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 ntdb
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 /* If we were threaded, we could wait for unlock, but we're not, so fail. */
33 enum NTDB_ERROR owner_conflict(struct ntdb_context *ntdb, const char *call)
35 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
36 "%s: lock owned by another ntdb in this process.",
40 /* If we fork, we no longer really own locks. */
41 bool check_lock_pid(struct ntdb_context *ntdb, const char *call, bool log)
43 /* No locks? No problem! */
44 if (ntdb->file->allrecord_lock.count == 0
45 && ntdb->file->num_lockrecs == 0) {
49 /* No fork? No problem! */
50 if (ntdb->file->locker == getpid()) {
55 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
56 "%s: fork() detected after lock acquisition!"
57 " (%u vs %u)", call, ntdb->file->locker, getpid());
62 int ntdb_fcntl_lock(int fd, int rw, off_t off, off_t len, bool waitflag,
70 fl.l_whence = SEEK_SET;
75 ret = fcntl(fd, F_SETLKW, &fl);
77 ret = fcntl(fd, F_SETLK, &fl);
78 } while (ret != 0 && errno == EINTR);
82 int ntdb_fcntl_unlock(int fd, int rw, off_t off, off_t len, void *unused)
89 fl.l_whence = SEEK_SET;
93 ret = fcntl(fd, F_SETLKW, &fl);
94 } while (ret != 0 && errno == EINTR);
98 static int lock(struct ntdb_context *ntdb,
99 int rw, off_t off, off_t len, bool waitflag)
102 if (ntdb->file->allrecord_lock.count == 0
103 && ntdb->file->num_lockrecs == 0) {
104 ntdb->file->locker = getpid();
107 ntdb->stats.lock_lowlevel++;
108 ret = ntdb->lock_fn(ntdb->file->fd, rw, off, len, waitflag,
111 ntdb->stats.lock_nonblock++;
113 ntdb->stats.lock_nonblock_fail++;
118 static int unlock(struct ntdb_context *ntdb, int rw, off_t off, off_t len)
120 #if 0 /* Check they matched up locks and unlocks correctly. */
125 locks = fopen("/proc/locks", "r");
127 while (fgets(line, 80, locks)) {
131 /* eg. 1: FLOCK ADVISORY WRITE 2440 08:01:2180826 0 EOF */
132 p = strchr(line, ':') + 1;
133 if (strncmp(p, " POSIX ADVISORY ", strlen(" POSIX ADVISORY ")))
135 p += strlen(" FLOCK ADVISORY ");
136 if (strncmp(p, "READ ", strlen("READ ")) == 0)
138 else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
143 if (atoi(p) != getpid())
145 p = strchr(strchr(p, ' ') + 1, ' ') + 1;
147 p = strchr(p, ' ') + 1;
148 if (strncmp(p, "EOF", 3) == 0)
151 l = atoi(p) - start + 1;
155 fprintf(stderr, "Len %u should be %u: %s",
160 fprintf(stderr, "Type %s wrong: %s",
161 rw == F_RDLCK ? "READ" : "WRITE", line);
170 fprintf(stderr, "Unlock on %u@%u not found!",
178 return ntdb->unlock_fn(ntdb->file->fd, rw, off, len, ntdb->lock_data);
181 /* a byte range locking function - return 0 on success
182 this functions locks len bytes at the specified offset.
184 note that a len of zero means lock to end of file
186 static enum NTDB_ERROR ntdb_brlock(struct ntdb_context *ntdb,
187 int rw_type, ntdb_off_t offset, ntdb_off_t len,
188 enum ntdb_lock_flags flags)
192 if (ntdb->flags & NTDB_NOLOCK) {
196 if (rw_type == F_WRLCK && (ntdb->flags & NTDB_RDONLY)) {
197 return ntdb_logerr(ntdb, NTDB_ERR_RDONLY, NTDB_LOG_USE_ERROR,
198 "Write lock attempted on read-only database");
201 /* A 32 bit system cannot open a 64-bit file, but it could have
202 * expanded since then: check here. */
203 if ((size_t)(offset + len) != offset + len) {
204 return ntdb_logerr(ntdb, NTDB_ERR_IO, NTDB_LOG_ERROR,
205 "ntdb_brlock: lock on giant offset %llu",
206 (long long)(offset + len));
209 ret = lock(ntdb, rw_type, offset, len, flags & NTDB_LOCK_WAIT);
211 /* Generic lock error. errno set by fcntl.
