pidl: for wireshark use only the major of the version
[Samba/gebeck_regimport.git] / source3 / locking / brlock.c
blobb2e8f5fe15d1a6494d8a023445101f71cb2498e1
1 /*
2 Unix SMB/CIFS implementation.
3 byte range locking code
4 Updated to handle range splits/merges.
6 Copyright (C) Andrew Tridgell 1992-2000
7 Copyright (C) Jeremy Allison 1992-2000
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 3 of the License, or
12 (at your option) any later version.
14 This program is distributed in the hope that it will be useful,
15 but WITHOUT ANY WARRANTY; without even the implied warranty of
16 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 GNU General Public License for more details.
19 You should have received a copy of the GNU General Public License
20 along with this program. If not, see <http://www.gnu.org/licenses/>.
23 /* This module implements a tdb based byte range locking service,
24 replacing the fcntl() based byte range locking previously
25 used. This allows us to provide the same semantics as NT */
27 #include "includes.h"
28 #include "system/filesys.h"
29 #include "locking/proto.h"
30 #include "smbd/globals.h"
31 #include "dbwrap/dbwrap.h"
32 #include "dbwrap/dbwrap_open.h"
33 #include "serverid.h"
34 #include "messages.h"
36 #undef DBGC_CLASS
37 #define DBGC_CLASS DBGC_LOCKING
39 #define ZERO_ZERO 0
41 /* The open brlock.tdb database. */
43 static struct db_context *brlock_db;
45 /****************************************************************************
46 Debug info at level 10 for lock struct.
47 ****************************************************************************/
49 static void print_lock_struct(unsigned int i, struct lock_struct *pls)
51 DEBUG(10,("[%u]: smblctx = %llu, tid = %u, pid = %s, ",
53 (unsigned long long)pls->context.smblctx,
54 (unsigned int)pls->context.tid,
55 server_id_str(talloc_tos(), &pls->context.pid) ));
57 DEBUG(10,("start = %.0f, size = %.0f, fnum = %d, %s %s\n",
58 (double)pls->start,
59 (double)pls->size,
60 pls->fnum,
61 lock_type_name(pls->lock_type),
62 lock_flav_name(pls->lock_flav) ));
65 /****************************************************************************
66 See if two locking contexts are equal.
67 ****************************************************************************/
69 bool brl_same_context(const struct lock_context *ctx1,
70 const struct lock_context *ctx2)
72 return (procid_equal(&ctx1->pid, &ctx2->pid) &&
73 (ctx1->smblctx == ctx2->smblctx) &&
74 (ctx1->tid == ctx2->tid));
77 /****************************************************************************
78 See if lck1 and lck2 overlap.
79 ****************************************************************************/
81 static bool brl_overlap(const struct lock_struct *lck1,
82 const struct lock_struct *lck2)
84 /* XXX Remove for Win7 compatibility. */
85 /* this extra check is not redundent - it copes with locks
86 that go beyond the end of 64 bit file space */
87 if (lck1->size != 0 &&
88 lck1->start == lck2->start &&
89 lck1->size == lck2->size) {
90 return True;
93 if (lck1->start >= (lck2->start+lck2->size) ||
94 lck2->start >= (lck1->start+lck1->size)) {
95 return False;
97 return True;
100 /****************************************************************************
101 See if lock2 can be added when lock1 is in place.
102 ****************************************************************************/
104 static bool brl_conflict(const struct lock_struct *lck1,
105 const struct lock_struct *lck2)
107 /* Ignore PENDING locks. */
108 if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
109 return False;
111 /* Read locks never conflict. */
112 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
113 return False;
116 /* A READ lock can stack on top of a WRITE lock if they have the same
117 * context & fnum. */
118 if (lck1->lock_type == WRITE_LOCK && lck2->lock_type == READ_LOCK &&
119 brl_same_context(&lck1->context, &lck2->context) &&
120 lck1->fnum == lck2->fnum) {
121 return False;
124 return brl_overlap(lck1, lck2);
127 /****************************************************************************
128 See if lock2 can be added when lock1 is in place - when both locks are POSIX
129 flavour. POSIX locks ignore fnum - they only care about dev/ino which we
130 know already match.
131 ****************************************************************************/
133 static bool brl_conflict_posix(const struct lock_struct *lck1,
134 const struct lock_struct *lck2)
136 #if defined(DEVELOPER)
137 SMB_ASSERT(lck1->lock_flav == POSIX_LOCK);
138 SMB_ASSERT(lck2->lock_flav == POSIX_LOCK);
139 #endif
141 /* Ignore PENDING locks. */
142 if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
143 return False;
145 /* Read locks never conflict. */
146 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
147 return False;
150 /* Locks on the same context con't conflict. Ignore fnum. */
151 if (brl_same_context(&lck1->context, &lck2->context)) {
152 return False;
155 /* One is read, the other write, or the context is different,
156 do they overlap ? */
157 return brl_overlap(lck1, lck2);
160 #if ZERO_ZERO
161 static bool brl_conflict1(const struct lock_struct *lck1,
162 const struct lock_struct *lck2)
164 if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
165 return False;
167 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
168 return False;
171 if (brl_same_context(&lck1->context, &lck2->context) &&
172 lck2->lock_type == READ_LOCK && lck1->fnum == lck2->fnum) {
173 return False;
176 if (lck2->start == 0 && lck2->size == 0 && lck1->size != 0) {
177 return True;
180 if (lck1->start >= (lck2->start + lck2->size) ||
181 lck2->start >= (lck1->start + lck1->size)) {
182 return False;
185 return True;
187 #endif
189 /****************************************************************************
190 Check to see if this lock conflicts, but ignore our own locks on the
191 same fnum only. This is the read/write lock check code path.
192 This is never used in the POSIX lock case.
193 ****************************************************************************/
195 static bool brl_conflict_other(const struct lock_struct *lck1, const struct lock_struct *lck2)
197 if (IS_PENDING_LOCK(lck1->lock_type) || IS_PENDING_LOCK(lck2->lock_type))
198 return False;
200 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK)
201 return False;
203 /* POSIX flavour locks never conflict here - this is only called
204 in the read/write path. */
206 if (lck1->lock_flav == POSIX_LOCK && lck2->lock_flav == POSIX_LOCK)
207 return False;
210 * Incoming WRITE locks conflict with existing READ locks even
211 * if the context is the same. JRA. See LOCKTEST7 in smbtorture.
214 if (!(lck2->lock_type == WRITE_LOCK && lck1->lock_type == READ_LOCK)) {
215 if (brl_same_context(&lck1->context, &lck2->context) &&
216 lck1->fnum == lck2->fnum)
217 return False;
220 return brl_overlap(lck1, lck2);
223 /****************************************************************************
224 Check if an unlock overlaps a pending lock.
225 ****************************************************************************/
227 static bool brl_pending_overlap(const struct lock_struct *lock, const struct lock_struct *pend_lock)
229 if ((lock->start <= pend_lock->start) && (lock->start + lock->size > pend_lock->start))
230 return True;
231 if ((lock->start >= pend_lock->start) && (lock->start <= pend_lock->start + pend_lock->size))
232 return True;
233 return False;
236 /****************************************************************************
237 Amazingly enough, w2k3 "remembers" whether the last lock failure on a fnum
238 is the same as this one and changes its error code. I wonder if any
239 app depends on this ?
