s3:brlock: use NUMERIC_CMP in #ifdef-zeroed lock_compare
[samba.git] / source3 / locking / brlock.c
blob328a9bfba3d410eca2786c99f85597987e9d0056
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 "lib/util/server_id.h"
30 #include "locking/proto.h"
31 #include "smbd/globals.h"
32 #include "dbwrap/dbwrap.h"
33 #include "dbwrap/dbwrap_open.h"
34 #include "serverid.h"
35 #include "messages.h"
36 #include "util_tdb.h"
38 #undef DBGC_CLASS
39 #define DBGC_CLASS DBGC_LOCKING
41 #define ZERO_ZERO 0
43 /* The open brlock.tdb database. */
45 static struct db_context *brlock_db;
47 struct byte_range_lock {
48 struct files_struct *fsp;
49 TALLOC_CTX *req_mem_ctx;
50 const struct GUID *req_guid;
51 unsigned int num_locks;
52 bool modified;
53 struct lock_struct *lock_data;
54 struct db_record *record;
57 /****************************************************************************
58 Debug info at level 10 for lock struct.
59 ****************************************************************************/
61 static void print_lock_struct(unsigned int i, const struct lock_struct *pls)
63 struct server_id_buf tmp;
65 DBG_DEBUG("[%u]: smblctx = %"PRIu64", tid = %"PRIu32", pid = %s, "
66 "start = %"PRIu64", size = %"PRIu64", fnum = %"PRIu64", "
67 "%s %s\n",
69 pls->context.smblctx,
70 pls->context.tid,
71 server_id_str_buf(pls->context.pid, &tmp),
72 pls->start,
73 pls->size,
74 pls->fnum,
75 lock_type_name(pls->lock_type),
76 lock_flav_name(pls->lock_flav));
79 unsigned int brl_num_locks(const struct byte_range_lock *brl)
81 return brl->num_locks;
84 struct files_struct *brl_fsp(struct byte_range_lock *brl)
86 return brl->fsp;
89 TALLOC_CTX *brl_req_mem_ctx(const struct byte_range_lock *brl)
91 if (brl->req_mem_ctx == NULL) {
92 return talloc_get_type_abort(brl, struct byte_range_lock);
95 return brl->req_mem_ctx;
98 const struct GUID *brl_req_guid(const struct byte_range_lock *brl)
100 if (brl->req_guid == NULL) {
101 static const struct GUID brl_zero_req_guid;
102 return &brl_zero_req_guid;
105 return brl->req_guid;
108 /****************************************************************************
109 See if two locking contexts are equal.
110 ****************************************************************************/
112 static bool brl_same_context(const struct lock_context *ctx1,
113 const struct lock_context *ctx2)
115 return (server_id_equal(&ctx1->pid, &ctx2->pid) &&
116 (ctx1->smblctx == ctx2->smblctx) &&
117 (ctx1->tid == ctx2->tid));
120 bool byte_range_valid(uint64_t ofs, uint64_t len)
122 uint64_t max_len = UINT64_MAX - ofs;
123 uint64_t effective_len;
126 * [MS-FSA] specifies this:
128 * If (((FileOffset + Length - 1) < FileOffset) && Length != 0) {
129 * return STATUS_INVALID_LOCK_RANGE
132 * We avoid integer wrapping and calculate
133 * max and effective len instead.
136 if (len == 0) {
137 return true;
140 effective_len = len - 1;
141 if (effective_len <= max_len) {
142 return true;
145 return false;
148 bool byte_range_overlap(uint64_t ofs1,
149 uint64_t len1,
150 uint64_t ofs2,
151 uint64_t len2)
153 uint64_t last1;
154 uint64_t last2;
155 bool valid;
158 * This is based on [MS-FSA] 2.1.4.10
159 * Algorithm for Determining If a Range Access
160 * Conflicts with Byte-Range Locks
164 * The {0, 0} range doesn't conflict with any byte-range lock
166 if (ofs1 == 0 && len1 == 0) {
167 return false;
169 if (ofs2 == 0 && len2 == 0) {
170 return false;
174 * The caller should have checked that the ranges are
175 * valid. But currently we gracefully handle
176 * the overflow of a read/write check.
178 valid = byte_range_valid(ofs1, len1);
179 if (valid) {
180 last1 = ofs1 + len1 - 1;
181 } else {
182 last1 = UINT64_MAX;
184 valid = byte_range_valid(ofs2, len2);
185 if (valid) {
186 last2 = ofs2 + len2 - 1;
187 } else {
188 last2 = UINT64_MAX;
192 * If one range starts after the last
193 * byte of the other range there's
194 * no conflict.
196 if (ofs1 > last2) {
197 return false;
199 if (ofs2 > last1) {
200 return false;
203 return true;
206 /****************************************************************************
207 See if lck1 and lck2 overlap.
208 ****************************************************************************/
210 static bool brl_overlap(const struct lock_struct *lck1,
211 const struct lock_struct *lck2)
213 return byte_range_overlap(lck1->start,
214 lck1->size,
215 lck2->start,
216 lck2->size);
219 /****************************************************************************
220 See if lock2 can be added when lock1 is in place.
221 ****************************************************************************/
223 static bool brl_conflict(const struct lock_struct *lck1,
224 const struct lock_struct *lck2)
226 /* Read locks never conflict. */
227 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
228 return False;
231 /* A READ lock can stack on top of a WRITE lock if they have the same
232 * context & fnum. */
233 if (lck1->lock_type == WRITE_LOCK && lck2->lock_type == READ_LOCK &&
234 brl_same_context(&lck1->context, &lck2->context) &&
235 lck1->fnum == lck2->fnum) {
236 return False;
239 return brl_overlap(lck1, lck2);
242 /****************************************************************************
243 See if lock2 can be added when lock1 is in place - when both locks are POSIX
244 flavour. POSIX locks ignore fnum - they only care about dev/ino which we
245 know already match.
246 ****************************************************************************/
248 static bool brl_conflict_posix(const struct lock_struct *lck1,
249 const struct lock_struct *lck2)
251 #if defined(DEVELOPER)
252 SMB_ASSERT(lck1->lock_flav == POSIX_LOCK);
253 SMB_ASSERT(lck2->lock_flav == POSIX_LOCK);
254 #endif
256 /* Read locks never conflict. */
257 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
258 return False;
261 /* Locks on the same context don't conflict. Ignore fnum. */
262 if (brl_same_context(&lck1->context, &lck2->context)) {
263 return False;
266 /* One is read, the other write, or the context is different,
267 do they overlap ? */
268 return brl_overlap(lck1, lck2);
271 #if ZERO_ZERO
272 static bool brl_conflict1(const struct lock_struct *lck1,
273 const struct lock_struct *lck2)
275 if (lck1->lock_type == READ_LOCK && lck2->lock_type == READ_LOCK) {
276 return False;
279 if (brl_same_context(&lck1->context, &lck2->context) &&
280 lck2->lock_type == READ_LOCK && lck1->fnum == lck2->fnum) {
281 return False;
284 if (lck2->start == 0 && lck2->size == 0 && lck1->size != 0) {
285 return True;
288 if (lck1->start >= (lck2->start + lck2->size) ||
289 lck2->start >= (lck1->start + lck1->size)) {
290 return False;
293 return True;
295 #endif
297 /****************************************************************************
298 Check to see if this lock conflicts, but ignore our own locks on the
299 same fnum only. This is the read/write lock check code path.
