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
*lck1
, const struct lock_struct
*lck2
)
232 if (IS_PENDING_LOCK(lck1
->lock_type
) || IS_PENDING_LOCK(lck2
->lock_type
))
235 if (lck1
->lock_type
== READ_LOCK
&& lck2
->lock_type
== READ_LOCK
)
238 /* POSIX flavour locks never conflict here - this is only called
239 in the read/write path. */
241 if (lck1
->lock_flav
== POSIX_LOCK
&& lck2
->lock_flav
== POSIX_LOCK
)
245 * Incoming WRITE locks conflict with existing READ locks even
246 * if the context is the same. JRA. See LOCKTEST7 in smbtorture.
249 if (!(lck2
->lock_type
== WRITE_LOCK
&& lck1
->lock_type
== READ_LOCK
)) {
250 if (brl_same_context(&lck1
->context
, &lck2
->context
) &&
251 lck1
->fnum
== lck2
->fnum
)
255 return brl_overlap(lck1
, lck2
);
258 /****************************************************************************
259 Check if an unlock overlaps a pending lock.
260 ****************************************************************************/
262 static bool brl_pending_overlap(const struct lock_struct
*lock
, const struct lock_struct
*pend_lock
)
264 if ((lock
->start
<= pend_lock
->start
) && (lock
->start
+ lock
->size
> pend_lock
->start
))
266 if ((lock
->start
>= pend_lock
->start
) && (lock
->start
< pend_lock
->start
+ pend_lock
->size
))
271 /****************************************************************************
272 Amazingly enough, w2k3 "remembers" whether the last lock failure on a fnum
273 is the same as this one and changes its error code. I wonder if any
274 app depends on this ?
275 ****************************************************************************/
277 static NTSTATUS
brl_lock_failed(files_struct
*fsp
,
278 const struct lock_struct
*lock
,
281 if (lock
->start
>= 0xEF000000 && (lock
->start
>> 63) == 0) {
282 /* amazing the little things you learn with a test
283 suite. Locks beyond this offset (as a 64 bit
284 number!) always generate the conflict error code,
285 unless the top bit is set */
286 if (!blocking_lock
) {
287 fsp
->last_lock_failure
= *lock
;
289 return NT_STATUS_FILE_LOCK_CONFLICT
;
292 if (serverid_equal(&lock
->context
.pid
, &fsp
->last_lock_failure
.context
.pid
) &&
293 lock
->context
.tid
== fsp
->last_lock_failure
.context
.tid
&&
294 lock
->fnum
== fsp
->last_lock_failure
.fnum
&&
295 lock
->start
== fsp
->last_lock_failure
.start
) {
296 return NT_STATUS_FILE_LOCK_CONFLICT
;
299 if (!blocking_lock
) {
300 fsp
->last_lock_failure
= *lock
;
302 return NT_STATUS_LOCK_NOT_GRANTED
;
305 /****************************************************************************
306 Open up the brlock.tdb database.
307 ****************************************************************************/
309 void brl_init(bool read_only
)
317 tdb_flags
= TDB_DEFAULT
|TDB_VOLATILE
|TDB_CLEAR_IF_FIRST
|TDB_INCOMPATIBLE_HASH
;
319 if (!lp_clustering()) {
321 * We can't use the SEQNUM trick to cache brlock
322 * entries in the clustering case because ctdb seqnum
323 * propagation has a delay.
325 tdb_flags
|= TDB_SEQNUM
;
328 brlock_db
= db_open(NULL
, lock_path("brlock.tdb"),
329 SMB_OPEN_DATABASE_TDB_HASH_SIZE
, tdb_flags
,
330 read_only
?O_RDONLY
:(O_RDWR
|O_CREAT
), 0644,
331 DBWRAP_LOCK_ORDER_2
, DBWRAP_FLAG_NONE
);
333 DEBUG(0,("Failed to open byte range locking database %s\n",
334 lock_path("brlock.tdb")));
339 /****************************************************************************
340 Close down the brlock.tdb database.
341 ****************************************************************************/
343 void brl_shutdown(void)
345 TALLOC_FREE(brlock_db
);
349 /****************************************************************************
350 Compare two locks for sorting.
351 ****************************************************************************/
353 static int lock_compare(const struct lock_struct
*lck1
,
354 const struct lock_struct
*lck2
)
356 if (lck1
->start
!= lck2
->start
) {
357 return (lck1
->start
- lck2
->start
);
359 if (lck2
->size
!= lck1
->size
) {
360 return ((int)lck1
->size
- (int)lck2
->size
);
366 /****************************************************************************
367 Lock a range of bytes - Windows lock semantics.
368 ****************************************************************************/
370 NTSTATUS
brl_lock_windows_default(struct byte_range_lock
*br_lck
,
371 struct lock_struct
*plock
, bool blocking_lock
)
374 files_struct
*fsp
= br_lck
->fsp
;
375 struct lock_struct
*locks
= br_lck
->lock_data
;
378 SMB_ASSERT(plock
->lock_type
!= UNLOCK_LOCK
);
380 if ((plock
->start
+ plock
->size
- 1 < plock
->start
) &&
382 return NT_STATUS_INVALID_LOCK_RANGE
;
385 for (i
=0; i
< br_lck
->num_locks
; i
++) {
386 /* Do any Windows or POSIX locks conflict ? */
387 if (brl_conflict(&locks
[i
], plock
)) {
388 /* Remember who blocked us. */
389 plock
->context
.smblctx
= locks
[i
].context
.smblctx
;
390 return brl_lock_failed(fsp
,plock
,blocking_lock
);
393 if (plock
->start
== 0 && plock
->size
== 0 &&
394 locks
[i
].size
== 0) {
400 if (!IS_PENDING_LOCK(plock
->lock_type
)) {
401 contend_level2_oplocks_begin(fsp
, LEVEL2_CONTEND_WINDOWS_BRL
);
404 /* We can get the Windows lock, now see if it needs to
405 be mapped into a lower level POSIX one, and if so can
408 if (!IS_PENDING_LOCK(plock
->lock_type
) && lp_posix_locking(fsp
->conn
->params
)) {
410 if (!set_posix_lock_windows_flavour(fsp
,
419 /* We don't know who blocked us. */
420 plock
->context
.smblctx
= 0xFFFFFFFFFFFFFFFFLL
;
422 if (errno_ret
== EACCES
|| errno_ret
== EAGAIN
) {
423 status
= NT_STATUS_FILE_LOCK_CONFLICT
;
426 status
= map_nt_error_from_unix(errno
);
432 /* no conflicts - add it to the list of locks */
433 locks
= talloc_realloc(br_lck
, locks
, struct lock_struct
,
434 (br_lck
->num_locks
+ 1));
436 status
= NT_STATUS_NO_MEMORY
;
440 memcpy(&locks
[br_lck
->num_locks
], plock
, sizeof(struct lock_struct
));
441 br_lck
->num_locks
+= 1;
442 br_lck
->lock_data
= locks
;
443 br_lck
->modified
= True
;
447 if (!IS_PENDING_LOCK(plock
->lock_type
)) {
448 contend_level2_oplocks_end(fsp
, LEVEL2_CONTEND_WINDOWS_BRL
);
453 /****************************************************************************
454 Cope with POSIX range splits and merges.
455 ****************************************************************************/
457 static unsigned int brlock_posix_split_merge(struct lock_struct
*lck_arr
, /* Output array. */
458 struct lock_struct
*ex
, /* existing lock. */
459 struct lock_struct
*plock
) /* proposed lock. */
461 bool lock_types_differ
= (ex
->lock_type
!= plock
->lock_type
);
463 /* We can't merge non-conflicting locks on different context - ignore fnum. */
465 if (!brl_same_context(&ex
->context
, &plock
->context
)) {
467 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
471 /* We now know we have the same context. */
473 /* Did we overlap ? */
475 /*********************************************
486 **********************************************/
488 if ( (ex
->start
> (plock
->start
+ plock
->size
)) ||
489 (plock
->start
> (ex
->start
+ ex
->size
))) {
491 /* No overlap with this lock - copy existing. */
493 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
497 /*********************************************
498 +---------------------------+
500 +---------------------------+
501 +---------------------------+
502 | plock | -> replace with plock.
