2 Unix SMB/Netbios implementation.
5 Copyright (C) Jeremy Allison 1992-2000
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000.
27 extern int DEBUGLEVEL;
30 * The POSIX locking database handle.
33 static TDB_CONTEXT *posix_lock_tdb;
36 * The pending close database handle.
39 static TDB_CONTEXT *posix_pending_close_tdb;
42 * The data in POSIX lock records is an unsorted linear array of these
43 * records. It is unnecessary to store the count as tdb provides the
55 * The data in POSIX pending close records is an unsorted linear array of int
56 * records. It is unnecessary to store the count as tdb provides the
60 /* The key used in both the POSIX databases. */
62 struct posix_lock_key {
67 /*******************************************************************
68 Form a static locking key for a dev/inode pair.
69 ******************************************************************/
71 static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
73 static struct posix_lock_key key;
76 memset(&key, '\0', sizeof(key));
79 kbuf.dptr = (char *)&key;
80 kbuf.dsize = sizeof(key);
84 /*******************************************************************
85 Convenience function to get a key from an fsp.
86 ******************************************************************/
88 static TDB_DATA locking_key_fsp(files_struct *fsp)
90 return locking_key(fsp->dev, fsp->inode);
93 /****************************************************************************
94 Add an fd to the pending close tdb.
95 ****************************************************************************/
97 static BOOL add_fd_to_close_entry(files_struct *fsp)
99 TDB_DATA kbuf = locking_key_fsp(fsp);
105 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
107 tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(int));
109 DEBUG(0,("add_fd_to_close_entry: Realloc fail !\n"));
116 memcpy(dbuf.dptr + dbuf.dsize, &fsp->fd, sizeof(int));
117 dbuf.dsize += sizeof(int);
119 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
120 DEBUG(0,("add_fd_to_close_entry: tdb_store fail !\n"));
127 /****************************************************************************
128 Remove all fd entries for a specific dev/inode pair from the tdb.
129 ****************************************************************************/
131 static void delete_close_entries(files_struct *fsp)
133 TDB_DATA kbuf = locking_key_fsp(fsp);
135 if (tdb_delete(posix_pending_close_tdb, kbuf) == -1)
136 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
139 /****************************************************************************
140 Get the array of POSIX pending close records for an open fsp. Caller must
141 free. Returns number of entries.
142 ****************************************************************************/
144 static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
146 TDB_DATA kbuf = locking_key_fsp(fsp);
153 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
159 *entries = (int *)dbuf.dptr;
160 count = (size_t)(dbuf.dsize / sizeof(int));
165 /****************************************************************************
166 Get the array of POSIX locks for an fsp. Caller must free. Returns
168 ****************************************************************************/
170 static size_t get_posix_lock_entries(files_struct *fsp, struct posix_lock **entries)
172 TDB_DATA kbuf = locking_key_fsp(fsp);
180 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
186 *entries = (struct posix_lock *)dbuf.dptr;
187 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
192 /****************************************************************************
193 Deal with pending closes needed by POSIX locking support.
194 Note that posix_locking_close_file() is expected to have been called
195 to delete all locks on this fsp before this function is called.
196 ****************************************************************************/
198 int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
203 struct posix_lock *entries = NULL;
204 int *fd_array = NULL;
205 BOOL locks_on_other_fds = False;
207 if (!lp_posix_locking(SNUM(conn))) {
209 * No POSIX to worry about, just close.
211 ret = conn->vfs_ops.close(fsp,fsp->fd);
217 * Get the number of outstanding POSIX locks on this dev/inode pair.
220 count = get_posix_lock_entries(fsp, &entries);
223 * Check if there are any outstanding locks belonging to
224 * other fd's. This should never be the case if posix_locking_close_file()
225 * has been called first, but it never hurts to be *sure*.
