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 */
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"
38 #define DBGC_CLASS DBGC_LOCKING
42 /* The open brlock.tdb database. */
44 static struct db_context
*brlock_db
;
46 struct byte_range_lock
{
47 struct files_struct
*fsp
;
48 unsigned int num_locks
;
50 bool have_read_oplocks
;
51 struct lock_struct
*lock_data
;
52 struct db_record
*record
;
55 /****************************************************************************
56 Debug info at level 10 for lock struct.
57 ****************************************************************************/
59 static void print_lock_struct(unsigned int i
, const struct lock_struct
*pls
)
61 DEBUG(10,("[%u]: smblctx = %llu, tid = %u, pid = %s, ",
63 (unsigned long long)pls
->context
.smblctx
,
64 (unsigned int)pls
->context
.tid
,
65 server_id_str(talloc_tos(), &pls
->context
.pid
) ));
67 DEBUG(10, ("start = %ju, size = %ju, fnum = %ju, %s %s\n",
68 (uintmax_t)pls
->start
,
71 lock_type_name(pls
->lock_type
),
72 lock_flav_name(pls
->lock_flav
)));
75 unsigned int brl_num_locks(const struct byte_range_lock
*brl
)
77 return brl
->num_locks
;
80 struct files_struct
*brl_fsp(struct byte_range_lock
*brl
)
85 bool brl_have_read_oplocks(const struct byte_range_lock
*brl
)
87 return brl
->have_read_oplocks
;
90 void brl_set_have_read_oplocks(struct byte_range_lock
*brl
,
91 bool have_read_oplocks
)
93 DEBUG(10, ("Setting have_read_oplocks to %s\n",
94 have_read_oplocks
? "true" : "false"));
95 SMB_ASSERT(brl
->record
!= NULL
); /* otherwise we're readonly */
96 brl
->have_read_oplocks
= have_read_oplocks
;
100 /****************************************************************************
101 See if two locking contexts are equal.
102 ****************************************************************************/
104 static bool brl_same_context(const struct lock_context
*ctx1
,
105 const struct lock_context
*ctx2
)
107 return (serverid_equal(&ctx1
->pid
, &ctx2
->pid
) &&
108 (ctx1
->smblctx
== ctx2
->smblctx
) &&
109 (ctx1
->tid
== ctx2
->tid
));
112 /****************************************************************************
113 See if lck1 and lck2 overlap.
114 ****************************************************************************/
116 static bool brl_overlap(const struct lock_struct
*lck1
,
117 const struct lock_struct
*lck2
)
119 /* XXX Remove for Win7 compatibility. */
120 /* this extra check is not redundant - it copes with locks
121 that go beyond the end of 64 bit file space */
122 if (lck1
->size
!= 0 &&
123 lck1
->start
== lck2
->start
&&
124 lck1
->size
== lck2
->size
) {
128 if (lck1
->start
>= (lck2
->start
+lck2
->size
) ||
129 lck2
->start
>= (lck1
->start
+lck1
->size
)) {
135 /****************************************************************************
136 See if lock2 can be added when lock1 is in place.
137 ****************************************************************************/
139 static bool brl_conflict(const struct lock_struct
*lck1
,
140 const struct lock_struct
*lck2
)
142 /* Ignore PENDING locks. */
143 if (IS_PENDING_LOCK(lck1
->lock_type
) || IS_PENDING_LOCK(lck2
->lock_type
))
146 /* Read locks never conflict. */
147 if (lck1
->lock_type
== READ_LOCK
&& lck2
->lock_type
== READ_LOCK
) {
151 /* A READ lock can stack on top of a WRITE lock if they have the same
153 if (lck1
->lock_type
== WRITE_LOCK
&& lck2
->lock_type
== READ_LOCK
&&
154 brl_same_context(&lck1
->context
, &lck2
->context
) &&
155 lck1
->fnum
== lck2
->fnum
) {
159 return brl_overlap(lck1
, lck2
);
162 /****************************************************************************
163 See if lock2 can be added when lock1 is in place - when both locks are POSIX
164 flavour. POSIX locks ignore fnum - they only care about dev/ino which we
166 ****************************************************************************/
168 static bool brl_conflict_posix(const struct lock_struct
*lck1
,
169 const struct lock_struct
*lck2
)
171 #if defined(DEVELOPER)
172 SMB_ASSERT(lck1
->lock_flav
== POSIX_LOCK
);
173 SMB_ASSERT(lck2
->lock_flav
== POSIX_LOCK
);
176 /* Ignore PENDING locks. */
177 if (IS_PENDING_LOCK(lck1
->lock_type
) || IS_PENDING_LOCK(lck2
->lock_type
))
180 /* Read locks never conflict. */
181 if (lck1
->lock_type
== READ_LOCK
&& lck2
->lock_type
== READ_LOCK
) {
185 /* Locks on the same context don't conflict. Ignore fnum. */
186 if (brl_same_context(&lck1
->context
, &lck2
->context
)) {
190 /* One is read, the other write, or the context is different,
192 return brl_overlap(lck1
, lck2
);
196 static bool brl_conflict1(const struct lock_struct
*lck1
,
197 const struct lock_struct
*lck2
)
199 if (IS_PENDING_LOCK(lck1
->lock_type
) || IS_PENDING_LOCK(lck2
->lock_type
))
202 if (lck1
->lock_type
== READ_LOCK
&& lck2
->lock_type
== READ_LOCK
) {
206 if (brl_same_context(&lck1
->context
, &lck2
->context
) &&
207 lck2
->lock_type
== READ_LOCK
&& lck1
->fnum
== lck2
->fnum
) {
211 if (lck2
->start
== 0 && lck2
->size
== 0 && lck1
->size
!= 0) {
215 if (lck1
->start
>= (lck2
->start
+ lck2
->size
) ||
216 lck2
->start
>= (lck1
->start
+ lck1
->size
)) {
224 /****************************************************************************
225 Check to see if this lock conflicts, but ignore our own locks on the
226 same fnum only. This is the read/write lock check code path.
227 This is never used in the POSIX lock case.
228 ****************************************************************************/
230 static bool brl_conflict_other(const struct lock_struct
*lock
,
231 const struct lock_struct
*rw_probe
)
233 if (IS_PENDING_LOCK(lock
->lock_type
) ||
234 IS_PENDING_LOCK(rw_probe
->lock_type
)) {
238 if (lock
->lock_type
== READ_LOCK
&& rw_probe
->lock_type
== READ_LOCK
) {
242 if (lock
->lock_flav
== POSIX_LOCK
&&
243 rw_probe
->lock_flav
== POSIX_LOCK
) {
245 * POSIX flavour locks never conflict here - this is only called
246 * in the read/write path.
251 if (!brl_overlap(lock
, rw_probe
)) {
253 * I/O can only conflict when overlapping a lock, thus let it
259 if (!brl_same_context(&lock
->context
, &rw_probe
->context
)) {
261 * Different process, conflict
266 if (lock
->fnum
!= rw_probe
->fnum
) {
268 * Different file handle, conflict
273 if ((lock
->lock_type
== READ_LOCK
) &&
274 (rw_probe
->lock_type
== WRITE_LOCK
)) {
276 * Incoming WRITE locks conflict with existing READ locks even
277 * if the context is the same. JRA. See LOCKTEST7 in
284 * I/O request compatible with existing lock, let it pass without
291 /****************************************************************************
292 Check if an unlock overlaps a pending lock.
293 ****************************************************************************/
295 static bool brl_pending_overlap(const struct lock_struct
*lock
, const struct lock_struct
*pend_lock
)
297 if ((lock
->start
<= pend_lock
->start
) && (lock
->start
+ lock
->size
> pend_lock
->start
))
299 if ((lock
->start
>= pend_lock
->start
) && (lock
->start
< pend_lock
->start
+ pend_lock
->size
))
304 /****************************************************************************
305 Amazingly enough, w2k3 "remembers" whether the last lock failure on a fnum
306 is the same as this one and changes its error code. I wonder if any
307 app depends on this ?