300 This is never used in the POSIX lock case.
301 ****************************************************************************/
303 static bool brl_conflict_other(const struct lock_struct *lock,
304 const struct lock_struct *rw_probe)
306 if (lock->lock_type == READ_LOCK && rw_probe->lock_type == READ_LOCK) {
307 return False;
310 if (lock->lock_flav == POSIX_LOCK &&
311 rw_probe->lock_flav == POSIX_LOCK) {
313 * POSIX flavour locks never conflict here - this is only called
314 * in the read/write path.
316 return False;
319 if (!brl_overlap(lock, rw_probe)) {
321 * I/O can only conflict when overlapping a lock, thus let it
322 * pass
324 return false;
327 if (!brl_same_context(&lock->context, &rw_probe->context)) {
329 * Different process, conflict
331 return true;
334 if (lock->fnum != rw_probe->fnum) {
336 * Different file handle, conflict
338 return true;
341 if ((lock->lock_type == READ_LOCK) &&
342 (rw_probe->lock_type == WRITE_LOCK)) {
344 * Incoming WRITE locks conflict with existing READ locks even
345 * if the context is the same. JRA. See LOCKTEST7 in
346 * smbtorture.
348 return true;
352 * I/O request compatible with existing lock, let it pass without
353 * conflict
356 return false;
359 /****************************************************************************
360 Open up the brlock.tdb database.
361 ****************************************************************************/
363 void brl_init(bool read_only)
365 int tdb_flags;
366 char *db_path;
368 if (brlock_db) {
369 return;
372 tdb_flags = SMBD_VOLATILE_TDB_FLAGS | TDB_SEQNUM;
374 db_path = lock_path(talloc_tos(), "brlock.tdb");
375 if (db_path == NULL) {
376 DEBUG(0, ("out of memory!\n"));
377 return;
380 brlock_db = db_open(NULL, db_path,
381 SMBD_VOLATILE_TDB_HASH_SIZE, tdb_flags,
382 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644,
383 DBWRAP_LOCK_ORDER_2, DBWRAP_FLAG_NONE);
384 if (!brlock_db) {
385 DEBUG(0,("Failed to open byte range locking database %s\n",
386 db_path));
387 TALLOC_FREE(db_path);
388 return;
390 TALLOC_FREE(db_path);
393 /****************************************************************************
394 Close down the brlock.tdb database.
395 ****************************************************************************/
397 void brl_shutdown(void)
399 TALLOC_FREE(brlock_db);
402 #if ZERO_ZERO
403 /****************************************************************************
404 Compare two locks for sorting.
405 ****************************************************************************/
407 static int lock_compare(const struct lock_struct *lck1,
408 const struct lock_struct *lck2)
410 if (lck1->start != lck2->start) {
411 return NUMERIC_CMP(lck1->start, lck2->start);
413 return NUMERIC_CMP(lck1->size, lck2->size);
415 #endif
417 /****************************************************************************
418 Lock a range of bytes - Windows lock semantics.
419 ****************************************************************************/
421 NTSTATUS brl_lock_windows_default(struct byte_range_lock *br_lck,
422 struct lock_struct *plock)
424 unsigned int i;
425 files_struct *fsp = br_lck->fsp;
426 struct lock_struct *locks = br_lck->lock_data;
427 NTSTATUS status;
428 bool valid;
430 SMB_ASSERT(plock->lock_type != UNLOCK_LOCK);
432 valid = byte_range_valid(plock->start, plock->size);
433 if (!valid) {
434 return NT_STATUS_INVALID_LOCK_RANGE;
437 for (i=0; i < br_lck->num_locks; i++) {
438 /* Do any Windows or POSIX locks conflict ? */
439 if (brl_conflict(&locks[i], plock)) {
440 if (!serverid_exists(&locks[i].context.pid)) {
441 locks[i].context.pid.pid = 0;
442 br_lck->modified = true;
443 continue;
445 /* Remember who blocked us. */
446 plock->context.smblctx = locks[i].context.smblctx;
447 return NT_STATUS_LOCK_NOT_GRANTED;
449 #if ZERO_ZERO
450 if (plock->start == 0 && plock->size == 0 &&
451 locks[i].size == 0) {
452 break;
454 #endif
457 contend_level2_oplocks_begin(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
459 /* We can get the Windows lock, now see if it needs to
460 be mapped into a lower level POSIX one, and if so can
461 we get it ? */
463 if (lp_posix_locking(fsp->conn->params)) {
464 int errno_ret;
465 if (!set_posix_lock_windows_flavour(fsp,
466 plock->start,
467 plock->size,
468 plock->lock_type,
469 &plock->context,
470 locks,
471 br_lck->num_locks,
472 &errno_ret)) {
474 /* We don't know who blocked us. */
475 plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
477 if (errno_ret == EACCES || errno_ret == EAGAIN) {
478 status = NT_STATUS_LOCK_NOT_GRANTED;
479 goto fail;
480 } else {
481 status = map_nt_error_from_unix(errno);
482 goto fail;
487 /* no conflicts - add it to the list of locks */
488 locks = talloc_realloc(br_lck, locks, struct lock_struct,
489 (br_lck->num_locks + 1));
490 if (!locks) {
491 status = NT_STATUS_NO_MEMORY;
492 goto fail;
495 memcpy(&locks[br_lck->num_locks], plock, sizeof(struct lock_struct));
496 br_lck->num_locks += 1;
497 br_lck->lock_data = locks;
498 br_lck->modified = True;
500 return NT_STATUS_OK;
501 fail:
502 contend_level2_oplocks_end(fsp, LEVEL2_CONTEND_WINDOWS_BRL);
503 return status;
506 /****************************************************************************
507 Cope with POSIX range splits and merges.
508 ****************************************************************************/
510 static unsigned int brlock_posix_split_merge(struct lock_struct *lck_arr, /* Output array. */
511 struct lock_struct *ex, /* existing lock. */
512 struct lock_struct *plock) /* proposed lock. */
514 bool lock_types_differ = (ex->lock_type != plock->lock_type);
516 /* We can't merge non-conflicting locks on different context - ignore fnum. */
518 if (!brl_same_context(&ex->context, &plock->context)) {
519 /* Just copy. */
520 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
521 return 1;
524 /* We now know we have the same context. */
526 /* Did we overlap ? */
528 /*********************************************
529 +---------+
530 | ex |
531 +---------+
532 +-------+
533 | plock |
534 +-------+
535 OR....
536 +---------+
537 | ex |
538 +---------+
539 **********************************************/
541 if ( (ex->start > (plock->start + plock->size)) ||
542 (plock->start > (ex->start + ex->size))) {
544 /* No overlap with this lock - copy existing. */
546 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
547 return 1;
550 /*********************************************
551 +---------------------------+
552 | ex |
553 +---------------------------+
554 +---------------------------+
555 | plock | -> replace with plock.
556 +---------------------------+
558 +---------------+
559 | ex |
560 +---------------+
561 +---------------------------+
562 | plock | -> replace with plock.