228 for (i = 0; i < count; i++) {
229 if (entries[i].fd != fsp->fd) {
230 locks_on_other_fds = True;
235 if (locks_on_other_fds) {
238 * There are outstanding locks on this dev/inode pair on other fds.
239 * Add our fd to the pending close tdb and set fsp->fd to -1.
242 if (!add_fd_to_close_entry(fsp)) {
243 free((char *)entries);
247 free((char *)entries);
253 free((char *)entries);
256 * No outstanding POSIX locks. Get the pending close fd's
257 * from the tdb and close them all.
260 count = get_posix_pending_close_entries(fsp, &fd_array);
263 DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
265 for(i = 0; i < count; i++) {
266 if (conn->vfs_ops.close(fsp,fd_array[i]) == -1) {
272 * Delete all fd's stored in the tdb
273 * for this dev/inode pair.
276 delete_close_entries(fsp);
280 free((char *)fd_array);
283 * Finally close the fd associated with this fsp.
286 ret = conn->vfs_ops.close(fsp,fsp->fd);
288 if (saved_errno != 0) {
298 /****************************************************************************
300 ****************************************************************************/
302 static const char *posix_lock_type_name(int lock_type)
304 return (lock_type == F_RDLCK) ? "READ" : "WRITE";
307 /****************************************************************************
308 Delete a POSIX lock entry by index number. Used if the tdb add succeeds, but
309 then the POSIX fcntl lock fails.
310 ****************************************************************************/
312 static BOOL delete_posix_lock_entry_by_index(files_struct *fsp, size_t entry)
314 TDB_DATA kbuf = locking_key_fsp(fsp);
316 struct posix_lock *locks;
321 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
324 DEBUG(10,("delete_posix_lock_entry_by_index: tdb_fetch failed !\n"));
328 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
329 locks = (struct posix_lock *)dbuf.dptr;
332 tdb_delete(posix_lock_tdb, kbuf);
334 if (entry < count-1) {
335 memmove(&locks[entry], &locks[entry+1], sizeof(*locks)*((count-1) - entry));
337 dbuf.dsize -= sizeof(*locks);
338 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
351 /****************************************************************************
352 Add an entry into the POSIX locking tdb. We return the index number of the
353 added lock (used in case we need to delete *exactly* this entry). Returns
354 False on fail, True on success.
355 ****************************************************************************/
357 static BOOL add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *pentry_num)
359 TDB_DATA kbuf = locking_key_fsp(fsp);
361 struct posix_lock pl;
366 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
368 *pentry_num = (size_t)(dbuf.dsize / sizeof(pl));
377 pl.lock_type = lock_type;
379 tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(pl));
381 DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
386 memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
387 dbuf.dsize += sizeof(pl);
389 if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
390 DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
396 DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: dev=%.0f inode=%.0f\n",
397 fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size,
398 (double)fsp->dev, (double)fsp->inode ));
408 /****************************************************************************
409 Calculate if locks have any overlap at all.
410 ****************************************************************************/
412 static BOOL does_lock_overlap(SMB_OFF_T start1, SMB_OFF_T size1, SMB_OFF_T start2, SMB_OFF_T size2)
414 if (start1 >= start2 && start1 <= start2 + size2)
417 if (start1 < start2 && start1 + size1 > start2)
423 /****************************************************************************
424 Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
425 deleted and the number of records that are overlapped by this one, or -1 on error.
426 ****************************************************************************/
428 static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
430 TDB_DATA kbuf = locking_key_fsp(fsp);
432 struct posix_lock *locks;
435 int num_overlapping_records = 0;
439 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
442 DEBUG(10,("delete_posix_lock_entry: tdb_fetch failed !\n"));
446 /* There are existing locks - find a match. */
447 locks = (struct posix_lock *)dbuf.dptr;
448 count = (size_t)(dbuf.dsize / sizeof(*locks));
451 * Search for and delete the first record that matches the
455 for (i=0; i<count; i++) {
456 struct posix_lock *entry = &locks[i];
458 if (entry->fd == fsp->fd &&
459 entry->start == start &&
460 entry->size == size) {
462 /* Make a copy if requested. */
466 /* Found it - delete it. */
468 tdb_delete(posix_lock_tdb, kbuf);
471 memmove(&locks[i], &locks[i+1], sizeof(*locks)*((count-1) - i));
473 dbuf.dsize -= sizeof(*locks);
474 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
486 * Count the number of entries that are
487 * overlapped by this unlock request.
