2 Unix SMB/CIFS implementation.
4 Copyright (C) Jeremy Allison 1992-2000
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
22 POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000.
28 #define DBGC_CLASS DBGC_LOCKING
31 * The POSIX locking database handle.
34 static TDB_CONTEXT *posix_lock_tdb;
37 * The pending close database handle.
40 static TDB_CONTEXT *posix_pending_close_tdb;
43 * The data in POSIX lock records is an unsorted linear array of these
44 * records. It is unnecessary to store the count as tdb provides the
56 * The data in POSIX pending close records is an unsorted linear array of int
57 * records. It is unnecessary to store the count as tdb provides the
61 /* The key used in both the POSIX databases. */
63 struct posix_lock_key {
68 /*******************************************************************
69 Form a static locking key for a dev/inode pair.
70 ******************************************************************/
72 static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
74 static struct posix_lock_key key;
77 memset(&key, '\0', sizeof(key));
80 kbuf.dptr = (char *)&key;
81 kbuf.dsize = sizeof(key);
85 /*******************************************************************
86 Convenience function to get a key from an fsp.
87 ******************************************************************/
89 static TDB_DATA locking_key_fsp(files_struct *fsp)
91 return locking_key(fsp->dev, fsp->inode);
94 /****************************************************************************
95 Add an fd to the pending close tdb.
96 ****************************************************************************/
98 static BOOL add_fd_to_close_entry(files_struct *fsp)
100 TDB_DATA kbuf = locking_key_fsp(fsp);
106 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
108 dbuf.dptr = SMB_REALLOC(dbuf.dptr, dbuf.dsize + sizeof(int));
110 DEBUG(0,("add_fd_to_close_entry: Realloc fail !\n"));
114 memcpy(dbuf.dptr + dbuf.dsize, &fsp->fh->fd, sizeof(int));
115 dbuf.dsize += sizeof(int);
117 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
118 DEBUG(0,("add_fd_to_close_entry: tdb_store fail !\n"));
121 SAFE_FREE(dbuf.dptr);
125 /****************************************************************************
126 Remove all fd entries for a specific dev/inode pair from the tdb.
127 ****************************************************************************/
129 static void delete_close_entries(files_struct *fsp)
131 TDB_DATA kbuf = locking_key_fsp(fsp);
133 if (tdb_delete(posix_pending_close_tdb, kbuf) == -1)
134 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
137 /****************************************************************************
138 Get the array of POSIX pending close records for an open fsp. Caller must
139 free. Returns number of entries.
140 ****************************************************************************/
142 static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
144 TDB_DATA kbuf = locking_key_fsp(fsp);
151 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
157 *entries = (int *)dbuf.dptr;
158 count = (size_t)(dbuf.dsize / sizeof(int));
163 /****************************************************************************
164 Get the array of POSIX locks for an fsp. Caller must free. Returns
166 ****************************************************************************/
168 static size_t get_posix_lock_entries(files_struct *fsp, struct posix_lock **entries)
170 TDB_DATA kbuf = locking_key_fsp(fsp);
178 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
184 *entries = (struct posix_lock *)dbuf.dptr;
185 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
190 /****************************************************************************
191 Deal with pending closes needed by POSIX locking support.
192 Note that posix_locking_close_file() is expected to have been called
193 to delete all locks on this fsp before this function is called.
194 ****************************************************************************/
196 int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
201 struct posix_lock *entries = NULL;
202 int *fd_array = NULL;
203 BOOL locks_on_other_fds = False;
205 if (!lp_posix_locking(SNUM(conn))) {
207 * No POSIX to worry about, just close.
209 ret = SMB_VFS_CLOSE(fsp,fsp->fh->fd);
215 * Get the number of outstanding POSIX locks on this dev/inode pair.
218 count = get_posix_lock_entries(fsp, &entries);
221 * Check if there are any outstanding locks belonging to
222 * other fd's. This should never be the case if posix_locking_close_file()
223 * has been called first, but it never hurts to be *sure*.
226 for (i = 0; i < count; i++) {
227 if (entries[i].fd != fsp->fh->fd) {
228 locks_on_other_fds = True;
233 if (locks_on_other_fds) {
236 * There are outstanding locks on this dev/inode pair on other fds.
237 * Add our fd to the pending close tdb and set fsp->fh->fd to -1.