563 +---------------------------+
565 **********************************************/
567 if ( (ex->start >= plock->start) &&
568 (ex->start + ex->size <= plock->start + plock->size) ) {
570 /* Replace - discard existing lock. */
572 return 0;
575 /*********************************************
576 Adjacent after.
577 +-------+
578 | ex |
579 +-------+
580 +---------------+
581 | plock |
582 +---------------+
584 BECOMES....
585 +---------------+-------+
586 | plock | ex | - different lock types.
587 +---------------+-------+
588 OR.... (merge)
589 +-----------------------+
590 | plock | - same lock type.
591 +-----------------------+
592 **********************************************/
594 if (plock->start + plock->size == ex->start) {
596 /* If the lock types are the same, we merge, if different, we
597 add the remainder of the old lock. */
599 if (lock_types_differ) {
600 /* Add existing. */
601 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
602 return 1;
603 } else {
604 /* Merge - adjust incoming lock as we may have more
605 * merging to come. */
606 plock->size += ex->size;
607 return 0;
611 /*********************************************
612 Adjacent before.
613 +-------+
614 | ex |
615 +-------+
616 +---------------+
617 | plock |
618 +---------------+
619 BECOMES....
620 +-------+---------------+
621 | ex | plock | - different lock types
622 +-------+---------------+
624 OR.... (merge)
625 +-----------------------+
626 | plock | - same lock type.
627 +-----------------------+
629 **********************************************/
631 if (ex->start + ex->size == plock->start) {
633 /* If the lock types are the same, we merge, if different, we
634 add the existing lock. */
636 if (lock_types_differ) {
637 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
638 return 1;
639 } else {
640 /* Merge - adjust incoming lock as we may have more
641 * merging to come. */
642 plock->start = ex->start;
643 plock->size += ex->size;
644 return 0;
648 /*********************************************
649 Overlap after.
650 +-----------------------+
651 | ex |
652 +-----------------------+
653 +---------------+
654 | plock |
655 +---------------+
657 +----------------+
658 | ex |
659 +----------------+
660 +---------------+
661 | plock |
662 +---------------+
664 BECOMES....
665 +---------------+-------+
666 | plock | ex | - different lock types.
667 +---------------+-------+
668 OR.... (merge)
669 +-----------------------+
670 | plock | - same lock type.
671 +-----------------------+
672 **********************************************/
674 if ( (ex->start >= plock->start) &&
675 (ex->start <= plock->start + plock->size) &&
676 (ex->start + ex->size > plock->start + plock->size) ) {
678 /* If the lock types are the same, we merge, if different, we
679 add the remainder of the old lock. */
681 if (lock_types_differ) {
682 /* Add remaining existing. */
683 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
684 /* Adjust existing start and size. */
685 lck_arr[0].start = plock->start + plock->size;
686 lck_arr[0].size = (ex->start + ex->size) - (plock->start + plock->size);
687 return 1;
688 } else {
689 /* Merge - adjust incoming lock as we may have more
690 * merging to come. */
691 plock->size += (ex->start + ex->size) - (plock->start + plock->size);
692 return 0;
696 /*********************************************
697 Overlap before.
698 +-----------------------+
699 | ex |
700 +-----------------------+
701 +---------------+
702 | plock |
703 +---------------+
705 +-------------+
706 | ex |
707 +-------------+
708 +---------------+
709 | plock |
710 +---------------+
712 BECOMES....
713 +-------+---------------+
714 | ex | plock | - different lock types
715 +-------+---------------+
717 OR.... (merge)
718 +-----------------------+
719 | plock | - same lock type.
720 +-----------------------+
722 **********************************************/
724 if ( (ex->start < plock->start) &&
725 (ex->start + ex->size >= plock->start) &&
726 (ex->start + ex->size <= plock->start + plock->size) ) {
728 /* If the lock types are the same, we merge, if different, we
729 add the truncated old lock. */
731 if (lock_types_differ) {
732 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
733 /* Adjust existing size. */
734 lck_arr[0].size = plock->start - ex->start;
735 return 1;
736 } else {
737 /* Merge - adjust incoming lock as we may have more
738 * merging to come. MUST ADJUST plock SIZE FIRST ! */
739 plock->size += (plock->start - ex->start);
740 plock->start = ex->start;
741 return 0;
745 /*********************************************
746 Complete overlap.
747 +---------------------------+
748 | ex |
749 +---------------------------+
750 +---------+
751 | plock |
752 +---------+
753 BECOMES.....
754 +-------+---------+---------+
755 | ex | plock | ex | - different lock types.
756 +-------+---------+---------+
758 +---------------------------+
759 | plock | - same lock type.
760 +---------------------------+
761 **********************************************/
763 if ( (ex->start < plock->start) && (ex->start + ex->size > plock->start + plock->size) ) {
765 if (lock_types_differ) {
767 /* We have to split ex into two locks here. */
769 memcpy(&lck_arr[0], ex, sizeof(struct lock_struct));
770 memcpy(&lck_arr[1], ex, sizeof(struct lock_struct));
772 /* Adjust first existing size. */
773 lck_arr[0].size = plock->start - ex->start;
775 /* Adjust second existing start and size. */
776 lck_arr[1].start = plock->start + plock->size;
777 lck_arr[1].size = (ex->start + ex->size) - (plock->start + plock->size);
778 return 2;
779 } else {
780 /* Just eat the existing locks, merge them into plock. */
781 plock->start = ex->start;
782 plock->size = ex->size;
783 return 0;
787 /* Never get here. */
788 smb_panic("brlock_posix_split_merge");
789 /* Notreached. */
791 /* Keep some compilers happy. */
792 return 0;
795 /****************************************************************************