490 for (i = 0; i < count; i++) {
491 struct posix_lock *entry = &locks[i];
493 if (fsp->fd == entry->fd &&
494 does_lock_overlap( start, size, entry->start, entry->size))
495 num_overlapping_records++;
498 DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
499 posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
500 (unsigned int)num_overlapping_records ));
505 return num_overlapping_records;
513 /****************************************************************************
514 Utility function to map a lock type correctly depending on the open
516 ****************************************************************************/
518 static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
520 if((lock_type == WRITE_LOCK) && !fsp->can_write) {
522 * Many UNIX's cannot get a write lock on a file opened read-only.
523 * Win32 locking semantics allow this.
524 * Do the best we can and attempt a read-only lock.
526 DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
528 } else if((lock_type == READ_LOCK) && !fsp->can_read) {
530 * Ditto for read locks on write only files.
532 DEBUG(10,("map_posix_lock_type: Changing read lock to write due to write-only file.\n"));
537 * This return should be the most normal, as we attempt
538 * to always open files read/write.
541 return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
544 /****************************************************************************
545 Check to see if the given unsigned lock range is within the possible POSIX
546 range. Modifies the given args to be in range if possible, just returns
548 ****************************************************************************/
550 static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
551 SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
553 SMB_OFF_T offset = (SMB_OFF_T)u_offset;
554 SMB_OFF_T count = (SMB_OFF_T)u_count;
557 * For the type of system we are, attempt to
558 * find the maximum positive lock offset as an SMB_OFF_T.
561 #if defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
564 * In this case SMB_OFF_T is 64 bits,
565 * and the underlying system can handle 64 bit signed locks.
568 SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
569 SMB_OFF_T mask = (mask2<<1);
570 SMB_OFF_T max_positive_lock_offset = ~mask;
572 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
575 * In this case either SMB_OFF_T is 32 bits,
576 * or the underlying system cannot handle 64 bit signed locks.
577 * All offsets & counts must be 2^31 or less.
580 SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
582 #endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
585 * POSIX locks of length zero mean lock to end-of-file.
586 * Win32 locks of length zero are point probes. Ignore
587 * any Win32 locks of length zero. JRA.
590 if (count == (SMB_OFF_T)0) {
591 DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n"));
596 * If the given offset was > max_positive_lock_offset then we cannot map this at all
600 if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
601 DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
602 (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
607 * We must truncate the offset and count to less than max_positive_lock_offset.
610 offset &= max_positive_lock_offset;
611 count &= max_positive_lock_offset;
615 * Deal with a very common case of count of all ones.
616 * (lock entire file).
619 if(count == (SMB_OFF_T)-1)
620 count = max_positive_lock_offset;
623 * Truncate count to end at max lock offset.
626 if (offset + count < 0 || offset + count > max_positive_lock_offset)
627 count = max_positive_lock_offset - offset;
630 * If we ate all the count, ignore this lock.
634 DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
635 (double)u_offset, (double)u_count ));
640 * The mapping was successful.
643 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
644 (double)offset, (double)count ));
646 *offset_out = offset;
652 /****************************************************************************
653 Pathetically try and map a 64 bit lock offset into 31 bits. I hate Windows :-).
654 ****************************************************************************/
656 uint32 map_lock_offset(uint32 high, uint32 low)
660 uint32 highcopy = high;
663 * Try and find out how many significant bits there are in high.
