source: trunk/server/lib/tdb/common/lock.c

Last change on this file was 987, checked in by Silvan Scherrer, 8 years ago

samba server: fix crlf in tdb trunk code
File size: 21.4 KB
Line 
1 /*
2 Unix SMB/CIFS implementation.
3
4 trivial database library
5
6 Copyright (C) Andrew Tridgell 1999-2005
7 Copyright (C) Paul `Rusty' Russell 2000
8 Copyright (C) Jeremy Allison 2000-2003
9
10 ** NOTE! The following LGPL license applies to the tdb
11 ** library. This does NOT imply that all of Samba is released
12 ** under the LGPL
13
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.
18
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.
23
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/>.
26*/
27
28#include "tdb_private.h"
29
30_PUBLIC_ void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
31{
32 tdb->interrupt_sig_ptr = ptr;
33}
34
35static int fcntl_lock(struct tdb_context *tdb,
36 int rw, off_t off, off_t len, bool waitflag)
37{
38 struct flock fl;
39
40 fl.l_type = rw;
41 fl.l_whence = SEEK_SET;
42 fl.l_start = off;
43 fl.l_len = len;
44 fl.l_pid = 0;
45
46 if (waitflag)
47 return fcntl(tdb->fd, F_SETLKW, &fl);
48 else
49 return fcntl(tdb->fd, F_SETLK, &fl);
50}
51
52static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
53{
54 struct flock fl;
55#if 0 /* Check they matched up locks and unlocks correctly. */
56 char line[80];
57 FILE *locks;
58 bool found = false;
59
60 locks = fopen("/proc/locks", "r");
61
62 while (fgets(line, 80, locks)) {
63 char *p;
64 int type, start, l;
65
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 ")))
69 continue;
70 p += strlen(" FLOCK ADVISORY ");
71 if (strncmp(p, "READ ", strlen("READ ")) == 0)
72 type = F_RDLCK;
73 else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
74 type = F_WRLCK;
75 else
76 abort();
77 p += 6;
78 if (atoi(p) != getpid())
79 continue;
80 p = strchr(strchr(p, ' ') + 1, ' ') + 1;
81 start = atoi(p);
82 p = strchr(p, ' ') + 1;
83 if (strncmp(p, "EOF", 3) == 0)
84 l = 0;
85 else
86 l = atoi(p) - start + 1;
87
88 if (off == start) {
89 if (len != l) {
90 fprintf(stderr, "Len %u should be %u: %s",
91 (int)len, l, line);
92 abort();
93 }
94 if (type != rw) {
95 fprintf(stderr, "Type %s wrong: %s",
96 rw == F_RDLCK ? "READ" : "WRITE", line);
97 abort();
98 }
99 found = true;
100 break;
101 }
102 }
103
104 if (!found) {
105 fprintf(stderr, "Unlock on %u@%u not found!\n",
106 (int)off, (int)len);
107 abort();
108 }
109
110 fclose(locks);
111#endif
112
113 fl.l_type = F_UNLCK;
114 fl.l_whence = SEEK_SET;
115 fl.l_start = off;
116 fl.l_len = len;
117 fl.l_pid = 0;
118
119 return fcntl(tdb->fd, F_SETLKW, &fl);
120}
121
122/* list -1 is the alloc list, otherwise a hash chain. */
123static tdb_off_t lock_offset(int list)
124{
125 return FREELIST_TOP + 4*list;
126}
127
128/* a byte range locking function - return 0 on success
129 this functions locks/unlocks 1 byte at the specified offset.
130
131 On error, errno is also set so that errors are passed back properly
132 through tdb_open().