503 +---------------------------+
508 +---------------------------+
509 | plock | -> replace with plock.
510 +---------------------------+
512 **********************************************/
514 if ( (ex
->start
>= plock
->start
) &&
515 (ex
->start
+ ex
->size
<= plock
->start
+ plock
->size
) ) {
517 /* Replace - discard existing lock. */
522 /*********************************************
532 +---------------+-------+
533 | plock | ex | - different lock types.
534 +---------------+-------+
536 +-----------------------+
537 | plock | - same lock type.
538 +-----------------------+
539 **********************************************/
541 if (plock
->start
+ plock
->size
== ex
->start
) {
543 /* If the lock types are the same, we merge, if different, we
544 add the remainder of the old lock. */
546 if (lock_types_differ
) {
548 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
551 /* Merge - adjust incoming lock as we may have more
552 * merging to come. */
553 plock
->size
+= ex
->size
;
558 /*********************************************
567 +-------+---------------+
568 | ex | plock | - different lock types
569 +-------+---------------+
572 +-----------------------+
573 | plock | - same lock type.
574 +-----------------------+
576 **********************************************/
578 if (ex
->start
+ ex
->size
== plock
->start
) {
580 /* If the lock types are the same, we merge, if different, we
581 add the existing lock. */
583 if (lock_types_differ
) {
584 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
587 /* Merge - adjust incoming lock as we may have more
588 * merging to come. */
589 plock
->start
= ex
->start
;
590 plock
->size
+= ex
->size
;
595 /*********************************************
597 +-----------------------+
599 +-----------------------+
612 +---------------+-------+
613 | plock | ex | - different lock types.
614 +---------------+-------+
616 +-----------------------+
617 | plock | - same lock type.
618 +-----------------------+
619 **********************************************/
621 if ( (ex
->start
>= plock
->start
) &&
622 (ex
->start
<= plock
->start
+ plock
->size
) &&
623 (ex
->start
+ ex
->size
> plock
->start
+ plock
->size
) ) {
625 /* If the lock types are the same, we merge, if different, we
626 add the remainder of the old lock. */
628 if (lock_types_differ
) {
629 /* Add remaining existing. */
630 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
631 /* Adjust existing start and size. */
632 lck_arr
[0].start
= plock
->start
+ plock
->size
;
633 lck_arr
[0].size
= (ex
->start
+ ex
->size
) - (plock
->start
+ plock
->size
);
636 /* Merge - adjust incoming lock as we may have more
637 * merging to come. */
638 plock
->size
+= (ex
->start
+ ex
->size
) - (plock
->start
+ plock
->size
);
643 /*********************************************
645 +-----------------------+
647 +-----------------------+
660 +-------+---------------+
661 | ex | plock | - different lock types
662 +-------+---------------+
665 +-----------------------+
666 | plock | - same lock type.
667 +-----------------------+
669 **********************************************/
671 if ( (ex
->start
< plock
->start
) &&
672 (ex
->start
+ ex
->size
>= plock
->start
) &&
673 (ex
->start
+ ex
->size
<= plock
->start
+ plock
->size
) ) {
675 /* If the lock types are the same, we merge, if different, we
676 add the truncated old lock. */
678 if (lock_types_differ
) {
679 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
680 /* Adjust existing size. */
681 lck_arr
[0].size
= plock
->start
- ex
->start
;
684 /* Merge - adjust incoming lock as we may have more
685 * merging to come. MUST ADJUST plock SIZE FIRST ! */
686 plock
->size
+= (plock
->start
- ex
->start
);
687 plock
->start
= ex
->start
;
692 /*********************************************
694 +---------------------------+
696 +---------------------------+
701 +-------+---------+---------+
702 | ex | plock | ex | - different lock types.
703 +-------+---------+---------+
705 +---------------------------+
706 | plock | - same lock type.
707 +---------------------------+
708 **********************************************/
710 if ( (ex
->start
< plock
->start
) && (ex
->start
+ ex
->size
> plock
->start
+ plock
->size
) ) {
712 if (lock_types_differ
) {
714 /* We have to split ex into two locks here. */
716 memcpy(&lck_arr
[0], ex
, sizeof(struct lock_struct
));
717 memcpy(&lck_arr
[1], ex
, sizeof(struct lock_struct
));
719 /* Adjust first existing size. */
720 lck_arr
[0].size
= plock
->start
- ex
->start
;
722 /* Adjust second existing start and size. */
723 lck_arr
[1].start
= plock
->start
+ plock
->size
;
724 lck_arr
[1].size
= (ex
->start
+ ex
->size
) - (plock
->start
+ plock
->size
);
727 /* Just eat the existing locks, merge them into plock. */
728 plock
->start
= ex
->start
;
729 plock
->size
= ex
->size
;
734 /* Never get here. */
735 smb_panic("brlock_posix_split_merge");
738 /* Keep some compilers happy. */
742 /****************************************************************************