666 for(i = 0; highcopy; i++)
670 * We use 31 bits not 32 here as POSIX
671 * lock offsets may not be negative.
674 mask = (~0) << (31 - i);
677 return 0; /* Fail. */
684 /****************************************************************************
685 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
686 broken NFS implementations.
687 ****************************************************************************/
689 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
692 struct connection_struct *conn = fsp->conn;
694 #if defined(LARGE_SMB_OFF_T)
696 * In the 64 bit locking case we store the original
697 * values in case we have to map to a 32 bit lock on
698 * a filesystem that doesn't support 64 bit locks.
700 SMB_OFF_T orig_offset = offset;
701 SMB_OFF_T orig_count = count;
702 #endif /* LARGE_SMB_OFF_T */
704 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
706 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
708 if (!ret && (errno == EFBIG)) {
710 dbgtext("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n", (double)offset,(double)count);
711 dbgtext("a 'file too large' error. This can happen when using 64 bit lock offsets\n");
712 dbgtext("on 32 bit NFS mounted file systems. Retrying with 32 bit truncated length.\n");
714 /* 32 bit NFS file system, retry with smaller offset */
717 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
720 /* A lock query - just return. */
721 if (op == SMB_F_GETLK)
724 /* A lock set or unset. */
726 DEBUG(3,("posix_fcntl_lock: lock failed at offset %.0f count %.0f op %d type %d (%s)\n",
727 (double)offset,(double)count,op,type,strerror(errno)));
729 /* Perhaps it doesn't support this sort of locking ? */
730 if (errno == EINVAL) {
731 #if defined(LARGE_SMB_OFF_T)
734 * Ok - if we get here then we have a 64 bit lock request
735 * that has returned EINVAL. Try and map to 31 bits for offset
736 * and length and try again. This may happen if a filesystem
737 * doesn't support 64 bit offsets (efs/ufs) although the underlying
740 uint32 off_low = (orig_offset & 0xFFFFFFFF);
741 uint32 off_high = ((orig_offset >> 32) & 0xFFFFFFFF);
743 count = (orig_count & 0x7FFFFFFF);
744 offset = (SMB_OFF_T)map_lock_offset(off_high, off_low);
745 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
747 if (errno == EINVAL) {
748 DEBUG(3,("posix_fcntl_lock: locking not supported? returning True\n"));
753 DEBUG(3,("posix_fcntl_lock: 64 -> 32 bit modified lock call successful\n"));
756 #else /* LARGE_SMB_OFF_T */
757 DEBUG(3,("locking not supported? returning True\n"));
759 #endif /* LARGE_SMB_OFF_T */
765 DEBUG(8,("posix_fcntl_lock: Lock call successful\n"));
770 /****************************************************************************
771 POSIX function to see if a file region is locked. Returns True if the
772 region is locked, False otherwise.
773 ****************************************************************************/
775 BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
779 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
781 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
782 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
785 * If the requested lock won't fit in the POSIX range, we will
786 * never set it, so presume it is not locked.
789 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
793 * Note that most UNIX's can *test* for a write lock on
794 * a read-only fd, just not *set* a write lock on a read-only
795 * fd. So we don't need to use map_lock_type here.
798 return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
802 * Structure used when splitting a lock range
803 * into a POSIX lock range. Doubly linked list.
807 struct lock_list *next;
808 struct lock_list *prev;
813 /****************************************************************************
814 Create a list of lock ranges that don't overlap a given range. Used in calculating
815 POSIX locks and unlocks. This is a difficult function that requires ASCII art to
817 ****************************************************************************/
819 static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
821 TDB_DATA kbuf = locking_key_fsp(fsp);
823 struct posix_lock *locks;
828 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
833 locks = (struct posix_lock *)dbuf.dptr;
834 num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
837 * Check the current lock list on this dev/inode pair.
838 * Quit if the list is deleted.