240 if (!add_fd_to_close_entry(fsp)) {
253 * No outstanding POSIX locks. Get the pending close fd's
254 * from the tdb and close them all.
257 count = get_posix_pending_close_entries(fsp, &fd_array);
260 DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
262 for(i = 0; i < count; i++) {
263 if (SMB_VFS_CLOSE(fsp,fd_array[i]) == -1) {
269 * Delete all fd's stored in the tdb
270 * for this dev/inode pair.
273 delete_close_entries(fsp);
279 * Finally close the fd associated with this fsp.
282 ret = SMB_VFS_CLOSE(fsp,fsp->fh->fd);
284 if (saved_errno != 0) {
294 /****************************************************************************
296 ****************************************************************************/
298 static const char *posix_lock_type_name(int lock_type)
300 return (lock_type == F_RDLCK) ? "READ" : "WRITE";
303 /****************************************************************************
304 Delete a POSIX lock entry by index number. Used if the tdb add succeeds, but
305 then the POSIX fcntl lock fails.
306 ****************************************************************************/
308 static BOOL delete_posix_lock_entry_by_index(files_struct *fsp, size_t entry)
310 TDB_DATA kbuf = locking_key_fsp(fsp);
312 struct posix_lock *locks;
317 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
320 DEBUG(10,("delete_posix_lock_entry_by_index: tdb_fetch failed !\n"));
324 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
325 locks = (struct posix_lock *)dbuf.dptr;
328 tdb_delete(posix_lock_tdb, kbuf);
330 if (entry < count-1) {
331 memmove(&locks[entry], &locks[entry+1], sizeof(struct posix_lock)*((count-1) - entry));
333 dbuf.dsize -= sizeof(struct posix_lock);
334 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
337 SAFE_FREE(dbuf.dptr);
343 SAFE_FREE(dbuf.dptr);
347 /****************************************************************************
348 Add an entry into the POSIX locking tdb. We return the index number of the
349 added lock (used in case we need to delete *exactly* this entry). Returns
350 False on fail, True on success.
351 ****************************************************************************/
353 static BOOL add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *pentry_num)
355 TDB_DATA kbuf = locking_key_fsp(fsp);
357 struct posix_lock pl;
362 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
364 *pentry_num = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
373 pl.lock_type = lock_type;
375 dbuf.dptr = SMB_REALLOC(dbuf.dptr, dbuf.dsize + sizeof(struct posix_lock));
377 DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
381 memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(struct posix_lock));
382 dbuf.dsize += sizeof(struct posix_lock);
384 if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
385 DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
389 SAFE_FREE(dbuf.dptr);
391 DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: dev=%.0f inode=%.0f\n",
392 fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size,
393 (double)fsp->dev, (double)fsp->inode ));
399 SAFE_FREE(dbuf.dptr);
403 /****************************************************************************
404 Calculate if locks have any overlap at all.
405 ****************************************************************************/
407 static BOOL does_lock_overlap(SMB_OFF_T start1, SMB_OFF_T size1, SMB_OFF_T start2, SMB_OFF_T size2)
409 if (start1 >= start2 && start1 <= start2 + size2)
412 if (start1 < start2 && start1 + size1 > start2)
418 /****************************************************************************
419 Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
420 deleted and the number of records that are overlapped by this one, or -1 on error.
421 ****************************************************************************/
423 static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
425 TDB_DATA kbuf = locking_key_fsp(fsp);
427 struct posix_lock *locks;
430 int num_overlapping_records = 0;
434 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
437 DEBUG(10,("delete_posix_lock_entry: tdb_fetch failed !\n"));
441 /* There are existing locks - find a match. */
442 locks = (struct posix_lock *)dbuf.dptr;
443 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
446 * Search for and delete the first record that matches the
450 for (i=0; i<count; i++) {
451 struct posix_lock *entry = &locks[i];
453 if (entry->fd == fsp->fh->fd &&
454 entry->start == start &&
455 entry->size == size) {
457 /* Make a copy if requested. */
461 /* Found it - delete it. */
463 tdb_delete(posix_lock_tdb, kbuf);
466 memmove(&locks[i], &locks[i+1], sizeof(struct posix_lock)*((count-1) - i));
468 dbuf.dsize -= sizeof(struct posix_lock);
469 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
481 * Count the number of entries that are
482 * overlapped by this unlock request.