133
134 note that a len of zero means lock to end of file
135*/
136int tdb_brlock(struct tdb_context *tdb,
137 int rw_type, tdb_off_t offset, size_t len,
138 enum tdb_lock_flags flags)
139{
140 int ret;
141
142 if (tdb->flags & TDB_NOLOCK) {
143 return 0;
144 }
145
146 if (flags & TDB_LOCK_MARK_ONLY) {
147 return 0;
148 }
149
150 if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
151 tdb->ecode = TDB_ERR_RDONLY;
152 return -1;
153 }
154
155 do {
156 ret = fcntl_lock(tdb, rw_type, offset, len,
157 flags & TDB_LOCK_WAIT);
158 /* Check for a sigalarm break. */
159 if (ret == -1 && errno == EINTR &&
160 tdb->interrupt_sig_ptr &&
161 *tdb->interrupt_sig_ptr) {
162 break;
163 }
164 } while (ret == -1 && errno == EINTR);
165
166 if (ret == -1) {
167 tdb->ecode = TDB_ERR_LOCK;
168 /* Generic lock error. errno set by fcntl.
169 * EAGAIN is an expected return from non-blocking
170 * locks. */
171 if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
172 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d flags=%d len=%d\n",
173 tdb->fd, offset, rw_type, flags, (int)len));
174 }
175 return -1;
176 }
177 return 0;
178}
179
180int tdb_brunlock(struct tdb_context *tdb,
181 int rw_type, tdb_off_t offset, size_t len)
182{
183 int ret;
184
185 if (tdb->flags & TDB_NOLOCK) {
186 return 0;
187 }
188
189 do {
190 ret = fcntl_unlock(tdb, rw_type, offset, len);
191 } while (ret == -1 && errno == EINTR);
192
193 if (ret == -1) {
194 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brunlock failed (fd=%d) at offset %d rw_type=%d len=%d\n",
195 tdb->fd, offset, rw_type, (int)len));
196 }
197 return ret;
198}
199
200/*
201 upgrade a read lock to a write lock. This needs to be handled in a
202 special way as some OSes (such as solaris) have too conservative
203 deadlock detection and claim a deadlock when progress can be
204 made. For those OSes we may loop for a while.
205*/
206int tdb_allrecord_upgrade(struct tdb_context *tdb)
207{
208 int count = 1000;
209
210 if (tdb->allrecord_lock.count != 1) {
211 TDB_LOG((tdb, TDB_DEBUG_ERROR,
212 "tdb_allrecord_upgrade failed: count %u too high\n",
213 tdb->allrecord_lock.count));
214 return -1;
215 }
216
217 if (tdb->allrecord_lock.off != 1) {
218 TDB_LOG((tdb, TDB_DEBUG_ERROR,
219 "tdb_allrecord_upgrade failed: already upgraded?\n"));
220 return -1;
221 }
222
223 while (count--) {
224 struct timeval tv;
225 if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
226 TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
227 tdb->allrecord_lock.ltype = F_WRLCK;
228 tdb->allrecord_lock.off = 0;
229 return 0;
230 }
231 if (errno != EDEADLK) {
232 break;
233 }
234 /* sleep for as short a time as we can - more portable than usleep() */
235 tv.tv_sec = 0;
236 tv.tv_usec = 1;
237 select(0, NULL, NULL, NULL, &tv);
238 }
239 TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_allrecord_upgrade failed\n"));
240 return -1;
241}
242
243static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
244 tdb_off_t offset)
245{
246 unsigned int i;
247
248 for (i=0; i<tdb->num_lockrecs; i++) {
249 if (tdb->lockrecs[i].off == offset) {
250 return &tdb->lockrecs[i];
251 }
252 }
253 return NULL;
254}
255
256/* lock an offset in the database. */
257int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
258 enum tdb_lock_flags flags)
259{
260 struct tdb_lock_type *new_lck;
261
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",
265 offset, ltype));
266 return -1;
267 }
268 if (tdb->flags & TDB_NOLOCK)
269 return 0;
270
271 new_lck = find_nestlock(tdb, offset);
272 if (new_lck) {
273 /*
274 * Just increment the in-memory struct, posix locks
275 * don't stack.
