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/>.
28 #include "tdb_private.h"
30 void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
32 tdb->interrupt_sig_ptr = ptr;
35 static int fcntl_lock(struct tdb_context *tdb,
36 int rw, off_t off, off_t len, bool waitflag)
41 fl.l_whence = SEEK_SET;
47 return fcntl(tdb->fd, F_SETLKW, &fl);
49 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!\n",
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 On error, errno is also set so that errors are passed back properly
128 note that a len of zero means lock to end of file
130 int tdb_brlock(struct tdb_context *tdb,
131 int rw_type, tdb_off_t offset, size_t len,
132 enum tdb_lock_flags flags)
136 if (tdb->flags & TDB_NOLOCK) {
140 if (flags & TDB_LOCK_MARK_ONLY) {
144 if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
145 tdb->ecode = TDB_ERR_RDONLY;
150 ret = fcntl_lock(tdb, rw_type, offset, len,
151 flags & TDB_LOCK_WAIT);
152 /* Check for a sigalarm break. */
153 if (ret == -1 && errno == EINTR &&
154 tdb->interrupt_sig_ptr &&
155 *tdb->interrupt_sig_ptr) {
158 } while (ret == -1 && errno == EINTR);
161 tdb->ecode = TDB_ERR_LOCK;
162 /* Generic lock error. errno set by fcntl.
163 * EAGAIN is an expected return from non-blocking
165 if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
166 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d flags=%d len=%d\n",
167 tdb->fd, offset, rw_type, flags, (int)len));
174 int tdb_brunlock(struct tdb_context *tdb,
175 int rw_type, tdb_off_t offset, size_t len)
179 if (tdb->flags & TDB_NOLOCK) {
184 ret = fcntl_unlock(tdb, rw_type, offset, len);
185 } while (ret == -1 && errno == EINTR);
188 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brunlock failed (fd=%d) at offset %d rw_type=%d len=%d\n",
189 tdb->fd, offset, rw_type, (int)len));
195 upgrade a read lock to a write lock. This needs to be handled in a
196 special way as some OSes (such as solaris) have too conservative
197 deadlock detection and claim a deadlock when progress can be
198 made. For those OSes we may loop for a while.
200 int tdb_allrecord_upgrade(struct tdb_context *tdb)
204 if (tdb->allrecord_lock.count != 1) {
205 TDB_LOG((tdb, TDB_DEBUG_ERROR,
206 "tdb_allrecord_upgrade failed: count %u too high\n",
207 tdb->allrecord_lock.count));
211 if (tdb->allrecord_lock.off != 1) {
212 TDB_LOG((tdb, TDB_DEBUG_ERROR,
213 "tdb_allrecord_upgrade failed: already upgraded?\n"));
219 if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 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_TRACE,"tdb_allrecord_upgrade failed\n"));
237 /* list -1 is the alloc list, otherwise a hash chain. */
238 static tdb_off_t lock_offset(int list)
240 return FREELIST_TOP + 4*list;
243 static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
248 for (i=0; i<tdb->num_lockrecs; i++) {
249 if (tdb->lockrecs[i].off == offset) {
250 return &tdb->lockrecs[i];
256 /* lock an offset in the database. */
257 int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
258 enum tdb_lock_flags flags)
260 struct tdb_lock_type *new_lck;
262 if (offset >= lock_offset(tdb->header.hash_size)) {
263 tdb->ecode = TDB_ERR_LOCK;
264 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid offset %u for ltype=%d\n",
268 if (tdb->flags & TDB_NOLOCK)
271 new_lck = find_nestlock(tdb, offset);
274 * Just increment the in-memory struct, posix locks
281 new_lck = (struct tdb_lock_type *)realloc(
283 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
284 if (new_lck == NULL) {
288 tdb->lockrecs = new_lck;
290 /* Since fcntl locks don't nest, we do a lock for the first one,
291 and simply bump the count for future ones */
292 if (tdb->methods->brlock(tdb, ltype, offset, 1, flags)) {
296 tdb->lockrecs[tdb->num_lockrecs].off = offset;
297 tdb->lockrecs[tdb->num_lockrecs].count = 1;
298 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
304 /* lock a list in the database. list -1 is the alloc list */
305 int tdb_lock(struct tdb_context *tdb, int list, int ltype)
309 /* a allrecord lock allows us to avoid per chain locks */
310 if (tdb->allrecord_lock.count &&
311 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
315 if (tdb->allrecord_lock.count) {
316 tdb->ecode = TDB_ERR_LOCK;
319 ret = tdb_nest_lock(tdb, lock_offset(list), ltype,
323 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
324 "ltype=%d (%s)\n", list, ltype, strerror(errno)));
329 /* lock a list in the database. list -1 is the alloc list. non-blocking lock */
330 int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
332 /* a allrecord lock allows us to avoid per chain locks */
333 if (tdb->allrecord_lock.count &&
334 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
338 if (tdb->allrecord_lock.count) {
339 tdb->ecode = TDB_ERR_LOCK;
343 return tdb_nest_lock(tdb, lock_offset(list), ltype, TDB_LOCK_NOWAIT);
347 int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
351 struct tdb_lock_type *lck;
353 if (tdb->flags & TDB_NOLOCK)
357 if (offset >= lock_offset(tdb->header.hash_size)) {
358 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: offset %u invalid (%d)\n", offset, tdb->header.hash_size));
362 lck = find_nestlock(tdb, offset);
363 if ((lck == NULL) || (lck->count == 0)) {
364 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
368 if (lck->count > 1) {
374 * This lock has count==1 left, so we need to unlock it in the
375 * kernel. We don't bother with decrementing the in-memory array
376 * element, we're about to overwrite it with the last array element
383 ret = tdb->methods->brunlock(tdb, ltype, offset, 1);
387 * Shrink the array by overwriting the element just unlocked with the
388 * last array element.