743 Lock a range of bytes - POSIX lock semantics.
744 We must cope with range splits and merges.
745 ****************************************************************************/
747 static NTSTATUS
brl_lock_posix(struct messaging_context
*msg_ctx
,
748 struct byte_range_lock
*br_lck
,
749 struct lock_struct
*plock
)
751 unsigned int i
, count
, posix_count
;
752 struct lock_struct
*locks
= br_lck
->lock_data
;
753 struct lock_struct
*tp
;
754 bool signal_pending_read
= False
;
755 bool break_oplocks
= false;
758 /* No zero-zero locks for POSIX. */
759 if (plock
->start
== 0 && plock
->size
== 0) {
760 return NT_STATUS_INVALID_PARAMETER
;
763 /* Don't allow 64-bit lock wrap. */
764 if (plock
->start
+ plock
->size
- 1 < plock
->start
) {
765 return NT_STATUS_INVALID_PARAMETER
;
768 /* The worst case scenario here is we have to split an
769 existing POSIX lock range into two, and add our lock,
770 so we need at most 2 more entries. */
772 tp
= talloc_array(br_lck
, struct lock_struct
, br_lck
->num_locks
+ 2);
774 return NT_STATUS_NO_MEMORY
;
777 count
= posix_count
= 0;
779 for (i
=0; i
< br_lck
->num_locks
; i
++) {
780 struct lock_struct
*curr_lock
= &locks
[i
];
782 /* If we have a pending read lock, a lock downgrade should
783 trigger a lock re-evaluation. */
784 if (curr_lock
->lock_type
== PENDING_READ_LOCK
&&
785 brl_pending_overlap(plock
, curr_lock
)) {
786 signal_pending_read
= True
;
789 if (curr_lock
->lock_flav
== WINDOWS_LOCK
) {
790 /* Do any Windows flavour locks conflict ? */
791 if (brl_conflict(curr_lock
, plock
)) {
792 /* No games with error messages. */
794 /* Remember who blocked us. */
795 plock
->context
.smblctx
= curr_lock
->context
.smblctx
;
796 return NT_STATUS_FILE_LOCK_CONFLICT
;
798 /* Just copy the Windows lock into the new array. */
799 memcpy(&tp
[count
], curr_lock
, sizeof(struct lock_struct
));
802 unsigned int tmp_count
= 0;
804 /* POSIX conflict semantics are different. */
805 if (brl_conflict_posix(curr_lock
, plock
)) {
806 /* Can't block ourselves with POSIX locks. */
807 /* No games with error messages. */
809 /* Remember who blocked us. */
810 plock
->context
.smblctx
= curr_lock
->context
.smblctx
;
811 return NT_STATUS_FILE_LOCK_CONFLICT
;
814 /* Work out overlaps. */
815 tmp_count
+= brlock_posix_split_merge(&tp
[count
], curr_lock
, plock
);
816 posix_count
+= tmp_count
;
822 * Break oplocks while we hold a brl. Since lock() and unlock() calls
823 * are not symetric with POSIX semantics, we cannot guarantee our
824 * contend_level2_oplocks_begin/end calls will be acquired and
825 * released one-for-one as with Windows semantics. Therefore we only
826 * call contend_level2_oplocks_begin if this is the first POSIX brl on
829 break_oplocks
= (!IS_PENDING_LOCK(plock
->lock_type
) &&
832 contend_level2_oplocks_begin(br_lck
->fsp
,
833 LEVEL2_CONTEND_POSIX_BRL
);
836 /* Try and add the lock in order, sorted by lock start. */
837 for (i
=0; i
< count
; i
++) {
838 struct lock_struct
*curr_lock
= &tp
[i
];
840 if (curr_lock
->start
<= plock
->start
) {
846 memmove(&tp
[i
+1], &tp
[i
],
847 (count
- i
)*sizeof(struct lock_struct
));
849 memcpy(&tp
[i
], plock
, sizeof(struct lock_struct
));
852 /* We can get the POSIX lock, now see if it needs to
853 be mapped into a lower level POSIX one, and if so can
856 if (!IS_PENDING_LOCK(plock
->lock_type
) && lp_posix_locking(br_lck
->fsp
->conn
->params
)) {
859 /* The lower layer just needs to attempt to
860 get the system POSIX lock. We've weeded out
861 any conflicts above. */
863 if (!set_posix_lock_posix_flavour(br_lck
->fsp
,
869 /* We don't know who blocked us. */
870 plock
->context
.smblctx
= 0xFFFFFFFFFFFFFFFFLL
;
872 if (errno_ret
== EACCES
|| errno_ret
== EAGAIN
) {
874 status
= NT_STATUS_FILE_LOCK_CONFLICT
;
878 status
= map_nt_error_from_unix(errno
);
884 /* If we didn't use all the allocated size,
885 * Realloc so we don't leak entries per lock call. */
886 if (count
< br_lck
->num_locks
+ 2) {
887 tp
= talloc_realloc(br_lck
, tp
, struct lock_struct
, count
);
889 status
= NT_STATUS_NO_MEMORY
;
894 br_lck
->num_locks
= count
;
895 TALLOC_FREE(br_lck
->lock_data
);
896 br_lck
->lock_data
= tp
;
898 br_lck
->modified
= True
;
900 /* A successful downgrade from write to read lock can trigger a lock
901 re-evalutation where waiting readers can now proceed. */
903 if (signal_pending_read
) {
904 /* Send unlock messages to any pending read waiters that overlap. */
905 for (i
=0; i
< br_lck
->num_locks
; i
++) {
906 struct lock_struct
*pend_lock
= &locks
[i
];
908 /* Ignore non-pending locks. */
909 if (!IS_PENDING_LOCK(pend_lock
->lock_type
)) {
913 if (pend_lock
->lock_type
== PENDING_READ_LOCK
&&
914 brl_pending_overlap(plock
, pend_lock
)) {
915 DEBUG(10,("brl_lock_posix: sending unlock message to pid %s\n",
916 procid_str_static(&pend_lock
->context
.pid
)));
918 messaging_send(msg_ctx
, pend_lock
->context
.pid
,
919 MSG_SMB_UNLOCK
, &data_blob_null
);
927 contend_level2_oplocks_end(br_lck
->fsp
,
928 LEVEL2_CONTEND_POSIX_BRL
);
933 NTSTATUS
smb_vfs_call_brl_lock_windows(struct vfs_handle_struct
*handle
,
934 struct byte_range_lock
*br_lck
,
935 struct lock_struct
*plock
,
938 VFS_FIND(brl_lock_windows
);
939 return handle
->fns
->brl_lock_windows_fn(handle
, br_lck
, plock
,
943 /****************************************************************************
944 Lock a range of bytes.
945 ****************************************************************************/
947 NTSTATUS
brl_lock(struct messaging_context
*msg_ctx
,
948 struct byte_range_lock
*br_lck
,
950 struct server_id pid
,
953 enum brl_type lock_type
,
954 enum brl_flavour lock_flav
,
959 struct lock_struct lock
;
962 if (start
== 0 && size
== 0) {
963 DEBUG(0,("client sent 0/0 lock - please report this\n"));
968 /* Quieten valgrind on test. */
972 lock
.context
.smblctx
= smblctx
;
973 lock
.context
.pid
= pid
;
974 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
977 lock
.fnum
= br_lck
->fsp
->fnum
;
978 lock
.lock_type
= lock_type
;
979 lock
.lock_flav
= lock_flav
;
981 if (lock_flav
== WINDOWS_LOCK
) {
982 ret
= SMB_VFS_BRL_LOCK_WINDOWS(br_lck
->fsp
->conn
, br_lck
,
983 &lock
, blocking_lock
);
985 ret
= brl_lock_posix(msg_ctx
, br_lck
, &lock
);
989 /* sort the lock list */
990 TYPESAFE_QSORT(br_lck
->lock_data
, (size_t)br_lck
->num_locks
, lock_compare
);
993 /* If we're returning an error, return who blocked us. */
994 if (!NT_STATUS_IS_OK(ret
) && psmblctx
) {
995 *psmblctx
= lock
.context
.smblctx
;
1000 static void brl_delete_lock_struct(struct lock_struct
*locks
,
1004 if (del_idx
>= num_locks
) {
1007 memmove(&locks
[del_idx
], &locks
[del_idx
+1],
1008 sizeof(*locks
) * (num_locks
- del_idx
- 1));
1011 /****************************************************************************
1012 Unlock a range of bytes - Windows semantics.
1013 ****************************************************************************/
1015 bool brl_unlock_windows_default(struct messaging_context
*msg_ctx
,
1016 struct byte_range_lock
*br_lck
,
1017 const struct lock_struct
*plock
)
1020 struct lock_struct
*locks
= br_lck
->lock_data
;
1021 enum brl_type deleted_lock_type
= READ_LOCK
; /* shut the compiler up.... */
1023 SMB_ASSERT(plock
->lock_type
== UNLOCK_LOCK
);
1026 /* Delete write locks by preference... The lock list
1027 is sorted in the zero zero case. */
1029 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1030 struct lock_struct
*lock
= &locks
[i
];
1032 if (lock
->lock_type
== WRITE_LOCK
&&
1033 brl_same_context(&lock
->context
, &plock
->context
) &&
1034 lock
->fnum
== plock
->fnum
&&
1035 lock
->lock_flav
== WINDOWS_LOCK
&&
1036 lock
->start
== plock
->start
&&
1037 lock
->size
== plock
->size
) {
1039 /* found it - delete it */
1040 deleted_lock_type
= lock
->lock_type
;
1045 if (i
!= br_lck
->num_locks
) {
1046 /* We found it - don't search again. */
1047 goto unlock_continue
;
1051 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1052 struct lock_struct
*lock
= &locks
[i
];
1054 if (IS_PENDING_LOCK(lock
->lock_type
)) {
1058 /* Only remove our own locks that match in start, size, and flavour. */
1059 if (brl_same_context(&lock
->context
, &plock
->context
) &&
1060 lock
->fnum
== plock
->fnum
&&
1061 lock
->lock_flav
== WINDOWS_LOCK
&&
1062 lock
->start
== plock
->start
&&
1063 lock
->size
== plock
->size
) {
1064 deleted_lock_type
= lock
->lock_type
;
1069 if (i
== br_lck
->num_locks
) {
1070 /* we didn't find it */
1078 brl_delete_lock_struct(locks
, br_lck
->num_locks
, i
);
1079 br_lck
->num_locks
-= 1;
1080 br_lck
->modified
= True
;
1082 /* Unlock the underlying POSIX regions. */
1083 if(lp_posix_locking(br_lck
->fsp
->conn
->params
)) {
1084 release_posix_lock_windows_flavour(br_lck
->fsp
,
1093 /* Send unlock messages to any pending waiters that overlap. */
1094 for (j
=0; j
< br_lck
->num_locks
; j
++) {
1095 struct lock_struct
*pend_lock
= &locks
[j
];
1097 /* Ignore non-pending locks. */
1098 if (!IS_PENDING_LOCK(pend_lock
->lock_type
)) {
1102 /* We could send specific lock info here... */
1103 if (brl_pending_overlap(plock
, pend_lock
)) {
1104 DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1105 procid_str_static(&pend_lock
->context
.pid
)));
1107 messaging_send(msg_ctx
, pend_lock
->context
.pid
,
1108 MSG_SMB_UNLOCK
, &data_blob_null
);
1112 contend_level2_oplocks_end(br_lck
->fsp
, LEVEL2_CONTEND_WINDOWS_BRL
);
1116 /****************************************************************************