276 */
277 new_lck->count++;
278 return 0;
279 }
280
281 new_lck = (struct tdb_lock_type *)realloc(
282 tdb->lockrecs,
283 sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
284 if (new_lck == NULL) {
285 errno = ENOMEM;
286 return -1;
287 }
288 tdb->lockrecs = new_lck;
289
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_brlock(tdb, ltype, offset, 1, flags)) {
293 return -1;
294 }
295
296 tdb->lockrecs[tdb->num_lockrecs].off = offset;
297 tdb->lockrecs[tdb->num_lockrecs].count = 1;
298 tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
299 tdb->num_lockrecs++;
300
301 return 0;
302}
303
304static int tdb_lock_and_recover(struct tdb_context *tdb)
305{
306 int ret;
307
308 /* We need to match locking order in transaction commit. */
309 if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
310 return -1;
311 }
312
313 if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
314 tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
315 return -1;
316 }
317
318 ret = tdb_transaction_recover(tdb);
319
320 tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
321 tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
322
323 return ret;
324}
325
326static bool have_data_locks(const struct tdb_context *tdb)
327{
328 unsigned int i;
329
330 for (i = 0; i < tdb->num_lockrecs; i++) {
331 if (tdb->lockrecs[i].off >= lock_offset(-1))
332 return true;
333 }
334 return false;
335}
336
337static int tdb_lock_list(struct tdb_context *tdb, int list, int ltype,
338 enum tdb_lock_flags waitflag)
339{
340 int ret;
341 bool check = false;
342
343 /* a allrecord lock allows us to avoid per chain locks */
344 if (tdb->allrecord_lock.count &&
345 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
346 return 0;
347 }
348
349 if (tdb->allrecord_lock.count) {
350 tdb->ecode = TDB_ERR_LOCK;
351 ret = -1;
352 } else {
353 /* Only check when we grab first data lock. */
354 check = !have_data_locks(tdb);
355 ret = tdb_nest_lock(tdb, lock_offset(list), ltype, waitflag);
356
357 if (ret == 0 && check && tdb_needs_recovery(tdb)) {
358 tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
359
360 if (tdb_lock_and_recover(tdb) == -1) {
361 return -1;
362 }
363 return tdb_lock_list(tdb, list, ltype, waitflag);
364 }
365 }
366 return ret;
367}
368
369/* lock a list in the database. list -1 is the alloc list */
370int tdb_lock(struct tdb_context *tdb, int list, int ltype)
371{
372 int ret;
373
374 ret = tdb_lock_list(tdb, list, ltype, TDB_LOCK_WAIT);
375 if (ret) {
376 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
377 "ltype=%d (%s)\n", list, ltype, strerror(errno)));
378 }
379 return ret;
380}
381
382/* lock a list in the database. list -1 is the alloc list. non-blocking lock */
383int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
384{
385 return tdb_lock_list(tdb, list, ltype, TDB_LOCK_NOWAIT);
386}
387
388
389int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
390 bool mark_lock)
391{
392 int ret = -1;
393 struct tdb_lock_type *lck;
394
395 if (tdb->flags & TDB_NOLOCK)
396 return 0;
397
398 /* Sanity checks */
399 if (offset >= lock_offset(tdb->header.hash_size)) {
400 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: offset %u invalid (%d)\n", offset, tdb->header.hash_size));
401 return ret;
402 }
403
404 lck = find_nestlock(tdb, offset);
405 if ((lck == NULL) || (lck->count == 0)) {
406 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
407 return -1;
408 }
409
410 if (lck->count > 1) {
411 lck->count--;
412 return 0;
413 }
414
415 /*
416 * This lock has count==1 left, so we need to unlock it in the
417 * kernel. We don't bother with decrementing the in-memory array
418 * element, we're about to overwrite it with the last array element
419 * anyway.
420 */
421
422 if (mark_lock) {
423 ret = 0;
424 } else {
425 ret = tdb_brunlock(tdb, ltype, offset, 1);
426 }
427
428 /*
429 * Shrink the array by overwriting the element just unlocked with the
430 * last array element.
431 */
432 *lck = tdb->lockrecs[--tdb->num_lockrecs];
433
434 /*
435 * We don't bother with realloc when the array shrinks, but if we have
436 * a completely idle tdb we should get rid of the locked array.