841 DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
842 (double)lhead->start, (double)lhead->size ));
844 for (i=0; i<num_locks && lhead; i++) {
846 struct posix_lock *lock = &locks[i];
847 struct lock_list *l_curr;
850 * Walk the lock list, checking for overlaps. Note that
851 * the lock list can expand within this loop if the current
852 * range being examined needs to be split.
855 for (l_curr = lhead; l_curr;) {
857 DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
858 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
860 if ( (l_curr->start >= (lock->start + lock->size)) ||
861 (lock->start >= (l_curr->start + l_curr->size))) {
863 /* No overlap with this lock - leave this range alone. */
864 /*********************************************
875 **********************************************/
877 DEBUG(10,("no overlap case.\n" ));
879 l_curr = l_curr->next;
881 } else if ( (l_curr->start >= lock->start) &&
882 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
885 * This unlock is completely overlapped by this existing lock range
886 * and thus should have no effect (not be unlocked). Delete it from the list.
888 /*********************************************
892 +---------------------------+
894 +---------------------------+
895 **********************************************/
896 /* Save the next pointer */
897 struct lock_list *ul_next = l_curr->next;
899 DEBUG(10,("delete case.\n" ));
901 DLIST_REMOVE(lhead, l_curr);
903 break; /* No more list... */
907 } else if ( (l_curr->start >= lock->start) &&
908 (l_curr->start < lock->start + lock->size) &&
909 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
912 * This unlock overlaps the existing lock range at the high end.
913 * Truncate by moving start to existing range end and reducing size.
915 /*********************************************
926 **********************************************/
928 l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
929 l_curr->start = lock->start + lock->size;
931 DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
932 (double)l_curr->start, (double)l_curr->size ));
934 l_curr = l_curr->next;
936 } else if ( (l_curr->start < lock->start) &&
937 (l_curr->start + l_curr->size > lock->start) &&
938 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
941 * This unlock overlaps the existing lock range at the low end.
942 * Truncate by reducing size.
944 /*********************************************
955 **********************************************/
957 l_curr->size = lock->start - l_curr->start;
959 DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
960 (double)l_curr->start, (double)l_curr->size ));
962 l_curr = l_curr->next;
964 } else if ( (l_curr->start < lock->start) &&
965 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
967 * Worst case scenario. Unlock request completely overlaps an existing
968 * lock range. Split the request into two, push the new (upper) request
969 * into the dlink list, and continue with the entry after ul_new (as we
970 * know that ul_new will not overlap with this lock).
972 /*********************************************
973 +---------------------------+
975 +---------------------------+
980 +-------+ +---------+
982 +-------+ +---------+
983 **********************************************/
984 struct lock_list *l_new = (struct lock_list *)talloc(ctx,
985 sizeof(struct lock_list));
988 DEBUG(0,("posix_lock_list: talloc fail.\n"));
989 return NULL; /* The talloc_destroy takes care of cleanup. */
993 l_new->start = lock->start + lock->size;
994 l_new->size = l_curr->start + l_curr->size - l_new->start;
996 /* Truncate the l_curr. */
997 l_curr->size = lock->start - l_curr->start;
999 DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
1000 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
1001 (double)l_new->start, (double)l_new->size ));
1004 * Add into the dlink list after the l_curr point - NOT at lhead.
1005 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
1008 l_new->prev = l_curr;
1009 l_new->next = l_curr->next;
1010 l_curr->next = l_new;
1012 /* And move after the link we added. */
1013 l_curr = l_new->next;
1018 * This logic case should never happen. Ensure this is the
1019 * case by forcing an abort.... Remove in production.
1023 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
1024 lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
1028 } /* end for ( l_curr = lhead; l_curr;) */
1029 } /* end for (i=0; i<num_locks && ul_head; i++) */
1037 /****************************************************************************
1038 POSIX function to acquire a lock. Returns True if the
1039 lock could be granted, False if not.