390 *lck = tdb->lockrecs[--tdb->num_lockrecs];
393 * We don't bother with realloc when the array shrinks, but if we have
394 * a completely idle tdb we should get rid of the locked array.
397 if (tdb->num_lockrecs == 0) {
398 SAFE_FREE(tdb->lockrecs);
402 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
406 int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
408 /* a global lock allows us to avoid per chain locks */
409 if (tdb->allrecord_lock.count &&
410 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
414 if (tdb->allrecord_lock.count) {
415 tdb->ecode = TDB_ERR_LOCK;
419 return tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
423 get the transaction lock
425 int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
427 return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
431 release the transaction lock
433 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
435 return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
439 /* lock/unlock entire database. It can only be upgradable if you have some
440 * other way of guaranteeing exclusivity (ie. transaction write lock). */
441 int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
442 enum tdb_lock_flags flags, bool upgradable)
444 /* There are no locks on read-only dbs */
445 if (tdb->read_only || tdb->traverse_read) {
446 tdb->ecode = TDB_ERR_LOCK;
450 if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
451 tdb->allrecord_lock.count++;
455 if (tdb->allrecord_lock.count) {
456 /* a global lock of a different type exists */
457 tdb->ecode = TDB_ERR_LOCK;
461 if (tdb_have_extra_locks(tdb)) {
462 /* can't combine global and chain locks */
463 tdb->ecode = TDB_ERR_LOCK;
467 if (upgradable && ltype != F_RDLCK) {
468 /* tdb error: you can't upgrade a write lock! */
469 tdb->ecode = TDB_ERR_LOCK;
473 if (tdb->methods->brlock(tdb, ltype, FREELIST_TOP, 0, flags)) {
474 if (flags & TDB_LOCK_WAIT) {
475 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
480 tdb->allrecord_lock.count = 1;
481 /* If it's upgradable, it's actually exclusive so we can treat
482 * it as a write lock. */
483 tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
484 tdb->allrecord_lock.off = upgradable;
491 /* unlock entire db */
492 int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock)
494 /* There are no locks on read-only dbs */
495 if (tdb->read_only || tdb->traverse_read) {
496 tdb->ecode = TDB_ERR_LOCK;
500 if (tdb->allrecord_lock.count == 0) {
501 tdb->ecode = TDB_ERR_LOCK;
505 /* Upgradable locks are marked as write locks. */
506 if (tdb->allrecord_lock.ltype != ltype
507 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
508 tdb->ecode = TDB_ERR_LOCK;
512 if (tdb->allrecord_lock.count > 1) {
513 tdb->allrecord_lock.count--;
518 tdb->methods->brunlock(tdb, ltype, FREELIST_TOP, 0)) {
519 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
523 tdb->allrecord_lock.count = 0;
524 tdb->allrecord_lock.ltype = 0;
529 /* lock entire database with write lock */
530 int tdb_lockall(struct tdb_context *tdb)
532 tdb_trace(tdb, "tdb_lockall");
533 return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
536 /* lock entire database with write lock - mark only */
537 int tdb_lockall_mark(struct tdb_context *tdb)
539 tdb_trace(tdb, "tdb_lockall_mark");
540 return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY, false);
543 /* unlock entire database with write lock - unmark only */
544 int tdb_lockall_unmark(struct tdb_context *tdb)
546 tdb_trace(tdb, "tdb_lockall_unmark");
547 return tdb_allrecord_unlock(tdb, F_WRLCK, true);
550 /* lock entire database with write lock - nonblocking varient */
551 int tdb_lockall_nonblock(struct tdb_context *tdb)
553 int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
554 tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
558 /* unlock entire database with write lock */
559 int tdb_unlockall(struct tdb_context *tdb)
561 tdb_trace(tdb, "tdb_unlockall");
562 return tdb_allrecord_unlock(tdb, F_WRLCK, false);
565 /* lock entire database with read lock */
566 int tdb_lockall_read(struct tdb_context *tdb)
568 tdb_trace(tdb, "tdb_lockall_read");
569 return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
572 /* lock entire database with read lock - nonblock varient */
573 int tdb_lockall_read_nonblock(struct tdb_context *tdb)
575 int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