796 Lock a range of bytes - POSIX lock semantics.
797 We must cope with range splits and merges.
798 ****************************************************************************/
800 static NTSTATUS brl_lock_posix(struct byte_range_lock *br_lck,
801 struct lock_struct *plock)
803 unsigned int i, count, posix_count;
804 struct lock_struct *locks = br_lck->lock_data;
805 struct lock_struct *tp;
806 bool break_oplocks = false;
807 NTSTATUS status;
809 /* No zero-zero locks for POSIX. */
810 if (plock->start == 0 && plock->size == 0) {
811 return NT_STATUS_INVALID_PARAMETER;
814 /* Don't allow 64-bit lock wrap. */
815 if (plock->start + plock->size - 1 < plock->start) {
816 return NT_STATUS_INVALID_PARAMETER;
819 /* The worst case scenario here is we have to split an
820 existing POSIX lock range into two, and add our lock,
821 so we need at most 2 more entries. */
823 tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 2);
824 if (!tp) {
825 return NT_STATUS_NO_MEMORY;
828 count = posix_count = 0;
830 for (i=0; i < br_lck->num_locks; i++) {
831 struct lock_struct *curr_lock = &locks[i];
833 if (curr_lock->lock_flav == WINDOWS_LOCK) {
834 /* Do any Windows flavour locks conflict ? */
835 if (brl_conflict(curr_lock, plock)) {
836 if (!serverid_exists(&curr_lock->context.pid)) {
837 curr_lock->context.pid.pid = 0;
838 br_lck->modified = true;
839 continue;
841 /* No games with error messages. */
842 TALLOC_FREE(tp);
843 /* Remember who blocked us. */
844 plock->context.smblctx = curr_lock->context.smblctx;
845 return NT_STATUS_LOCK_NOT_GRANTED;
847 /* Just copy the Windows lock into the new array. */
848 memcpy(&tp[count], curr_lock, sizeof(struct lock_struct));
849 count++;
850 } else {
851 unsigned int tmp_count = 0;
853 /* POSIX conflict semantics are different. */
854 if (brl_conflict_posix(curr_lock, plock)) {
855 if (!serverid_exists(&curr_lock->context.pid)) {
856 curr_lock->context.pid.pid = 0;
857 br_lck->modified = true;
858 continue;
860 /* Can't block ourselves with POSIX locks. */
861 /* No games with error messages. */
862 TALLOC_FREE(tp);
863 /* Remember who blocked us. */
864 plock->context.smblctx = curr_lock->context.smblctx;
865 return NT_STATUS_LOCK_NOT_GRANTED;
868 /* Work out overlaps. */
869 tmp_count += brlock_posix_split_merge(&tp[count], curr_lock, plock);
870 posix_count += tmp_count;
871 count += tmp_count;
876 * Break oplocks while we hold a brl. Since lock() and unlock() calls
877 * are not symmetric with POSIX semantics, we cannot guarantee our
878 * contend_level2_oplocks_begin/end calls will be acquired and
879 * released one-for-one as with Windows semantics. Therefore we only
880 * call contend_level2_oplocks_begin if this is the first POSIX brl on
881 * the file.
883 break_oplocks = (posix_count == 0);
884 if (break_oplocks) {
885 contend_level2_oplocks_begin(br_lck->fsp,
886 LEVEL2_CONTEND_POSIX_BRL);
889 /* Try and add the lock in order, sorted by lock start. */
890 for (i=0; i < count; i++) {
891 struct lock_struct *curr_lock = &tp[i];
893 if (curr_lock->start <= plock->start) {
894 continue;
898 if (i < count) {
899 memmove(&tp[i+1], &tp[i],
900 (count - i)*sizeof(struct lock_struct));
902 memcpy(&tp[i], plock, sizeof(struct lock_struct));
903 count++;
905 /* We can get the POSIX lock, now see if it needs to
906 be mapped into a lower level POSIX one, and if so can
907 we get it ? */
909 if (lp_posix_locking(br_lck->fsp->conn->params)) {
910 int errno_ret;
912 /* The lower layer just needs to attempt to
913 get the system POSIX lock. We've weeded out
914 any conflicts above. */
916 if (!set_posix_lock_posix_flavour(br_lck->fsp,
917 plock->start,
918 plock->size,
919 plock->lock_type,
920 &plock->context,
921 &errno_ret)) {
923 /* We don't know who blocked us. */
924 plock->context.smblctx = 0xFFFFFFFFFFFFFFFFLL;
926 if (errno_ret == EACCES || errno_ret == EAGAIN) {
927 TALLOC_FREE(tp);
928 status = NT_STATUS_LOCK_NOT_GRANTED;
929 goto fail;
930 } else {
931 TALLOC_FREE(tp);
932 status = map_nt_error_from_unix(errno);
933 goto fail;
938 /* If we didn't use all the allocated size,
939 * Realloc so we don't leak entries per lock call. */
940 if (count < br_lck->num_locks + 2) {
941 tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
942 if (!tp) {
943 status = NT_STATUS_NO_MEMORY;
944 goto fail;
948 br_lck->num_locks = count;
949 TALLOC_FREE(br_lck->lock_data);
950 br_lck->lock_data = tp;
951 locks = tp;
952 br_lck->modified = True;
954 /* A successful downgrade from write to read lock can trigger a lock
955 re-evalutation where waiting readers can now proceed. */
957 return NT_STATUS_OK;
958 fail:
959 if (break_oplocks) {
960 contend_level2_oplocks_end(br_lck->fsp,
961 LEVEL2_CONTEND_POSIX_BRL);
963 return status;
966 /****************************************************************************
967 Lock a range of bytes.
968 ****************************************************************************/
970 NTSTATUS brl_lock(
971 struct byte_range_lock *br_lck,
972 uint64_t smblctx,
973 struct server_id pid,
974 br_off start,
975 br_off size,
976 enum brl_type lock_type,
977 enum brl_flavour lock_flav,
978 struct server_id *blocker_pid,
979 uint64_t *psmblctx)
981 NTSTATUS ret;
982 struct lock_struct lock;
984 ZERO_STRUCT(lock);
986 #if !ZERO_ZERO
987 if (start == 0 && size == 0) {
988 DEBUG(0,("client sent 0/0 lock - please report this\n"));
990 #endif
992 lock = (struct lock_struct) {
993 .context.smblctx = smblctx,
994 .context.pid = pid,
995 .context.tid = br_lck->fsp->conn->cnum,
996 .start = start,
997 .size = size,
998 .fnum = br_lck->fsp->fnum,
999 .lock_type = lock_type,
1000 .lock_flav = lock_flav
1003 if (lock_flav == WINDOWS_LOCK) {
1004 ret = SMB_VFS_BRL_LOCK_WINDOWS(
1005 br_lck->fsp->conn, br_lck, &lock);
1006 } else {
1007 ret = brl_lock_posix(br_lck, &lock);
1010 #if ZERO_ZERO
1011 /* sort the lock list */
1012 TYPESAFE_QSORT(br_lck->lock_data, (size_t)br_lck->num_locks, lock_compare);
1013 #endif
1014 /* If we're returning an error, return who blocked us. */
1015 if (!NT_STATUS_IS_OK(ret) && psmblctx) {
1016 *blocker_pid = lock.context.pid;
1017 *psmblctx = lock.context.smblctx;
1019 return ret;
1022 /****************************************************************************