437 */
438
439 if (tdb->num_lockrecs == 0) {
440 SAFE_FREE(tdb->lockrecs);
441 }
442
443 if (ret)
444 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
445 return ret;
446}
447
448int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
449{
450 /* a global lock allows us to avoid per chain locks */
451 if (tdb->allrecord_lock.count &&
452 (ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
453 return 0;
454 }
455
456 if (tdb->allrecord_lock.count) {
457 tdb->ecode = TDB_ERR_LOCK;
458 return -1;
459 }
460
461 return tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
462}
463
464/*
465 get the transaction lock
466 */
467int tdb_transaction_lock(struct tdb_context *tdb, int ltype,
468 enum tdb_lock_flags lockflags)
469{
470 return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, lockflags);
471}
472
473/*
474 release the transaction lock
475 */
476int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
477{
478 return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
479}
480
481/* Returns 0 if all done, -1 if error, 1 if ok. */
482static int tdb_allrecord_check(struct tdb_context *tdb, int ltype,
483 enum tdb_lock_flags flags, bool upgradable)
484{
485 /* There are no locks on read-only dbs */
486 if (tdb->read_only || tdb->traverse_read) {
487 tdb->ecode = TDB_ERR_LOCK;
488 return -1;
489 }
490
491 if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
492 tdb->allrecord_lock.count++;
493 return 0;
494 }
495
496 if (tdb->allrecord_lock.count) {
497 /* a global lock of a different type exists */
498 tdb->ecode = TDB_ERR_LOCK;
499 return -1;
500 }
501
502 if (tdb_have_extra_locks(tdb)) {
503 /* can't combine global and chain locks */
504 tdb->ecode = TDB_ERR_LOCK;
505 return -1;
506 }
507
508 if (upgradable && ltype != F_RDLCK) {
509 /* tdb error: you can't upgrade a write lock! */
510 tdb->ecode = TDB_ERR_LOCK;
511 return -1;
512 }
513 return 1;
514}
515
516/* We only need to lock individual bytes, but Linux merges consecutive locks
517 * so we lock in contiguous ranges. */
518static int tdb_chainlock_gradual(struct tdb_context *tdb,
519 int ltype, enum tdb_lock_flags flags,
520 size_t off, size_t len)
521{
522 int ret;
523 enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
524
525 if (len <= 4) {
526 /* Single record. Just do blocking lock. */
527 return tdb_brlock(tdb, ltype, off, len, flags);
528 }
529
530 /* First we try non-blocking. */
531 ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
532 if (ret == 0) {
533 return 0;
534 }
535
536 /* Try locking first half, then second. */
537 ret = tdb_chainlock_gradual(tdb, ltype, flags, off, len / 2);
538 if (ret == -1)
539 return -1;
540
541 ret = tdb_chainlock_gradual(tdb, ltype, flags,
542 off + len / 2, len - len / 2);
543 if (ret == -1) {
544 tdb_brunlock(tdb, ltype, off, len / 2);
545 return -1;
546 }
547 return 0;
548}
549
550/* lock/unlock entire database. It can only be upgradable if you have some
551 * other way of guaranteeing exclusivity (ie. transaction write lock).
552 * We do the locking gradually to avoid being starved by smaller locks. */
553int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
554 enum tdb_lock_flags flags, bool upgradable)
555{
556 switch (tdb_allrecord_check(tdb, ltype, flags, upgradable)) {
557 case -1:
558 return -1;
559 case 0:
560 return 0;
561 }
562
563 /* We cover two kinds of locks:
564 * 1) Normal chain locks. Taken for almost all operations.
565 * 3) Individual records locks. Taken after normal or free
566 * chain locks.