212 * EAGAIN is an expected return from non-blocking
214 if (!(flags & NTDB_LOCK_PROBE)
215 && (errno != EAGAIN && errno != EINTR)) {
216 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
217 "ntdb_brlock failed (fd=%d) at"
218 " offset %zu rw_type=%d flags=%d len=%zu:"
220 ntdb->file->fd, (size_t)offset, rw_type,
221 flags, (size_t)len, strerror(errno));
223 return NTDB_ERR_LOCK;
228 static enum NTDB_ERROR ntdb_brunlock(struct ntdb_context *ntdb,
229 int rw_type, ntdb_off_t offset, size_t len)
231 if (ntdb->flags & NTDB_NOLOCK) {
235 if (!check_lock_pid(ntdb, "ntdb_brunlock", false))
236 return NTDB_ERR_LOCK;
238 if (unlock(ntdb, rw_type, offset, len) == -1) {
239 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
240 "ntdb_brunlock failed (fd=%d) at offset %zu"
241 " rw_type=%d len=%zu: %s",
242 ntdb->file->fd, (size_t)offset, rw_type,
243 (size_t)len, strerror(errno));
249 upgrade a read lock to a write lock. This needs to be handled in a
250 special way as some OSes (such as solaris) have too conservative
251 deadlock detection and claim a deadlock when progress can be
252 made. For those OSes we may loop for a while.
254 enum NTDB_ERROR ntdb_allrecord_upgrade(struct ntdb_context *ntdb, off_t start)
258 if (!check_lock_pid(ntdb, "ntdb_transaction_prepare_commit", true))
259 return NTDB_ERR_LOCK;
261 if (ntdb->file->allrecord_lock.count != 1) {
262 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
263 "ntdb_allrecord_upgrade failed:"
264 " count %u too high",
265 ntdb->file->allrecord_lock.count);
268 if (ntdb->file->allrecord_lock.off != 1) {
269 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
270 "ntdb_allrecord_upgrade failed:"
271 " already upgraded?");
274 if (ntdb->file->allrecord_lock.owner != ntdb) {
275 return owner_conflict(ntdb, "ntdb_allrecord_upgrade");
280 if (ntdb_brlock(ntdb, F_WRLCK, start, 0,
281 NTDB_LOCK_WAIT|NTDB_LOCK_PROBE) == NTDB_SUCCESS) {
282 ntdb->file->allrecord_lock.ltype = F_WRLCK;
283 ntdb->file->allrecord_lock.off = 0;
286 if (errno != EDEADLK) {
289 /* sleep for as short a time as we can - more portable than usleep() */
292 select(0, NULL, NULL, NULL, &tv);
295 if (errno != EAGAIN && errno != EINTR)
296 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
297 "ntdb_allrecord_upgrade failed");
298 return NTDB_ERR_LOCK;
301 static struct ntdb_lock *find_nestlock(struct ntdb_context *ntdb, ntdb_off_t offset,
302 const struct ntdb_context *owner)
306 for (i=0; i<ntdb->file->num_lockrecs; i++) {
307 if (ntdb->file->lockrecs[i].off == offset) {
308 if (owner && ntdb->file->lockrecs[i].owner != owner)
310 return &ntdb->file->lockrecs[i];
316 enum NTDB_ERROR ntdb_lock_and_recover(struct ntdb_context *ntdb)
318 enum NTDB_ERROR ecode;
320 if (!check_lock_pid(ntdb, "ntdb_transaction_prepare_commit", true))
321 return NTDB_ERR_LOCK;
323 ecode = ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK,
325 if (ecode != NTDB_SUCCESS) {
329 ecode = ntdb_lock_open(ntdb, F_WRLCK, NTDB_LOCK_WAIT|NTDB_LOCK_NOCHECK);
330 if (ecode != NTDB_SUCCESS) {
331 ntdb_allrecord_unlock(ntdb, F_WRLCK);
334 ecode = ntdb_transaction_recover(ntdb);
335 ntdb_unlock_open(ntdb, F_WRLCK);
336 ntdb_allrecord_unlock(ntdb, F_WRLCK);