240 ****************************************************************************/
242 NTSTATUS brl_lock_failed(files_struct *fsp, const struct lock_struct *lock, bool blocking_lock)
244 if (lock->start >= 0xEF000000 && (lock->start >> 63) == 0) {
245 /* amazing the little things you learn with a test
246 suite. Locks beyond this offset (as a 64 bit
247 number!) always generate the conflict error code,
248 unless the top bit is set */
249 if (!blocking_lock) {
250 fsp->last_lock_failure = *lock;
252 return NT_STATUS_FILE_LOCK_CONFLICT;
255 if (procid_equal(&lock->context.pid, &fsp->last_lock_failure.context.pid) &&
256 lock->context.tid == fsp->last_lock_failure.context.tid &&
257 lock->fnum == fsp->last_lock_failure.fnum &&
258 lock->start == fsp->last_lock_failure.start) {
259 return NT_STATUS_FILE_LOCK_CONFLICT;
262 if (!blocking_lock) {
263 fsp->last_lock_failure = *lock;
265 return NT_STATUS_LOCK_NOT_GRANTED;
268 /****************************************************************************
269 Open up the brlock.tdb database.
270 ****************************************************************************/
272 void brl_init(bool read_only)
274 int tdb_flags;
276 if (brlock_db) {
277 return;
280 tdb_flags = TDB_DEFAULT|TDB_VOLATILE|TDB_CLEAR_IF_FIRST|TDB_INCOMPATIBLE_HASH;
282 if (!lp_clustering()) {
284 * We can't use the SEQNUM trick to cache brlock
285 * entries in the clustering case because ctdb seqnum
286 * propagation has a delay.
288 tdb_flags |= TDB_SEQNUM;
291 brlock_db = db_open(NULL, lock_path("brlock.tdb"),
292 lp_open_files_db_hash_size(), tdb_flags,
293 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644 );
294 if (!brlock_db) {
295 DEBUG(0,("Failed to open byte range locking database %s\n",
296 lock_path("brlock.tdb")));
297 return;
301 /****************************************************************************
302 Close down the brlock.tdb database.
303 ****************************************************************************/
305 void brl_shutdown(void)
307 TALLOC_FREE(brlock_db);
310 #if ZERO_ZERO
311 /****************************************************************************
312 Compare two locks for sorting.
313 ****************************************************************************/
315 static int lock_compare(const struct lock_struct *lck1,
316 const struct lock_struct *lck2)
318 if (lck1->start != lck2->start) {
319 return (lck1->start - lck2->start);
321 if (lck2->size != lck1->size) {
322 return ((int)lck1->size - (int)lck2->size);
324 return 0;
326 #endif
328 /****************************************************************************
329 Lock a range of bytes - Windows lock semantics.
330 ****************************************************************************/
332 NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
333 struct lock_struct *plock, bool blocking_lock)
335 unsigned int i;
336 files_struct *fsp = br_lck->fsp;
337 struct lock_struct *locks = br_lck->lock_data;
338 NTSTATUS status;
340 SMB_ASSERT(plock->lock_type != UNLOCK_LOCK);
342 if ((plock->start + plock->size - 1 < plock->start) &&
343 plock->size != 0) {
344 return NT_STATUS_INVALID_LOCK_RANGE;
347 for (i=0; i < br_lck->num_locks; i++) {
348 /* Do any Windows or POSIX locks conflict ? */
349 if (brl_conflict(&locks[i], plock)) {
350 /* Remember who blocked us. */
351 plock->context.smblctx = locks[i].context.smblctx;
352 return brl_lock_failed(fsp,plock,blocking_lock);
354 #if ZERO_ZERO
355 if (plock->start == 0 && plock->size == 0 &&
356 locks[i].size == 0) {
357 break;
359 #endif
362 if (!IS_PENDING_LOCK(plock->lock_type)) {
363 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
366 /* We can get the Windows lock, now see if it needs to
367 be mapped into a lower level POSIX one, and if so can
368 we get it ? */
370 if (!IS_PENDING_LOCK(plock->lock_type) && lp_posix_locking(fsp->conn->params)) {
371 int errno_ret;
372 if (!set_posix_lock_windows_flavour(fsp,
373 plock->start,
374 plock->size,
375 plock->lock_type,
376 &plock->context,
377 locks,
378 br_lck->num_locks,
379 &errno_ret)) {
381 /* We don't know who blocked us. */
382 plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
384 if (errno_ret == EACCES || errno_ret == EAGAIN) {
385 status = NT_STATUS_FILE_LOCK_CONFLICT;
386 goto fail;
387 } else {
388 status = map_nt_error_from_unix(errno);
389 goto fail;
394 /* no conflicts - add it to the list of locks */
395 locks = (struct lock_struct *)SMB_REALLOC(locks, (br_lck->num_locks + 1) * sizeof(*locks));
396 if (!locks) {
397 status = NT_STATUS_NO_MEMORY;
398 goto fail;
401 memcpy(&locks[br_lck->num_locks], plock, sizeof(struct lock_struct));
402 br_lck->num_locks += 1;
403 br_lck->lock_data = locks;
404 br_lck->modified = True;
406 return NT_STATUS_OK;
407 fail:
408 if (!IS_PENDING_LOCK(plock->lock_type)) {
409 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
411 return status;
414 /****************************************************************************
415 Cope with POSIX range splits and merges.
416 ****************************************************************************/
418 static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr, /* Output array. */
419 struct lock_struct *ex, /* existing lock. */
420 struct lock_struct *plock) /* proposed lock. */
422 bool lock_types_differ = (ex->lock_type != plock->lock_type);
424 /* We can't merge non-conflicting locks on different context - ignore fnum. */
426 if (!brl_same_context(&ex->context, &plock->context)) {
427 /* Just copy. */
428 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
429 return 1;
432 /* We now know we have the same context. */
434 /* Did we overlap ? */
436 /*********************************************
437 +---------+
438 | ex |
439 +---------+
440 +-------+
441 | plock |
442 +-------+
443 OR....
444 +---------+
445 | ex |
446 +---------+
447 **********************************************/
449 if ( (ex->start > (plock->start + plock->size)) ||
450 (plock->start > (ex->start + ex->size))) {
452 /* No overlap with this lock - copy existing. */
454 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
455 return 1;
458 /*********************************************
459 +---------------------------+
460 | ex |
461 +---------------------------+
462 +---------------------------+
463 | plock | -> replace with plock.
464 +---------------------------+
466 +---------------+
467 | ex |
468 +---------------+
469 +---------------------------+
470 | plock | -> replace with plock.
471 +---------------------------+
473 **********************************************/
475 if ( (ex->start >= plock->start) &&
476 (ex->start + ex->size <= plock->start + plock->size) ) {
478 /* Replace - discard existing lock. */
480 return 0;
483 /*********************************************
484 Adjacent after.
485 +-------+
486 | ex |
487 +-------+
488 +---------------+
489 | plock |
490 +---------------+
492 BECOMES....
493 +---------------+-------+
494 | plock | ex | - different lock types.
495 +---------------+-------+
496 OR.... (merge)
497 +-----------------------+
498 | plock | - same lock type.
499 +-----------------------+
500 **********************************************/
502 if (plock->start + plock->size == ex->start) {
504 /* If the lock types are the same, we merge, if different, we
505 add the remainder of the old lock. */
507 if (lock_types_differ) {
508 /* Add existing. */
509 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
510 return 1;
511 } else {
512 /* Merge - adjust incoming lock as we may have more
513 * merging to come. */
514 plock->size += ex->size;
515 return 0;
519 /*********************************************
520 Adjacent before.
521 +-------+
522 | ex |
523 +-------+
524 +---------------+
525 | plock |
526 +---------------+
527 BECOMES....
528 +-------+---------------+
529 | ex | plock | - different lock types
530 +-------+---------------+
532 OR.... (merge)
533 +-----------------------+
534 | plock | - same lock type.