308 ****************************************************************************/
310 static NTSTATUS
brl_lock_failed(files_struct
*fsp
,
311 const struct lock_struct
*lock
,
314 if (lock
->start
>= 0xEF000000 && (lock
->start
>> 63) == 0) {
315 /* amazing the little things you learn with a test
316 suite. Locks beyond this offset (as a 64 bit
317 number!) always generate the conflict error code,
318 unless the top bit is set */
319 if (!blocking_lock
) {
320 fsp
->last_lock_failure
= *lock
;
322 return NT_STATUS_FILE_LOCK_CONFLICT
;
325 if (serverid_equal(&lock
->context
.pid
, &fsp
->last_lock_failure
.context
.pid
) &&
326 lock
->context
.tid
== fsp
->last_lock_failure
.context
.tid
&&
327 lock
->fnum
== fsp
->last_lock_failure
.fnum
&&
328 lock
->start
== fsp
->last_lock_failure
.start
) {
329 return NT_STATUS_FILE_LOCK_CONFLICT
;
332 if (!blocking_lock
) {
333 fsp
->last_lock_failure
= *lock
;
335 return NT_STATUS_LOCK_NOT_GRANTED
;
338 /****************************************************************************
339 Open up the brlock.tdb database.
340 ****************************************************************************/
342 void brl_init(bool read_only
)
350 tdb_flags
= TDB_DEFAULT
|TDB_VOLATILE
|TDB_CLEAR_IF_FIRST
|TDB_INCOMPATIBLE_HASH
;
352 if (!lp_clustering()) {
354 * We can't use the SEQNUM trick to cache brlock
355 * entries in the clustering case because ctdb seqnum
356 * propagation has a delay.
358 tdb_flags
|= TDB_SEQNUM
;
361 brlock_db
= db_open(NULL
, lock_path("brlock.tdb"),
362 SMB_OPEN_DATABASE_TDB_HASH_SIZE
, tdb_flags
,
363 read_only
?O_RDONLY
:(O_RDWR
|O_CREAT
), 0644,
364 DBWRAP_LOCK_ORDER_2
, DBWRAP_FLAG_NONE
);
366 DEBUG(0,("Failed to open byte range locking database %s\n",
367 lock_path("brlock.tdb")));
372 /****************************************************************************
373 Close down the brlock.tdb database.
374 ****************************************************************************/
376 void brl_shutdown(void)
378 TALLOC_FREE(brlock_db
);
382 /****************************************************************************
383 Compare two locks for sorting.
384 ****************************************************************************/
386 static int lock_compare(const struct lock_struct
*lck1
,
387 const struct lock_struct
*lck2
)
389 if (lck1
->start
!= lck2
->start
) {
390 return (lck1
->start
- lck2
->start
);
392 if (lck2
->size
!= lck1
->size
) {
393 return ((int)lck1
->size
- (int)lck2
->size
);
399 /****************************************************************************
400 Lock a range of bytes - Windows lock semantics.
401 ****************************************************************************/
403 NTSTATUS
brl_lock_windows_default(struct byte_range_lock
*br_lck
,
404 struct lock_struct
*plock
, bool blocking_lock
)
407 files_struct
*fsp
= br_lck
->fsp
;
408 struct lock_struct
*locks
= br_lck
->lock_data
;
411 SMB_ASSERT(plock
->lock_type
!= UNLOCK_LOCK
);
413 if ((plock
->start
+ plock
->size
- 1 < plock
->start
) &&
415 return NT_STATUS_INVALID_LOCK_RANGE
;
418 for (i
=0; i
< br_lck
->num_locks
; i
++) {
419 /* Do any Windows or POSIX locks conflict ? */
420 if (brl_conflict(&locks
[i
], plock
)) {
421 /* Remember who blocked us. */
422 plock
->context
.smblctx
= locks
[i
].context
.smblctx
;
423 return brl_lock_failed(fsp
,plock
,blocking_lock
);
426 if (plock
->start
== 0 && plock
->size
== 0 &&
427 locks
[i
].size
== 0) {
433 if (!IS_PENDING_LOCK(plock
->lock_type
)) {
434 contend_level2_oplocks_begin(fsp
, LEVEL2_CONTEND_WINDOWS_BRL
);
437 /* We can get the Windows lock, now see if it needs to
438 be mapped into a lower level POSIX one, and if so can
441 if (!IS_PENDING_LOCK(plock
->lock_type
) && lp_posix_locking(fsp
->conn
->params
)) {
443 if (!set_posix_lock_windows_flavour(fsp
,
452 /* We don't know who blocked us. */
453 plock
->context
.smblctx
= 0xFFFFFFFFFFFFFFFFLL
;
455 if (errno_ret
== EACCES
|| errno_ret
== EAGAIN
) {
456 status
= NT_STATUS_FILE_LOCK_CONFLICT
;
459 status
= map_nt_error_from_unix(errno
);
465 /* no conflicts - add it to the list of locks */
466 locks
= talloc_realloc(br_lck
, locks
, struct lock_struct
,
467 (br_lck
->num_locks
+ 1));
469 status
= NT_STATUS_NO_MEMORY
;
473 memcpy(&locks
[br_lck
->num_locks
], plock
, sizeof(struct lock_struct
));
474 br_lck
->num_locks
+= 1;
475 br_lck
->lock_data
= locks
;
476 br_lck
->modified
= True
;
480 if (!IS_PENDING_LOCK(plock
->lock_type
)) {
481 contend_level2_oplocks_end(fsp
, LEVEL2_CONTEND_WINDOWS_BRL
);
486 /****************************************************************************
487 Cope with POSIX range splits and merges.
488 ****************************************************************************/
490 static unsigned int brlock_posix_split_merge(struct lock_struct
*lck_arr
, /* Output array. */
491 struct lock_struct
*ex
, /* existing lock. */
492 struct lock_struct
*plock
) /* proposed lock. */
494 bool lock_types_differ
= (ex
->lock_type
!= plock
->lock_type
);
496 /* We can't merge non-conflicting locks on different context - ignore fnum. */
498 if (!brl_same_context(&ex
->context
, &plock
->context
)) {
500 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
504 /* We now know we have the same context. */
506 /* Did we overlap ? */
508 /*********************************************
519 **********************************************/
521 if ( (ex
->start
> (plock
->start
+ plock
->size
)) ||
522 (plock
->start
> (ex
->start
+ ex
->size
))) {
524 /* No overlap with this lock - copy existing. */
526 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
530 /*********************************************
531 +---------------------------+
533 +---------------------------+
534 +---------------------------+
535 | plock | -> replace with plock.
536 +---------------------------+
541 +---------------------------+
542 | plock | -> replace with plock.
543 +---------------------------+
545 **********************************************/
547 if ( (ex
->start
>= plock
->start
) &&
548 (ex
->start
+ ex
->size
<= plock
->start
+ plock
->size
) ) {
550 /* Replace - discard existing lock. */
555 /*********************************************
565 +---------------+-------+
566 | plock | ex | - different lock types.
567 +---------------+-------+
569 +-----------------------+
570 | plock | - same lock type.
571 +-----------------------+
572 **********************************************/
574 if (plock
->start
+ plock
->size
== ex
->start
) {
576 /* If the lock types are the same, we merge, if different, we
577 add the remainder of the old lock. */
579 if (lock_types_differ
) {
581 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
584 /* Merge - adjust incoming lock as we may have more
585 * merging to come. */
586 plock
->size
+= ex
->size
;
591 /*********************************************
600 +-------+---------------+
601 | ex | plock | - different lock types
602 +-------+---------------+
605 +-----------------------+
606 | plock | - same lock type.
607 +-----------------------+
609 **********************************************/
611 if (ex
->start
+ ex
->size
== plock
->start
) {
613 /* If the lock types are the same, we merge, if different, we
614 add the existing lock. */
616 if (lock_types_differ
) {
617 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
620 /* Merge - adjust incoming lock as we may have more
621 * merging to come. */
622 plock
->start
= ex
->start
;
623 plock
->size
+= ex
->size
;
628 /*********************************************
630 +-----------------------+
632 +-----------------------+
645 +---------------+-------+
646 | plock | ex | - different lock types.
647 +---------------+-------+
649 +-----------------------+
650 | plock | - same lock type.
651 +-----------------------+
652 **********************************************/
654 if ( (ex
->start
>= plock
->start
) &&
655 (ex
->start
<= plock
->start
+ plock
->size
) &&
656 (ex
->start
+ ex
->size
> plock
->start
+ plock
->size
) ) {
658 /* If the lock types are the same, we merge, if different, we
659 add the remainder of the old lock. */
661 if (lock_types_differ
) {
662 /* Add remaining existing. */
663 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
664 /* Adjust existing start and size. */
665 lck_arr
[0].start
= plock
->start
+ plock
->size
;
666 lck_arr
[0].size
= (ex
->start
+ ex
->size
) - (plock
->start
+ plock
->size
);
669 /* Merge - adjust incoming lock as we may have more
670 * merging to come. */
671 plock
->size
+= (ex
->start
+ ex
->size
) - (plock
->start
+ plock
->size
);
676 /*********************************************
678 +-----------------------+
680 +-----------------------+
693 +-------+---------------+
694 | ex | plock | - different lock types
695 +-------+---------------+
698 +-----------------------+
699 | plock | - same lock type.
