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 wait)
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 /* Generic lock error. errno set by fcntl.
162 * EAGAIN is an expected return from non-blocking
164 if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
165 /* Ensure error code is set for log fun to examine. */
166 tdb->ecode = TDB_ERR_LOCK;
167 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d flags=%d len=%d\n",
168 tdb->fd, offset, rw_type, flags, (int)len));
175 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_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brunlock failed (fd=%d) at offset %d rw_type=%d len=%d\n",
190 tdb->fd, offset, rw_type, (int)len));
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 int tdb_brlock_upgrade(struct tdb_context *tdb, tdb_off_t offset, size_t len)
206 if (tdb_brlock(tdb, F_WRLCK, offset, len,
207 TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
210 if (errno != EDEADLK) {
213 /* sleep for as short a time as we can - more portable than usleep() */
216 select(0, NULL, NULL, NULL, &tv);
218 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock_upgrade failed at offset %d\n", offset));
223 /* lock a list in the database. list -1 is the alloc list */
224 static int _tdb_lock(struct tdb_context *tdb, int list, int ltype,
225 enum tdb_lock_flags flags)
227 struct tdb_lock_type *new_lck;
230 /* a global lock allows us to avoid per chain locks */
231 if (tdb->global_lock.count &&
232 (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
236 if (tdb->global_lock.count) {
237 tdb->ecode = TDB_ERR_LOCK;
241 if (list < -1 || list >= (int)tdb->header.hash_size) {
242 tdb->ecode = TDB_ERR_LOCK;
243 TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid list %d for ltype=%d\n",
247 if (tdb->flags & TDB_NOLOCK)
250 for (i=0; i<tdb->num_lockrecs; i++) {
251 if (tdb->lockrecs[i].list == list) {
252 if (tdb->lockrecs[i].count == 0) {
254 * Can't happen, see tdb_unlock(). It should
257 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock: "
258 "lck->count == 0 for list %d", list));
261 * Just increment the in-memory struct, posix locks
264 tdb->lockrecs[i].count++;
269 new_lck = (struct tdb_lock_type *)realloc(
271 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
272 if (new_lck == NULL) {
276 tdb->lockrecs = new_lck;
278 /* Since fcntl locks don't nest, we do a lock for the first one,
279 and simply bump the count for future ones */
280 if (tdb->methods->brlock(tdb, ltype, FREELIST_TOP+4*list, 1, flags)) {
286 tdb->lockrecs[tdb->num_lockrecs].list = list;
287 tdb->lockrecs[tdb->num_lockrecs].count = 1;
288 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
289 tdb->num_lockrecs += 1;
294 /* lock a list in the database. list -1 is the alloc list */
295 int tdb_lock(struct tdb_context *tdb, int list, int ltype)
299 ret = _tdb_lock(tdb, list, ltype, TDB_LOCK_WAIT);
301 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
302 "ltype=%d (%s)\n", list, ltype, strerror(errno)));
307 /* lock a list in the database. list -1 is the alloc list. non-blocking lock */
308 int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
310 return _tdb_lock(tdb, list, ltype, TDB_LOCK_NOWAIT);
314 static int _tdb_unlock(struct tdb_context *tdb, int list, int ltype,
319 struct tdb_lock_type *lck = NULL;
321 /* a global lock allows us to avoid per chain locks */
322 if (tdb->global_lock.count &&
323 (ltype == tdb->global_lock.ltype || ltype == F_RDLCK)) {
327 if (tdb->global_lock.count) {
328 tdb->ecode = TDB_ERR_LOCK;
332 if (tdb->flags & TDB_NOLOCK)
336 if (list < -1 || list >= (int)tdb->header.hash_size) {
337 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: list %d invalid (%d)\n", list, tdb->header.hash_size));
341 for (i=0; i<tdb->num_lockrecs; i++) {
342 if (tdb->lockrecs[i].list == list) {
343 lck = &tdb->lockrecs[i];
348 if ((lck == NULL) || (lck->count == 0)) {
349 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
353 if (lck->count > 1) {
359 * This lock has count==1 left, so we need to unlock it in the
360 * kernel. We don't bother with decrementing the in-memory array
361 * element, we're about to overwrite it with the last array element
368 ret = tdb->methods->brunlock(tdb, ltype,
369 FREELIST_TOP+4*list, 1);
374 * Shrink the array by overwriting the element just unlocked with the
375 * last array element.
378 if (tdb->num_lockrecs > 1) {
379 *lck = tdb->lockrecs[tdb->num_lockrecs-1];
381 tdb->num_lockrecs -= 1;
384 * We don't bother with realloc when the array shrinks, but if we have
385 * a completely idle tdb we should get rid of the locked array.