1117 Unlock a range of bytes - POSIX semantics.
1118 ****************************************************************************/
1120 static bool brl_unlock_posix(struct messaging_context
*msg_ctx
,
1121 struct byte_range_lock
*br_lck
,
1122 struct lock_struct
*plock
)
1124 unsigned int i
, j
, count
;
1125 struct lock_struct
*tp
;
1126 struct lock_struct
*locks
= br_lck
->lock_data
;
1127 bool overlap_found
= False
;
1129 /* No zero-zero locks for POSIX. */
1130 if (plock
->start
== 0 && plock
->size
== 0) {
1134 /* Don't allow 64-bit lock wrap. */
1135 if (plock
->start
+ plock
->size
< plock
->start
||
1136 plock
->start
+ plock
->size
< plock
->size
) {
1137 DEBUG(10,("brl_unlock_posix: lock wrap\n"));
1141 /* The worst case scenario here is we have to split an
1142 existing POSIX lock range into two, so we need at most
1145 tp
= talloc_array(br_lck
, struct lock_struct
, br_lck
->num_locks
+ 1);
1147 DEBUG(10,("brl_unlock_posix: malloc fail\n"));
1152 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1153 struct lock_struct
*lock
= &locks
[i
];
1154 unsigned int tmp_count
;
1156 /* Only remove our own locks - ignore fnum. */
1157 if (IS_PENDING_LOCK(lock
->lock_type
) ||
1158 !brl_same_context(&lock
->context
, &plock
->context
)) {
1159 memcpy(&tp
[count
], lock
, sizeof(struct lock_struct
));
1164 if (lock
->lock_flav
== WINDOWS_LOCK
) {
1165 /* Do any Windows flavour locks conflict ? */
1166 if (brl_conflict(lock
, plock
)) {
1170 /* Just copy the Windows lock into the new array. */
1171 memcpy(&tp
[count
], lock
, sizeof(struct lock_struct
));
1176 /* Work out overlaps. */
1177 tmp_count
= brlock_posix_split_merge(&tp
[count
], lock
, plock
);
1179 if (tmp_count
== 0) {
1180 /* plock overlapped the existing lock completely,
1181 or replaced it. Don't copy the existing lock. */
1182 overlap_found
= true;
1183 } else if (tmp_count
== 1) {
1184 /* Either no overlap, (simple copy of existing lock) or
1185 * an overlap of an existing lock. */
1186 /* If the lock changed size, we had an overlap. */
1187 if (tp
[count
].size
!= lock
->size
) {
1188 overlap_found
= true;
1191 } else if (tmp_count
== 2) {
1192 /* We split a lock range in two. */
1193 overlap_found
= true;
1196 /* Optimisation... */
1197 /* We know we're finished here as we can't overlap any
1198 more POSIX locks. Copy the rest of the lock array. */
1200 if (i
< br_lck
->num_locks
- 1) {
1201 memcpy(&tp
[count
], &locks
[i
+1],
1202 sizeof(*locks
)*((br_lck
->num_locks
-1) - i
));
1203 count
+= ((br_lck
->num_locks
-1) - i
);
1210 if (!overlap_found
) {
1211 /* Just ignore - no change. */
1213 DEBUG(10,("brl_unlock_posix: No overlap - unlocked.\n"));
1217 /* Unlock any POSIX regions. */
1218 if(lp_posix_locking(br_lck
->fsp
->conn
->params
)) {
1219 release_posix_lock_posix_flavour(br_lck
->fsp
,
1227 /* Realloc so we don't leak entries per unlock call. */
1229 tp
= talloc_realloc(br_lck
, tp
, struct lock_struct
, count
);
1231 DEBUG(10,("brl_unlock_posix: realloc fail\n"));
1235 /* We deleted the last lock. */
1240 contend_level2_oplocks_end(br_lck
->fsp
,
1241 LEVEL2_CONTEND_POSIX_BRL
);
1243 br_lck
->num_locks
= count
;
1244 TALLOC_FREE(br_lck
->lock_data
);
1246 br_lck
->lock_data
= tp
;
1247 br_lck
->modified
= True
;
1249 /* Send unlock messages to any pending waiters that overlap. */
1251 for (j
=0; j
< br_lck
->num_locks
; j
++) {
1252 struct lock_struct
*pend_lock
= &locks
[j
];
1254 /* Ignore non-pending locks. */
1255 if (!IS_PENDING_LOCK(pend_lock
->lock_type
)) {
1259 /* We could send specific lock info here... */
1260 if (brl_pending_overlap(plock
, pend_lock
)) {
1261 DEBUG(10,("brl_unlock: sending unlock message to pid %s\n",
1262 procid_str_static(&pend_lock
->context
.pid
)));
1264 messaging_send(msg_ctx
, pend_lock
->context
.pid
,
1265 MSG_SMB_UNLOCK
, &data_blob_null
);
1272 bool smb_vfs_call_brl_unlock_windows(struct vfs_handle_struct
*handle
,
1273 struct messaging_context
*msg_ctx
,
1274 struct byte_range_lock
*br_lck
,
1275 const struct lock_struct
*plock
)
1277 VFS_FIND(brl_unlock_windows
);
1278 return handle
->fns
->brl_unlock_windows_fn(handle
, msg_ctx
, br_lck
,
1282 /****************************************************************************
1283 Unlock a range of bytes.
1284 ****************************************************************************/
1286 bool brl_unlock(struct messaging_context
*msg_ctx
,
1287 struct byte_range_lock
*br_lck
,
1289 struct server_id pid
,
1292 enum brl_flavour lock_flav
)
1294 struct lock_struct lock
;
1296 lock
.context
.smblctx
= smblctx
;
1297 lock
.context
.pid
= pid
;
1298 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1301 lock
.fnum
= br_lck
->fsp
->fnum
;
1302 lock
.lock_type
= UNLOCK_LOCK
;
1303 lock
.lock_flav
= lock_flav
;
1305 if (lock_flav
== WINDOWS_LOCK
) {
1306 return SMB_VFS_BRL_UNLOCK_WINDOWS(br_lck
->fsp
->conn
, msg_ctx
,
1309 return brl_unlock_posix(msg_ctx
, br_lck
, &lock
);
1313 /****************************************************************************
1314 Test if we could add a lock if we wanted to.
1315 Returns True if the region required is currently unlocked, False if locked.
1316 ****************************************************************************/
1318 bool brl_locktest(struct byte_range_lock
*br_lck
,
1320 struct server_id pid
,
1323 enum brl_type lock_type
,
1324 enum brl_flavour lock_flav
)
1328 struct lock_struct lock
;
1329 const struct lock_struct
*locks
= br_lck
->lock_data
;
1330 files_struct
*fsp
= br_lck
->fsp
;
1332 lock
.context
.smblctx
= smblctx
;
1333 lock
.context
.pid
= pid
;
1334 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1337 lock
.fnum
= fsp
->fnum
;
1338 lock
.lock_type
= lock_type
;
1339 lock
.lock_flav
= lock_flav
;
1341 /* Make sure existing locks don't conflict */
1342 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1344 * Our own locks don't conflict.