700 +-----------------------+
702 **********************************************/
704 if ( (ex
->start
< plock
->start
) &&
705 (ex
->start
+ ex
->size
>= plock
->start
) &&
706 (ex
->start
+ ex
->size
<= plock
->start
+ plock
->size
) ) {
708 /* If the lock types are the same, we merge, if different, we
709 add the truncated old lock. */
711 if (lock_types_differ
) {
712 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
713 /* Adjust existing size. */
714 lck_arr
[0].size
= plock
->start
- ex
->start
;
717 /* Merge - adjust incoming lock as we may have more
718 * merging to come. MUST ADJUST plock SIZE FIRST ! */
719 plock
->size
+= (plock
->start
- ex
->start
);
720 plock
->start
= ex
->start
;
725 /*********************************************
727 +---------------------------+
729 +---------------------------+
734 +-------+---------+---------+
735 | ex | plock | ex | - different lock types.
736 +-------+---------+---------+
738 +---------------------------+
739 | plock | - same lock type.
740 +---------------------------+
741 **********************************************/
743 if ( (ex
->start
< plock
->start
) && (ex
->start
+ ex
->size
> plock
->start
+ plock
->size
) ) {
745 if (lock_types_differ
) {
747 /* We have to split ex into two locks here. */
749 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
750 memcpy(&lck_arr
[1], ex
, sizeof(struct lock_struct
));
752 /* Adjust first existing size. */
753 lck_arr
[0].size
= plock
->start
- ex
->start
;
755 /* Adjust second existing start and size. */
756 lck_arr
[1].start
= plock
->start
+ plock
->size
;
757 lck_arr
[1].size
= (ex
->start
+ ex
->size
) - (plock
->start
+ plock
->size
);
760 /* Just eat the existing locks, merge them into plock. */
761 plock
->start
= ex
->start
;
762 plock
->size
= ex
->size
;
767 /* Never get here. */
768 smb_panic("brlock_posix_split_merge");
771 /* Keep some compilers happy. */
775 /****************************************************************************
776 Lock a range of bytes - POSIX lock semantics.
777 We must cope with range splits and merges.
778 ****************************************************************************/
780 static NTSTATUS
brl_lock_posix(struct messaging_context
*msg_ctx
,
781 struct byte_range_lock
*br_lck
,
782 struct lock_struct
*plock
)
784 unsigned int i
, count
, posix_count
;
785 struct lock_struct
*locks
= br_lck
->lock_data
;
786 struct lock_struct
*tp
;
787 bool signal_pending_read
= False
;
788 bool break_oplocks
= false;
791 /* No zero-zero locks for POSIX. */
792 if (plock
->start
== 0 && plock
->size
== 0) {
793 return NT_STATUS_INVALID_PARAMETER
;
796 /* Don't allow 64-bit lock wrap. */
797 if (plock
->start
+ plock
->size
- 1 < plock
->start
) {
798 return NT_STATUS_INVALID_PARAMETER
;
801 /* The worst case scenario here is we have to split an
802 existing POSIX lock range into two, and add our lock,
803 so we need at most 2 more entries. */
805 tp
= talloc_array(br_lck
, struct lock_struct
, br_lck
->num_locks
+ 2);
807 return NT_STATUS_NO_MEMORY
;
810 count
= posix_count
= 0;
812 for (i
=0; i
< br_lck
->num_locks
; i
++) {
813 struct lock_struct
*curr_lock
= &locks
[i
];
815 /* If we have a pending read lock, a lock downgrade should
816 trigger a lock re-evaluation. */
817 if (curr_lock
->lock_type
== PENDING_READ_LOCK
&&
818 brl_pending_overlap(plock
, curr_lock
)) {
819 signal_pending_read
= True
;
822 if (curr_lock
->lock_flav
== WINDOWS_LOCK
) {
823 /* Do any Windows flavour locks conflict ? */
824 if (brl_conflict(curr_lock
, plock
)) {
825 /* No games with error messages. */
827 /* Remember who blocked us. */
828 plock
->context
.smblctx
= curr_lock
->context
.smblctx
;
829 return NT_STATUS_FILE_LOCK_CONFLICT
;
831 /* Just copy the Windows lock into the new array. */
832 memcpy(&tp
[count
], curr_lock
, sizeof(struct lock_struct
));
835 unsigned int tmp_count
= 0;
837 /* POSIX conflict semantics are different. */
838 if (brl_conflict_posix(curr_lock
, plock
)) {
839 /* Can't block ourselves with POSIX locks. */
840 /* No games with error messages. */
842 /* Remember who blocked us. */
843 plock
->context
.smblctx
= curr_lock
->context
.smblctx
;
844 return NT_STATUS_FILE_LOCK_CONFLICT
;
847 /* Work out overlaps. */
848 tmp_count
+= brlock_posix_split_merge(&tp
[count
], curr_lock
, plock
);
849 posix_count
+= tmp_count
;
855 * Break oplocks while we hold a brl. Since lock() and unlock() calls
856 * are not symetric with POSIX semantics, we cannot guarantee our
857 * contend_level2_oplocks_begin/end calls will be acquired and
858 * released one-for-one as with Windows semantics. Therefore we only
859 * call contend_level2_oplocks_begin if this is the first POSIX brl on
862 break_oplocks
= (!IS_PENDING_LOCK(plock
->lock_type
) &&
865 contend_level2_oplocks_begin(br_lck
->fsp
,
866 LEVEL2_CONTEND_POSIX_BRL
);
869 /* Try and add the lock in order, sorted by lock start. */
870 for (i
=0; i
< count
; i
++) {
871 struct lock_struct
*curr_lock
= &tp
[i
];
873 if (curr_lock
->start
<= plock
->start
) {
879 memmove(&tp
[i
+1], &tp
[i
],
880 (count
- i
)*sizeof(struct lock_struct
));
882 memcpy(&tp
[i
], plock
, sizeof(struct lock_struct
));
885 /* We can get the POSIX lock, now see if it needs to
886 be mapped into a lower level POSIX one, and if so can
889 if (!IS_PENDING_LOCK(plock
->lock_type
) && lp_posix_locking(br_lck
->fsp
->conn
->params
)) {
892 /* The lower layer just needs to attempt to
893 get the system POSIX lock. We've weeded out
894 any conflicts above. */
896 if (!set_posix_lock_posix_flavour(br_lck
->fsp
,
902 /* We don't know who blocked us. */
903 plock
->context
.smblctx
= 0xFFFFFFFFFFFFFFFFLL
;
905 if (errno_ret
== EACCES
|| errno_ret
== EAGAIN
) {
907 status
= NT_STATUS_FILE_LOCK_CONFLICT
;
911 status
= map_nt_error_from_unix(errno
);
917 /* If we didn't use all the allocated size,
918 * Realloc so we don't leak entries per lock call. */
919 if (count
< br_lck
->num_locks
+ 2) {
920 tp
= talloc_realloc(br_lck
, tp
, struct lock_struct
, count
);
922 status
= NT_STATUS_NO_MEMORY
;
927 br_lck
->num_locks
= count
;
928 TALLOC_FREE(br_lck
->lock_data
);
929 br_lck
->lock_data
= tp
;
931 br_lck
->modified
= True
;
933 /* A successful downgrade from write to read lock can trigger a lock
934 re-evalutation where waiting readers can now proceed. */
936 if (signal_pending_read
) {
937 /* Send unlock messages to any pending read waiters that overlap. */
938 for (i
=0; i
< br_lck
->num_locks
; i
++) {
939 struct lock_struct
*pend_lock
= &locks
[i
];
941 /* Ignore non-pending locks. */
942 if (!IS_PENDING_LOCK(pend_lock
->lock_type
)) {
946 if (pend_lock
->lock_type
== PENDING_READ_LOCK
&&
947 brl_pending_overlap(plock
, pend_lock
)) {
948 DEBUG(10,("brl_lock_posix: sending unlock message to pid %s\n",
949 procid_str_static(&pend_lock
->context
.pid
)));
951 messaging_send(msg_ctx
, pend_lock
->context
.pid
,
952 MSG_SMB_UNLOCK
, &data_blob_null
);
960 contend_level2_oplocks_end(br_lck
->fsp
,
961 LEVEL2_CONTEND_POSIX_BRL
);
966 NTSTATUS
smb_vfs_call_brl_lock_windows(struct vfs_handle_struct
*handle
,
967 struct byte_range_lock
*br_lck
,
968 struct lock_struct
*plock
,
971 VFS_FIND(brl_lock_windows
);
972 return handle
->fns
->brl_lock_windows_fn(handle
, br_lck
, plock
,
976 /****************************************************************************