1023 Unlock a range of bytes - Windows semantics.
1024 ****************************************************************************/
1026 bool brl_unlock_windows_default(struct byte_range_lock *br_lck,
1027 const struct lock_struct *plock)
1029 unsigned int i;
1030 struct lock_struct *locks = br_lck->lock_data;
1031 enum brl_type deleted_lock_type = READ_LOCK; /* shut the compiler up.... */
1033 SMB_ASSERT(plock->lock_type == UNLOCK_LOCK);
1035 #if ZERO_ZERO
1036 /* Delete write locks by preference... The lock list
1037 is sorted in the zero zero case. */
1039 for (i = 0; i < br_lck->num_locks; i++) {
1040 struct lock_struct *lock = &locks[i];
1042 if (lock->lock_type == WRITE_LOCK &&
1043 brl_same_context(&lock->context, &plock->context) &&
1044 lock->fnum == plock->fnum &&
1045 lock->lock_flav == WINDOWS_LOCK &&
1046 lock->start == plock->start &&
1047 lock->size == plock->size) {
1049 /* found it - delete it */
1050 deleted_lock_type = lock->lock_type;
1051 break;
1055 if (i != br_lck->num_locks) {
1056 /* We found it - don't search again. */
1057 goto unlock_continue;
1059 #endif
1061 for (i = 0; i < br_lck->num_locks; i++) {
1062 struct lock_struct *lock = &locks[i];
1064 /* Only remove our own locks that match in start, size, and flavour. */
1065 if (brl_same_context(&lock->context, &plock->context) &&
1066 lock->fnum == plock->fnum &&
1067 lock->lock_flav == WINDOWS_LOCK &&
1068 lock->start == plock->start &&
1069 lock->size == plock->size ) {
1070 deleted_lock_type = lock->lock_type;
1071 break;
1075 if (i == br_lck->num_locks) {
1076 /* we didn't find it */
1077 return False;
1080 #if ZERO_ZERO
1081 unlock_continue:
1082 #endif
1084 ARRAY_DEL_ELEMENT(locks, i, br_lck->num_locks);
1085 br_lck->num_locks -= 1;
1086 br_lck->modified = True;
1088 /* Unlock the underlying POSIX regions. */
1089 if(lp_posix_locking(br_lck->fsp->conn->params)) {
1090 release_posix_lock_windows_flavour(br_lck->fsp,
1091 plock->start,
1092 plock->size,
1093 deleted_lock_type,
1094 &plock->context,
1095 locks,
1096 br_lck->num_locks);
1099 contend_level2_oplocks_end(br_lck->fsp, LEVEL2_CONTEND_WINDOWS_BRL);
1100 return True;
1103 /****************************************************************************
1104 Unlock a range of bytes - POSIX semantics.
1105 ****************************************************************************/
1107 static bool brl_unlock_posix(struct byte_range_lock *br_lck,
1108 struct lock_struct *plock)
1110 unsigned int i, count;
1111 struct lock_struct *tp;
1112 struct lock_struct *locks = br_lck->lock_data;
1113 bool overlap_found = False;
1115 /* No zero-zero locks for POSIX. */
1116 if (plock->start == 0 && plock->size == 0) {
1117 return False;
1120 /* Don't allow 64-bit lock wrap. */
1121 if (plock->start + plock->size < plock->start ||
1122 plock->start + plock->size < plock->size) {
1123 DEBUG(10,("brl_unlock_posix: lock wrap\n"));
1124 return False;
1127 /* The worst case scenario here is we have to split an
1128 existing POSIX lock range into two, so we need at most
1129 1 more entry. */
1131 tp = talloc_array(br_lck, struct lock_struct, br_lck->num_locks + 1);
1132 if (!tp) {
1133 DEBUG(10,("brl_unlock_posix: malloc fail\n"));
1134 return False;
1137 count = 0;
1138 for (i = 0; i < br_lck->num_locks; i++) {
1139 struct lock_struct *lock = &locks[i];
1140 unsigned int tmp_count;
1142 /* Only remove our own locks - ignore fnum. */
1143 if (!brl_same_context(&lock->context, &plock->context)) {
1144 memcpy(&tp[count], lock, sizeof(struct lock_struct));
1145 count++;
1146 continue;
1149 if (lock->lock_flav == WINDOWS_LOCK) {
1150 /* Do any Windows flavour locks conflict ? */
1151 if (brl_conflict(lock, plock)) {
1152 TALLOC_FREE(tp);
1153 return false;
1155 /* Just copy the Windows lock into the new array. */
1156 memcpy(&tp[count], lock, sizeof(struct lock_struct));
1157 count++;
1158 continue;
1161 /* Work out overlaps. */
1162 tmp_count = brlock_posix_split_merge(&tp[count], lock, plock);
1164 if (tmp_count == 0) {
1165 /* plock overlapped the existing lock completely,
1166 or replaced it. Don't copy the existing lock. */
1167 overlap_found = true;
1168 } else if (tmp_count == 1) {
1169 /* Either no overlap, (simple copy of existing lock) or
1170 * an overlap of an existing lock. */
1171 /* If the lock changed size, we had an overlap. */
1172 if (tp[count].size != lock->size) {
1173 overlap_found = true;
1175 count += tmp_count;
1176 } else if (tmp_count == 2) {
1177 /* We split a lock range in two. */
1178 overlap_found = true;
1179 count += tmp_count;
1181 /* Optimisation... */
1182 /* We know we're finished here as we can't overlap any
1183 more POSIX locks. Copy the rest of the lock array. */
1185 if (i < br_lck->num_locks - 1) {
1186 memcpy(&tp[count], &locks[i+1],
1187 sizeof(*locks)*((br_lck->num_locks-1) - i));
1188 count += ((br_lck->num_locks-1) - i);
1190 break;
1195 if (!overlap_found) {
1196 /* Just ignore - no change. */
1197 TALLOC_FREE(tp);
1198 DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
1199 return True;
1202 /* Unlock any POSIX regions. */
1203 if(lp_posix_locking(br_lck->fsp->conn->params)) {
1204 release_posix_lock_posix_flavour(br_lck->fsp,
1205 plock->start,
1206 plock->size,
1207 &plock->context,
1209 count);
1212 /* Realloc so we don't leak entries per unlock call. */
1213 if (count) {
1214 tp = talloc_realloc(br_lck, tp, struct lock_struct, count);
1215 if (!tp) {
1216 DEBUG(10,("brl_unlock_posix: realloc fail\n"));
1217 return False;
1219 } else {
1220 /* We deleted the last lock. */
1221 TALLOC_FREE(tp);
1222 tp = NULL;
1225 contend_level2_oplocks_end(br_lck->fsp,
1226 LEVEL2_CONTEND_POSIX_BRL);
1228 br_lck->num_locks = count;
1229 TALLOC_FREE(br_lck->lock_data);
1230 locks = tp;
1231 br_lck->lock_data = tp;
1232 br_lck->modified = True;
1234 return True;
1237 /****************************************************************************
1238 Unlock a range of bytes.
1239 ****************************************************************************/
1241 bool brl_unlock(struct byte_range_lock *br_lck,
1242 uint64_t smblctx,
1243 struct server_id pid,
1244 br_off start,
1245 br_off size,
1246 enum brl_flavour lock_flav)
1248 struct lock_struct lock;
1250 lock.context.smblctx = smblctx;
1251 lock.context.pid = pid;
1252 lock.context.tid = br_lck->fsp->conn->cnum;
1253 lock.start = start;
1254 lock.size = size;
1255 lock.fnum = br_lck->fsp->fnum;
1256 lock.lock_type = UNLOCK_LOCK;
1257 lock.lock_flav = lock_flav;
1259 if (lock_flav == WINDOWS_LOCK) {
1260 return SMB_VFS_BRL_UNLOCK_WINDOWS(
1261 br_lck->fsp->conn, br_lck, &lock);
1262 } else {
1263 return brl_unlock_posix(br_lck, &lock);
1267 /****************************************************************************
1268 Test if we could add a lock if we wanted to.
1269 Returns True if the region required is currently unlocked, False if locked.
1270 ****************************************************************************/
1272 bool brl_locktest(struct byte_range_lock *br_lck,
1273 const struct lock_struct *rw_probe)
1275 bool ret = True;
1276 unsigned int i;
1277 struct lock_struct *locks = br_lck->lock_data;
1278 files_struct *fsp = br_lck->fsp;
1280 /* Make sure existing locks don't conflict */
1281 for (i=0; i < br_lck->num_locks; i++) {
1283 * Our own locks don't conflict.