485 for (i = 0; i < count; i++) {
486 struct posix_lock *entry = &locks[i];
488 if (fsp->fh->fd == entry->fd &&
489 does_lock_overlap( start, size, entry->start, entry->size))
490 num_overlapping_records++;
493 DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
494 posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
495 (unsigned int)num_overlapping_records ));
497 SAFE_FREE(dbuf.dptr);
499 return num_overlapping_records;
503 SAFE_FREE(dbuf.dptr);
507 /****************************************************************************
508 Utility function to map a lock type correctly depending on the open
510 ****************************************************************************/
512 static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
514 if((lock_type == WRITE_LOCK) && !fsp->can_write) {
516 * Many UNIX's cannot get a write lock on a file opened read-only.
517 * Win32 locking semantics allow this.
518 * Do the best we can and attempt a read-only lock.
520 DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
524 /* We no longer open files write-only. */
525 else if((lock_type == READ_LOCK) && !fsp->can_read) {
527 * Ditto for read locks on write only files.
529 DEBUG(10,("map_posix_lock_type: Changing read lock to write due to write-only file.\n"));
535 * This return should be the most normal, as we attempt
536 * to always open files read/write.
539 return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
542 /****************************************************************************
543 Check to see if the given unsigned lock range is within the possible POSIX
544 range. Modifies the given args to be in range if possible, just returns
546 ****************************************************************************/
548 static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
549 SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
551 SMB_OFF_T offset = (SMB_OFF_T)u_offset;
552 SMB_OFF_T count = (SMB_OFF_T)u_count;
555 * For the type of system we are, attempt to
556 * find the maximum positive lock offset as an SMB_OFF_T.
559 #if defined(MAX_POSITIVE_LOCK_OFFSET) /* Some systems have arbitrary limits. */
561 SMB_OFF_T max_positive_lock_offset = (MAX_POSITIVE_LOCK_OFFSET);
563 #elif defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
566 * In this case SMB_OFF_T is 64 bits,
567 * and the underlying system can handle 64 bit signed locks.
570 SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
571 SMB_OFF_T mask = (mask2<<1);
572 SMB_OFF_T max_positive_lock_offset = ~mask;
574 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
577 * In this case either SMB_OFF_T is 32 bits,
578 * or the underlying system cannot handle 64 bit signed locks.
579 * All offsets & counts must be 2^31 or less.
582 SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
584 #endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
587 * POSIX locks of length zero mean lock to end-of-file.
588 * Win32 locks of length zero are point probes. Ignore
589 * any Win32 locks of length zero. JRA.
592 if (count == (SMB_OFF_T)0) {
593 DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n"));
598 * If the given offset was > max_positive_lock_offset then we cannot map this at all
602 if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
603 DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
604 (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
609 * We must truncate the count to less than max_positive_lock_offset.
612 if (u_count & ~((SMB_BIG_UINT)max_positive_lock_offset))
613 count = max_positive_lock_offset;
616 * Truncate count to end at max lock offset.
619 if (offset + count < 0 || offset + count > max_positive_lock_offset)
620 count = max_positive_lock_offset - offset;
623 * If we ate all the count, ignore this lock.
627 DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
628 (double)u_offset, (double)u_count ));
633 * The mapping was successful.
636 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
637 (double)offset, (double)count ));
639 *offset_out = offset;
645 /****************************************************************************
646 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
647 broken NFS implementations.
648 ****************************************************************************/
650 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
654 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fh->fd,op,(double)offset,(double)count,type));
656 ret = SMB_VFS_LOCK(fsp,fsp->fh->fd,op,offset,count,type);
658 if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno == EINVAL))) {
660 DEBUG(0,("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n",
661 (double)offset,(double)count));
662 DEBUG(0,("an %s error. This can happen when using 64 bit lock offsets\n", strerror(errno)));
663 DEBUG(0,("on 32 bit NFS mounted file systems.\n"));
666 * If the offset is > 0x7FFFFFFF then this will cause problems on
667 * 32 bit NFS mounted filesystems. Just ignore it.