977 Lock a range of bytes.
978 ****************************************************************************/
980 NTSTATUS
brl_lock(struct messaging_context
*msg_ctx
,
981 struct byte_range_lock
*br_lck
,
983 struct server_id pid
,
986 enum brl_type lock_type
,
987 enum brl_flavour lock_flav
,
992 struct lock_struct lock
;
995 if (start
== 0 && size
== 0) {
996 DEBUG(0,("client sent 0/0 lock - please report this\n"));
1000 lock
= (struct lock_struct
) {
1001 .context
.smblctx
= smblctx
,
1003 .context
.tid
= br_lck
->fsp
->conn
->cnum
,
1006 .fnum
= br_lck
->fsp
->fnum
,
1007 .lock_type
= lock_type
,
1008 .lock_flav
= lock_flav
1011 if (lock_flav
== WINDOWS_LOCK
) {
1012 ret
= SMB_VFS_BRL_LOCK_WINDOWS(br_lck
->fsp
->conn
, br_lck
,
1013 &lock
, blocking_lock
);
1015 ret
= brl_lock_posix(msg_ctx
, br_lck
, &lock
);
1019 /* sort the lock list */
1020 TYPESAFE_QSORT(br_lck
->lock_data
, (size_t)br_lck
->num_locks
, lock_compare
);
1023 /* If we're returning an error, return who blocked us. */
1024 if (!NT_STATUS_IS_OK(ret
) && psmblctx
) {
1025 *psmblctx
= lock
.context
.smblctx
;
1030 static void brl_delete_lock_struct(struct lock_struct
*locks
,
1034 if (del_idx
>= num_locks
) {
1037 memmove(&locks
[del_idx
], &locks
[del_idx
+1],
1038 sizeof(*locks
) * (num_locks
- del_idx
- 1));
1041 /****************************************************************************
1042 Unlock a range of bytes - Windows semantics.
1043 ****************************************************************************/
1045 bool brl_unlock_windows_default(struct messaging_context
*msg_ctx
,
1046 struct byte_range_lock
*br_lck
,
1047 const struct lock_struct
*plock
)
1050 struct lock_struct
*locks
= br_lck
->lock_data
;
1051 enum brl_type deleted_lock_type
= READ_LOCK
; /* shut the compiler up.... */
1053 SMB_ASSERT(plock
->lock_type
== UNLOCK_LOCK
);
1056 /* Delete write locks by preference... The lock list
1057 is sorted in the zero zero case. */
1059 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1060 struct lock_struct
*lock
= &locks
[i
];
1062 if (lock
->lock_type
== WRITE_LOCK
&&
1063 brl_same_context(&lock
->context
, &plock
->context
) &&
1064 lock
->fnum
== plock
->fnum
&&
1065 lock
->lock_flav
== WINDOWS_LOCK
&&
1066 lock
->start
== plock
->start
&&
1067 lock
->size
== plock
->size
) {
1069 /* found it - delete it */
1070 deleted_lock_type
= lock
->lock_type
;
1075 if (i
!= br_lck
->num_locks
) {
1076 /* We found it - don't search again. */
1077 goto unlock_continue
;
1081 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1082 struct lock_struct
*lock
= &locks
[i
];
1084 if (IS_PENDING_LOCK(lock
->lock_type
)) {
1088 /* Only remove our own locks that match in start, size, and flavour. */
1089 if (brl_same_context(&lock
->context
, &plock
->context
) &&
1090 lock
->fnum
== plock
->fnum
&&
1091 lock
->lock_flav
== WINDOWS_LOCK
&&
1092 lock
->start
== plock
->start
&&
1093 lock
->size
== plock
->size
) {
1094 deleted_lock_type
= lock
->lock_type
;
1099 if (i
== br_lck
->num_locks
) {
1100 /* we didn't find it */
1108 brl_delete_lock_struct(locks
, br_lck
->num_locks
, i
);
1109 br_lck
->num_locks
-= 1;
1110 br_lck
->modified
= True
;
1112 /* Unlock the underlying POSIX regions. */
1113 if(lp_posix_locking(br_lck
->fsp
->conn
->params
)) {
1114 release_posix_lock_windows_flavour(br_lck
->fsp
,
1123 /* Send unlock messages to any pending waiters that overlap. */
1124 for (j
=0; j
< br_lck
->num_locks
; j
++) {
1125 struct lock_struct
*pend_lock
= &locks
[j
];
1127 /* Ignore non-pending locks. */
1128 if (!IS_PENDING_LOCK(pend_lock
->lock_type
)) {
1132 /* We could send specific lock info here... */
1133 if (brl_pending_overlap(plock
, pend_lock
)) {
1134 DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1135 procid_str_static(&pend_lock
->context
.pid
)));
1137 messaging_send(msg_ctx
, pend_lock
->context
.pid
,
1138 MSG_SMB_UNLOCK
, &data_blob_null
);
1142 contend_level2_oplocks_end(br_lck
->fsp
, LEVEL2_CONTEND_WINDOWS_BRL
);
1146 /****************************************************************************
1147 Unlock a range of bytes - POSIX semantics.
1148 ****************************************************************************/
1150 static bool brl_unlock_posix(struct messaging_context
*msg_ctx
,
1151 struct byte_range_lock
*br_lck
,
1152 struct lock_struct
*plock
)
1154 unsigned int i
, j
, count
;
1155 struct lock_struct
*tp
;
1156 struct lock_struct
*locks
= br_lck
->lock_data
;
1157 bool overlap_found
= False
;
1159 /* No zero-zero locks for POSIX. */
1160 if (plock
->start
== 0 && plock
->size
== 0) {
1164 /* Don't allow 64-bit lock wrap. */
1165 if (plock
->start
+ plock
->size
< plock
->start
||
1166 plock
->start
+ plock
->size
< plock
->size
) {
1167 DEBUG(10,("brl_unlock_posix: lock wrap\n"));
1171 /* The worst case scenario here is we have to split an
1172 existing POSIX lock range into two, so we need at most
1175 tp
= talloc_array(br_lck
, struct lock_struct
, br_lck
->num_locks
+ 1);
1177 DEBUG(10,("brl_unlock_posix: malloc fail\n"));
1182 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1183 struct lock_struct
*lock
= &locks
[i
];
1184 unsigned int tmp_count
;
1186 /* Only remove our own locks - ignore fnum. */
1187 if (IS_PENDING_LOCK(lock
->lock_type
) ||
1188 !brl_same_context(&lock
->context
, &plock
->context
)) {
1189 memcpy(&tp
[count
], lock
, sizeof(struct lock_struct
));
1194 if (lock
->lock_flav
== WINDOWS_LOCK
) {
1195 /* Do any Windows flavour locks conflict ? */
1196 if (brl_conflict(lock
, plock
)) {
1200 /* Just copy the Windows lock into the new array. */
1201 memcpy(&tp
[count
], lock
, sizeof(struct lock_struct
));
1206 /* Work out overlaps. */
1207 tmp_count
= brlock_posix_split_merge(&tp
[count
], lock
, plock
);
1209 if (tmp_count
== 0) {
1210 /* plock overlapped the existing lock completely,
1211 or replaced it. Don't copy the existing lock. */
1212 overlap_found
= true;
1213 } else if (tmp_count
== 1) {
1214 /* Either no overlap, (simple copy of existing lock) or
1215 * an overlap of an existing lock. */
1216 /* If the lock changed size, we had an overlap. */
1217 if (tp
[count
].size
!= lock
->size
) {
1218 overlap_found
= true;
1221 } else if (tmp_count
== 2) {
1222 /* We split a lock range in two. */
1223 overlap_found
= true;
1226 /* Optimisation... */
1227 /* We know we're finished here as we can't overlap any
1228 more POSIX locks. Copy the rest of the lock array. */
1230 if (i
< br_lck
->num_locks
- 1) {
1231 memcpy(&tp
[count
], &locks
[i
+1],
1232 sizeof(*locks
)*((br_lck
->num_locks
-1) - i
));
1233 count
+= ((br_lck
->num_locks
-1) - i
);
1240 if (!overlap_found
) {
1241 /* Just ignore - no change. */
1243 DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
1247 /* Unlock any POSIX regions. */
1248 if(lp_posix_locking(br_lck
->fsp
->conn
->params
)) {
1249 release_posix_lock_posix_flavour(br_lck
->fsp
,
1257 /* Realloc so we don't leak entries per unlock call. */
1259 tp
= talloc_realloc(br_lck
, tp
, struct lock_struct
, count
);
1261 DEBUG(10,("brl_unlock_posix: realloc fail\n"));
1265 /* We deleted the last lock. */
1270 contend_level2_oplocks_end(br_lck
->fsp
,
1271 LEVEL2_CONTEND_POSIX_BRL
);
1273 br_lck
->num_locks
= count
;
1274 TALLOC_FREE(br_lck
->lock_data
);
1276 br_lck
->lock_data
= tp
;
1277 br_lck
->modified
= True
;
1279 /* Send unlock messages to any pending waiters that overlap. */
1281 for (j
=0; j
< br_lck
->num_locks
; j
++) {
1282 struct lock_struct
*pend_lock
= &locks
[j
];
1284 /* Ignore non-pending locks. */
1285 if (!IS_PENDING_LOCK(pend_lock
->lock_type
)) {
1289 /* We could send specific lock info here... */
1290 if (brl_pending_overlap(plock
, pend_lock
)) {
1291 DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1292 procid_str_static(&pend_lock
->context
.pid
)));
1294 messaging_send(msg_ctx
, pend_lock
->context
.pid
,
1295 MSG_SMB_UNLOCK
, &data_blob_null
);
1302 bool smb_vfs_call_brl_unlock_windows(struct vfs_handle_struct
*handle
,
1303 struct messaging_context
*msg_ctx
,
1304 struct byte_range_lock
*br_lck
,
1305 const struct lock_struct
*plock
)
1307 VFS_FIND(brl_unlock_windows
);
1308 return handle
->fns
->brl_unlock_windows_fn(handle
, msg_ctx
, br_lck
,
1312 /****************************************************************************