535 +-----------------------+
537 **********************************************/
539 if (ex->start + ex->size == plock->start) {
541 /* If the lock types are the same, we merge, if different, we
542 add the existing lock. */
544 if (lock_types_differ) {
545 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
546 return 1;
547 } else {
548 /* Merge - adjust incoming lock as we may have more
549 * merging to come. */
550 plock->start = ex->start;
551 plock->size += ex->size;
552 return 0;
556 /*********************************************
557 Overlap after.
558 +-----------------------+
559 | ex |
560 +-----------------------+
561 +---------------+
562 | plock |
563 +---------------+
565 +----------------+
566 | ex |
567 +----------------+
568 +---------------+
569 | plock |
570 +---------------+
572 BECOMES....
573 +---------------+-------+
574 | plock | ex | - different lock types.
575 +---------------+-------+
576 OR.... (merge)
577 +-----------------------+
578 | plock | - same lock type.
579 +-----------------------+
580 **********************************************/
582 if ( (ex->start >= plock->start) &&
583 (ex->start <= plock->start + plock->size) &&
584 (ex->start + ex->size > plock->start + plock->size) ) {
586 /* If the lock types are the same, we merge, if different, we
587 add the remainder of the old lock. */
589 if (lock_types_differ) {
590 /* Add remaining existing. */
591 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
592 /* Adjust existing start and size. */
593 lck_arr[0].start = plock->start + plock->size;
594 lck_arr[0].size = (ex->start + ex->size) - (plock->start + plock->size);
595 return 1;
596 } else {
597 /* Merge - adjust incoming lock as we may have more
598 * merging to come. */
599 plock->size += (ex->start + ex->size) - (plock->start + plock->size);
600 return 0;
604 /*********************************************
605 Overlap before.
606 +-----------------------+
607 | ex |
608 +-----------------------+
609 +---------------+
610 | plock |
611 +---------------+
613 +-------------+
614 | ex |
615 +-------------+
616 +---------------+
617 | plock |
618 +---------------+
620 BECOMES....
621 +-------+---------------+
622 | ex | plock | - different lock types
623 +-------+---------------+
625 OR.... (merge)
626 +-----------------------+
627 | plock | - same lock type.
628 +-----------------------+
630 **********************************************/
632 if ( (ex->start < plock->start) &&
633 (ex->start + ex->size >= plock->start) &&
634 (ex->start + ex->size <= plock->start + plock->size) ) {
636 /* If the lock types are the same, we merge, if different, we
637 add the truncated old lock. */
639 if (lock_types_differ) {
640 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
641 /* Adjust existing size. */
642 lck_arr[0].size = plock->start - ex->start;
643 return 1;
644 } else {
645 /* Merge - adjust incoming lock as we may have more
646 * merging to come. MUST ADJUST plock SIZE FIRST ! */
647 plock->size += (plock->start - ex->start);
648 plock->start = ex->start;
649 return 0;
653 /*********************************************
654 Complete overlap.
655 +---------------------------+
656 | ex |
657 +---------------------------+
658 +---------+
659 | plock |
660 +---------+
661 BECOMES.....
662 +-------+---------+---------+
663 | ex | plock | ex | - different lock types.
664 +-------+---------+---------+
666 +---------------------------+
667 | plock | - same lock type.
668 +---------------------------+
669 **********************************************/
671 if ( (ex->start < plock->start) && (ex->start + ex->size > plock->start + plock->size) ) {
673 if (lock_types_differ) {
675 /* We have to split ex into two locks here. */
677 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
678 memcpy(&lck_arr[1], ex, sizeof(struct lock_struct));
680 /* Adjust first existing size. */
681 lck_arr[0].size = plock->start - ex->start;
683 /* Adjust second existing start and size. */
684 lck_arr[1].start = plock->start + plock->size;
685 lck_arr[1].size = (ex->start + ex->size) - (plock->start + plock->size);
686 return 2;
687 } else {
688 /* Just eat the existing locks, merge them into plock. */
689 plock->start = ex->start;
690 plock->size = ex->size;
691 return 0;
695 /* Never get here. */
696 smb_panic("brlock_posix_split_merge");
697 /* Notreached. */
699 /* Keep some compilers happy. */
700 return 0;
703 /****************************************************************************
704 Lock a range of bytes - POSIX lock semantics.
705 We must cope with range splits and merges.
706 ****************************************************************************/
708 static NTSTATUS brl_lock_posix(struct messaging_context *msg_ctx,
709 struct byte_range_lock *br_lck,
710 struct lock_struct *plock)
712 unsigned int i, count, posix_count;
713 struct lock_struct *locks = br_lck->lock_data;
714 struct lock_struct *tp;
715 bool signal_pending_read = False;
716 bool break_oplocks = false;
717 NTSTATUS status;
719 /* No zero-zero locks for POSIX. */
720 if (plock->start == 0 && plock->size == 0) {
721 return NT_STATUS_INVALID_PARAMETER;
724 /* Don't allow 64-bit lock wrap. */
725 if (plock->start + plock->size - 1 < plock->start) {
726 return NT_STATUS_INVALID_PARAMETER;
729 /* The worst case scenario here is we have to split an
730 existing POSIX lock range into two, and add our lock,
731 so we need at most 2 more entries. */
733 tp = SMB_MALLOC_ARRAY(struct lock_struct, (br_lck->num_locks + 2));
734 if (!tp) {
735 return NT_STATUS_NO_MEMORY;
738 count = posix_count = 0;
740 for (i=0; i < br_lck->num_locks; i++) {
741 struct lock_struct *curr_lock = &locks[i];
743 /* If we have a pending read lock, a lock downgrade should
744 trigger a lock re-evaluation. */
745 if (curr_lock->lock_type == PENDING_READ_LOCK &&
746 brl_pending_overlap(plock, curr_lock)) {
747 signal_pending_read = True;
750 if (curr_lock->lock_flav == WINDOWS_LOCK) {
751 /* Do any Windows flavour locks conflict ? */
752 if (brl_conflict(curr_lock, plock)) {
753 /* No games with error messages. */
754 SAFE_FREE(tp);
755 /* Remember who blocked us. */
756 plock->context.smblctx = curr_lock->context.smblctx;
757 return NT_STATUS_FILE_LOCK_CONFLICT;
759 /* Just copy the Windows lock into the new array. */
760 memcpy(&tp[count], curr_lock, sizeof(struct lock_struct));
761 count++;
762 } else {
763 unsigned int tmp_count = 0;
765 /* POSIX conflict semantics are different. */
766 if (brl_conflict_posix(curr_lock, plock)) {
767 /* Can't block ourselves with POSIX locks. */
768 /* No games with error messages. */
769 SAFE_FREE(tp);
770 /* Remember who blocked us. */
771 plock->context.smblctx = curr_lock->context.smblctx;
772 return NT_STATUS_FILE_LOCK_CONFLICT;
775 /* Work out overlaps. */
776 tmp_count += brlock_posix_split_merge(&tp[count], curr_lock, plock);
777 posix_count += tmp_count;
778 count += tmp_count;
783 * Break oplocks while we hold a brl. Since lock() and unlock() calls
784 * are not symetric with POSIX semantics, we cannot guarantee our
785 * contend_level2_oplocks_begin/end calls will be acquired and
786 * released one-for-one as with Windows semantics. Therefore we only
787 * call contend_level2_oplocks_begin if this is the first POSIX brl on
788 * the file.