576 tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
580 /* unlock entire database with read lock */
581 int tdb_unlockall_read(struct tdb_context *tdb)
583 tdb_trace(tdb, "tdb_unlockall_read");
584 return tdb_allrecord_unlock(tdb, F_RDLCK, false);
587 /* lock/unlock one hash chain. This is meant to be used to reduce
588 contention - it cannot guarantee how many records will be locked */
589 int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
591 int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
592 tdb_trace_1rec(tdb, "tdb_chainlock", key);
596 /* lock/unlock one hash chain, non-blocking. This is meant to be used
597 to reduce contention - it cannot guarantee how many records will be
599 int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
601 int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
602 tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
606 /* mark a chain as locked without actually locking it. Warning! use with great caution! */
607 int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
609 int ret = tdb_nest_lock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
610 F_WRLCK, TDB_LOCK_MARK_ONLY);
611 tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
615 /* unmark a chain as locked without actually locking it. Warning! use with great caution! */
616 int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
618 tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
619 return tdb_nest_unlock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
623 int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
625 tdb_trace_1rec(tdb, "tdb_chainunlock", key);
626 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
629 int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
632 ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
633 tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
637 int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
639 tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
640 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
645 /* record lock stops delete underneath */
646 int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
648 if (tdb->allrecord_lock.count) {
651 return off ? tdb->methods->brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
655 Write locks override our own fcntl readlocks, so check it here.
656 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
657 an error to fail to get the lock here.
659 int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
661 struct tdb_traverse_lock *i;
662 for (i = &tdb->travlocks; i; i = i->next)
665 return tdb->methods->brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
668 int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
670 return tdb->methods->brunlock(tdb, F_WRLCK, off, 1);
673 /* fcntl locks don't stack: avoid unlocking someone else's */
674 int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
676 struct tdb_traverse_lock *i;
679 if (tdb->allrecord_lock.count) {
685 for (i = &tdb->travlocks; i; i = i->next)
688 return (count == 1 ? tdb->methods->brunlock(tdb, F_RDLCK, off, 1) : 0);
691 bool tdb_have_extra_locks(struct tdb_context *tdb)
693 unsigned int extra = tdb->num_lockrecs;
695 /* A transaction holds the lock for all records. */
696 if (!tdb->transaction && tdb->allrecord_lock.count) {
700 /* We always hold the active lock if CLEAR_IF_FIRST. */
701 if (find_nestlock(tdb, ACTIVE_LOCK)) {
705 /* In a transaction, we expect to hold the transaction lock */
706 if (tdb->transaction && find_nestlock(tdb, TRANSACTION_LOCK)) {
713 /* The transaction code uses this to remove all locks. Note that this
714 may include OPEN_LOCK. */
715 void tdb_release_extra_locks(struct tdb_context *tdb)
717 unsigned int i, extra = 0;
719 if (tdb->allrecord_lock.count != 0) {
720 tdb_brunlock(tdb, tdb->allrecord_lock.ltype, FREELIST_TOP, 0);
721 tdb->allrecord_lock.count = 0;
724 for (i=0;i<tdb->num_lockrecs;i++) {
725 struct tdb_lock_type *lck = &tdb->lockrecs[i];
727 /* Don't release transaction or active locks! */
728 if (tdb->transaction && lck->off == TRANSACTION_LOCK) {
729 tdb->lockrecs[extra++] = *lck;
730 } else if (lck->off == ACTIVE_LOCK) {
731 tdb->lockrecs[extra++] = *lck;
733 tdb_brunlock(tdb, lck->ltype, lck->off, 1);
736 tdb->num_lockrecs = extra;
737 if (tdb->num_lockrecs == 0) {
738 SAFE_FREE(tdb->lockrecs);