670 if (offset & ~((SMB_OFF_T)0x7fffffff)) {
671 DEBUG(0,("Offset greater than 31 bits. Returning success.\n"));
675 if (count & ~((SMB_OFF_T)0x7fffffff)) {
676 /* 32 bit NFS file system, retry with smaller offset */
677 DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n"));
680 ret = SMB_VFS_LOCK(fsp,fsp->fh->fd,op,offset,count,type);
684 DEBUG(8,("posix_fcntl_lock: Lock call %s\n", ret ? "successful" : "failed"));
689 /****************************************************************************
690 POSIX function to see if a file region is locked. Returns True if the
691 region is locked, False otherwise.
692 ****************************************************************************/
694 BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
698 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
700 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
701 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
704 * If the requested lock won't fit in the POSIX range, we will
705 * never set it, so presume it is not locked.
708 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
712 * Note that most UNIX's can *test* for a write lock on
713 * a read-only fd, just not *set* a write lock on a read-only
714 * fd. So we don't need to use map_lock_type here.
717 return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
721 * Structure used when splitting a lock range
722 * into a POSIX lock range. Doubly linked list.
726 struct lock_list *next;
727 struct lock_list *prev;
732 /****************************************************************************
733 Create a list of lock ranges that don't overlap a given range. Used in calculating
734 POSIX locks and unlocks. This is a difficult function that requires ASCII art to
736 ****************************************************************************/
738 static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
740 TDB_DATA kbuf = locking_key_fsp(fsp);
742 struct posix_lock *locks;
747 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
752 locks = (struct posix_lock *)dbuf.dptr;
753 num_locks = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
756 * Check the current lock list on this dev/inode pair.
757 * Quit if the list is deleted.
760 DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
761 (double)lhead->start, (double)lhead->size ));
763 for (i=0; i<num_locks && lhead; i++) {
765 struct posix_lock *lock = &locks[i];
766 struct lock_list *l_curr;
769 * Walk the lock list, checking for overlaps. Note that
770 * the lock list can expand within this loop if the current
771 * range being examined needs to be split.
774 for (l_curr = lhead; l_curr;) {
776 DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
777 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
779 if ( (l_curr->start >= (lock->start + lock->size)) ||
780 (lock->start >= (l_curr->start + l_curr->size))) {
782 /* No overlap with this lock - leave this range alone. */
783 /*********************************************
794 **********************************************/
796 DEBUG(10,("no overlap case.\n" ));
798 l_curr = l_curr->next;
800 } else if ( (l_curr->start >= lock->start) &&
801 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
804 * This unlock is completely overlapped by this existing lock range
805 * and thus should have no effect (not be unlocked). Delete it from the list.
807 /*********************************************
811 +---------------------------+
813 +---------------------------+
814 **********************************************/
815 /* Save the next pointer */
816 struct lock_list *ul_next = l_curr->next;
818 DEBUG(10,("delete case.\n" ));
820 DLIST_REMOVE(lhead, l_curr);
822 break; /* No more list... */
826 } else if ( (l_curr->start >= lock->start) &&
827 (l_curr->start < lock->start + lock->size) &&
828 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
831 * This unlock overlaps the existing lock range at the high end.
832 * Truncate by moving start to existing range end and reducing size.
834 /*********************************************
845 **********************************************/
847 l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
848 l_curr->start = lock->start + lock->size;
850 DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
851 (double)l_curr->start, (double)l_curr->size ));
853 l_curr = l_curr->next;
855 } else if ( (l_curr->start < lock->start) &&
856 (l_curr->start + l_curr->size > lock->start) &&
857 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
860 * This unlock overlaps the existing lock range at the low end.
861 * Truncate by reducing size.
863 /*********************************************
874 **********************************************/
876 l_curr->size = lock->start - l_curr->start;
878 DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
879 (double)l_curr->start, (double)l_curr->size ));
881 l_curr = l_curr->next;
883 } else if ( (l_curr->start < lock->start) &&
884 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
886 * Worst case scenario. Unlock request completely overlaps an existing
887 * lock range. Split the request into two, push the new (upper) request
888 * into the dlink list, and continue with the entry after ul_new (as we
889 * know that ul_new will not overlap with this lock).