341 /* lock an offset in the database. */
342 static enum NTDB_ERROR ntdb_nest_lock(struct ntdb_context *ntdb,
343 ntdb_off_t offset, int ltype,
344 enum ntdb_lock_flags flags)
346 struct ntdb_lock *new_lck;
347 enum NTDB_ERROR ecode;
349 if (offset > (NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE
350 + ntdb->file->map_size / 8)) {
351 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
352 "ntdb_nest_lock: invalid offset %zu ltype=%d",
353 (size_t)offset, ltype);
356 if (ntdb->flags & NTDB_NOLOCK)
359 if (!check_lock_pid(ntdb, "ntdb_nest_lock", true)) {
360 return NTDB_ERR_LOCK;
365 new_lck = find_nestlock(ntdb, offset, NULL);
367 if (new_lck->owner != ntdb) {
368 return owner_conflict(ntdb, "ntdb_nest_lock");
371 if (new_lck->ltype == F_RDLCK && ltype == F_WRLCK) {
372 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
374 " offset %zu has read lock",
377 /* Just increment the struct, posix locks don't stack. */
383 if (ntdb->file->num_lockrecs
384 && offset >= NTDB_HASH_LOCK_START
385 && offset < NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE) {
386 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
387 "ntdb_nest_lock: already have a hash lock?");
391 new_lck = (struct ntdb_lock *)realloc(
392 ntdb->file->lockrecs,
393 sizeof(*ntdb->file->lockrecs) * (ntdb->file->num_lockrecs+1));
394 if (new_lck == NULL) {
395 return ntdb_logerr(ntdb, NTDB_ERR_OOM, NTDB_LOG_ERROR,
397 " unable to allocate %zu lock struct",
398 ntdb->file->num_lockrecs + 1);
400 ntdb->file->lockrecs = new_lck;
402 /* Since fcntl locks don't nest, we do a lock for the first one,
403 and simply bump the count for future ones */
404 ecode = ntdb_brlock(ntdb, ltype, offset, 1, flags);
405 if (ecode != NTDB_SUCCESS) {
409 /* First time we grab a lock, perhaps someone died in commit? */
410 if (!(flags & NTDB_LOCK_NOCHECK)
411 && ntdb->file->num_lockrecs == 0) {
412 ntdb_bool_err berr = ntdb_needs_recovery(ntdb);
414 ntdb_brunlock(ntdb, ltype, offset, 1);
417 return NTDB_OFF_TO_ERR(berr);
418 ecode = ntdb_lock_and_recover(ntdb);
419 if (ecode == NTDB_SUCCESS) {
420 ecode = ntdb_brlock(ntdb, ltype, offset, 1,
423 if (ecode != NTDB_SUCCESS) {
429 ntdb->file->lockrecs[ntdb->file->num_lockrecs].owner = ntdb;
430 ntdb->file->lockrecs[ntdb->file->num_lockrecs].off = offset;
431 ntdb->file->lockrecs[ntdb->file->num_lockrecs].count = 1;
432 ntdb->file->lockrecs[ntdb->file->num_lockrecs].ltype = ltype;
433 ntdb->file->num_lockrecs++;
438 static enum NTDB_ERROR ntdb_nest_unlock(struct ntdb_context *ntdb,
439 ntdb_off_t off, int ltype)
441 struct ntdb_lock *lck;
442 enum NTDB_ERROR ecode;
444 if (ntdb->flags & NTDB_NOLOCK)
447 lck = find_nestlock(ntdb, off, ntdb);
448 if ((lck == NULL) || (lck->count == 0)) {
449 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
450 "ntdb_nest_unlock: no lock for %zu",
454 if (lck->count > 1) {
460 * This lock has count==1 left, so we need to unlock it in the
461 * kernel. We don't bother with decrementing the in-memory array
462 * element, we're about to overwrite it with the last array element
465 ecode = ntdb_brunlock(ntdb, ltype, off, 1);
468 * Shrink the array by overwriting the element just unlocked with the
469 * last array element.