1285 if (brl_conflict_other(&locks[i], rw_probe)) {
1286 if (br_lck->record == NULL) {
1287 /* readonly */
1288 return false;
1291 if (!serverid_exists(&locks[i].context.pid)) {
1292 locks[i].context.pid.pid = 0;
1293 br_lck->modified = true;
1294 continue;
1297 return False;
1302 * There is no lock held by an SMB daemon, check to
1303 * see if there is a POSIX lock from a UNIX or NFS process.
1304 * This only conflicts with Windows locks, not POSIX locks.
1307 if(lp_posix_locking(fsp->conn->params) &&
1308 (rw_probe->lock_flav == WINDOWS_LOCK)) {
1310 * Make copies -- is_posix_locked might modify the values
1313 br_off start = rw_probe->start;
1314 br_off size = rw_probe->size;
1315 enum brl_type lock_type = rw_probe->lock_type;
1317 ret = is_posix_locked(fsp, &start, &size, &lock_type, WINDOWS_LOCK);
1319 DEBUG(10, ("brl_locktest: posix start=%ju len=%ju %s for %s "
1320 "file %s\n", (uintmax_t)start, (uintmax_t)size,
1321 ret ? "locked" : "unlocked",
1322 fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
1324 /* We need to return the inverse of is_posix_locked. */
1325 ret = !ret;
1328 /* no conflicts - we could have added it */
1329 return ret;
1332 /****************************************************************************
1333 Query for existing locks.
1334 ****************************************************************************/
1336 NTSTATUS brl_lockquery(struct byte_range_lock *br_lck,
1337 uint64_t *psmblctx,
1338 struct server_id pid,
1339 br_off *pstart,
1340 br_off *psize,
1341 enum brl_type *plock_type,
1342 enum brl_flavour lock_flav)
1344 unsigned int i;
1345 struct lock_struct lock;
1346 const struct lock_struct *locks = br_lck->lock_data;
1347 files_struct *fsp = br_lck->fsp;
1349 lock.context.smblctx = *psmblctx;
1350 lock.context.pid = pid;
1351 lock.context.tid = br_lck->fsp->conn->cnum;
1352 lock.start = *pstart;
1353 lock.size = *psize;
1354 lock.fnum = fsp->fnum;
1355 lock.lock_type = *plock_type;
1356 lock.lock_flav = lock_flav;
1358 /* Make sure existing locks don't conflict */
1359 for (i=0; i < br_lck->num_locks; i++) {
1360 const struct lock_struct *exlock = &locks[i];
1361 bool conflict = False;
1363 if (exlock->lock_flav == WINDOWS_LOCK) {
1364 conflict = brl_conflict(exlock, &lock);
1365 } else {
1366 conflict = brl_conflict_posix(exlock, &lock);
1369 if (conflict) {
1370 *psmblctx = exlock->context.smblctx;
1371 *pstart = exlock->start;
1372 *psize = exlock->size;
1373 *plock_type = exlock->lock_type;
1374 return NT_STATUS_LOCK_NOT_GRANTED;
1379 * There is no lock held by an SMB daemon, check to
1380 * see if there is a POSIX lock from a UNIX or NFS process.
1383 if(lp_posix_locking(fsp->conn->params)) {
1384 bool ret = is_posix_locked(fsp, pstart, psize, plock_type, POSIX_LOCK);
1386 DEBUG(10, ("brl_lockquery: posix start=%ju len=%ju %s for %s "
1387 "file %s\n", (uintmax_t)*pstart,
1388 (uintmax_t)*psize, ret ? "locked" : "unlocked",
1389 fsp_fnum_dbg(fsp), fsp_str_dbg(fsp)));
1391 if (ret) {
1392 /* Hmmm. No clue what to set smblctx to - use -1. */
1393 *psmblctx = 0xFFFFFFFFFFFFFFFFLL;
1394 return NT_STATUS_LOCK_NOT_GRANTED;
1398 return NT_STATUS_OK;
1402 /****************************************************************************
1403 Remove any locks associated with a open file.
1404 We return True if this process owns any other Windows locks on this
1405 fd and so we should not immediately close the fd.
1406 ****************************************************************************/
1408 void brl_close_fnum(struct byte_range_lock *br_lck)
1410 files_struct *fsp = br_lck->fsp;
1411 uint32_t tid = fsp->conn->cnum;
1412 uint64_t fnum = fsp->fnum;
1413 unsigned int i;
1414 struct lock_struct *locks = br_lck->lock_data;
1415 struct server_id pid = messaging_server_id(fsp->conn->sconn->msg_ctx);
1416 struct lock_struct *locks_copy;
1417 unsigned int num_locks_copy;
1419 /* Copy the current lock array. */
1420 if (br_lck->num_locks) {
1421 locks_copy = (struct lock_struct *)talloc_memdup(br_lck, locks, br_lck->num_locks * sizeof(struct lock_struct));
1422 if (!locks_copy) {
1423 smb_panic("brl_close_fnum: talloc failed");
1425 } else {
1426 locks_copy = NULL;
1429 num_locks_copy = br_lck->num_locks;
1431 for (i=0; i < num_locks_copy; i++) {
1432 struct lock_struct *lock = &locks_copy[i];
1434 if (lock->context.tid == tid &&
1435 server_id_equal(&lock->context.pid, &pid) &&
1436 (lock->fnum == fnum)) {
1437 brl_unlock(
1438 br_lck,
1439 lock->context.smblctx,
1440 pid,
1441 lock->start,
1442 lock->size,
1443 lock->lock_flav);
1448 bool brl_mark_disconnected(struct files_struct *fsp)
1450 uint32_t tid = fsp->conn->cnum;
1451 uint64_t smblctx;
1452 uint64_t fnum = fsp->fnum;
1453 unsigned int i;
1454 struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
1455 struct byte_range_lock *br_lck = NULL;
1457 if (fsp->op == NULL) {
1458 return false;
1461 smblctx = fsp->op->global->open_persistent_id;
1463 if (!fsp->op->global->durable) {
1464 return false;
1467 if (fsp->current_lock_count == 0) {
1468 return true;
1471 br_lck = brl_get_locks(talloc_tos(), fsp);
1472 if (br_lck == NULL) {
1473 return false;
1476 for (i=0; i < br_lck->num_locks; i++) {
1477 struct lock_struct *lock = &br_lck->lock_data[i];
1480 * as this is a durable handle, we only expect locks
1481 * of the current file handle!