790 break_oplocks = (!IS_PENDING_LOCK(plock->lock_type) &&
791 posix_count == 0);
792 if (break_oplocks) {
793 contend_level2_oplocks_begin(br_lck->fsp,
794 LEVEL2_CONTEND_POSIX_BRL);
797 /* Try and add the lock in order, sorted by lock start. */
798 for (i=0; i < count; i++) {
799 struct lock_struct *curr_lock = &tp[i];
801 if (curr_lock->start <= plock->start) {
802 continue;
806 if (i < count) {
807 memmove(&tp[i+1], &tp[i],
808 (count - i)*sizeof(struct lock_struct));
810 memcpy(&tp[i], plock, sizeof(struct lock_struct));
811 count++;
813 /* We can get the POSIX lock, now see if it needs to
814 be mapped into a lower level POSIX one, and if so can
815 we get it ? */
817 if (!IS_PENDING_LOCK(plock->lock_type) && lp_posix_locking(br_lck->fsp->conn->params)) {
818 int errno_ret;
820 /* The lower layer just needs to attempt to
821 get the system POSIX lock. We've weeded out
822 any conflicts above. */
824 if (!set_posix_lock_posix_flavour(br_lck->fsp,
825 plock->start,
826 plock->size,
827 plock->lock_type,
828 &errno_ret)) {
830 /* We don't know who blocked us. */
831 plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
833 if (errno_ret == EACCES || errno_ret == EAGAIN) {
834 SAFE_FREE(tp);
835 status = NT_STATUS_FILE_LOCK_CONFLICT;
836 goto fail;
837 } else {
838 SAFE_FREE(tp);
839 status = map_nt_error_from_unix(errno);
840 goto fail;
845 /* If we didn't use all the allocated size,
846 * Realloc so we don't leak entries per lock call. */
847 if (count < br_lck->num_locks + 2) {
848 tp = (struct lock_struct *)SMB_REALLOC(tp, count * sizeof(*locks));
849 if (!tp) {
850 status = NT_STATUS_NO_MEMORY;
851 goto fail;
855 br_lck->num_locks = count;
856 SAFE_FREE(br_lck->lock_data);
857 br_lck->lock_data = tp;
858 locks = tp;
859 br_lck->modified = True;
861 /* A successful downgrade from write to read lock can trigger a lock
862 re-evalutation where waiting readers can now proceed. */
864 if (signal_pending_read) {
865 /* Send unlock messages to any pending read waiters that overlap. */
866 for (i=0; i < br_lck->num_locks; i++) {
867 struct lock_struct *pend_lock = &locks[i];
869 /* Ignore non-pending locks. */
870 if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
871 continue;
874 if (pend_lock->lock_type == PENDING_READ_LOCK &&
875 brl_pending_overlap(plock, pend_lock)) {
876 DEBUG(10,("brl_lock_posix: sending unlock message to pid %s\n",
877 procid_str_static(&pend_lock->context.pid )));
879 messaging_send(msg_ctx, pend_lock->context.pid,
880 MSG_SMB_UNLOCK, &data_blob_null);
885 return NT_STATUS_OK;
886 fail:
887 if (break_oplocks) {
888 contend_level2_oplocks_end(br_lck->fsp,
889 LEVEL2_CONTEND_POSIX_BRL);
891 return status;
894 NTSTATUS smb_vfs_call_brl_lock_windows(struct vfs_handle_struct *handle,
895 struct byte_range_lock *br_lck,
896 struct lock_struct *plock,
897 bool blocking_lock,
898 struct blocking_lock_record *blr)
900 VFS_FIND(brl_lock_windows);
901 return handle->fns->brl_lock_windows(handle, br_lck, plock,
902 blocking_lock, blr);
905 /****************************************************************************
906 Lock a range of bytes.
907 ****************************************************************************/
909 NTSTATUS brl_lock(struct messaging_context *msg_ctx,
910 struct byte_range_lock *br_lck,
911 uint64_t smblctx,
912 struct server_id pid,
913 br_off start,
914 br_off size,
915 enum brl_type lock_type,
916 enum brl_flavour lock_flav,
917 bool blocking_lock,
918 uint64_t *psmblctx,
919 struct blocking_lock_record *blr)
921 NTSTATUS ret;
922 struct lock_struct lock;
924 #if !ZERO_ZERO
925 if (start == 0 && size == 0) {
926 DEBUG(0,("client sent 0/0 lock - please report this\n"));
928 #endif
930 #ifdef DEVELOPER
931 /* Quieten valgrind on test. */
932 memset(&lock, '\0', sizeof(lock));
933 #endif
935 lock.context.smblctx = smblctx;
936 lock.context.pid = pid;
937 lock.context.tid = br_lck->fsp->conn->cnum;
938 lock.start = start;
939 lock.size = size;
940 lock.fnum = br_lck->fsp->fnum;
941 lock.lock_type = lock_type;
942 lock.lock_flav = lock_flav;
944 if (lock_flav == WINDOWS_LOCK) {
945 ret = SMB_VFS_BRL_LOCK_WINDOWS(br_lck->fsp->conn, br_lck,
946 &lock, blocking_lock, blr);
947 } else {
948 ret = brl_lock_posix(msg_ctx, br_lck, &lock);
951 #if ZERO_ZERO
952 /* sort the lock list */
953 TYPESAFE_QSORT(br_lck->lock_data, (size_t)br_lck->num_locks, lock_compare);
954 #endif
956 /* If we're returning an error, return who blocked us. */
957 if (!NT_STATUS_IS_OK(ret) && psmblctx) {
958 *psmblctx = lock.context.smblctx;
960 return ret;
963 /****************************************************************************
964 Unlock a range of bytes - Windows semantics.
965 ****************************************************************************/
967 bool brl_unlock_windows_default(struct messaging_context *msg_ctx,
968 struct byte_range_lock *br_lck,
969 const struct lock_struct *plock)
971 unsigned int i, j;
972 struct lock_struct *locks = br_lck->lock_data;
973 enum brl_type deleted_lock_type = READ_LOCK; /* shut the compiler up.... */
975 SMB_ASSERT(plock->lock_type == UNLOCK_LOCK);
977 #if ZERO_ZERO
978 /* Delete write locks by preference... The lock list
979 is sorted in the zero zero case. */
981 for (i = 0; i < br_lck->num_locks; i++) {
982 struct lock_struct *lock = &locks[i];
984 if (lock->lock_type == WRITE_LOCK &&
985 brl_same_context(&lock->context, &plock->context) &&
986 lock->fnum == plock->fnum &&
987 lock->lock_flav == WINDOWS_LOCK &&
988 lock->start == plock->start &&
989 lock->size == plock->size) {
991 /* found it - delete it */
992 deleted_lock_type = lock->lock_type;
993 break;
997 if (i != br_lck->num_locks) {
998 /* We found it - don't search again. */
999 goto unlock_continue;
1001 #endif
1003 for (i = 0; i < br_lck->num_locks; i++) {
1004 struct lock_struct *lock = &locks[i];
1006 if (IS_PENDING_LOCK(lock->lock_type)) {
1007 continue;
1010 /* Only remove our own locks that match in start, size, and flavour. */
1011 if (brl_same_context(&lock->context, &plock->context) &&
1012 lock->fnum == plock->fnum &&
1013 lock->lock_flav == WINDOWS_LOCK &&
1014 lock->start == plock->start &&
1015 lock->size == plock->size ) {
1016 deleted_lock_type = lock->lock_type;
1017 break;
1021 if (i == br_lck->num_locks) {
1022 /* we didn't find it */
1023 return False;
1026 #if ZERO_ZERO
1027 unlock_continue:
1028 #endif
1030 /* Actually delete the lock. */
1031 if (i < br_lck->num_locks - 1) {
1032 memmove(&locks[i], &locks[i+1],
1033 sizeof(*locks)*((br_lck->num_locks-1) - i));
1036 br_lck->num_locks -= 1;
1037 br_lck->modified = True;
1039 /* Unlock the underlying POSIX regions. */
1040 if(lp_posix_locking(br_lck->fsp->conn->params)) {
1041 release_posix_lock_windows_flavour(br_lck->fsp,
1042 plock->start,
1043 plock->size,
1044 deleted_lock_type,
1045 &plock->context,
1046 locks,
1047 br_lck->num_locks);
1050 /* Send unlock messages to any pending waiters that overlap. */
1051 for (j=0; j < br_lck->num_locks; j++) {
1052 struct lock_struct *pend_lock = &locks[j];
1054 /* Ignore non-pending locks. */
1055 if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
1056 continue;
1059 /* We could send specific lock info here... */
1060 if (brl_pending_overlap(plock, pend_lock)) {
1061 DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1062 procid_str_static(&pend_lock->context.pid )));
1064 messaging_send(msg_ctx, pend_lock->context.pid,
1065 MSG_SMB_UNLOCK, &data_blob_null);
1069 contend_level2_oplocks_end(br_lck->fsp, LEVEL2_CONTEND_WINDOWS_BRL);
1070 return True;
1073 /****************************************************************************