567 *
568 * It is (1) which cause the starvation problem, so we're only
569 * gradual for that. */
570 if (tdb_chainlock_gradual(tdb, ltype, flags, FREELIST_TOP,
571 tdb->header.hash_size * 4) == -1) {
572 return -1;
573 }
574
575 /* Grab individual record locks. */
576 if (tdb_brlock(tdb, ltype, lock_offset(tdb->header.hash_size), 0,
577 flags) == -1) {
578 tdb_brunlock(tdb, ltype, FREELIST_TOP,
579 tdb->header.hash_size * 4);
580 return -1;
581 }
582
583 tdb->allrecord_lock.count = 1;
584 /* If it's upgradable, it's actually exclusive so we can treat
585 * it as a write lock. */
586 tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
587 tdb->allrecord_lock.off = upgradable;
588
589 if (tdb_needs_recovery(tdb)) {
590 bool mark = flags & TDB_LOCK_MARK_ONLY;
591 tdb_allrecord_unlock(tdb, ltype, mark);
592 if (mark) {
593 tdb->ecode = TDB_ERR_LOCK;
594 TDB_LOG((tdb, TDB_DEBUG_ERROR,
595 "tdb_lockall_mark cannot do recovery\n"));
596 return -1;
597 }
598 if (tdb_lock_and_recover(tdb) == -1) {
599 return -1;
600 }
601 return tdb_allrecord_lock(tdb, ltype, flags, upgradable);
602 }
603
604 return 0;
605}
606
607
608
609/* unlock entire db */
610int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock)
611{
612 /* There are no locks on read-only dbs */
613 if (tdb->read_only || tdb->traverse_read) {
614 tdb->ecode = TDB_ERR_LOCK;
615 return -1;
616 }
617
618 if (tdb->allrecord_lock.count == 0) {
619 tdb->ecode = TDB_ERR_LOCK;
620 return -1;
621 }
622
623 /* Upgradable locks are marked as write locks. */
624 if (tdb->allrecord_lock.ltype != ltype
625 && (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
626 tdb->ecode = TDB_ERR_LOCK;
627 return -1;
628 }
629
630 if (tdb->allrecord_lock.count > 1) {
631 tdb->allrecord_lock.count--;
632 return 0;
633 }
634
635 if (!mark_lock && tdb_brunlock(tdb, ltype, FREELIST_TOP, 0)) {
636 TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
637 return -1;
638 }
639
640 tdb->allrecord_lock.count = 0;
641 tdb->allrecord_lock.ltype = 0;
642
643 return 0;
644}
645
646/* lock entire database with write lock */
647_PUBLIC_ int tdb_lockall(struct tdb_context *tdb)
648{
649 tdb_trace(tdb, "tdb_lockall");
650 return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
651}
652
653/* lock entire database with write lock - mark only */
654_PUBLIC_ int tdb_lockall_mark(struct tdb_context *tdb)
655{
656 tdb_trace(tdb, "tdb_lockall_mark");
657 return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY, false);
658}
659
660/* unlock entire database with write lock - unmark only */
661_PUBLIC_ int tdb_lockall_unmark(struct tdb_context *tdb)
662{
663 tdb_trace(tdb, "tdb_lockall_unmark");
664 return tdb_allrecord_unlock(tdb, F_WRLCK, true);
665}
666
667/* lock entire database with write lock - nonblocking varient */
668_PUBLIC_ int tdb_lockall_nonblock(struct tdb_context *tdb)
669{
670 int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
671 tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
672 return ret;
673}
674
675/* unlock entire database with write lock */
676_PUBLIC_ int tdb_unlockall(struct tdb_context *tdb)
677{
678 tdb_trace(tdb, "tdb_unlockall");
679 return tdb_allrecord_unlock(tdb, F_WRLCK, false);
680}
681
682/* lock entire database with read lock */
683_PUBLIC_ int tdb_lockall_read(struct tdb_context *tdb)
684{
685 tdb_trace(tdb, "tdb_lockall_read");
686 return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
687}
688
689/* lock entire database with read lock - nonblock varient */
690_PUBLIC_ int tdb_lockall_read_nonblock(struct tdb_context *tdb)
691{
692 int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
693 tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
694 return ret;
695}
696
697/* unlock entire database with read lock */
698_PUBLIC_ int tdb_unlockall_read(struct tdb_context *tdb)
699{
700 tdb_trace(tdb, "tdb_unlockall_read");
701 return tdb_allrecord_unlock(tdb, F_RDLCK, false);
702}
703
704/* lock/unlock one hash chain. This is meant to be used to reduce
705 contention - it cannot guarantee how many records will be locked */
706_PUBLIC_ int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
707{
708 int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
709 tdb_trace_1rec(tdb, "tdb_chainlock", key);
710 return ret;
711}
712
713/* lock/unlock one hash chain, non-blocking. This is meant to be used
714 to reduce contention - it cannot guarantee how many records will be
715 locked */
716_PUBLIC_ int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
717{
718 int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
719 tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
720 return ret;
721}
722
723/* mark a chain as locked without actually locking it. Warning! use with great caution! */
724_PUBLIC_ int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
725{
726 int ret = tdb_nest_lock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
727 F_WRLCK, TDB_LOCK_MARK_ONLY);
728 tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
729 return ret;
730}
731
732/* unmark a chain as locked without actually locking it. Warning! use with great caution! */
733_PUBLIC_ int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
734{
735 tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
736 return tdb_nest_unlock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
737 F_WRLCK, true);
738}
739
740_PUBLIC_ int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
741{
742 tdb_trace_1rec(tdb, "tdb_chainunlock", key);
743 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
744}
745
746_PUBLIC_ int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
747{
748 int ret;
749 ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
750 tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
751 return ret;
752}
753
754_PUBLIC_ int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
755{
756 tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
757 return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
758}
759
760/* record lock stops delete underneath */
761int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
762{
763 if (tdb->allrecord_lock.count) {
764 return 0;