1313 Unlock a range of bytes.
1314 ****************************************************************************/
1316 bool brl_unlock(struct messaging_context
*msg_ctx
,
1317 struct byte_range_lock
*br_lck
,
1319 struct server_id pid
,
1322 enum brl_flavour lock_flav
)
1324 struct lock_struct lock
;
1326 lock
.context
.smblctx
= smblctx
;
1327 lock
.context
.pid
= pid
;
1328 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1331 lock
.fnum
= br_lck
->fsp
->fnum
;
1332 lock
.lock_type
= UNLOCK_LOCK
;
1333 lock
.lock_flav
= lock_flav
;
1335 if (lock_flav
== WINDOWS_LOCK
) {
1336 return SMB_VFS_BRL_UNLOCK_WINDOWS(br_lck
->fsp
->conn
, msg_ctx
,
1339 return brl_unlock_posix(msg_ctx
, br_lck
, &lock
);
1343 /****************************************************************************
1344 Test if we could add a lock if we wanted to.
1345 Returns True if the region required is currently unlocked, False if locked.
1346 ****************************************************************************/
1348 bool brl_locktest(const struct byte_range_lock
*br_lck
,
1349 const struct lock_struct
*rw_probe
)
1353 const struct lock_struct
*locks
= br_lck
->lock_data
;
1354 files_struct
*fsp
= br_lck
->fsp
;
1356 /* Make sure existing locks don't conflict */
1357 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1359 * Our own locks don't conflict.
1361 if (brl_conflict_other(&locks
[i
], rw_probe
)) {
1367 * There is no lock held by an SMB daemon, check to
1368 * see if there is a POSIX lock from a UNIX or NFS process.
1369 * This only conflicts with Windows locks, not POSIX locks.
1372 if(lp_posix_locking(fsp
->conn
->params
) &&
1373 (rw_probe
->lock_flav
== WINDOWS_LOCK
)) {
1375 * Make copies -- is_posix_locked might modify the values
1378 br_off start
= rw_probe
->start
;
1379 br_off size
= rw_probe
->size
;
1380 enum brl_type lock_type
= rw_probe
->lock_type
;
1382 ret
= is_posix_locked(fsp
, &start
, &size
, &lock_type
, WINDOWS_LOCK
);
1384 DEBUG(10, ("brl_locktest: posix start=%ju len=%ju %s for %s "
1385 "file %s\n", (uintmax_t)start
, (uintmax_t)size
,
1386 ret
? "locked" : "unlocked",
1387 fsp_fnum_dbg(fsp
), fsp_str_dbg(fsp
)));
1389 /* We need to return the inverse of is_posix_locked. */
1393 /* no conflicts - we could have added it */
1397 /****************************************************************************
1398 Query for existing locks.
1399 ****************************************************************************/
1401 NTSTATUS
brl_lockquery(struct byte_range_lock
*br_lck
,
1403 struct server_id pid
,
1406 enum brl_type
*plock_type
,
1407 enum brl_flavour lock_flav
)
1410 struct lock_struct lock
;
1411 const struct lock_struct
*locks
= br_lck
->lock_data
;
1412 files_struct
*fsp
= br_lck
->fsp
;
1414 lock
.context
.smblctx
= *psmblctx
;
1415 lock
.context
.pid
= pid
;
1416 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1417 lock
.start
= *pstart
;
1419 lock
.fnum
= fsp
->fnum
;
1420 lock
.lock_type
= *plock_type
;
1421 lock
.lock_flav
= lock_flav
;
1423 /* Make sure existing locks don't conflict */
1424 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1425 const struct lock_struct
*exlock
= &locks
[i
];
1426 bool conflict
= False
;
1428 if (exlock
->lock_flav
== WINDOWS_LOCK
) {
1429 conflict
= brl_conflict(exlock
, &lock
);
1431 conflict
= brl_conflict_posix(exlock
, &lock
);
1435 *psmblctx
= exlock
->context
.smblctx
;
1436 *pstart
= exlock
->start
;
1437 *psize
= exlock
->size
;
1438 *plock_type
= exlock
->lock_type
;
1439 return NT_STATUS_LOCK_NOT_GRANTED
;
1444 * There is no lock held by an SMB daemon, check to
1445 * see if there is a POSIX lock from a UNIX or NFS process.
1448 if(lp_posix_locking(fsp
->conn
->params
)) {
1449 bool ret
= is_posix_locked(fsp
, pstart
, psize
, plock_type
, POSIX_LOCK
);
1451 DEBUG(10, ("brl_lockquery: posix start=%ju len=%ju %s for %s "
1452 "file %s\n", (uintmax_t)*pstart
,
1453 (uintmax_t)*psize
, ret
? "locked" : "unlocked",
1454 fsp_fnum_dbg(fsp
), fsp_str_dbg(fsp
)));
1457 /* Hmmm. No clue what to set smblctx to - use -1. */
1458 *psmblctx
= 0xFFFFFFFFFFFFFFFFLL
;
1459 return NT_STATUS_LOCK_NOT_GRANTED
;
1463 return NT_STATUS_OK
;
1467 bool smb_vfs_call_brl_cancel_windows(struct vfs_handle_struct
*handle
,
1468 struct byte_range_lock
*br_lck
,
1469 struct lock_struct
*plock
)
1471 VFS_FIND(brl_cancel_windows
);
1472 return handle
->fns
->brl_cancel_windows_fn(handle
, br_lck
, plock
);
1475 /****************************************************************************