1484 if (lock->context.smblctx != smblctx) {
1485 TALLOC_FREE(br_lck);
1486 return false;
1489 if (lock->context.tid != tid) {
1490 TALLOC_FREE(br_lck);
1491 return false;
1494 if (!server_id_equal(&lock->context.pid, &self)) {
1495 TALLOC_FREE(br_lck);
1496 return false;
1499 if (lock->fnum != fnum) {
1500 TALLOC_FREE(br_lck);
1501 return false;
1504 server_id_set_disconnected(&lock->context.pid);
1505 lock->context.tid = TID_FIELD_INVALID;
1506 lock->fnum = FNUM_FIELD_INVALID;
1509 br_lck->modified = true;
1510 TALLOC_FREE(br_lck);
1511 return true;
1514 bool brl_reconnect_disconnected(struct files_struct *fsp)
1516 uint32_t tid = fsp->conn->cnum;
1517 uint64_t smblctx;
1518 uint64_t fnum = fsp->fnum;
1519 unsigned int i;
1520 struct server_id self = messaging_server_id(fsp->conn->sconn->msg_ctx);
1521 struct byte_range_lock *br_lck = NULL;
1523 if (fsp->op == NULL) {
1524 return false;
1527 smblctx = fsp->op->global->open_persistent_id;
1529 if (!fsp->op->global->durable) {
1530 return false;
1534 * When reconnecting, we do not want to validate the brlock entries
1535 * and thereby remove our own (disconnected) entries but reactivate
1536 * them instead.
1539 br_lck = brl_get_locks(talloc_tos(), fsp);
1540 if (br_lck == NULL) {
1541 return false;
1544 if (br_lck->num_locks == 0) {
1545 TALLOC_FREE(br_lck);
1546 return true;
1549 for (i=0; i < br_lck->num_locks; i++) {
1550 struct lock_struct *lock = &br_lck->lock_data[i];
1553 * as this is a durable handle we only expect locks
1554 * of the current file handle!
1557 if (lock->context.smblctx != smblctx) {
1558 TALLOC_FREE(br_lck);
1559 return false;
1562 if (lock->context.tid != TID_FIELD_INVALID) {
1563 TALLOC_FREE(br_lck);
1564 return false;
1567 if (!server_id_is_disconnected(&lock->context.pid)) {
1568 TALLOC_FREE(br_lck);
1569 return false;
1572 if (lock->fnum != FNUM_FIELD_INVALID) {
1573 TALLOC_FREE(br_lck);
1574 return false;
1577 lock->context.pid = self;
1578 lock->context.tid = tid;
1579 lock->fnum = fnum;
1582 fsp->current_lock_count = br_lck->num_locks;
1583 br_lck->modified = true;
1584 TALLOC_FREE(br_lck);
1585 return true;
1588 struct brl_forall_cb {
1589 void (*fn)(struct file_id id, struct server_id pid,
1590 enum brl_type lock_type,
1591 enum brl_flavour lock_flav,
1592 br_off start, br_off size,
1593 void *private_data);
1594 void *private_data;
1597 /****************************************************************************
1598 Traverse the whole database with this function, calling traverse_callback
1599 on each lock.
1600 ****************************************************************************/
1602 static int brl_traverse_fn(struct db_record *rec, void *state)
1604 struct brl_forall_cb *cb = (struct brl_forall_cb *)state;
1605 struct lock_struct *locks;
1606 struct file_id *key;
1607 unsigned int i;
1608 unsigned int num_locks = 0;
1609 TDB_DATA dbkey;
1610 TDB_DATA value;
1612 dbkey = dbwrap_record_get_key(rec);
1613 value = dbwrap_record_get_value(rec);
1615 /* In a traverse function we must make a copy of
1616 dbuf before modifying it. */
1618 locks = (struct lock_struct *)talloc_memdup(
1619 talloc_tos(), value.dptr, value.dsize);
1620 if (!locks) {
1621 return -1; /* Terminate traversal. */
1624 key = (struct file_id *)dbkey.dptr;
1625 num_locks = value.dsize/sizeof(*locks);
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 TALLOC_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;
1655 NTSTATUS status;
1656 int count = 0;
1658 if (!brlock_db) {
1659 return 0;
1661 cb.fn = fn;
1662 cb.private_data = private_data;
1663 status = dbwrap_traverse(brlock_db, brl_traverse_fn, &cb, &count);
1665 if (!NT_STATUS_IS_OK(status)) {
1666 return -1;
1667 } else {
1668 return count;
1672 /*******************************************************************
1673 Store a potentially modified set of byte range lock data back into
1674 the database.
1675 Unlock the record.
1676 ********************************************************************/
1678 static void byte_range_lock_flush(struct byte_range_lock *br_lck)
1680 unsigned i;
1681 struct lock_struct *locks = br_lck->lock_data;
1683 if (!br_lck->modified) {
1684 DEBUG(10, ("br_lck not modified\n"));
1685 goto done;
1688 i = 0;
1690 while (i < br_lck->num_locks) {
1691 if (locks[i].context.pid.pid == 0) {
1693 * Autocleanup, the process conflicted and does not
1694 * exist anymore.
1696 locks[i] = locks[br_lck->num_locks-1];
1697 br_lck->num_locks -= 1;
1698 } else {
1699 i += 1;
1703 if (br_lck->num_locks == 0) {
1704 /* No locks - delete this entry. */
1705 NTSTATUS status = dbwrap_record_delete(br_lck->record);
1706 if (!NT_STATUS_IS_OK(status)) {
1707 DEBUG(0, ("delete_rec returned %s\n",
1708 nt_errstr(status)));
1709 smb_panic("Could not delete byte range lock entry");
1711 } else {
1712 TDB_DATA data = {
1713 .dsize = br_lck->num_locks * sizeof(struct lock_struct),
1714 .dptr = (uint8_t *)br_lck->lock_data,
1716 NTSTATUS status;
1718 status = dbwrap_record_store(br_lck->record, data, TDB_REPLACE);
1719 if (!NT_STATUS_IS_OK(status)) {
1720 DEBUG(0, ("store returned %s\n", nt_errstr(status)));
1721 smb_panic("Could not store byte range mode entry");
1725 DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db)));
1727 done:
1728 br_lck->modified = false;
1729 TALLOC_FREE(br_lck->record);
1732 static int byte_range_lock_destructor(struct byte_range_lock *br_lck)
1734 byte_range_lock_flush(br_lck);
1735 return 0;
1738 static bool brl_parse_data(struct byte_range_lock *br_lck, TDB_DATA data)
1740 size_t data_len;
1742 if (data.dsize == 0) {
1743 return true;
1745 if (data.dsize % sizeof(struct lock_struct) != 0) {
1746 DEBUG(1, ("Invalid data size: %u\n", (unsigned)data.dsize));
1747 return false;
1750 br_lck->num_locks = data.dsize / sizeof(struct lock_struct);
1751 data_len = br_lck->num_locks * sizeof(struct lock_struct);
1753 br_lck->lock_data = talloc_memdup(br_lck, data.dptr, data_len);
1754 if (br_lck->lock_data == NULL) {
1755 DEBUG(1, ("talloc_memdup failed\n"));
1756 return false;
1758 return true;
