 tdb_len_t hash_range,
618 int ltype, enum tdb_lock_flags waitflag)
620 /* FIXME: Do this properly, using hlock_range */
621 unsigned lock = TDB_HASH_LOCK_START
622 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
624 /* a allrecord lock allows us to avoid per chain locks */
625 if (tdb->allrecord_lock.count &&
626 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
630 if (tdb->allrecord_lock.count) {
631 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
632 "tdb_lock_hashes: already have %s allrecordlock",
633 tdb->allrecord_lock.ltype == F_RDLCK
638 if (tdb_has_free_lock(tdb)) {
639 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
640 "tdb_lock_hashes: already have free lock");
644 if (tdb_has_expansion_lock(tdb)) {
645 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
646 "tdb_lock_hashes: already have expansion lock");
650 return tdb_nest_lock(tdb, lock, ltype, waitflag);
653 int tdb_unlock_hashes(struct tdb_context *tdb,
655 tdb_len_t hash_range, int ltype)
657 unsigned lock = TDB_HASH_LOCK_START
658 + (hash_lock >> (64 - TDB_HASH_LOCK_RANGE_BITS));
660 /* a allrecord lock allows us to avoid per chain locks */
661 if (tdb->allrecord_lock.count) {
662 if (tdb->allrecord_lock.ltype == F_RDLCK
663 && ltype == F_WRLCK) {
664 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_FATAL,
665 "tdb_unlock_hashes RO allrecord!");
671 return tdb_nest_unlock(tdb, lock, ltype);
674 /* Hash locks use TDB_HASH_LOCK_START + the next 30 bits.
675 * Then we begin; bucket offsets are sizeof(tdb_len_t) apart, so we divide.
676 * The result is that on 32 bit systems we don't use lock values > 2^31 on
677 * files that are less than 4GB.
679 static tdb_off_t free_lock_off(tdb_off_t b_off)
681 return TDB_HASH_LOCK_START + TDB_HASH_LOCK_RANGE
682 + b_off / sizeof(tdb_off_t);
685 int tdb_lock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off,
686 enum tdb_lock_flags waitflag)
688 assert(b_off >= sizeof(struct tdb_header));
690 /* a allrecord lock allows us to avoid per chain locks */
691 if (tdb->allrecord_lock.count) {
692 if (tdb->allrecord_lock.ltype == F_WRLCK)
694 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_FATAL,
695 "tdb_lock_free_bucket with RO allrecordlock!");
700 if (tdb_has_expansion_lock(tdb)) {
701 tdb_logerr(tdb, TDB_ERR_LOCK, TDB_DEBUG_ERROR,
702 "tdb_lock_free_bucket: already have expansion lock");
707 return tdb_nest_lock(tdb, free_lock_off(b_off), F_WRLCK, waitflag);
710 void tdb_unlock_free_bucket(struct tdb_context *tdb, tdb_off_t b_off)
712 if (tdb->allrecord_lock.count)
715 tdb_nest_unlock(tdb, free_lock_off(b_off), F_WRLCK);
718 void tdb_lock_init(struct tdb_context *tdb)
720 tdb->num_lockrecs = 0;
721 tdb->lockrecs = NULL;
722 tdb->allrecord_lock.count = 0;