388 if (tdb->num_lockrecs == 0) {
389 SAFE_FREE(tdb->lockrecs);
393 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
397 int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
399 return _tdb_unlock(tdb, list, ltype, false);
403 get the transaction lock
405 int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
407 if (tdb->global_lock.count) {
410 if (tdb->transaction_lock_count > 0) {
411 tdb->transaction_lock_count++;
415 if (tdb->methods->brlock(tdb, ltype, TRANSACTION_LOCK, 1, TDB_LOCK_WAIT) == -1) {
416 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_lock: failed to get transaction lock\n"));
417 tdb->ecode = TDB_ERR_LOCK;
420 tdb->transaction_lock_count++;
425 release the transaction lock
427 int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
430 if (tdb->global_lock.count) {
433 if (tdb->transaction_lock_count > 1) {
434 tdb->transaction_lock_count--;
437 ret = tdb->methods->brunlock(tdb, ltype, TRANSACTION_LOCK, 1);
439 tdb->transaction_lock_count = 0;
447 /* lock/unlock entire database */
448 static int _tdb_lockall(struct tdb_context *tdb, int ltype,
449 enum tdb_lock_flags flags)
451 /* There are no locks on read-only dbs */
452 if (tdb->read_only || tdb->traverse_read) {
453 tdb->ecode = TDB_ERR_LOCK;
457 if (tdb->global_lock.count && tdb->global_lock.ltype == ltype) {
458 tdb->global_lock.count++;
462 if (tdb->global_lock.count) {
463 /* a global lock of a different type exists */
464 tdb->ecode = TDB_ERR_LOCK;
468 if (tdb->num_locks != 0) {
469 /* can't combine global and chain locks */
470 tdb->ecode = TDB_ERR_LOCK;
474 if (tdb->methods->brlock(tdb, ltype,
475 FREELIST_TOP, 4*tdb->header.hash_size,
477 if (flags & TDB_LOCK_WAIT) {
478 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lockall failed (%s)\n", strerror(errno)));
483 tdb->global_lock.count = 1;
484 tdb->global_lock.ltype = ltype;
491 /* unlock entire db */
492 static int _tdb_unlockall(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->global_lock.ltype != ltype || tdb->global_lock.count == 0) {
501 tdb->ecode = TDB_ERR_LOCK;
505 if (tdb->global_lock.count > 1) {
506 tdb->global_lock.count--;
511 tdb->methods->brunlock(tdb, F_WRLCK,
512 FREELIST_TOP, 4*tdb->header.hash_size)) {
513 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
517 tdb->global_lock.count = 0;
518 tdb->global_lock.ltype = 0;
523 /* lock entire database with write lock */
524 int tdb_lockall(struct tdb_context *tdb)
526 tdb_trace(tdb, "tdb_lockall");
527 return _tdb_lockall(tdb, F_WRLCK, TDB_LOCK_WAIT);
530 /* lock entire database with write lock - mark only */
531 int tdb_lockall_mark(struct tdb_context *tdb)
533 tdb_trace(tdb, "tdb_lockall_mark");
534 return _tdb_lockall(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY);
537 /* unlock entire database with write lock - unmark only */
538 int tdb_lockall_unmark(struct tdb_context *tdb)
540 tdb_trace(tdb, "tdb_lockall_unmark");
541 return _tdb_unlockall(tdb, F_WRLCK, true);
544 /* lock entire database with write lock - nonblocking varient */
545 int tdb_lockall_nonblock(struct tdb_context *tdb)
547 int ret = _tdb_lockall(tdb, F_WRLCK, TDB_LOCK_NOWAIT);
548 tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
552 /* unlock entire database with write lock */
553 int tdb_unlockall(struct tdb_context *tdb)
555 tdb_trace(tdb, "tdb_unlockall");
556 return _tdb_unlockall(tdb, F_WRLCK, false);
559 /* lock entire database with read lock */
560 int tdb_lockall_read(struct tdb_context *tdb)
562 tdb_trace(tdb, "tdb_lockall_read");
563 return _tdb_lockall(tdb, F_RDLCK, TDB_LOCK_WAIT);
566 /* lock entire database with read lock - nonblock varient */
567 int tdb_lockall_read_nonblock(struct tdb_context *tdb)
569 int ret = _tdb_lockall(tdb, F_RDLCK, TDB_LOCK_NOWAIT);
570 tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
574 /* unlock entire database with read lock */
575 int tdb_unlockall_read(struct tdb_context *tdb)
577 tdb_trace(tdb, "tdb_unlockall_read");
578 return _tdb_unlockall(tdb, F_RDLCK, false);
581 /* lock/unlock one hash chain. This is meant to be used to reduce
582 contention - it cannot guarantee how many records will be locked */
583 int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
585 int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
586 tdb_trace_1rec(tdb, "tdb_chainlock", key);
590 /* lock/unlock one hash chain, non-blocking. This is meant to be used
591 to reduce contention - it cannot guarantee how many records will be
593 int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
595 int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
596 tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
600 /* mark a chain as locked without actually locking it. Warning! use with great caution! */
601 int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
603 int ret = _tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK,
605 tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
609 /* unmark a chain as locked without actually locking it. Warning! use with great caution! */
610 int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
612 tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
613 return _tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK, true);
616 int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
618 tdb_trace_1rec(tdb, "tdb_chainunlock", key);
619 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
622 int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
625 ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
626 tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
630 int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
632 tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
633 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
636 /* record lock stops delete underneath */
637 int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
639 if (tdb->global_lock.count) {
642 return off ? tdb->methods->brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
646 Write locks override our own fcntl readlocks, so check it here.
647 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
648 an error to fail to get the lock here.
650 int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
652 struct tdb_traverse_lock *i;
653 for (i = &tdb->travlocks; i; i = i->next)
656 return tdb->methods->brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
659 int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
661 return tdb->methods->brunlock(tdb, F_WRLCK, off, 1);
664 /* fcntl locks don't stack: avoid unlocking someone else's */
665 int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
667 struct tdb_traverse_lock *i;
670 if (tdb->global_lock.count) {
676 for (i = &tdb->travlocks; i; i = i->next)
679 return (count == 1 ? tdb->methods->brunlock(tdb, F_RDLCK, off, 1) : 0);