471 *lck = ntdb->file->lockrecs[--ntdb->file->num_lockrecs];
477 get the transaction lock
479 enum NTDB_ERROR ntdb_transaction_lock(struct ntdb_context *ntdb, int ltype)
481 return ntdb_nest_lock(ntdb, NTDB_TRANSACTION_LOCK, ltype, NTDB_LOCK_WAIT);
485 release the transaction lock
487 void ntdb_transaction_unlock(struct ntdb_context *ntdb, int ltype)
489 ntdb_nest_unlock(ntdb, NTDB_TRANSACTION_LOCK, ltype);
492 /* We only need to lock individual bytes, but Linux merges consecutive locks
493 * so we lock in contiguous ranges. */
494 static enum NTDB_ERROR ntdb_lock_gradual(struct ntdb_context *ntdb,
495 int ltype, enum ntdb_lock_flags flags,
496 ntdb_off_t off, ntdb_off_t len)
498 enum NTDB_ERROR ecode;
499 enum ntdb_lock_flags nb_flags = (flags & ~NTDB_LOCK_WAIT);
502 /* 0 would mean to end-of-file... */
504 /* Single hash. Just do blocking lock. */
505 return ntdb_brlock(ntdb, ltype, off, len, flags);
508 /* First we try non-blocking. */
509 ecode = ntdb_brlock(ntdb, ltype, off, len, nb_flags);
510 if (ecode != NTDB_ERR_LOCK) {
514 /* Try locking first half, then second. */
515 ecode = ntdb_lock_gradual(ntdb, ltype, flags, off, len / 2);
516 if (ecode != NTDB_SUCCESS)
519 ecode = ntdb_lock_gradual(ntdb, ltype, flags,
520 off + len / 2, len - len / 2);
521 if (ecode != NTDB_SUCCESS) {
522 ntdb_brunlock(ntdb, ltype, off, len / 2);
527 /* lock/unlock entire database. It can only be upgradable if you have some
528 * other way of guaranteeing exclusivity (ie. transaction write lock). */
529 enum NTDB_ERROR ntdb_allrecord_lock(struct ntdb_context *ntdb, int ltype,
530 enum ntdb_lock_flags flags, bool upgradable)
532 enum NTDB_ERROR ecode;
535 if (ntdb->flags & NTDB_NOLOCK)
538 if (!check_lock_pid(ntdb, "ntdb_allrecord_lock", true)) {
539 return NTDB_ERR_LOCK;
542 if (ntdb->file->allrecord_lock.count) {
543 if (ntdb->file->allrecord_lock.owner != ntdb) {
544 return owner_conflict(ntdb, "ntdb_allrecord_lock");
548 || ntdb->file->allrecord_lock.ltype == F_WRLCK) {
549 ntdb->file->allrecord_lock.count++;
553 /* a global lock of a different type exists */
554 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
555 "ntdb_allrecord_lock: already have %s lock",
556 ntdb->file->allrecord_lock.ltype == F_RDLCK
560 if (ntdb_has_hash_locks(ntdb)) {
561 /* can't combine global and chain locks */
562 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
563 "ntdb_allrecord_lock:"
564 " already have chain lock");
567 if (upgradable && ltype != F_RDLCK) {
568 /* ntdb error: you can't upgrade a write lock! */
569 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
570 "ntdb_allrecord_lock:"
571 " can't upgrade a write lock");
576 /* Lock hashes, gradually. */
577 ecode = ntdb_lock_gradual(ntdb, ltype, flags, NTDB_HASH_LOCK_START,
578 NTDB_HASH_LOCK_RANGE);
579 if (ecode != NTDB_SUCCESS)
582 /* Lock free tables: there to end of file. */
583 ecode = ntdb_brlock(ntdb, ltype,
584 NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE,
586 if (ecode != NTDB_SUCCESS) {
587 ntdb_brunlock(ntdb, ltype, NTDB_HASH_LOCK_START,
588 NTDB_HASH_LOCK_RANGE);
592 ntdb->file->allrecord_lock.owner = ntdb;
593 ntdb->file->allrecord_lock.count = 1;
594 /* If it's upgradable, it's actually exclusive so we can treat
595 * it as a write lock. */
596 ntdb->file->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
597 ntdb->file->allrecord_lock.off = upgradable;
599 /* Now check for needing recovery. */
600 if (flags & NTDB_LOCK_NOCHECK)
603 berr = ntdb_needs_recovery(ntdb);
604 if (likely(berr == false))
607 ntdb_allrecord_unlock(ntdb, ltype);
609 return NTDB_OFF_TO_ERR(berr);
610 ecode = ntdb_lock_and_recover(ntdb);
611 if (ecode != NTDB_SUCCESS) {
617 enum NTDB_ERROR ntdb_lock_open(struct ntdb_context *ntdb,
618 int ltype, enum ntdb_lock_flags flags)
620 return ntdb_nest_lock(ntdb, NTDB_OPEN_LOCK, ltype, flags);
623 void ntdb_unlock_open(struct ntdb_context *ntdb, int ltype)
625 ntdb_nest_unlock(ntdb, NTDB_OPEN_LOCK, ltype);
628 bool ntdb_has_open_lock(struct ntdb_context *ntdb)
630 return !(ntdb->flags & NTDB_NOLOCK)