1476 Remove a particular pending lock.
1477 ****************************************************************************/
1478 bool brl_lock_cancel(struct byte_range_lock
*br_lck
,
1480 struct server_id pid
,
1483 enum brl_flavour lock_flav
)
1486 struct lock_struct lock
;
1488 lock
.context
.smblctx
= smblctx
;
1489 lock
.context
.pid
= pid
;
1490 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1493 lock
.fnum
= br_lck
->fsp
->fnum
;
1494 lock
.lock_flav
= lock_flav
;
1495 /* lock.lock_type doesn't matter */
1497 if (lock_flav
== WINDOWS_LOCK
) {
1498 ret
= SMB_VFS_BRL_CANCEL_WINDOWS(br_lck
->fsp
->conn
, br_lck
,
1501 ret
= brl_lock_cancel_default(br_lck
, &lock
);
1507 bool brl_lock_cancel_default(struct byte_range_lock
*br_lck
,
1508 struct lock_struct
*plock
)
1511 struct lock_struct
*locks
= br_lck
->lock_data
;
1515 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1516 struct lock_struct
*lock
= &locks
[i
];
1518 /* For pending locks we *always* care about the fnum. */
1519 if (brl_same_context(&lock
->context
, &plock
->context
) &&
1520 lock
->fnum
== plock
->fnum
&&
1521 IS_PENDING_LOCK(lock
->lock_type
) &&
1522 lock
->lock_flav
== plock
->lock_flav
&&
1523 lock
->start
== plock
->start
&&
1524 lock
->size
== plock
->size
) {
1529 if (i
== br_lck
->num_locks
) {
1530 /* Didn't find it. */
1534 brl_delete_lock_struct(locks
, br_lck
->num_locks
, i
);
1535 br_lck
->num_locks
-= 1;
1536 br_lck
->modified
= True
;
1540 /****************************************************************************
1541 Remove any locks associated with a open file.
1542 We return True if this process owns any other Windows locks on this
1543 fd and so we should not immediately close the fd.
1544 ****************************************************************************/
1546 void brl_close_fnum(struct messaging_context
*msg_ctx
,
1547 struct byte_range_lock
*br_lck
)
1549 files_struct
*fsp
= br_lck
->fsp
;
1550 uint32_t tid
= fsp
->conn
->cnum
;
1551 uint64_t fnum
= fsp
->fnum
;
1553 struct lock_struct
*locks
= br_lck
->lock_data
;
1554 struct server_id pid
= messaging_server_id(fsp
->conn
->sconn
->msg_ctx
);
1555 struct lock_struct
*locks_copy
;
1556 unsigned int num_locks_copy
;
1558 /* Copy the current lock array. */
1559 if (br_lck
->num_locks
) {
1560 locks_copy
= (struct lock_struct
*)talloc_memdup(br_lck
, locks
, br_lck
->num_locks
* sizeof(struct lock_struct
));
1562 smb_panic("brl_close_fnum: talloc failed");
1568 num_locks_copy
= br_lck
->num_locks
;
1570 for (i
=0; i
< num_locks_copy
; i
++) {
1571 struct lock_struct
*lock
= &locks_copy
[i
];
1573 if (lock
->context
.tid
== tid
&& serverid_equal(&lock
->context
.pid
, &pid
) &&
1574 (lock
->fnum
== fnum
)) {
1577 lock
->context
.smblctx
,
1586 bool brl_mark_disconnected(struct files_struct
*fsp
)
1588 uint32_t tid
= fsp
->conn
->cnum
;
1590 uint64_t fnum
= fsp
->fnum
;
1592 struct server_id self
= messaging_server_id(fsp
->conn
->sconn
->msg_ctx
);
1593 struct byte_range_lock
*br_lck
= NULL
;
1595 if (fsp
->op
== NULL
) {
1599 smblctx
= fsp
->op
->global
->open_persistent_id
;
1601 if (!fsp
->op
->global
->durable
) {
1605 if (fsp
->current_lock_count
== 0) {
1609 br_lck
= brl_get_locks(talloc_tos(), fsp
);
1610 if (br_lck
== NULL
) {
1614 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1615 struct lock_struct
*lock
= &br_lck
->lock_data
[i
];
1618 * as this is a durable handle, we only expect locks
1619 * of the current file handle!
1622 if (lock
->context
.smblctx
!= smblctx
) {
1623 TALLOC_FREE(br_lck
);
1627 if (lock
->context
.tid
!= tid
) {
1628 TALLOC_FREE(br_lck
);
1632 if (!serverid_equal(&lock
->context
.pid
, &self
)) {
1633 TALLOC_FREE(br_lck
);
1637 if (lock
->fnum
!= fnum
) {
1638 TALLOC_FREE(br_lck
);
1642 server_id_set_disconnected(&lock
->context
.pid
);
1643 lock
->context
.tid
= TID_FIELD_INVALID
;
1644 lock
->fnum
= FNUM_FIELD_INVALID
;
1647 br_lck
->modified
= true;
1648 TALLOC_FREE(br_lck
);
1652 bool brl_reconnect_disconnected(struct files_struct
*fsp
)
1654 uint32_t tid
= fsp
->conn
->cnum
;
1656 uint64_t fnum
= fsp
->fnum
;
1658 struct server_id self
= messaging_server_id(fsp
->conn
->sconn
->msg_ctx
);
1659 struct byte_range_lock
*br_lck
= NULL
;
1661 if (fsp
->op
== NULL
) {
1665 smblctx
= fsp
->op
->global
->open_persistent_id
;
1667 if (!fsp
->op
->global
->durable
) {
1672 * When reconnecting, we do not want to validate the brlock entries
1673 * and thereby remove our own (disconnected) entries but reactivate
1676 fsp
->lockdb_clean
= true;
1678 br_lck
= brl_get_locks(talloc_tos(), fsp
);
1679 if (br_lck
== NULL
) {
1683 if (br_lck
->num_locks
== 0) {
1684 TALLOC_FREE(br_lck
);
1688 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1689 struct lock_struct
*lock
= &br_lck
->lock_data
[i
];
1692 * as this is a durable handle we only expect locks
1693 * of the current file handle!
1696 if (lock
->context
.smblctx
!= smblctx
) {
1697 TALLOC_FREE(br_lck
);
1701 if (lock
->context
.tid
!= TID_FIELD_INVALID
) {
1702 TALLOC_FREE(br_lck
);
1706 if (!server_id_is_disconnected(&lock
->context
.pid
)) {
1707 TALLOC_FREE(br_lck
);
1711 if (lock
->fnum
!= FNUM_FIELD_INVALID
) {
1712 TALLOC_FREE(br_lck
);
1716 lock
->context
.pid
= self
;
1717 lock
->context
.tid
= tid
;
1721 fsp
->current_lock_count
= br_lck
->num_locks
;
1722 br_lck
->modified
= true;
1723 TALLOC_FREE(br_lck
);
1727 /****************************************************************************
1728 Ensure this set of lock entries is valid.
1729 ****************************************************************************/
1730 static bool validate_lock_entries(unsigned int *pnum_entries
, struct lock_struct
**pplocks
,
1731 bool keep_disconnected
)
1734 struct lock_struct
*locks
= *pplocks
;
1735 unsigned int num_entries
= *pnum_entries
;
1737 struct server_id
*ids
;
1740 if (num_entries
== 0) {
1744 frame
= talloc_stackframe();
1746 ids
= talloc_array(frame
, struct server_id
, num_entries
);
1748 DEBUG(0, ("validate_lock_entries: "
1749 "talloc_array(struct server_id, %u) failed\n",
1755 exists
= talloc_array(frame
, bool, num_entries
);
1756 if (exists
== NULL
) {
1757 DEBUG(0, ("validate_lock_entries: "
1758 "talloc_array(bool, %u) failed\n",
1764 for (i
= 0; i
< num_entries
; i
++) {
1765 ids
[i
] = locks
[i
].context
.pid
;
1768 if (!serverids_exist(ids
, num_entries
, exists
)) {
1769 DEBUG(3, ("validate_lock_entries: serverids_exists failed\n"));
1776 while (i
< num_entries
) {
1782 if (keep_disconnected
&&
1783 server_id_is_disconnected(&ids
[i
]))
1789 /* This process no longer exists */
1791 brl_delete_lock_struct(locks
, num_entries
, i
);
1796 *pnum_entries
= num_entries
;
1801 struct brl_forall_cb
{
1802 void (*fn
)(struct file_id id
, struct server_id pid
,
1803 enum brl_type lock_type
,
1804 enum brl_flavour lock_flav
,
1805 br_off start
, br_off size
,
1806 void *private_data
);
1810 /****************************************************************************
1811 Traverse the whole database with this function, calling traverse_callback
1813 ****************************************************************************/
1815 static int brl_traverse_fn(struct db_record
*rec
, void *state
)
1817 struct brl_forall_cb
*cb
= (struct brl_forall_cb
*)state
;
1818 struct lock_struct
*locks
;
1819 struct file_id
*key
;
1821 unsigned int num_locks
= 0;
1822 unsigned int orig_num_locks
= 0;
1826 dbkey
= dbwrap_record_get_key(rec
);
1827 value
= dbwrap_record_get_value(rec
);
1829 /* In a traverse function we must make a copy of
1830 dbuf before modifying it. */
1832 locks
= (struct lock_struct
*)talloc_memdup(
1833 talloc_tos(), value
.dptr
, value
.dsize
);
1835 return -1; /* Terminate traversal. */
1838 key
= (struct file_id
*)dbkey
.dptr
;
1839 orig_num_locks
= num_locks
= value
.dsize
/sizeof(*locks
);
1841 /* Ensure the lock db is clean of entries from invalid processes. */
1843 if (!validate_lock_entries(&num_locks
, &locks
, true)) {
1845 return -1; /* Terminate traversal */
1848 if (orig_num_locks
!= num_locks
) {
1851 data
.dptr
= (uint8_t *)locks
;
1852 data
.dsize
= num_locks
*sizeof(struct lock_struct
);
1853 dbwrap_record_store(rec
, data
, TDB_REPLACE
);
1855 dbwrap_record_delete(rec
);