765 }
766 return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
767}
768
769/*
770 Write locks override our own fcntl readlocks, so check it here.
771 Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
772 an error to fail to get the lock here.
773*/
774int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
775{
776 struct tdb_traverse_lock *i;
777 for (i = &tdb->travlocks; i; i = i->next)
778 if (i->off == off)
779 return -1;
780 if (tdb->allrecord_lock.count) {
781 if (tdb->allrecord_lock.ltype == F_WRLCK) {
782 return 0;
783 }
784 return -1;
785 }
786 return tdb_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
787}
788
789int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
790{
791 if (tdb->allrecord_lock.count) {
792 return 0;
793 }
794 return tdb_brunlock(tdb, F_WRLCK, off, 1);
795}
796
797/* fcntl locks don't stack: avoid unlocking someone else's */
798int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
799{
800 struct tdb_traverse_lock *i;
801 uint32_t count = 0;
802
803 if (tdb->allrecord_lock.count) {
804 return 0;
805 }
806
807 if (off == 0)
808 return 0;
809 for (i = &tdb->travlocks; i; i = i->next)
810 if (i->off == off)
811 count++;
812 return (count == 1 ? tdb_brunlock(tdb, F_RDLCK, off, 1) : 0);
813}
814
815bool tdb_have_extra_locks(struct tdb_context *tdb)
816{
817 unsigned int extra = tdb->num_lockrecs;
818
819 /* A transaction holds the lock for all records. */
820 if (!tdb->transaction && tdb->allrecord_lock.count) {
821 return true;
822 }
823
824 /* We always hold the active lock if CLEAR_IF_FIRST. */
825 if (find_nestlock(tdb, ACTIVE_LOCK)) {
826 extra--;
827 }
828
829 /* In a transaction, we expect to hold the transaction lock */
830 if (tdb->transaction && find_nestlock(tdb, TRANSACTION_LOCK)) {
831 extra--;
832 }
833
834 return extra;
835}
836
837/* The transaction code uses this to remove all locks. */
838void tdb_release_transaction_locks(struct tdb_context *tdb)
839{
840 unsigned int i, active = 0;
841
842 if (tdb->allrecord_lock.count != 0) {
843 tdb_brunlock(tdb, tdb->allrecord_lock.ltype, FREELIST_TOP, 0);
844 tdb->allrecord_lock.count = 0;
845 }
846
847 for (i=0;i<tdb->num_lockrecs;i++) {
848 struct tdb_lock_type *lck = &tdb->lockrecs[i];
849
850 /* Don't release the active lock! Copy it to first entry. */
851 if (lck->off == ACTIVE_LOCK) {
852 tdb->lockrecs[active++] = *lck;
853 } else {
854 tdb_brunlock(tdb, lck->ltype, lck->off, 1);
855 }
856 }
857 tdb->num_lockrecs = active;
858 if (tdb->num_lockrecs == 0) {
859 SAFE_FREE(tdb->lockrecs);
860 }
861}
Note: See TracBrowser for help on using the repository browser.