631 && find_nestlock(ntdb, NTDB_OPEN_LOCK, ntdb) != NULL;
634 enum NTDB_ERROR ntdb_lock_expand(struct ntdb_context *ntdb, int ltype)
636 /* Lock doesn't protect data, so don't check (we recurse if we do!) */
637 return ntdb_nest_lock(ntdb, NTDB_EXPANSION_LOCK, ltype,
638 NTDB_LOCK_WAIT | NTDB_LOCK_NOCHECK);
641 void ntdb_unlock_expand(struct ntdb_context *ntdb, int ltype)
643 ntdb_nest_unlock(ntdb, NTDB_EXPANSION_LOCK, ltype);
646 /* unlock entire db */
647 void ntdb_allrecord_unlock(struct ntdb_context *ntdb, int ltype)
649 if (ntdb->flags & NTDB_NOLOCK)
652 if (ntdb->file->allrecord_lock.count == 0) {
653 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
654 "ntdb_allrecord_unlock: not locked!");
658 if (ntdb->file->allrecord_lock.owner != ntdb) {
659 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
660 "ntdb_allrecord_unlock: not locked by us!");
664 /* Upgradable locks are marked as write locks. */
665 if (ntdb->file->allrecord_lock.ltype != ltype
666 && (!ntdb->file->allrecord_lock.off || ltype != F_RDLCK)) {
667 ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
668 "ntdb_allrecord_unlock: have %s lock",
669 ntdb->file->allrecord_lock.ltype == F_RDLCK
674 if (ntdb->file->allrecord_lock.count > 1) {
675 ntdb->file->allrecord_lock.count--;
679 ntdb->file->allrecord_lock.count = 0;
680 ntdb->file->allrecord_lock.ltype = 0;
682 ntdb_brunlock(ntdb, ltype, NTDB_HASH_LOCK_START, 0);
685 bool ntdb_has_expansion_lock(struct ntdb_context *ntdb)
687 return find_nestlock(ntdb, NTDB_EXPANSION_LOCK, ntdb) != NULL;
690 bool ntdb_has_hash_locks(struct ntdb_context *ntdb)
694 for (i=0; i<ntdb->file->num_lockrecs; i++) {
695 if (ntdb->file->lockrecs[i].off >= NTDB_HASH_LOCK_START
696 && ntdb->file->lockrecs[i].off < (NTDB_HASH_LOCK_START
697 + NTDB_HASH_LOCK_RANGE))
703 static bool ntdb_has_free_lock(struct ntdb_context *ntdb)
707 if (ntdb->flags & NTDB_NOLOCK)
710 for (i=0; i<ntdb->file->num_lockrecs; i++) {
711 if (ntdb->file->lockrecs[i].off
712 > NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE)
718 enum NTDB_ERROR ntdb_lock_hashes(struct ntdb_context *ntdb,
719 ntdb_off_t hash_lock,
720 ntdb_len_t hash_range,
721 int ltype, enum ntdb_lock_flags waitflag)
723 /* FIXME: Do this properly, using hlock_range */
724 unsigned l = NTDB_HASH_LOCK_START
725 + (hash_lock >> (64 - NTDB_HASH_LOCK_RANGE_BITS));
727 /* a allrecord lock allows us to avoid per chain locks */
728 if (ntdb->file->allrecord_lock.count) {
729 if (!check_lock_pid(ntdb, "ntdb_lock_hashes", true))
730 return NTDB_ERR_LOCK;
732 if (ntdb->file->allrecord_lock.owner != ntdb)
733 return owner_conflict(ntdb, "ntdb_lock_hashes");
734 if (ltype == ntdb->file->allrecord_lock.ltype
735 || ltype == F_RDLCK) {
739 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
741 " already have %s allrecordlock",
742 ntdb->file->allrecord_lock.ltype == F_RDLCK
746 if (ntdb_has_free_lock(ntdb)) {
747 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
748 "ntdb_lock_hashes: already have free lock");
751 if (ntdb_has_expansion_lock(ntdb)) {
752 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
754 " already have expansion lock");
757 return ntdb_nest_lock(ntdb, l, ltype, waitflag);
760 enum NTDB_ERROR ntdb_unlock_hashes(struct ntdb_context *ntdb,
761 ntdb_off_t hash_lock,
762 ntdb_len_t hash_range, int ltype)
764 unsigned l = NTDB_HASH_LOCK_START
765 + (hash_lock >> (64 - NTDB_HASH_LOCK_RANGE_BITS));
767 if (ntdb->flags & NTDB_NOLOCK)
770 /* a allrecord lock allows us to avoid per chain locks */
771 if (ntdb->file->allrecord_lock.count) {
772 if (ntdb->file->allrecord_lock.ltype == F_RDLCK
773 && ltype == F_WRLCK) {
774 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
775 "ntdb_unlock_hashes RO allrecord!");
777 if (ntdb->file->allrecord_lock.owner != ntdb) {
778 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_USE_ERROR,
779 "ntdb_unlock_hashes:"
780 " not locked by us!");
785 return ntdb_nest_unlock(ntdb, l, ltype);
788 /* Hash locks use NTDB_HASH_LOCK_START + the next 30 bits.