1860 for ( i
=0; i
<num_locks
; i
++) {
1862 locks
[i
].context
.pid
,
1875 /*******************************************************************
1876 Call the specified function on each lock in the database.
1877 ********************************************************************/
1879 int brl_forall(void (*fn
)(struct file_id id
, struct server_id pid
,
1880 enum brl_type lock_type
,
1881 enum brl_flavour lock_flav
,
1882 br_off start
, br_off size
,
1883 void *private_data
),
1886 struct brl_forall_cb cb
;
1894 cb
.private_data
= private_data
;
1895 status
= dbwrap_traverse(brlock_db
, brl_traverse_fn
, &cb
, &count
);
1897 if (!NT_STATUS_IS_OK(status
)) {
1904 /*******************************************************************
1905 Store a potentially modified set of byte range lock data back into
1908 ********************************************************************/
1910 static void byte_range_lock_flush(struct byte_range_lock
*br_lck
)
1913 if (!br_lck
->modified
) {
1914 DEBUG(10, ("br_lck not modified\n"));
1918 data_len
= br_lck
->num_locks
* sizeof(struct lock_struct
);
1920 if (br_lck
->have_read_oplocks
) {
1924 DEBUG(10, ("data_len=%d\n", (int)data_len
));
1926 if (data_len
== 0) {
1927 /* No locks - delete this entry. */
1928 NTSTATUS status
= dbwrap_record_delete(br_lck
->record
);
1929 if (!NT_STATUS_IS_OK(status
)) {
1930 DEBUG(0, ("delete_rec returned %s\n",
1931 nt_errstr(status
)));
1932 smb_panic("Could not delete byte range lock entry");
1938 data
.dsize
= data_len
;
1939 data
.dptr
= talloc_array(talloc_tos(), uint8_t, data_len
);
1940 SMB_ASSERT(data
.dptr
!= NULL
);
1942 memcpy(data
.dptr
, br_lck
->lock_data
,
1943 br_lck
->num_locks
* sizeof(struct lock_struct
));
1945 if (br_lck
->have_read_oplocks
) {
1946 data
.dptr
[data_len
-1] = 1;
1949 status
= dbwrap_record_store(br_lck
->record
, data
, TDB_REPLACE
);
1950 TALLOC_FREE(data
.dptr
);
1951 if (!NT_STATUS_IS_OK(status
)) {
1952 DEBUG(0, ("store returned %s\n", nt_errstr(status
)));
1953 smb_panic("Could not store byte range mode entry");
1957 DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db
)));
1960 br_lck
->modified
= false;
1961 TALLOC_FREE(br_lck
->record
);
1964 static int byte_range_lock_destructor(struct byte_range_lock
*br_lck
)
1966 byte_range_lock_flush(br_lck
);
1970 /*******************************************************************
1971 Fetch a set of byte range lock data from the database.
1972 Leave the record locked.
1973 TALLOC_FREE(brl) will release the lock in the destructor.
1974 ********************************************************************/
1976 struct byte_range_lock
*brl_get_locks(TALLOC_CTX
*mem_ctx
, files_struct
*fsp
)
1979 struct byte_range_lock
*br_lck
= talloc(mem_ctx
, struct byte_range_lock
);
1981 if (br_lck
== NULL
) {
1986 br_lck
->num_locks
= 0;
1987 br_lck
->have_read_oplocks
= false;
1988 br_lck
->modified
= False
;
1990 key
.dptr
= (uint8
*)&fsp
->file_id
;
1991 key
.dsize
= sizeof(struct file_id
);
1993 br_lck
->record
= dbwrap_fetch_locked(brlock_db
, br_lck
, key
);
1995 if (br_lck
->record
== NULL
) {
1996 DEBUG(3, ("Could not lock byte range lock entry\n"));
1997 TALLOC_FREE(br_lck
);
2001 data
= dbwrap_record_get_value(br_lck
->record
);
2003 br_lck
->lock_data
= NULL
;
2005 talloc_set_destructor(br_lck
, byte_range_lock_destructor
);
2007 br_lck
->num_locks
= data
.dsize
/ sizeof(struct lock_struct
);
2009 if (br_lck
->num_locks
!= 0) {
2010 br_lck
->lock_data
= talloc_array(
2011 br_lck
, struct lock_struct
, br_lck
->num_locks
);
2012 if (br_lck
->lock_data
== NULL
) {
2013 DEBUG(0, ("malloc failed\n"));
2014 TALLOC_FREE(br_lck
);
2018 memcpy(br_lck
->lock_data
, data
.dptr
,
2019 talloc_get_size(br_lck
->lock_data
));
2022 DEBUG(10, ("data.dsize=%d\n", (int)data
.dsize
));
2024 if ((data
.dsize
% sizeof(struct lock_struct
)) == 1) {
2025 br_lck
->have_read_oplocks
= (data
.dptr
[data
.dsize
-1] == 1);
2028 if (!fsp
->lockdb_clean
) {
2029 int orig_num_locks
= br_lck
->num_locks
;
2032 * This is the first time we access the byte range lock
2033 * record with this fsp. Go through and ensure all entries
2034 * are valid - remove any that don't.
2035 * This makes the lockdb self cleaning at low cost.
2037 * Note: Disconnected entries belong to disconnected
2038 * durable handles. So at this point, we have a new
2039 * handle on the file and the disconnected durable has
2040 * already been closed (we are not a durable reconnect).
2041 * So we need to clean the disconnected brl entry.
2044 if (!validate_lock_entries(&br_lck
->num_locks
,
2045 &br_lck
->lock_data
, false)) {
2046 TALLOC_FREE(br_lck
);
2050 /* Ensure invalid locks are cleaned up in the destructor. */
2051 if (orig_num_locks
!= br_lck
->num_locks
) {
2052 br_lck
->modified
= True
;
2055 /* Mark the lockdb as "clean" as seen from this open file. */
2056 fsp
->lockdb_clean
= True
;
2059 if (DEBUGLEVEL
>= 10) {
2061 struct lock_struct
*locks
= br_lck
->lock_data
;
2062 DEBUG(10,("brl_get_locks_internal: %u current locks on file_id %s\n",
2064 file_id_string_tos(&fsp
->file_id
)));
2065 for( i
= 0; i
< br_lck
->num_locks
; i
++) {
2066 print_lock_struct(i
, &locks
[i
]);
2073 struct brl_get_locks_readonly_state
{
2074 TALLOC_CTX
*mem_ctx
;
2075 struct byte_range_lock
**br_lock
;
2078 static void brl_get_locks_readonly_parser(TDB_DATA key
, TDB_DATA data
,
2081 struct brl_get_locks_readonly_state
*state
=
2082 (struct brl_get_locks_readonly_state
*)private_data
;
2083 struct byte_range_lock
*br_lock
;
2085 br_lock
= talloc_pooled_object(
2086 state
->mem_ctx
, struct byte_range_lock
, 1, data
.dsize
);
2087 if (br_lock
== NULL
) {
2088 *state
->br_lock
= NULL
;
2091 br_lock
->lock_data
= (struct lock_struct
*)talloc_memdup(
2092 br_lock
, data
.dptr
, data
.dsize
);
2093 br_lock
->num_locks
= data
.dsize
/ sizeof(struct lock_struct
);
2095 if ((data
.dsize
% sizeof(struct lock_struct
)) == 1) {
2096 br_lock
->have_read_oplocks
= (data
.dptr
[data
.dsize
-1] == 1);
2098 br_lock
->have_read_oplocks
= false;
2101 DEBUG(10, ("Got %d bytes, have_read_oplocks: %s\n", (int)data
.dsize
,
2102 br_lock
->have_read_oplocks
? "true" : "false"));
2104 *state
->br_lock
= br_lock
;
2107 struct byte_range_lock
*brl_get_locks_readonly(files_struct
*fsp
)
2109 struct byte_range_lock
*br_lock
= NULL
;
2110 struct byte_range_lock
*rw
= NULL
;
2112 DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
2113 dbwrap_get_seqnum(brlock_db
), fsp
->brlock_seqnum
));
2115 if ((fsp
->brlock_rec
!= NULL
)
2116 && (dbwrap_get_seqnum(brlock_db
) == fsp
->brlock_seqnum
)) {
2118 * We have cached the brlock_rec and the database did not
2121 return fsp
->brlock_rec
;
2124 if (!fsp
->lockdb_clean
) {
2126 * Fetch the record in R/W mode to give validate_lock_entries
2127 * a chance to kick in once.