1074 Unlock a range of bytes - POSIX semantics.
1075 ****************************************************************************/
1077 static bool brl_unlock_posix(struct messaging_context *msg_ctx,
1078 struct byte_range_lock *br_lck,
1079 struct lock_struct *plock)
1081 unsigned int i, j, count;
1082 struct lock_struct *tp;
1083 struct lock_struct *locks = br_lck->lock_data;
1084 bool overlap_found = False;
1086 /* No zero-zero locks for POSIX. */
1087 if (plock->start == 0 && plock->size == 0) {
1088 return False;
1091 /* Don't allow 64-bit lock wrap. */
1092 if (plock->start + plock->size < plock->start ||
1093 plock->start + plock->size < plock->size) {
1094 DEBUG(10,("brl_unlock_posix: lock wrap\n"));
1095 return False;
1098 /* The worst case scenario here is we have to split an
1099 existing POSIX lock range into two, so we need at most
1100 1 more entry. */
1102 tp = SMB_MALLOC_ARRAY(struct lock_struct, (br_lck->num_locks + 1));
1103 if (!tp) {
1104 DEBUG(10,("brl_unlock_posix: malloc fail\n"));
1105 return False;
1108 count = 0;
1109 for (i = 0; i < br_lck->num_locks; i++) {
1110 struct lock_struct *lock = &locks[i];
1111 unsigned int tmp_count;
1113 /* Only remove our own locks - ignore fnum. */
1114 if (IS_PENDING_LOCK(lock->lock_type) ||
1115 !brl_same_context(&lock->context, &plock->context)) {
1116 memcpy(&tp[count], lock, sizeof(struct lock_struct));
1117 count++;
1118 continue;
1121 if (lock->lock_flav == WINDOWS_LOCK) {
1122 /* Do any Windows flavour locks conflict ? */
1123 if (brl_conflict(lock, plock)) {
1124 SAFE_FREE(tp);
1125 return false;
1127 /* Just copy the Windows lock into the new array. */
1128 memcpy(&tp[count], lock, sizeof(struct lock_struct));
1129 count++;
1130 continue;
1133 /* Work out overlaps. */
1134 tmp_count = brlock_posix_split_merge(&tp[count], lock, plock);
1136 if (tmp_count == 0) {
1137 /* plock overlapped the existing lock completely,
1138 or replaced it. Don't copy the existing lock. */
1139 overlap_found = true;
1140 } else if (tmp_count == 1) {
1141 /* Either no overlap, (simple copy of existing lock) or
1142 * an overlap of an existing lock. */
1143 /* If the lock changed size, we had an overlap. */
1144 if (tp[count].size != lock->size) {
1145 overlap_found = true;
1147 count += tmp_count;
1148 } else if (tmp_count == 2) {
1149 /* We split a lock range in two. */
1150 overlap_found = true;
1151 count += tmp_count;
1153 /* Optimisation... */
1154 /* We know we're finished here as we can't overlap any
1155 more POSIX locks. Copy the rest of the lock array. */
1157 if (i < br_lck->num_locks - 1) {
1158 memcpy(&tp[count], &locks[i+1],
1159 sizeof(*locks)*((br_lck->num_locks-1) - i));
1160 count += ((br_lck->num_locks-1) - i);
1162 break;
1167 if (!overlap_found) {
1168 /* Just ignore - no change. */
1169 SAFE_FREE(tp);
1170 DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
1171 return True;
1174 /* Unlock any POSIX regions. */
1175 if(lp_posix_locking(br_lck->fsp->conn->params)) {
1176 release_posix_lock_posix_flavour(br_lck->fsp,
1177 plock->start,
1178 plock->size,
1179 &plock->context,
1181 count);
1184 /* Realloc so we don't leak entries per unlock call. */
1185 if (count) {
1186 tp = (struct lock_struct *)SMB_REALLOC(tp, count * sizeof(*locks));
1187 if (!tp) {
1188 DEBUG(10,("brl_unlock_posix: realloc fail\n"));
1189 return False;
1191 } else {
1192 /* We deleted the last lock. */
1193 SAFE_FREE(tp);
1194 tp = NULL;
1197 contend_level2_oplocks_end(br_lck->fsp,
1198 LEVEL2_CONTEND_POSIX_BRL);
1200 br_lck->num_locks = count;
1201 SAFE_FREE(br_lck->lock_data);
1202 locks = tp;
1203 br_lck->lock_data = tp;
1204 br_lck->modified = True;
1206 /* Send unlock messages to any pending waiters that overlap. */
1208 for (j=0; j < br_lck->num_locks; j++) {
1209 struct lock_struct *pend_lock = &locks[j];
1211 /* Ignore non-pending locks. */
1212 if (!IS_PENDING_LOCK(pend_lock->lock_type)) {
1213 continue;
1216 /* We could send specific lock info here... */
1217 if (brl_pending_overlap(plock, pend_lock)) {
1218 DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1219 procid_str_static(&pend_lock->context.pid )));
1221 messaging_send(msg_ctx, pend_lock->context.pid,
1222 MSG_SMB_UNLOCK, &data_blob_null);
1226 return True;
1229 bool smb_vfs_call_brl_unlock_windows(struct vfs_handle_struct *handle,
1230 struct messaging_context *msg_ctx,
1231 struct byte_range_lock *br_lck,
1232 const struct lock_struct *plock)
1234 VFS_FIND(brl_unlock_windows);
1235 return handle->fns->brl_unlock_windows(handle, msg_ctx, br_lck, plock);
1238 /****************************************************************************