1346 if (brl_conflict_other(&locks
[i
], &lock
)) {
1352 * There is no lock held by an SMB daemon, check to
1353 * see if there is a POSIX lock from a UNIX or NFS process.
1354 * This only conflicts with Windows locks, not POSIX locks.
1357 if(lp_posix_locking(fsp
->conn
->params
) && (lock_flav
== WINDOWS_LOCK
)) {
1358 ret
= is_posix_locked(fsp
, &start
, &size
, &lock_type
, WINDOWS_LOCK
);
1360 DEBUG(10, ("brl_locktest: posix start=%ju len=%ju %s for %s "
1361 "file %s\n", (uintmax_t)start
, (uintmax_t)size
,
1362 ret
? "locked" : "unlocked",
1363 fsp_fnum_dbg(fsp
), fsp_str_dbg(fsp
)));
1365 /* We need to return the inverse of is_posix_locked. */
1369 /* no conflicts - we could have added it */
1373 /****************************************************************************
1374 Query for existing locks.
1375 ****************************************************************************/
1377 NTSTATUS
brl_lockquery(struct byte_range_lock
*br_lck
,
1379 struct server_id pid
,
1382 enum brl_type
*plock_type
,
1383 enum brl_flavour lock_flav
)
1386 struct lock_struct lock
;
1387 const struct lock_struct
*locks
= br_lck
->lock_data
;
1388 files_struct
*fsp
= br_lck
->fsp
;
1390 lock
.context
.smblctx
= *psmblctx
;
1391 lock
.context
.pid
= pid
;
1392 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1393 lock
.start
= *pstart
;
1395 lock
.fnum
= fsp
->fnum
;
1396 lock
.lock_type
= *plock_type
;
1397 lock
.lock_flav
= lock_flav
;
1399 /* Make sure existing locks don't conflict */
1400 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1401 const struct lock_struct
*exlock
= &locks
[i
];
1402 bool conflict
= False
;
1404 if (exlock
->lock_flav
== WINDOWS_LOCK
) {
1405 conflict
= brl_conflict(exlock
, &lock
);
1407 conflict
= brl_conflict_posix(exlock
, &lock
);
1411 *psmblctx
= exlock
->context
.smblctx
;
1412 *pstart
= exlock
->start
;
1413 *psize
= exlock
->size
;
1414 *plock_type
= exlock
->lock_type
;
1415 return NT_STATUS_LOCK_NOT_GRANTED
;
1420 * There is no lock held by an SMB daemon, check to
1421 * see if there is a POSIX lock from a UNIX or NFS process.
1424 if(lp_posix_locking(fsp
->conn
->params
)) {
1425 bool ret
= is_posix_locked(fsp
, pstart
, psize
, plock_type
, POSIX_LOCK
);
1427 DEBUG(10, ("brl_lockquery: posix start=%ju len=%ju %s for %s "
1428 "file %s\n", (uintmax_t)*pstart
,
1429 (uintmax_t)*psize
, ret
? "locked" : "unlocked",
1430 fsp_fnum_dbg(fsp
), fsp_str_dbg(fsp
)));
1433 /* Hmmm. No clue what to set smblctx to - use -1. */
1434 *psmblctx
= 0xFFFFFFFFFFFFFFFFLL
;
1435 return NT_STATUS_LOCK_NOT_GRANTED
;
1439 return NT_STATUS_OK
;
1443 bool smb_vfs_call_brl_cancel_windows(struct vfs_handle_struct
*handle
,
1444 struct byte_range_lock
*br_lck
,
1445 struct lock_struct
*plock
)
1447 VFS_FIND(brl_cancel_windows
);
1448 return handle
->fns
->brl_cancel_windows_fn(handle
, br_lck
, plock
);
1451 /****************************************************************************
1452 Remove a particular pending lock.
1453 ****************************************************************************/
1454 bool brl_lock_cancel(struct byte_range_lock
*br_lck
,
1456 struct server_id pid
,
1459 enum brl_flavour lock_flav
)
1462 struct lock_struct lock
;
1464 lock
.context
.smblctx
= smblctx
;
1465 lock
.context
.pid
= pid
;
1466 lock
.context
.tid
= br_lck
->fsp
->conn
->cnum
;
1469 lock
.fnum
= br_lck
->fsp
->fnum
;
1470 lock
.lock_flav
= lock_flav
;
1471 /* lock.lock_type doesn't matter */
1473 if (lock_flav
== WINDOWS_LOCK
) {
1474 ret
= SMB_VFS_BRL_CANCEL_WINDOWS(br_lck
->fsp
->conn
, br_lck
,
1477 ret
= brl_lock_cancel_default(br_lck
, &lock
);
1483 bool brl_lock_cancel_default(struct byte_range_lock
*br_lck
,
1484 struct lock_struct
*plock
)
1487 struct lock_struct
*locks
= br_lck
->lock_data
;
1491 for (i
= 0; i
< br_lck
->num_locks
; i
++) {
1492 struct lock_struct
*lock
= &locks
[i
];
1494 /* For pending locks we *always* care about the fnum. */
1495 if (brl_same_context(&lock
->context
, &plock
->context
) &&
1496 lock
->fnum
== plock
->fnum
&&
1497 IS_PENDING_LOCK(lock
->lock_type
) &&
1498 lock
->lock_flav
== plock
->lock_flav
&&
1499 lock
->start
== plock
->start
&&
1500 lock
->size
== plock
->size
) {
1505 if (i
== br_lck
->num_locks
) {
1506 /* Didn't find it. */
1510 brl_delete_lock_struct(locks
, br_lck
->num_locks
, i
);
1511 br_lck
->num_locks
-= 1;
1512 br_lck
->modified
= True
;
1516 /****************************************************************************
1517 Remove any locks associated with a open file.
1518 We return True if this process owns any other Windows locks on this
1519 fd and so we should not immediately close the fd.