891 /*********************************************
892 +---------------------------+
894 +---------------------------+
899 +-------+ +---------+
901 +-------+ +---------+
902 **********************************************/
903 struct lock_list *l_new = TALLOC_P(ctx, struct lock_list);
906 DEBUG(0,("posix_lock_list: talloc fail.\n"));
907 return NULL; /* The talloc_destroy takes care of cleanup. */
911 l_new->start = lock->start + lock->size;
912 l_new->size = l_curr->start + l_curr->size - l_new->start;
914 /* Truncate the l_curr. */
915 l_curr->size = lock->start - l_curr->start;
917 DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
918 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
919 (double)l_new->start, (double)l_new->size ));
922 * Add into the dlink list after the l_curr point - NOT at lhead.
923 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
926 l_new->prev = l_curr;
927 l_new->next = l_curr->next;
928 l_curr->next = l_new;
930 /* And move after the link we added. */
931 l_curr = l_new->next;
936 * This logic case should never happen. Ensure this is the
937 * case by forcing an abort.... Remove in production.
941 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
942 lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
946 } /* end for ( l_curr = lhead; l_curr;) */
947 } /* end for (i=0; i<num_locks && ul_head; i++) */
949 SAFE_FREE(dbuf.dptr);
954 /****************************************************************************
955 POSIX function to acquire a lock. Returns True if the
956 lock could be granted, False if not.
957 ****************************************************************************/
959 BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
964 size_t entry_num = 0;
966 TALLOC_CTX *l_ctx = NULL;
967 struct lock_list *llist = NULL;
968 struct lock_list *ll = NULL;
969 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
971 DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
972 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
975 * If the requested lock won't fit in the POSIX range, we will
976 * pretend it was successful.
979 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
983 * Windows is very strange. It allows read locks to be overlayed
984 * (even over a write lock), but leaves the write lock in force until the first
985 * unlock. It also reference counts the locks. This means the following sequence :
988 * ------------------------------------------------------------------------
989 * WRITE LOCK : start = 2, len = 10
990 * READ LOCK: start =0, len = 10 - FAIL
991 * READ LOCK : start = 0, len = 14
992 * READ LOCK: start =0, len = 10 - FAIL
993 * UNLOCK : start = 2, len = 10
994 * READ LOCK: start =0, len = 10 - OK
996 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
997 * would leave a single read lock over the 0-14 region. In order to
998 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
999 * entries, one for each overlayed lock request. We are guarenteed by the brlock
1000 * semantics that if a write lock is added, then it will be first in the array.
1003 if ((l_ctx = talloc_init("set_posix_lock")) == NULL) {
1004 DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
1005 return True; /* Not a fatal error. */
1008 if ((ll = TALLOC_P(l_ctx, struct lock_list)) == NULL) {
1009 DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
1010 talloc_destroy(l_ctx);
1011 return True; /* Not a fatal error. */
1015 * Create the initial list entry containing the
1016 * lock we want to add.
1023 DLIST_ADD(llist, ll);
1026 * The following call calculates if there are any
1027 * overlapping locks held by this process on
1028 * fd's open on the same file and splits this list
1029 * into a list of lock ranges that do not overlap with existing
1033 llist = posix_lock_list(l_ctx, llist, fsp);
1036 * Now we have the list of ranges to lock it is safe to add the
1037 * entry into the POSIX lock tdb. We take note of the entry we
1038 * added here in case we have to remove it on POSIX lock fail.
1041 if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
1042 DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
1043 talloc_destroy(l_ctx);
1048 * Add the POSIX locks on the list of ranges returned.
1049 * As the lock is supposed to be added atomically, we need to
1050 * back out all the locks if any one of these calls fail.
1053 for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
1057 DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
1058 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1060 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1061 DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
1062 posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
1071 * Back out all the POSIX locks we have on fail.
1074 for (ll = llist; lock_count; ll = ll->next, lock_count--) {
1078 DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
1079 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1081 posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
1085 * Remove the tdb entry for this lock.
1088 delete_posix_lock_entry_by_index(fsp,entry_num);
1091 talloc_destroy(l_ctx);
1095 /****************************************************************************
1096 POSIX function to release a lock. Returns True if the
1097 lock could be released, False if not.
1098 ****************************************************************************/
1100 BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
1105 TALLOC_CTX *ul_ctx = NULL;
1106 struct lock_list *ulist = NULL;
1107 struct lock_list *ul = NULL;
1108 struct posix_lock deleted_lock;
1109 int num_overlapped_entries;
1111 DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
1112 fsp->fsp_name, (double)u_offset, (double)u_count ));
1115 * If the requested lock won't fit in the POSIX range, we will
1116 * pretend it was successful.