1239 Unlock a range of bytes.
1240 ****************************************************************************/
1242 bool brl_unlock(struct messaging_context *msg_ctx,
1243 struct byte_range_lock *br_lck,
1244 uint64_t smblctx,
1245 struct server_id pid,
1246 br_off start,
1247 br_off size,
1248 enum brl_flavour lock_flav)
1250 struct lock_struct lock;
1252 lock.context.smblctx = smblctx;
1253 lock.context.pid = pid;
1254 lock.context.tid = br_lck->fsp->conn->cnum;
1255 lock.start = start;
1256 lock.size = size;
1257 lock.fnum = br_lck->fsp->fnum;
1258 lock.lock_type = UNLOCK_LOCK;
1259 lock.lock_flav = lock_flav;
1261 if (lock_flav == WINDOWS_LOCK) {
1262 return SMB_VFS_BRL_UNLOCK_WINDOWS(br_lck->fsp->conn, msg_ctx,
1263 br_lck, &lock);
1264 } else {
1265 return brl_unlock_posix(msg_ctx, br_lck, &lock);
1269 /****************************************************************************
1270 Test if we could add a lock if we wanted to.
1271 Returns True if the region required is currently unlocked, False if locked.
1272 ****************************************************************************/
1274 bool brl_locktest(struct byte_range_lock *br_lck,
1275 uint64_t smblctx,
1276 struct server_id pid,
1277 br_off start,
1278 br_off size,
1279 enum brl_type lock_type,
1280 enum brl_flavour lock_flav)
1282 bool ret = True;
1283 unsigned int i;
1284 struct lock_struct lock;
1285 const struct lock_struct *locks = br_lck->lock_data;
1286 files_struct *fsp = br_lck->fsp;
1288 lock.context.smblctx = smblctx;
1289 lock.context.pid = pid;
1290 lock.context.tid = br_lck->fsp->conn->cnum;
1291 lock.start = start;
1292 lock.size = size;
1293 lock.fnum = fsp->fnum;
1294 lock.lock_type = lock_type;
1295 lock.lock_flav = lock_flav;
1297 /* Make sure existing locks don't conflict */
1298 for (i=0; i < br_lck->num_locks; i++) {
1300 * Our own locks don't conflict.
1302 if (brl_conflict_other(&locks[i], &lock)) {
1303 return False;
1308 * There is no lock held by an SMB daemon, check to
1309 * see if there is a POSIX lock from a UNIX or NFS process.
1310 * This only conflicts with Windows locks, not POSIX locks.
1313 if(lp_posix_locking(fsp->conn->params) && (lock_flav == WINDOWS_LOCK)) {
1314 ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK);
1316 DEBUG(10,("brl_locktest: posix start=%.0f len=%.0f %s for fnum %d file %s\n",
1317 (double)start, (double)size, ret ? "locked" : "unlocked",
1318 fsp->fnum, fsp_str_dbg(fsp)));
1320 /* We need to return the inverse of is_posix_locked. */
1321 ret = !ret;
1324 /* no conflicts - we could have added it */
1325 return ret;
1328 /****************************************************************************
1329 Query for existing locks.
1330 ****************************************************************************/
1332 NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
1333 uint64_t *psmblctx,
1334 struct server_id pid,
1335 br_off *pstart,
1336 br_off *psize,
1337 enum brl_type *plock_type,
1338 enum brl_flavour lock_flav)
1340 unsigned int i;
1341 struct lock_struct lock;
1342 const struct lock_struct *locks = br_lck->lock_data;
1343 files_struct *fsp = br_lck->fsp;
1345 lock.context.smblctx = *psmblctx;
1346 lock.context.pid = pid;
1347 lock.context.tid = br_lck->fsp->conn->cnum;
1348 lock.start = *pstart;
1349 lock.size = *psize;
1350 lock.fnum = fsp->fnum;
1351 lock.lock_type = *plock_type;
1352 lock.lock_flav = lock_flav;
1354 /* Make sure existing locks don't conflict */
1355 for (i=0; i < br_lck->num_locks; i++) {
1356 const struct lock_struct *exlock = &locks[i];
1357 bool conflict = False;
1359 if (exlock->lock_flav == WINDOWS_LOCK) {
1360 conflict = brl_conflict(exlock, &lock);
1361 } else {
1362 conflict = brl_conflict_posix(exlock, &lock);
1365 if (conflict) {
1366 *psmblctx = exlock->context.smblctx;
1367 *pstart = exlock->start;
1368 *psize = exlock->size;
1369 *plock_type = exlock->lock_type;
1370 return NT_STATUS_LOCK_NOT_GRANTED;
1375 * There is no lock held by an SMB daemon, check to
1376 * see if there is a POSIX lock from a UNIX or NFS process.
1379 if(lp_posix_locking(fsp->conn->params)) {
1380 bool ret = is_posix_locked(fsp, pstart, psize, plock_type, POSIX_LOCK);
1382 DEBUG(10,("brl_lockquery: posix start=%.0f len=%.0f %s for fnum %d file %s\n",
1383 (double)*pstart, (double)*psize, ret ? "locked" : "unlocked",
1384 fsp->fnum, fsp_str_dbg(fsp)));
1386 if (ret) {
1387 /* Hmmm. No clue what to set smblctx to - use -1. */
1388 *psmblctx = 0xFFFFFFFFFFFFFFFFLL;
1389 return NT_STATUS_LOCK_NOT_GRANTED;
1393 return NT_STATUS_OK;
1397 bool smb_vfs_call_brl_cancel_windows(struct vfs_handle_struct *handle,
1398 struct byte_range_lock *br_lck,
1399 struct lock_struct *plock,
1400 struct blocking_lock_record *blr)
1402 VFS_FIND(brl_cancel_windows);
1403 return handle->fns->brl_cancel_windows(handle, br_lck, plock, blr);
1406 /****************************************************************************
1407 Remove a particular pending lock.
1408 ****************************************************************************/
1409 bool brl_lock_cancel(struct byte_range_lock *br_lck,
1410 uint64_t smblctx,
1411 struct server_id pid,
1412 br_off start,
1413 br_off size,
1414 enum brl_flavour lock_flav,
1415 struct blocking_lock_record *blr)
1417 bool ret;
1418 struct lock_struct lock;
1420 lock.context.smblctx = smblctx;
1421 lock.context.pid = pid;
1422 lock.context.tid = br_lck->fsp->conn->cnum;
1423 lock.start = start;
1424 lock.size = size;
1425 lock.fnum = br_lck->fsp->fnum;
1426 lock.lock_flav = lock_flav;
1427 /* lock.lock_type doesn't matter */
1429 if (lock_flav == WINDOWS_LOCK) {
1430 ret = SMB_VFS_BRL_CANCEL_WINDOWS(br_lck->fsp->conn, br_lck,
1431 &lock, blr);
1432 } else {
1433 ret = brl_lock_cancel_default(br_lck, &lock);
1436 return ret;
1439 bool brl_lock_cancel_default(struct byte_range_lock *br_lck,
1440 struct lock_struct *plock)
1442 unsigned int i;
1443 struct lock_struct *locks = br_lck->lock_data;
1445 SMB_ASSERT(plock);
1447 for (i = 0; i < br_lck->num_locks; i++) {
1448 struct lock_struct *lock = &locks[i];
1450 /* For pending locks we *always* care about the fnum. */
1451 if (brl_same_context(&lock->context, &plock->context) &&
1452 lock->fnum == plock->fnum &&
1453 IS_PENDING_LOCK(lock->lock_type) &&
1454 lock->lock_flav == plock->lock_flav &&
1455 lock->start == plock->start &&
1456 lock->size == plock->size) {
1457 break;
1461 if (i == br_lck->num_locks) {
1462 /* Didn't find it. */
1463 return False;
1466 if (i < br_lck->num_locks - 1) {
1467 /* Found this particular pending lock - delete it */
1468 memmove(&locks[i], &locks[i+1],
1469 sizeof(*locks)*((br_lck->num_locks-1) - i));
1472 br_lck->num_locks -= 1;
1473 br_lck->modified = True;
1474 return True;
1477 /****************************************************************************
1478 Remove any locks associated with a open file.
1479 We return True if this process owns any other Windows locks on this
1480 fd and so we should not immediately close the fd.