1040 ****************************************************************************/
1042 BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
1047 size_t entry_num = 0;
1049 TALLOC_CTX *l_ctx = NULL;
1050 struct lock_list *llist = NULL;
1051 struct lock_list *ll = NULL;
1052 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
1054 DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
1055 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
1058 * If the requested lock won't fit in the POSIX range, we will
1059 * pretend it was successful.
1062 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1066 * Windows is very strange. It allows read locks to be overlayed
1067 * (even over a write lock), but leaves the write lock in force until the first
1068 * unlock. It also reference counts the locks. This means the following sequence :
1071 * ------------------------------------------------------------------------
1072 * WRITE LOCK : start = 2, len = 10
1073 * READ LOCK: start =0, len = 10 - FAIL
1074 * READ LOCK : start = 0, len = 14
1075 * READ LOCK: start =0, len = 10 - FAIL
1076 * UNLOCK : start = 2, len = 10
1077 * READ LOCK: start =0, len = 10 - OK
1079 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
1080 * would leave a single read lock over the 0-14 region. In order to
1081 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
1082 * entries, one for each overlayed lock request. We are guarenteed by the brlock
1083 * semantics that if a write lock is added, then it will be first in the array.
1086 if ((l_ctx = talloc_init()) == NULL) {
1087 DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
1088 return True; /* Not a fatal error. */
1091 if ((ll = (struct lock_list *)talloc(l_ctx, sizeof(struct lock_list))) == NULL) {
1092 DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
1093 talloc_destroy(l_ctx);
1094 return True; /* Not a fatal error. */
1098 * Create the initial list entry containing the
1099 * lock we want to add.
1106 DLIST_ADD(llist, ll);
1109 * The following call calculates if there are any
1110 * overlapping locks held by this process on
1111 * fd's open on the same file and splits this list
1112 * into a list of lock ranges that do not overlap with existing
1116 llist = posix_lock_list(l_ctx, llist, fsp);
1119 * Now we have the list of ranges to lock it is safe to add the
1120 * entry into the POSIX lock tdb. We take note of the entry we
1121 * added here in case we have to remove it on POSIX lock fail.
1124 if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
1125 DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
1126 talloc_destroy(l_ctx);
1131 * Add the POSIX locks on the list of ranges returned.
1132 * As the lock is supposed to be added atomically, we need to
1133 * back out all the locks if any one of these calls fail.
1136 for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
1140 DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
1141 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1143 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1144 DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
1145 posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
1154 * Back out all the POSIX locks we have on fail.
1157 for (ll = llist; lock_count; ll = ll->next, lock_count--) {
1161 DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
1162 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1164 posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
1168 * Remove the tdb entry for this lock.
1171 delete_posix_lock_entry_by_index(fsp,entry_num);
1174 talloc_destroy(l_ctx);
1178 /****************************************************************************
1179 POSIX function to release a lock. Returns True if the
1180 lock could be released, False if not.
1181 ****************************************************************************/
1183 BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
1188 TALLOC_CTX *ul_ctx = NULL;
1189 struct lock_list *ulist = NULL;
1190 struct lock_list *ul = NULL;
1191 struct posix_lock deleted_lock;
1192 int num_overlapped_entries;
1194 DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
1195 fsp->fsp_name, (double)u_offset, (double)u_count ));
1198 * If the requested lock won't fit in the POSIX range, we will
1199 * pretend it was successful.
1202 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1206 * We treat this as one unlock request for POSIX accounting purposes even
1207 * if it may later be split into multiple smaller POSIX unlock ranges.
1208 * num_overlapped_entries is the number of existing locks that have any
1209 * overlap with this unlock request.
1212 num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
1214 if (num_overlapped_entries == -1) {
1215 smb_panic("release_posix_lock: unable find entry to delete !\n");
1219 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
1220 * a POSIX write lock, then before doing the unlock we need to downgrade
1221 * the POSIX lock to a read lock. This allows any overlapping read locks
1222 * to be atomically maintained.