1119 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1123 * We treat this as one unlock request for POSIX accounting purposes even
1124 * if it may later be split into multiple smaller POSIX unlock ranges.
1125 * num_overlapped_entries is the number of existing locks that have any
1126 * overlap with this unlock request.
1129 num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
1131 if (num_overlapped_entries == -1) {
1132 smb_panic("release_posix_lock: unable find entry to delete !\n");
1136 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
1137 * a POSIX write lock, then before doing the unlock we need to downgrade
1138 * the POSIX lock to a read lock. This allows any overlapping read locks
1139 * to be atomically maintained.
1142 if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
1143 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1144 DEBUG(0,("release_posix_lock: downgrade of lock failed with error %s !\n", strerror(errno) ));
1149 if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
1150 DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
1151 return True; /* Not a fatal error. */
1154 if ((ul = TALLOC_P(ul_ctx, struct lock_list)) == NULL) {
1155 DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
1156 talloc_destroy(ul_ctx);
1157 return True; /* Not a fatal error. */
1161 * Create the initial list entry containing the
1162 * lock we want to remove.
1169 DLIST_ADD(ulist, ul);
1172 * The following call calculates if there are any
1173 * overlapping locks held by this process on
1174 * fd's open on the same file and creates a
1175 * list of unlock ranges that will allow
1176 * POSIX lock ranges to remain on the file whilst the
1177 * unlocks are performed.
1180 ulist = posix_lock_list(ul_ctx, ulist, fsp);
1183 * Release the POSIX locks on the list of ranges returned.
1186 for(; ulist; ulist = ulist->next) {
1187 offset = ulist->start;
1188 count = ulist->size;
1190 DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
1191 (double)offset, (double)count ));
1193 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
1197 talloc_destroy(ul_ctx);
1202 /****************************************************************************
1203 Remove all lock entries for a specific dev/inode pair from the tdb.
1204 ****************************************************************************/
1206 static void delete_posix_lock_entries(files_struct *fsp)
1208 TDB_DATA kbuf = locking_key_fsp(fsp);
1210 if (tdb_delete(posix_lock_tdb, kbuf) == -1)
1211 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
1214 /****************************************************************************
1216 ****************************************************************************/
1218 static void dump_entry(struct posix_lock *pl)
1220 DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
1221 (double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
1224 /****************************************************************************
1225 Remove any locks on this fd. Called from file_close().
1226 ****************************************************************************/
1228 void posix_locking_close_file(files_struct *fsp)
1230 struct posix_lock *entries = NULL;
1234 * Optimization for the common case where we are the only
1235 * opener of a file. If all fd entries are our own, we don't
1236 * need to explicitly release all the locks via the POSIX functions,
1237 * we can just remove all the entries in the tdb and allow the
1238 * close to remove the real locks.
1241 count = get_posix_lock_entries(fsp, &entries);
1244 DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
1248 for (i = 0; i < count; i++) {
1249 if (entries[i].fd != fsp->fh->fd )
1252 dump_entry(&entries[i]);
1256 /* All locks are ours. */
1257 DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n",
1258 fsp->fsp_name, (unsigned int)count ));
1260 delete_posix_lock_entries(fsp);
1265 * Difficult case. We need to delete all our locks, whilst leaving
1266 * all other POSIX locks in place.
1269 for (i = 0; i < count; i++) {
1270 struct posix_lock *pl = &entries[i];
1271 if (pl->fd == fsp->fh->fd)
1272 release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
1277 /*******************************************************************
1278 Create the in-memory POSIX lock databases.
1279 ********************************************************************/
1281 BOOL posix_locking_init(int read_only)
1283 if (posix_lock_tdb && posix_pending_close_tdb)
1286 if (!posix_lock_tdb)
1287 posix_lock_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1288 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1289 if (!posix_lock_tdb) {
1290 DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
1293 if (!posix_pending_close_tdb)
1294 posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1295 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1296 if (!posix_pending_close_tdb) {
1297 DEBUG(0,("Failed to open POSIX pending close database.\n"));
1304 /*******************************************************************
1305 Delete the in-memory POSIX lock databases.
1306 ********************************************************************/
1308 BOOL posix_locking_end(void)
1310 if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
1312 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)