1520 ****************************************************************************/
1522 void brl_close_fnum(struct messaging_context
*msg_ctx
,
1523 struct byte_range_lock
*br_lck
)
1525 files_struct
*fsp
= br_lck
->fsp
;
1526 uint32_t tid
= fsp
->conn
->cnum
;
1527 uint64_t fnum
= fsp
->fnum
;
1529 struct lock_struct
*locks
= br_lck
->lock_data
;
1530 struct server_id pid
= messaging_server_id(fsp
->conn
->sconn
->msg_ctx
);
1531 struct lock_struct
*locks_copy
;
1532 unsigned int num_locks_copy
;
1534 /* Copy the current lock array. */
1535 if (br_lck
->num_locks
) {
1536 locks_copy
= (struct lock_struct
*)talloc_memdup(br_lck
, locks
, br_lck
->num_locks
* sizeof(struct lock_struct
));
1538 smb_panic("brl_close_fnum: talloc failed");
1544 num_locks_copy
= br_lck
->num_locks
;
1546 for (i
=0; i
< num_locks_copy
; i
++) {
1547 struct lock_struct
*lock
= &locks_copy
[i
];
1549 if (lock
->context
.tid
== tid
&& serverid_equal(&lock
->context
.pid
, &pid
) &&
1550 (lock
->fnum
== fnum
)) {
1553 lock
->context
.smblctx
,
1562 bool brl_mark_disconnected(struct files_struct
*fsp
)
1564 uint32_t tid
= fsp
->conn
->cnum
;
1566 uint64_t fnum
= fsp
->fnum
;
1568 struct server_id self
= messaging_server_id(fsp
->conn
->sconn
->msg_ctx
);
1569 struct byte_range_lock
*br_lck
= NULL
;
1571 if (fsp
->op
== NULL
) {
1575 smblctx
= fsp
->op
->global
->open_persistent_id
;
1577 if (!fsp
->op
->global
->durable
) {
1581 if (fsp
->current_lock_count
== 0) {
1585 br_lck
= brl_get_locks(talloc_tos(), fsp
);
1586 if (br_lck
== NULL
) {
1590 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1591 struct lock_struct
*lock
= &br_lck
->lock_data
[i
];
1594 * as this is a durable handle, we only expect locks
1595 * of the current file handle!
1598 if (lock
->context
.smblctx
!= smblctx
) {
1599 TALLOC_FREE(br_lck
);
1603 if (lock
->context
.tid
!= tid
) {
1604 TALLOC_FREE(br_lck
);
1608 if (!serverid_equal(&lock
->context
.pid
, &self
)) {
1609 TALLOC_FREE(br_lck
);
1613 if (lock
->fnum
!= fnum
) {
1614 TALLOC_FREE(br_lck
);
1618 server_id_set_disconnected(&lock
->context
.pid
);
1619 lock
->context
.tid
= TID_FIELD_INVALID
;
1620 lock
->fnum
= FNUM_FIELD_INVALID
;
1623 br_lck
->modified
= true;
1624 TALLOC_FREE(br_lck
);
1628 bool brl_reconnect_disconnected(struct files_struct
*fsp
)
1630 uint32_t tid
= fsp
->conn
->cnum
;
1632 uint64_t fnum
= fsp
->fnum
;
1634 struct server_id self
= messaging_server_id(fsp
->conn
->sconn
->msg_ctx
);
1635 struct byte_range_lock
*br_lck
= NULL
;
1637 if (fsp
->op
== NULL
) {
1641 smblctx
= fsp
->op
->global
->open_persistent_id
;
1643 if (!fsp
->op
->global
->durable
) {
1648 * When reconnecting, we do not want to validate the brlock entries
1649 * and thereby remove our own (disconnected) entries but reactivate
1652 fsp
->lockdb_clean
= true;
1654 br_lck
= brl_get_locks(talloc_tos(), fsp
);
1655 if (br_lck
== NULL
) {
1659 if (br_lck
->num_locks
== 0) {
1660 TALLOC_FREE(br_lck
);
1664 for (i
=0; i
< br_lck
->num_locks
; i
++) {
1665 struct lock_struct
*lock
= &br_lck
->lock_data
[i
];
1668 * as this is a durable handle we only expect locks
1669 * of the current file handle!
1672 if (lock
->context
.smblctx
!= smblctx
) {
1673 TALLOC_FREE(br_lck
);
1677 if (lock
->context
.tid
!= TID_FIELD_INVALID
) {
1678 TALLOC_FREE(br_lck
);
1682 if (!server_id_is_disconnected(&lock
->context
.pid
)) {
1683 TALLOC_FREE(br_lck
);
1687 if (lock
->fnum
!= FNUM_FIELD_INVALID
) {
1688 TALLOC_FREE(br_lck
);
1692 lock
->context
.pid
= self
;
1693 lock
->context
.tid
= tid
;
1697 fsp
->current_lock_count
= br_lck
->num_locks
;
1698 br_lck
->modified
= true;
1699 TALLOC_FREE(br_lck
);
1703 /****************************************************************************
1704 Ensure this set of lock entries is valid.
1705 ****************************************************************************/
1706 static bool validate_lock_entries(unsigned int *pnum_entries
, struct lock_struct
**pplocks
,
1707 bool keep_disconnected
)
1710 struct lock_struct
*locks
= *pplocks
;
1711 unsigned int num_entries
= *pnum_entries
;
1713 struct server_id
*ids
;
1716 if (num_entries
== 0) {
1720 frame
= talloc_stackframe();
1722 ids
= talloc_array(frame
, struct server_id
, num_entries
);
1724 DEBUG(0, ("validate_lock_entries: "
1725 "talloc_array(struct server_id, %u) failed\n",
1731 exists
= talloc_array(frame
, bool, num_entries
);
1732 if (exists
== NULL
) {
1733 DEBUG(0, ("validate_lock_entries: "
1734 "talloc_array(bool, %u) failed\n",
1740 for (i
= 0; i
< num_entries
; i
++) {
1741 ids
[i
] = locks
[i
].context
.pid
;
1744 if (!serverids_exist(ids
, num_entries
, exists
)) {
1745 DEBUG(3, ("validate_lock_entries: serverids_exists failed\n"));
1752 while (i
< num_entries
) {
1758 if (keep_disconnected
&&
1759 server_id_is_disconnected(&ids
[i
]))
1765 /* This process no longer exists */
1767 brl_delete_lock_struct(locks
, num_entries
, i
);
1772 *pnum_entries
= num_entries
;
1777 struct brl_forall_cb
{
1778 void (*fn
)(struct file_id id
, struct server_id pid
,
1779 enum brl_type lock_type
,
1780 enum brl_flavour lock_flav
,
1781 br_off start
, br_off size
,
1782 void *private_data
);
1786 /****************************************************************************
1787 Traverse the whole database with this function, calling traverse_callback
1789 ****************************************************************************/
1791 static int brl_traverse_fn(struct db_record
*rec
, void *state
)
1793 struct brl_forall_cb
*cb
= (struct brl_forall_cb
*)state
;
1794 struct lock_struct
*locks
;
1795 struct file_id
*key
;
1797 unsigned int num_locks
= 0;
1798 unsigned int orig_num_locks
= 0;
1802 dbkey
= dbwrap_record_get_key(rec
);
1803 value
= dbwrap_record_get_value(rec
);
1805 /* In a traverse function we must make a copy of
1806 dbuf before modifying it. */
1808 locks
= (struct lock_struct
*)talloc_memdup(
1809 talloc_tos(), value
.dptr
, value
.dsize
);
1811 return -1; /* Terminate traversal. */
1814 key
= (struct file_id
*)dbkey
.dptr
;
1815 orig_num_locks
= num_locks
= value
.dsize
/sizeof(*locks
);
1817 /* Ensure the lock db is clean of entries from invalid processes. */
1819 if (!validate_lock_entries(&num_locks
, &locks
, true)) {
1821 return -1; /* Terminate traversal */
1824 if (orig_num_locks
!= num_locks
) {
1827 data
.dptr
= (uint8_t *)locks
;
1828 data
.dsize
= num_locks
*sizeof(struct lock_struct
);
1829 dbwrap_record_store(rec
, data
, TDB_REPLACE
);
1831 dbwrap_record_delete(rec
);