1225 if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
1226 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1227 DEBUG(0,("release_posix_lock: downgrade of lock failed with error %s !\n", strerror(errno) ));
1232 if ((ul_ctx = talloc_init()) == NULL) {
1233 DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
1234 return True; /* Not a fatal error. */
1237 if ((ul = (struct lock_list *)talloc(ul_ctx, sizeof(struct lock_list))) == NULL) {
1238 DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
1239 talloc_destroy(ul_ctx);
1240 return True; /* Not a fatal error. */
1244 * Create the initial list entry containing the
1245 * lock we want to remove.
1252 DLIST_ADD(ulist, ul);
1255 * The following call calculates if there are any
1256 * overlapping locks held by this process on
1257 * fd's open on the same file and creates a
1258 * list of unlock ranges that will allow
1259 * POSIX lock ranges to remain on the file whilst the
1260 * unlocks are performed.
1263 ulist = posix_lock_list(ul_ctx, ulist, fsp);
1266 * Release the POSIX locks on the list of ranges returned.
1269 for(; ulist; ulist = ulist->next) {
1270 offset = ulist->start;
1271 count = ulist->size;
1273 DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
1274 (double)offset, (double)count ));
1276 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
1280 talloc_destroy(ul_ctx);
1285 /****************************************************************************
1286 Remove all lock entries for a specific dev/inode pair from the tdb.
1287 ****************************************************************************/
1289 static void delete_posix_lock_entries(files_struct *fsp)
1291 TDB_DATA kbuf = locking_key_fsp(fsp);
1293 if (tdb_delete(posix_lock_tdb, kbuf) == -1)
1294 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
1297 /****************************************************************************
1299 ****************************************************************************/
1301 static void dump_entry(struct posix_lock *pl)
1303 DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
1304 (double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
1307 /****************************************************************************
1308 Remove any locks on this fd. Called from file_close().
1309 ****************************************************************************/
1311 void posix_locking_close_file(files_struct *fsp)
1313 struct posix_lock *entries = NULL;
1317 * Optimization for the common case where we are the only
1318 * opener of a file. If all fd entries are our own, we don't
1319 * need to explicitly release all the locks via the POSIX functions,
1320 * we can just remove all the entries in the tdb and allow the
1321 * close to remove the real locks.
1324 count = get_posix_lock_entries(fsp, &entries);
1327 DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
1331 for (i = 0; i < count; i++) {
1332 if (entries[i].fd != fsp->fd )
1335 dump_entry(&entries[i]);
1339 /* All locks are ours. */
1340 DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n",
1341 fsp->fsp_name, (unsigned int)count ));
1342 free((char *)entries);
1343 delete_posix_lock_entries(fsp);
1348 * Difficult case. We need to delete all our locks, whilst leaving
1349 * all other POSIX locks in place.
1352 for (i = 0; i < count; i++) {
1353 struct posix_lock *pl = &entries[i];
1354 if (pl->fd == fsp->fd)
1355 release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
1357 free((char *)entries);
1360 /*******************************************************************
1361 Create the in-memory POSIX lock databases.
1362 ********************************************************************/
1364 BOOL posix_locking_init(int read_only)
1366 if (posix_lock_tdb && posix_pending_close_tdb)
1369 if (!posix_lock_tdb)
1370 posix_lock_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1371 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1372 if (!posix_lock_tdb) {
1373 DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
1376 if (!posix_pending_close_tdb)
1377 posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1378 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1379 if (!posix_pending_close_tdb) {
1380 DEBUG(0,("Failed to open POSIX pending close database.\n"));
1387 /*******************************************************************
1388 Delete the in-memory POSIX lock databases.
1389 ********************************************************************/
1391 BOOL posix_locking_end(void)
1393 if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
1395 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)