1481 ****************************************************************************/
1483 void brl_close_fnum(struct messaging_context *msg_ctx,
1484 struct byte_range_lock *br_lck)
1486 files_struct *fsp = br_lck->fsp;
1487 uint16 tid = fsp->conn->cnum;
1488 int fnum = fsp->fnum;
1489 unsigned int i;
1490 struct lock_struct *locks = br_lck->lock_data;
1491 struct server_id pid = sconn_server_id(fsp->conn->sconn);
1492 struct lock_struct *locks_copy;
1493 unsigned int num_locks_copy;
1495 /* Copy the current lock array. */
1496 if (br_lck->num_locks) {
1497 locks_copy = (struct lock_struct *)talloc_memdup(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
1498 if (!locks_copy) {
1499 smb_panic("brl_close_fnum: talloc failed");
1501 } else {
1502 locks_copy = NULL;
1505 num_locks_copy = br_lck->num_locks;
1507 for (i=0; i < num_locks_copy; i++) {
1508 struct lock_struct *lock = &locks_copy[i];
1510 if (lock->context.tid == tid && procid_equal(&lock->context.pid, &pid) &&
1511 (lock->fnum == fnum)) {
1512 brl_unlock(msg_ctx,
1513 br_lck,
1514 lock->context.smblctx,
1515 pid,
1516 lock->start,
1517 lock->size,
1518 lock->lock_flav);
1523 /****************************************************************************
1524 Ensure this set of lock entries is valid.
1525 ****************************************************************************/
1526 static bool validate_lock_entries(unsigned int *pnum_entries, struct lock_struct **pplocks)
1528 unsigned int i;
1529 unsigned int num_valid_entries = 0;
1530 struct lock_struct *locks = *pplocks;
1532 for (i = 0; i < *pnum_entries; i++) {
1533 struct lock_struct *lock_data = &locks[i];
1534 if (!serverid_exists(&lock_data->context.pid)) {
1535 /* This process no longer exists - mark this
1536 entry as invalid by zeroing it. */
1537 ZERO_STRUCTP(lock_data);
1538 } else {
1539 num_valid_entries++;
1543 if (num_valid_entries != *pnum_entries) {
1544 struct lock_struct *new_lock_data = NULL;
1546 if (num_valid_entries) {
1547 new_lock_data = SMB_MALLOC_ARRAY(struct lock_struct, num_valid_entries);
1548 if (!new_lock_data) {
1549 DEBUG(3, ("malloc fail\n"));
1550 return False;
1553 num_valid_entries = 0;
1554 for (i = 0; i < *pnum_entries; i++) {
1555 struct lock_struct *lock_data = &locks[i];
1556 if (lock_data->context.smblctx &&
1557 lock_data->context.tid) {
1558 /* Valid (nonzero) entry - copy it. */
1559 memcpy(&new_lock_data[num_valid_entries],
1560 lock_data, sizeof(struct lock_struct));
1561 num_valid_entries++;
1566 SAFE_FREE(*pplocks);
1567 *pplocks = new_lock_data;
1568 *pnum_entries = num_valid_entries;
1571 return True;
1574 struct brl_forall_cb {
1575 void (*fn)(struct file_id id, struct server_id pid,
1576 enum brl_type lock_type,
1577 enum brl_flavour lock_flav,
1578 br_off start, br_off size,
1579 void *private_data);
1580 void *private_data;
1583 /****************************************************************************
1584 Traverse the whole database with this function, calling traverse_callback
1585 on each lock.
1586 ****************************************************************************/
1588 static int traverse_fn(struct db_record *rec, void *state)
1590 struct brl_forall_cb *cb = (struct brl_forall_cb *)state;
1591 struct lock_struct *locks;
1592 struct file_id *key;
1593 unsigned int i;
1594 unsigned int num_locks = 0;
1595 unsigned int orig_num_locks = 0;
1597 /* In a traverse function we must make a copy of
1598 dbuf before modifying it. */
1600 locks = (struct lock_struct *)memdup(rec->value.dptr,
1601 rec->value.dsize);
1602 if (!locks) {
1603 return -1; /* Terminate traversal. */
1606 key = (struct file_id *)rec->key.dptr;
1607 orig_num_locks = num_locks = rec->value.dsize/sizeof(*locks);
1609 /* Ensure the lock db is clean of entries from invalid processes. */
1611 if (!validate_lock_entries(&num_locks, &locks)) {
1612 SAFE_FREE(locks);
1613 return -1; /* Terminate traversal */
1616 if (orig_num_locks != num_locks) {
1617 if (num_locks) {
1618 TDB_DATA data;
1619 data.dptr = (uint8_t *)locks;
1620 data.dsize = num_locks*sizeof(struct lock_struct);
1621 rec->store(rec, data, TDB_REPLACE);
1622 } else {
1623 rec->delete_rec(rec);
1627 if (cb->fn) {
1628 for ( i=0; i<num_locks; i++) {
1629 cb->fn(*key,
1630 locks[i].context.pid,
1631 locks[i].lock_type,
1632 locks[i].lock_flav,
1633 locks[i].start,
1634 locks[i].size,
1635 cb->private_data);
1639 SAFE_FREE(locks);
1640 return 0;
1643 /*******************************************************************
1644 Call the specified function on each lock in the database.
1645 ********************************************************************/
1647 int brl_forall(void (*fn)(struct file_id id, struct server_id pid,
1648 enum brl_type lock_type,
1649 enum brl_flavour lock_flav,
1650 br_off start, br_off size,
1651 void *private_data),
1652 void *private_data)
1654 struct brl_forall_cb cb;
1656 if (!brlock_db) {
1657 return 0;
1659 cb.fn = fn;
1660 cb.private_data = private_data;
1661 return brlock_db->traverse(brlock_db, traverse_fn, &cb);
1664 /*******************************************************************
1665 Store a potentially modified set of byte range lock data back into
1666 the database.
1667 Unlock the record.
1668 ********************************************************************/
1670 static void byte_range_lock_flush(struct byte_range_lock *br_lck)
1672 if (br_lck->read_only) {
1673 SMB_ASSERT(!br_lck->modified);
1676 if (!br_lck->modified) {
1677 goto done;
1680 if (br_lck->num_locks == 0) {
1681 /* No locks - delete this entry. */
1682 NTSTATUS status = br_lck->record->delete_rec(br_lck->record);
1683 if (!NT_STATUS_IS_OK(status)) {
1684 DEBUG(0, ("delete_rec returned %s\n",
1685 nt_errstr(status)));
1686 smb_panic("Could not delete byte range lock entry");
1688 } else {
1689 TDB_DATA data;
1690 NTSTATUS status;
1692 data.dptr = (uint8 *)br_lck->lock_data;
1693 data.dsize = br_lck->num_locks * sizeof(struct lock_struct);
1695 status = br_lck->record->store(br_lck->record, data,
1696 TDB_REPLACE);
1697 if (!NT_STATUS_IS_OK(status)) {
1698 DEBUG(0, ("store returned %s\n", nt_errstr(status)));
1699 smb_panic("Could not store byte range mode entry");
1703 done:
1705 br_lck->read_only = true;
1706 br_lck->modified = false;
1708 TALLOC_FREE(br_lck->record);
1711 static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
1713 byte_range_lock_flush(br_lck);
1714 SAFE_FREE(br_lck->lock_data);
1715 return 0;