1836 for ( i
=0; i
<num_locks
; i
++) {
1838 locks
[i
].context
.pid
,
1851 /*******************************************************************
1852 Call the specified function on each lock in the database.
1853 ********************************************************************/
1855 int brl_forall(void (*fn
)(struct file_id id
, struct server_id pid
,
1856 enum brl_type lock_type
,
1857 enum brl_flavour lock_flav
,
1858 br_off start
, br_off size
,
1859 void *private_data
),
1862 struct brl_forall_cb cb
;
1870 cb
.private_data
= private_data
;
1871 status
= dbwrap_traverse(brlock_db
, brl_traverse_fn
, &cb
, &count
);
1873 if (!NT_STATUS_IS_OK(status
)) {
1880 /*******************************************************************
1881 Store a potentially modified set of byte range lock data back into
1884 ********************************************************************/
1886 static void byte_range_lock_flush(struct byte_range_lock
*br_lck
)
1889 if (!br_lck
->modified
) {
1890 DEBUG(10, ("br_lck not modified\n"));
1894 data_len
= br_lck
->num_locks
* sizeof(struct lock_struct
);
1896 if (br_lck
->have_read_oplocks
) {
1900 DEBUG(10, ("data_len=%d\n", (int)data_len
));
1902 if (data_len
== 0) {
1903 /* No locks - delete this entry. */
1904 NTSTATUS status
= dbwrap_record_delete(br_lck
->record
);
1905 if (!NT_STATUS_IS_OK(status
)) {
1906 DEBUG(0, ("delete_rec returned %s\n",
1907 nt_errstr(status
)));
1908 smb_panic("Could not delete byte range lock entry");
1914 data
.dsize
= data_len
;
1915 data
.dptr
= talloc_array(talloc_tos(), uint8_t, data_len
);
1916 SMB_ASSERT(data
.dptr
!= NULL
);
1918 memcpy(data
.dptr
, br_lck
->lock_data
,
1919 br_lck
->num_locks
* sizeof(struct lock_struct
));
1921 if (br_lck
->have_read_oplocks
) {
1922 data
.dptr
[data_len
-1] = 1;
1925 status
= dbwrap_record_store(br_lck
->record
, data
, TDB_REPLACE
);
1926 TALLOC_FREE(data
.dptr
);
1927 if (!NT_STATUS_IS_OK(status
)) {
1928 DEBUG(0, ("store returned %s\n", nt_errstr(status
)));
1929 smb_panic("Could not store byte range mode entry");
1933 DEBUG(10, ("seqnum=%d\n", dbwrap_get_seqnum(brlock_db
)));
1936 br_lck
->modified
= false;
1937 TALLOC_FREE(br_lck
->record
);
1940 static int byte_range_lock_destructor(struct byte_range_lock
*br_lck
)
1942 byte_range_lock_flush(br_lck
);
1946 /*******************************************************************
1947 Fetch a set of byte range lock data from the database.
1948 Leave the record locked.
1949 TALLOC_FREE(brl) will release the lock in the destructor.
1950 ********************************************************************/
1952 struct byte_range_lock
*brl_get_locks(TALLOC_CTX
*mem_ctx
, files_struct
*fsp
)
1955 struct byte_range_lock
*br_lck
= talloc(mem_ctx
, struct byte_range_lock
);
1957 if (br_lck
== NULL
) {
1962 br_lck
->num_locks
= 0;
1963 br_lck
->have_read_oplocks
= false;
1964 br_lck
->modified
= False
;
1966 key
.dptr
= (uint8
*)&fsp
->file_id
;
1967 key
.dsize
= sizeof(struct file_id
);
1969 br_lck
->record
= dbwrap_fetch_locked(brlock_db
, br_lck
, key
);
1971 if (br_lck
->record
== NULL
) {
1972 DEBUG(3, ("Could not lock byte range lock entry\n"));
1973 TALLOC_FREE(br_lck
);
1977 data
= dbwrap_record_get_value(br_lck
->record
);
1979 br_lck
->lock_data
= NULL
;
1981 talloc_set_destructor(br_lck
, byte_range_lock_destructor
);
1983 br_lck
->num_locks
= data
.dsize
/ sizeof(struct lock_struct
);
1985 if (br_lck
->num_locks
!= 0) {
1986 br_lck
->lock_data
= talloc_array(
1987 br_lck
, struct lock_struct
, br_lck
->num_locks
);
1988 if (br_lck
->lock_data
== NULL
) {
1989 DEBUG(0, ("malloc failed\n"));
1990 TALLOC_FREE(br_lck
);
1994 memcpy(br_lck
->lock_data
, data
.dptr
,
1995 talloc_get_size(br_lck
->lock_data
));
1998 DEBUG(10, ("data.dsize=%d\n", (int)data
.dsize
));
2000 if ((data
.dsize
% sizeof(struct lock_struct
)) == 1) {
2001 br_lck
->have_read_oplocks
= (data
.dptr
[data
.dsize
-1] == 1);
2004 if (!fsp
->lockdb_clean
) {
2005 int orig_num_locks
= br_lck
->num_locks
;
2008 * This is the first time we access the byte range lock
2009 * record with this fsp. Go through and ensure all entries
2010 * are valid - remove any that don't.
2011 * This makes the lockdb self cleaning at low cost.
2013 * Note: Disconnected entries belong to disconnected
2014 * durable handles. So at this point, we have a new
2015 * handle on the file and the disconnected durable has
2016 * already been closed (we are not a durable reconnect).
2017 * So we need to clean the disconnected brl entry.
2020 if (!validate_lock_entries(&br_lck
->num_locks
,
2021 &br_lck
->lock_data
, false)) {
2022 TALLOC_FREE(br_lck
);
2026 /* Ensure invalid locks are cleaned up in the destructor. */
2027 if (orig_num_locks
!= br_lck
->num_locks
) {
2028 br_lck
->modified
= True
;
2031 /* Mark the lockdb as "clean" as seen from this open file. */
2032 fsp
->lockdb_clean
= True
;
2035 if (DEBUGLEVEL
>= 10) {
2037 struct lock_struct
*locks
= br_lck
->lock_data
;
2038 DEBUG(10,("brl_get_locks_internal: %u current locks on file_id %s\n",
2040 file_id_string_tos(&fsp
->file_id
)));
2041 for( i
= 0; i
< br_lck
->num_locks
; i
++) {
2042 print_lock_struct(i
, &locks
[i
]);
2049 struct brl_get_locks_readonly_state
{
2050 TALLOC_CTX
*mem_ctx
;
2051 struct byte_range_lock
**br_lock
;
2054 static void brl_get_locks_readonly_parser(TDB_DATA key
, TDB_DATA data
,
2057 struct brl_get_locks_readonly_state
*state
=
2058 (struct brl_get_locks_readonly_state
*)private_data
;
2059 struct byte_range_lock
*br_lock
;
2061 br_lock
= talloc_pooled_object(
2062 state
->mem_ctx
, struct byte_range_lock
, 1, data
.dsize
);
2063 if (br_lock
== NULL
) {
2064 *state
->br_lock
= NULL
;
2067 br_lock
->lock_data
= (struct lock_struct
*)talloc_memdup(
2068 br_lock
, data
.dptr
, data
.dsize
);
2069 br_lock
->num_locks
= data
.dsize
/ sizeof(struct lock_struct
);
2071 if ((data
.dsize
% sizeof(struct lock_struct
)) == 1) {
2072 br_lock
->have_read_oplocks
= (data
.dptr
[data
.dsize
-1] == 1);
2074 br_lock
->have_read_oplocks
= false;
2077 DEBUG(10, ("Got %d bytes, have_read_oplocks: %s\n", (int)data
.dsize
,
2078 br_lock
->have_read_oplocks
? "true" : "false"));
2080 *state
->br_lock
= br_lock
;
2083 struct byte_range_lock
*brl_get_locks_readonly(files_struct
*fsp
)
2085 struct byte_range_lock
*br_lock
= NULL
;
2086 struct byte_range_lock
*rw
= NULL
;
2088 DEBUG(10, ("seqnum=%d, fsp->brlock_seqnum=%d\n",
2089 dbwrap_get_seqnum(brlock_db
), fsp
->brlock_seqnum
));
2091 if ((fsp
->brlock_rec
!= NULL
)
2092 && (dbwrap_get_seqnum(brlock_db
) == fsp
->brlock_seqnum
)) {
2094 * We have cached the brlock_rec and the database did not
2097 return fsp
->brlock_rec
;
2100 if (!fsp
->lockdb_clean
) {
2102 * Fetch the record in R/W mode to give validate_lock_entries
2103 * a chance to kick in once.