1761 /*******************************************************************
1762 Fetch a set of byte range lock data from the database.
1763 Leave the record locked.
1764 TALLOC_FREE(brl) will release the lock in the destructor.
1765 ********************************************************************/
1767 struct byte_range_lock *brl_get_locks(TALLOC_CTX *mem_ctx, files_struct *fsp)
1769 TDB_DATA key, data;
1770 struct byte_range_lock *br_lck;
1772 br_lck = talloc_zero(mem_ctx, struct byte_range_lock);
1773 if (br_lck == NULL) {
1774 return NULL;
1777 br_lck->fsp = fsp;
1779 key.dptr = (uint8_t *)&fsp->file_id;
1780 key.dsize = sizeof(struct file_id);
1782 br_lck->record = dbwrap_fetch_locked(brlock_db, br_lck, key);
1784 if (br_lck->record == NULL) {
1785 DEBUG(3, ("Could not lock byte range lock entry\n"));
1786 TALLOC_FREE(br_lck);
1787 return NULL;
1790 data = dbwrap_record_get_value(br_lck->record);
1792 if (!brl_parse_data(br_lck, data)) {
1793 TALLOC_FREE(br_lck);
1794 return NULL;
1797 talloc_set_destructor(br_lck, byte_range_lock_destructor);
1799 if (DEBUGLEVEL >= 10) {
1800 unsigned int i;
1801 struct file_id_buf buf;
1802 struct lock_struct *locks = br_lck->lock_data;
1803 DBG_DEBUG("%u current locks on file_id %s\n",
1804 br_lck->num_locks,
1805 file_id_str_buf(fsp->file_id, &buf));
1806 for( i = 0; i < br_lck->num_locks; i++) {
1807 print_lock_struct(i, &locks[i]);
1811 return br_lck;
1814 struct byte_range_lock *brl_get_locks_for_locking(TALLOC_CTX *mem_ctx,
1815 files_struct *fsp,
1816 TALLOC_CTX *req_mem_ctx,
1817 const struct GUID *req_guid)
1819 struct byte_range_lock *br_lck = NULL;
1821 br_lck = brl_get_locks(mem_ctx, fsp);
1822 if (br_lck == NULL) {
1823 return NULL;
1825 SMB_ASSERT(req_mem_ctx != NULL);
1826 br_lck->req_mem_ctx = req_mem_ctx;
1827 SMB_ASSERT(req_guid != NULL);
1828 br_lck->req_guid = req_guid;
1830 return br_lck;
1833 struct brl_get_locks_readonly_state {
1834 TALLOC_CTX *mem_ctx;
1835 struct byte_range_lock **br_lock;
1838 static void brl_get_locks_readonly_parser(TDB_DATA key, TDB_DATA data,
1839 void *private_data)
1841 struct brl_get_locks_readonly_state *state =
1842 (struct brl_get_locks_readonly_state *)private_data;
1843 struct byte_range_lock *br_lck;
1845 br_lck = talloc_pooled_object(
1846 state->mem_ctx, struct byte_range_lock, 1, data.dsize);
1847 if (br_lck == NULL) {
1848 *state->br_lock = NULL;
1849 return;
1851 *br_lck = (struct byte_range_lock) { 0 };
1852 if (!brl_parse_data(br_lck, data)) {
1853 *state->br_lock = NULL;
1854 return;
1856 *state->br_lock = br_lck;
1859 struct byte_range_lock *brl_get_locks_readonly(files_struct *fsp)
1861 struct byte_range_lock *br_lock = NULL;
1862 struct brl_get_locks_readonly_state state;
1863 NTSTATUS status;
1865 DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
1866 dbwrap_get_seqnum(brlock_db), fsp->brlock_seqnum));
1868 if ((fsp->brlock_rec != NULL)
1869 && (dbwrap_get_seqnum(brlock_db) == fsp->brlock_seqnum)) {
1871 * We have cached the brlock_rec and the database did not
1872 * change.
1874 return fsp->brlock_rec;
1878 * Parse the record fresh from the database
1881 state.mem_ctx = fsp;
1882 state.br_lock = &br_lock;
1884 status = dbwrap_parse_record(
1885 brlock_db,
1886 make_tdb_data((uint8_t *)&fsp->file_id,
1887 sizeof(fsp->file_id)),
1888 brl_get_locks_readonly_parser, &state);
1890 if (NT_STATUS_EQUAL(status,NT_STATUS_NOT_FOUND)) {
1892 * No locks on this file. Return an empty br_lock.
1894 br_lock = talloc_zero(fsp, struct byte_range_lock);
1895 if (br_lock == NULL) {
1896 return NULL;
1899 } else if (!NT_STATUS_IS_OK(status)) {
1900 DEBUG(3, ("Could not parse byte range lock record: "
1901 "%s\n", nt_errstr(status)));
1902 return NULL;
1904 if (br_lock == NULL) {
1905 return NULL;
1908 br_lock->fsp = fsp;
1909 br_lock->modified = false;
1910 br_lock->record = NULL;
1913 * Cache the brlock struct, invalidated when the dbwrap_seqnum
1914 * changes. See beginning of this routine.
1916 TALLOC_FREE(fsp->brlock_rec);
1917 fsp->brlock_rec = br_lock;
1918 fsp->brlock_seqnum = dbwrap_get_seqnum(brlock_db);
1920 return br_lock;
1923 bool brl_cleanup_disconnected(struct file_id fid, uint64_t open_persistent_id)
1925 bool ret = false;
1926 TALLOC_CTX *frame = talloc_stackframe();
1927 TDB_DATA key, val;
1928 struct db_record *rec;
1929 struct lock_struct *lock;
1930 unsigned n, num;
1931 struct file_id_buf buf;
1932 NTSTATUS status;
1934 key = make_tdb_data((void*)&fid, sizeof(fid));
1936 rec = dbwrap_fetch_locked(brlock_db, frame, key);
1937 if (rec == NULL) {
1938 DBG_INFO("failed to fetch record for file %s\n",
1939 file_id_str_buf(fid, &buf));
1940 goto done;
1943 val = dbwrap_record_get_value(rec);
1944 lock = (struct lock_struct*)val.dptr;
1945 num = val.dsize / sizeof(struct lock_struct);
1946 if (lock == NULL) {
1947 DBG_DEBUG("no byte range locks for file %s\n",
1948 file_id_str_buf(fid, &buf));
1949 ret = true;
1950 goto done;
1953 for (n=0; n<num; n++) {
1954 struct lock_context *ctx = &lock[n].context;
1956 if (!server_id_is_disconnected(&ctx->pid)) {
1957 struct server_id_buf tmp;
1958 DBG_INFO("byte range lock "
1959 "%s used by server %s, do not cleanup\n",
1960 file_id_str_buf(fid, &buf),
1961 server_id_str_buf(ctx->pid, &tmp));
1962 goto done;
1965 if (ctx->smblctx != open_persistent_id) {
1966 DBG_INFO("byte range lock %s expected smblctx %"PRIu64" "
1967 "but found %"PRIu64", do not cleanup\n",
1968 file_id_str_buf(fid, &buf),
1969 open_persistent_id,
1970 ctx->smblctx);
1971 goto done;
1975 status = dbwrap_record_delete(rec);
1976 if (!NT_STATUS_IS_OK(status)) {
1977 DBG_INFO("failed to delete record "
1978 "for file %s from %s, open %"PRIu64": %s\n",
1979 file_id_str_buf(fid, &buf),
1980 dbwrap_name(brlock_db),
1981 open_persistent_id,
1982 nt_errstr(status));
1983 goto done;
1986 DBG_DEBUG("file %s cleaned up %u entries from open %"PRIu64"\n",
1987 file_id_str_buf(fid, &buf),
1988 num,
1989 open_persistent_id);
1991 ret = true;
1992 done:
1993 talloc_free(frame);
1994 return ret;