1718 /*******************************************************************
1719 Fetch a set of byte range lock data from the database.
1720 Leave the record locked.
1721 TALLOC_FREE(brl) will release the lock in the destructor.
1722 ********************************************************************/
1724 static struct byte_range_lock *brl_get_locks_internal(TALLOC_CTX *mem_ctx,
1725 files_struct *fsp, bool read_only)
1727 TDB_DATA key, data;
1728 struct byte_range_lock *br_lck = talloc(mem_ctx, struct byte_range_lock);
1729 bool do_read_only = read_only;
1731 if (br_lck == NULL) {
1732 return NULL;
1735 br_lck->fsp = fsp;
1736 br_lck->num_locks = 0;
1737 br_lck->modified = False;
1738 br_lck->key = fsp->file_id;
1740 key.dptr = (uint8 *)&br_lck->key;
1741 key.dsize = sizeof(struct file_id);
1743 if (!fsp->lockdb_clean) {
1744 /* We must be read/write to clean
1745 the dead entries. */
1746 do_read_only = false;
1749 if (do_read_only) {
1750 if (brlock_db->fetch(brlock_db, br_lck, key, &data) != 0) {
1751 DEBUG(3, ("Could not fetch byte range lock record\n"));
1752 TALLOC_FREE(br_lck);
1753 return NULL;
1755 br_lck->record = NULL;
1756 } else {
1757 br_lck->record = brlock_db->fetch_locked(brlock_db, br_lck, key);
1759 if (br_lck->record == NULL) {
1760 DEBUG(3, ("Could not lock byte range lock entry\n"));
1761 TALLOC_FREE(br_lck);
1762 return NULL;
1765 data = br_lck->record->value;
1768 br_lck->read_only = do_read_only;
1769 br_lck->lock_data = NULL;
1771 talloc_set_destructor(br_lck, byte_range_lock_destructor);
1773 br_lck->num_locks = data.dsize / sizeof(struct lock_struct);
1775 if (br_lck->num_locks != 0) {
1776 br_lck->lock_data = SMB_MALLOC_ARRAY(struct lock_struct,
1777 br_lck->num_locks);
1778 if (br_lck->lock_data == NULL) {
1779 DEBUG(0, ("malloc failed\n"));
1780 TALLOC_FREE(br_lck);
1781 return NULL;
1784 memcpy(br_lck->lock_data, data.dptr, data.dsize);
1787 if (!fsp->lockdb_clean) {
1788 int orig_num_locks = br_lck->num_locks;
1790 /* This is the first time we've accessed this. */
1791 /* Go through and ensure all entries exist - remove any that don't. */
1792 /* Makes the lockdb self cleaning at low cost. */
1794 if (!validate_lock_entries(&br_lck->num_locks,
1795 &br_lck->lock_data)) {
1796 SAFE_FREE(br_lck->lock_data);
1797 TALLOC_FREE(br_lck);
1798 return NULL;
1801 /* Ensure invalid locks are cleaned up in the destructor. */
1802 if (orig_num_locks != br_lck->num_locks) {
1803 br_lck->modified = True;
1806 /* Mark the lockdb as "clean" as seen from this open file. */
1807 fsp->lockdb_clean = True;
1810 if (DEBUGLEVEL >= 10) {
1811 unsigned int i;
1812 struct lock_struct *locks = br_lck->lock_data;
1813 DEBUG(10,("brl_get_locks_internal: %u current locks on file_id %s\n",
1814 br_lck->num_locks,
1815 file_id_string_tos(&fsp->file_id)));
1816 for( i = 0; i < br_lck->num_locks; i++) {
1817 print_lock_struct(i, &locks[i]);
1821 if (do_read_only != read_only) {
1823 * this stores the record and gets rid of
1824 * the write lock that is needed for a cleanup
1826 byte_range_lock_flush(br_lck);
1829 return br_lck;
1832 struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx,
1833 files_struct *fsp)
1835 return brl_get_locks_internal(mem_ctx, fsp, False);
1838 struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp)
1840 struct byte_range_lock *br_lock;
1842 if (lp_clustering()) {
1843 return brl_get_locks_internal(talloc_tos(), fsp, true);
1846 if ((fsp->brlock_rec != NULL)
1847 && (brlock_db->get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
1848 return fsp->brlock_rec;
1851 TALLOC_FREE(fsp->brlock_rec);
1853 br_lock = brl_get_locks_internal(talloc_tos(), fsp, true);
1854 if (br_lock == NULL) {
1855 return NULL;
1857 fsp->brlock_seqnum = brlock_db->get_seqnum(brlock_db);
1859 fsp->brlock_rec = talloc_move(fsp, &br_lock);
1861 return fsp->brlock_rec;
1864 struct brl_revalidate_state {
1865 ssize_t array_size;
1866 uint32 num_pids;
1867 struct server_id *pids;
1871 * Collect PIDs of all processes with pending entries
1874 static void brl_revalidate_collect(struct file_id id, struct server_id pid,
1875 enum brl_type lock_type,
1876 enum brl_flavour lock_flav,
1877 br_off start, br_off size,
1878 void *private_data)
1880 struct brl_revalidate_state *state =
1881 (struct brl_revalidate_state *)private_data;
1883 if (!IS_PENDING_LOCK(lock_type)) {
1884 return;
1887 add_to_large_array(state, sizeof(pid), (void *)&pid,
1888 &state->pids, &state->num_pids,
1889 &state->array_size);
1893 * qsort callback to sort the processes
1896 static int compare_procids(const void *p1, const void *p2)
1898 const struct server_id *i1 = (const struct server_id *)p1;
1899 const struct server_id *i2 = (const struct server_id *)p2;
1901 if (i1->pid < i2->pid) return -1;
1902 if (i2->pid > i2->pid) return 1;
1903 return 0;
1907 * Send a MSG_SMB_UNLOCK message to all processes with pending byte range
1908 * locks so that they retry. Mainly used in the cluster code after a node has
1909 * died.
1911 * Done in two steps to avoid double-sends: First we collect all entries in an
1912 * array, then qsort that array and only send to non-dupes.
1915 static void brl_revalidate(struct messaging_context *msg_ctx,
1916 void *private_data,
1917 uint32_t msg_type,
1918 struct server_id server_id,
1919 DATA_BLOB *data)
1921 struct brl_revalidate_state *state;
1922 uint32 i;
1923 struct server_id last_pid;
1925 if (!(state = talloc_zero(NULL, struct brl_revalidate_state))) {
1926 DEBUG(0, ("talloc failed\n"));
1927 return;
1930 brl_forall(brl_revalidate_collect, state);
1932 if (state->array_size == -1) {
1933 DEBUG(0, ("talloc failed\n"));
1934 goto done;
1937 if (state->num_pids == 0) {
1938 goto done;
1941 TYPESAFE_QSORT(state->pids, state->num_pids, compare_procids);
1943 ZERO_STRUCT(last_pid);
1945 for (i=0; i<state->num_pids; i++) {
1946 if (procid_equal(&last_pid, &state->pids[i])) {
1948 * We've seen that one already
1950 continue;
1953 messaging_send(msg_ctx, state->pids[i], MSG_SMB_UNLOCK,
1954 &data_blob_null);
1955 last_pid = state->pids[i];
1958 done:
1959 TALLOC_FREE(state);
1960 return;
1963 void brl_register_msgs(struct messaging_context *msg_ctx)
1965 messaging_register(msg_ctx, NULL, MSG_SMB_BRL_VALIDATE,
1966 brl_revalidate);