2105 rw
= brl_get_locks(talloc_tos(), fsp
);
2109 fsp
->lockdb_clean
= true;
2113 size_t lock_data_size
;
2116 * Make a copy of the already retrieved and sanitized rw record
2118 lock_data_size
= rw
->num_locks
* sizeof(struct lock_struct
);
2119 br_lock
= talloc_pooled_object(
2120 fsp
, struct byte_range_lock
, 1, lock_data_size
);
2121 if (br_lock
== NULL
) {
2124 br_lock
->have_read_oplocks
= rw
->have_read_oplocks
;
2125 br_lock
->num_locks
= rw
->num_locks
;
2126 br_lock
->lock_data
= (struct lock_struct
*)talloc_memdup(
2127 br_lock
, rw
->lock_data
, lock_data_size
);
2129 struct brl_get_locks_readonly_state state
;
2133 * Parse the record fresh from the database
2136 state
.mem_ctx
= fsp
;
2137 state
.br_lock
= &br_lock
;
2139 status
= dbwrap_parse_record(
2141 make_tdb_data((uint8_t *)&fsp
->file_id
,
2142 sizeof(fsp
->file_id
)),
2143 brl_get_locks_readonly_parser
, &state
);
2145 if (NT_STATUS_EQUAL(status
,NT_STATUS_NOT_FOUND
)) {
2147 * No locks on this file. Return an empty br_lock.
2149 br_lock
= talloc(fsp
, struct byte_range_lock
);
2150 if (br_lock
== NULL
) {
2154 br_lock
->have_read_oplocks
= false;
2155 br_lock
->num_locks
= 0;
2156 br_lock
->lock_data
= NULL
;
2158 } else if (!NT_STATUS_IS_OK(status
)) {
2159 DEBUG(3, ("Could not parse byte range lock record: "
2160 "%s\n", nt_errstr(status
)));
2163 if (br_lock
== NULL
) {
2169 br_lock
->modified
= false;
2170 br_lock
->record
= NULL
;
2172 if (lp_clustering()) {
2174 * In the cluster case we can't cache the brlock struct
2175 * because dbwrap_get_seqnum does not work reliably over
2176 * ctdb. Thus we have to throw away the brlock struct soon.
2178 talloc_steal(talloc_tos(), br_lock
);
2181 * Cache the brlock struct, invalidated when the dbwrap_seqnum
2182 * changes. See beginning of this routine.
2184 TALLOC_FREE(fsp
->brlock_rec
);
2185 fsp
->brlock_rec
= br_lock
;
2186 fsp
->brlock_seqnum
= dbwrap_get_seqnum(brlock_db
);
2194 struct brl_revalidate_state
{
2197 struct server_id
*pids
;
2201 * Collect PIDs of all processes with pending entries
2204 static void brl_revalidate_collect(struct file_id id
, struct server_id pid
,
2205 enum brl_type lock_type
,
2206 enum brl_flavour lock_flav
,
2207 br_off start
, br_off size
,
2210 struct brl_revalidate_state
*state
=
2211 (struct brl_revalidate_state
*)private_data
;
2213 if (!IS_PENDING_LOCK(lock_type
)) {
2217 add_to_large_array(state
, sizeof(pid
), (void *)&pid
,
2218 &state
->pids
, &state
->num_pids
,
2219 &state
->array_size
);
2223 * qsort callback to sort the processes
2226 static int compare_procids(const void *p1
, const void *p2
)
2228 const struct server_id
*i1
= (const struct server_id
*)p1
;
2229 const struct server_id
*i2
= (const struct server_id
*)p2
;
2231 if (i1
->pid
< i2
->pid
) return -1;
2232 if (i1
->pid
> i2
->pid
) return 1;
2237 * Send a MSG_SMB_UNLOCK message to all processes with pending byte range
2238 * locks so that they retry. Mainly used in the cluster code after a node has
2241 * Done in two steps to avoid double-sends: First we collect all entries in an
2242 * array, then qsort that array and only send to non-dupes.
2245 void brl_revalidate(struct messaging_context
*msg_ctx
,
2248 struct server_id server_id
,
2251 struct brl_revalidate_state
*state
;
2253 struct server_id last_pid
;
2255 if (!(state
= talloc_zero(NULL
, struct brl_revalidate_state
))) {
2256 DEBUG(0, ("talloc failed\n"));
2260 brl_forall(brl_revalidate_collect
, state
);
2262 if (state
->array_size
== -1) {
2263 DEBUG(0, ("talloc failed\n"));
2267 if (state
->num_pids
== 0) {
2271 TYPESAFE_QSORT(state
->pids
, state
->num_pids
, compare_procids
);
2273 ZERO_STRUCT(last_pid
);
2275 for (i
=0; i
<state
->num_pids
; i
++) {
2276 if (serverid_equal(&last_pid
, &state
->pids
[i
])) {
2278 * We've seen that one already
2283 messaging_send(msg_ctx
, state
->pids
[i
], MSG_SMB_UNLOCK
,
2285 last_pid
= state
->pids
[i
];
2293 bool brl_cleanup_disconnected(struct file_id fid
, uint64_t open_persistent_id
)
2296 TALLOC_CTX
*frame
= talloc_stackframe();
2298 struct db_record
*rec
;
2299 struct lock_struct
*lock
;
2303 key
= make_tdb_data((void*)&fid
, sizeof(fid
));
2305 rec
= dbwrap_fetch_locked(brlock_db
, frame
, key
);
2307 DEBUG(5, ("brl_cleanup_disconnected: failed to fetch record "
2308 "for file %s\n", file_id_string(frame
, &fid
)));
2312 val
= dbwrap_record_get_value(rec
);
2313 lock
= (struct lock_struct
*)val
.dptr
;
2314 num
= val
.dsize
/ sizeof(struct lock_struct
);
2316 DEBUG(10, ("brl_cleanup_disconnected: no byte range locks for "
2317 "file %s\n", file_id_string(frame
, &fid
)));
2322 for (n
=0; n
<num
; n
++) {
2323 struct lock_context
*ctx
= &lock
[n
].context
;
2325 if (!server_id_is_disconnected(&ctx
->pid
)) {
2326 DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
2327 "%s used by server %s, do not cleanup\n",
2328 file_id_string(frame
, &fid
),
2329 server_id_str(frame
, &ctx
->pid
)));
2333 if (ctx
->smblctx
!= open_persistent_id
) {
2334 DEBUG(5, ("brl_cleanup_disconnected: byte range lock "
2335 "%s expected smblctx %llu but found %llu"
2336 ", do not cleanup\n",
2337 file_id_string(frame
, &fid
),
2338 (unsigned long long)open_persistent_id
,
2339 (unsigned long long)ctx
->smblctx
));
2344 status
= dbwrap_record_delete(rec
);
2345 if (!NT_STATUS_IS_OK(status
)) {
2346 DEBUG(5, ("brl_cleanup_disconnected: failed to delete record "
2347 "for file %s from %s, open %llu: %s\n",
2348 file_id_string(frame
, &fid
), dbwrap_name(brlock_db
),
2349 (unsigned long long)open_persistent_id
,
2350 nt_errstr(status
)));
2354 DEBUG(10, ("brl_cleanup_disconnected: "
2355 "file %s cleaned up %u entries from open %llu\n",
2356 file_id_string(frame
, &fid
), num
,
2357 (unsigned long long)open_persistent_id
));