2129 rw
= brl_get_locks(talloc_tos(), fsp
);
2133 fsp
->lockdb_clean
= true;
2137 size_t lock_data_size
;
2140 * Make a copy of the already retrieved and sanitized rw record
2142 lock_data_size
= rw
->num_locks
* sizeof(struct lock_struct
);
2143 br_lock
= talloc_pooled_object(
2144 fsp
, struct byte_range_lock
, 1, lock_data_size
);
2145 if (br_lock
== NULL
) {
2148 br_lock
->have_read_oplocks
= rw
->have_read_oplocks
;
2149 br_lock
->num_locks
= rw
->num_locks
;
2150 br_lock
->lock_data
= (struct lock_struct
*)talloc_memdup(
2151 br_lock
, rw
->lock_data
, lock_data_size
);
2153 struct brl_get_locks_readonly_state state
;
2157 * Parse the record fresh from the database
2160 state
.mem_ctx
= fsp
;
2161 state
.br_lock
= &br_lock
;
2163 status
= dbwrap_parse_record(
2165 make_tdb_data((uint8_t *)&fsp
->file_id
,
2166 sizeof(fsp
->file_id
)),
2167 brl_get_locks_readonly_parser
, &state
);
2169 if (NT_STATUS_EQUAL(status
,NT_STATUS_NOT_FOUND
)) {
2171 * No locks on this file. Return an empty br_lock.
2173 br_lock
= talloc(fsp
, struct byte_range_lock
);
2174 if (br_lock
== NULL
) {
2178 br_lock
->have_read_oplocks
= false;
2179 br_lock
->num_locks
= 0;
2180 br_lock
->lock_data
= NULL
;
2182 } else if (!NT_STATUS_IS_OK(status
)) {
2183 DEBUG(3, ("Could not parse byte range lock record: "
2184 "%s\n", nt_errstr(status
)));
2187 if (br_lock
== NULL
) {
2193 br_lock
->modified
= false;
2194 br_lock
->record
= NULL
;
2196 if (lp_clustering()) {
2198 * In the cluster case we can't cache the brlock struct
2199 * because dbwrap_get_seqnum does not work reliably over
2200 * ctdb. Thus we have to throw away the brlock struct soon.
2202 talloc_steal(talloc_tos(), br_lock
);
2205 * Cache the brlock struct, invalidated when the dbwrap_seqnum
2206 * changes. See beginning of this routine.
2208 TALLOC_FREE(fsp
->brlock_rec
);
2209 fsp
->brlock_rec
= br_lock
;
2210 fsp
->brlock_seqnum
= dbwrap_get_seqnum(brlock_db
);
2218 struct brl_revalidate_state
{
2221 struct server_id
*pids
;
2225 * Collect PIDs of all processes with pending entries
2228 static void brl_revalidate_collect(struct file_id id
, struct server_id pid
,
2229 enum brl_type lock_type
,
2230 enum brl_flavour lock_flav
,
2231 br_off start
, br_off size
,
2234 struct brl_revalidate_state
*state
=
2235 (struct brl_revalidate_state
*)private_data
;
2237 if (!IS_PENDING_LOCK(lock_type
)) {
2241 add_to_large_array(state
, sizeof(pid
), (void *)&pid
,
2242 &state
->pids
, &state
->num_pids
,
2243 &state
->array_size
);
2247 * qsort callback to sort the processes
2250 static int compare_procids(const void *p1
, const void *p2
)
2252 const struct server_id
*i1
= (const struct server_id
*)p1
;
2253 const struct server_id
*i2
= (const struct server_id
*)p2
;
2255 if (i1
->pid
< i2
->pid
) return -1;
2256 if (i1
->pid
> i2
->pid
) return 1;
2261 * Send a MSG_SMB_UNLOCK message to all processes with pending byte range
2262 * locks so that they retry. Mainly used in the cluster code after a node has
2265 * Done in two steps to avoid double-sends: First we collect all entries in an
2266 * array, then qsort that array and only send to non-dupes.
2269 void brl_revalidate(struct messaging_context
*msg_ctx
,
2272 struct server_id server_id
,
2275 struct brl_revalidate_state
*state
;
2277 struct server_id last_pid
;
2279 if (!(state
= talloc_zero(NULL
, struct brl_revalidate_state
))) {
2280 DEBUG(0, ("talloc failed\n"));
2284 brl_forall(brl_revalidate_collect
, state
);
2286 if (state
->array_size
== -1) {
2287 DEBUG(0, ("talloc failed\n"));
2291 if (state
->num_pids
== 0) {
2295 TYPESAFE_QSORT(state
->pids
, state
->num_pids
, compare_procids
);
2297 ZERO_STRUCT(last_pid
);
2299 for (i
=0; i
<state
->num_pids
; i
++) {
2300 if (serverid_equal(&last_pid
, &state
->pids
[i
])) {
2302 * We've seen that one already
2307 messaging_send(msg_ctx
, state
->pids
[i
], MSG_SMB_UNLOCK
,
2309 last_pid
= state
->pids
[i
];
2317 bool brl_cleanup_disconnected(struct file_id fid
, uint64_t open_persistent_id
)
2320 TALLOC_CTX
*frame
= talloc_stackframe();
2322 struct db_record
*rec
;
2323 struct lock_struct
*lock
;
2327 key
= make_tdb_data((void*)&fid
, sizeof(fid
));
2329 rec
= dbwrap_fetch_locked(brlock_db
, frame
, key
);
2331 DEBUG(5, ("brl_cleanup_disconnected: failed to fetch record "
2332 "for file %s\n", file_id_string(frame
, &fid
)));
2336 val
= dbwrap_record_get_value(rec
);
2337 lock
= (struct lock_struct
*)val
.dptr
;
2338 num
= val
.dsize
/ sizeof(struct lock_struct
);
2340 DEBUG(10, ("brl_cleanup_disconnected: no byte range locks for "
2341 "file %s\n", file_id_string(frame
, &fid
)));
2346 for (n
=0; n
<num
; n
++) {
2347 struct lock_context
*ctx
= &lock
[n
].context
;
2349 if (!server_id_is_disconnected(&ctx
->pid
)) {
2350 DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
2351 "%s used by server %s, do not cleanup\n",
2352 file_id_string(frame
, &fid
),
2353 server_id_str(frame
, &ctx
->pid
)));
2357 if (ctx
->smblctx
!= open_persistent_id
) {
2358 DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
2359 "%s expected smblctx %llu but found %llu"
2360 ", do not cleanup\n",
2361 file_id_string(frame
, &fid
),
2362 (unsigned long long)open_persistent_id
,
2363 (unsigned long long)ctx
->smblctx
));
2368 status
= dbwrap_record_delete(rec
);
2369 if (!NT_STATUS_IS_OK(status
)) {
2370 DEBUG(5, ("brl_cleanup_disconnected: failed to delete record "
2371 "for file %s from %s, open %llu: %s\n",
2372 file_id_string(frame
, &fid
), dbwrap_name(brlock_db
),
2373 (unsigned long long)open_persistent_id
,
2374 nt_errstr(status
)));
2378 DEBUG(10, ("brl_cleanup_disconnected: "
2379 "file %s cleaned up %u entries from open %llu\n",
2380 file_id_string(frame
, &fid
), num
,
2381 (unsigned long long)open_persistent_id
));