789 * Then we begin; bucket offsets are sizeof(ntdb_len_t) apart, so we divide.
790 * The result is that on 32 bit systems we don't use lock values > 2^31 on
791 * files that are less than 4GB.
793 static ntdb_off_t free_lock_off(ntdb_off_t b_off)
795 return NTDB_HASH_LOCK_START + NTDB_HASH_LOCK_RANGE
796 + b_off / sizeof(ntdb_off_t);
799 enum NTDB_ERROR ntdb_lock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off,
800 enum ntdb_lock_flags waitflag)
802 assert(b_off >= sizeof(struct ntdb_header));
804 if (ntdb->flags & NTDB_NOLOCK)
807 /* a allrecord lock allows us to avoid per chain locks */
808 if (ntdb->file->allrecord_lock.count) {
809 if (!check_lock_pid(ntdb, "ntdb_lock_free_bucket", true))
810 return NTDB_ERR_LOCK;
812 if (ntdb->file->allrecord_lock.owner != ntdb) {
813 return owner_conflict(ntdb, "ntdb_lock_free_bucket");
816 if (ntdb->file->allrecord_lock.ltype == F_WRLCK)
818 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
819 "ntdb_lock_free_bucket with"
820 " read-only allrecordlock!");
824 if (ntdb_has_expansion_lock(ntdb)) {
825 return ntdb_logerr(ntdb, NTDB_ERR_LOCK, NTDB_LOG_ERROR,
826 "ntdb_lock_free_bucket:"
827 " already have expansion lock");
831 return ntdb_nest_lock(ntdb, free_lock_off(b_off), F_WRLCK, waitflag);
834 void ntdb_unlock_free_bucket(struct ntdb_context *ntdb, ntdb_off_t b_off)
836 if (ntdb->file->allrecord_lock.count)
839 ntdb_nest_unlock(ntdb, free_lock_off(b_off), F_WRLCK);
842 _PUBLIC_ enum NTDB_ERROR ntdb_lockall(struct ntdb_context *ntdb)
844 return ntdb_allrecord_lock(ntdb, F_WRLCK, NTDB_LOCK_WAIT, false);
847 _PUBLIC_ void ntdb_unlockall(struct ntdb_context *ntdb)
849 ntdb_allrecord_unlock(ntdb, F_WRLCK);
852 _PUBLIC_ enum NTDB_ERROR ntdb_lockall_read(struct ntdb_context *ntdb)
854 return ntdb_allrecord_lock(ntdb, F_RDLCK, NTDB_LOCK_WAIT, false);
857 _PUBLIC_ void ntdb_unlockall_read(struct ntdb_context *ntdb)
859 ntdb_allrecord_unlock(ntdb, F_RDLCK);
862 void ntdb_lock_cleanup(struct ntdb_context *ntdb)
866 /* We don't want to warn: they're allowed to close ntdb after fork. */
867 if (!check_lock_pid(ntdb, "ntdb_close", false))
870 while (ntdb->file->allrecord_lock.count
871 && ntdb->file->allrecord_lock.owner == ntdb) {
872 ntdb_allrecord_unlock(ntdb, ntdb->file->allrecord_lock.ltype);
875 for (i=0; i<ntdb->file->num_lockrecs; i++) {
876 if (ntdb->file->lockrecs[i].owner == ntdb) {
877 ntdb_nest_unlock(ntdb,
878 ntdb->file->lockrecs[i].off,
879 ntdb->file->lockrecs[i].ltype);