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 * The POSIX locking database handle.
31 static TDB_CONTEXT *posix_lock_tdb;
34 * The pending close database handle.
37 static TDB_CONTEXT *posix_pending_close_tdb;
40 * The data in POSIX lock records is an unsorted linear array of these
41 * records. It is unnecessary to store the count as tdb provides the
53 * The data in POSIX pending close records is an unsorted linear array of int
54 * records. It is unnecessary to store the count as tdb provides the
58 /* The key used in both the POSIX databases. */
60 struct posix_lock_key {
65 /*******************************************************************
66 Form a static locking key for a dev/inode pair.
67 ******************************************************************/
69 static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
71 static struct posix_lock_key key;
74 memset(&key, '\0', sizeof(key));
77 kbuf.dptr = (char *)&key;
78 kbuf.dsize = sizeof(key);
82 /*******************************************************************
83 Convenience function to get a key from an fsp.
84 ******************************************************************/
86 static TDB_DATA locking_key_fsp(files_struct *fsp)
88 return locking_key(fsp->dev, fsp->inode);
91 /****************************************************************************
92 Add an fd to the pending close tdb.
93 ****************************************************************************/
95 static BOOL add_fd_to_close_entry(files_struct *fsp)
97 TDB_DATA kbuf = locking_key_fsp(fsp);
103 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
105 tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(int));
107 DEBUG(0,("add_fd_to_close_entry: Realloc fail !\n"));
108 SAFE_FREE(dbuf.dptr);
113 memcpy(dbuf.dptr + dbuf.dsize, &fsp->fd, sizeof(int));
114 dbuf.dsize += sizeof(int);
116 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
117 DEBUG(0,("add_fd_to_close_entry: tdb_store fail !\n"));
120 SAFE_FREE(dbuf.dptr);
124 /****************************************************************************
125 Remove all fd entries for a specific dev/inode pair from the tdb.
126 ****************************************************************************/
128 static void delete_close_entries(files_struct *fsp)
130 TDB_DATA kbuf = locking_key_fsp(fsp);
132 if (tdb_delete(posix_pending_close_tdb, kbuf) == -1)
133 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
136 /****************************************************************************
137 Get the array of POSIX pending close records for an open fsp. Caller must
138 free. Returns number of entries.
139 ****************************************************************************/
141 static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
143 TDB_DATA kbuf = locking_key_fsp(fsp);
150 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
156 *entries = (int *)dbuf.dptr;
157 count = (size_t)(dbuf.dsize / sizeof(int));
162 /****************************************************************************
163 Get the array of POSIX locks for an fsp. Caller must free. Returns
165 ****************************************************************************/
167 static size_t get_posix_lock_entries(files_struct *fsp, struct posix_lock **entries)
169 TDB_DATA kbuf = locking_key_fsp(fsp);
177 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
183 *entries = (struct posix_lock *)dbuf.dptr;
184 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
189 /****************************************************************************
190 Deal with pending closes needed by POSIX locking support.
191 Note that posix_locking_close_file() is expected to have been called
192 to delete all locks on this fsp before this function is called.
193 ****************************************************************************/
195 int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
200 struct posix_lock *entries = NULL;
201 int *fd_array = NULL;
202 BOOL locks_on_other_fds = False;
204 if (!lp_posix_locking(SNUM(conn))) {
206 * No POSIX to worry about, just close.
208 ret = conn->vfs_ops.close(fsp,fsp->fd);
214 * Get the number of outstanding POSIX locks on this dev/inode pair.
217 count = get_posix_lock_entries(fsp, &entries);
220 * Check if there are any outstanding locks belonging to
221 * other fd's. This should never be the case if posix_locking_close_file()
222 * has been called first, but it never hurts to be *sure*.
225 for (i = 0; i < count; i++) {
226 if (entries[i].fd != fsp->fd) {
227 locks_on_other_fds = True;
232 if (locks_on_other_fds) {
235 * There are outstanding locks on this dev/inode pair on other fds.
236 * Add our fd to the pending close tdb and set fsp->fd to -1.
239 if (!add_fd_to_close_entry(fsp)) {
252 * No outstanding POSIX locks. Get the pending close fd's
253 * from the tdb and close them all.
256 count = get_posix_pending_close_entries(fsp, &fd_array);
259 DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
261 for(i = 0; i < count; i++) {
262 if (conn->vfs_ops.close(fsp,fd_array[i]) == -1) {
268 * Delete all fd's stored in the tdb
269 * for this dev/inode pair.
272 delete_close_entries(fsp);
278 * Finally close the fd associated with this fsp.
281 ret = conn->vfs_ops.close(fsp,fsp->fd);
283 if (saved_errno != 0) {
293 /****************************************************************************
295 ****************************************************************************/
297 static const char *posix_lock_type_name(int lock_type)
299 return (lock_type == F_RDLCK) ? "READ" : "WRITE";
302 /****************************************************************************
303 Delete a POSIX lock entry by index number. Used if the tdb add succeeds, but
304 then the POSIX fcntl lock fails.
305 ****************************************************************************/
307 static BOOL delete_posix_lock_entry_by_index(files_struct *fsp, size_t entry)
309 TDB_DATA kbuf = locking_key_fsp(fsp);
311 struct posix_lock *locks;
316 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
319 DEBUG(10,("delete_posix_lock_entry_by_index: tdb_fetch failed !\n"));
323 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
324 locks = (struct posix_lock *)dbuf.dptr;
327 tdb_delete(posix_lock_tdb, kbuf);
329 if (entry < count-1) {
330 memmove(&locks[entry], &locks[entry+1], sizeof(*locks)*((count-1) - entry));
332 dbuf.dsize -= sizeof(*locks);
333 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
336 SAFE_FREE(dbuf.dptr);
342 SAFE_FREE(dbuf.dptr);
346 /****************************************************************************
347 Add an entry into the POSIX locking tdb. We return the index number of the
348 added lock (used in case we need to delete *exactly* this entry). Returns
349 False on fail, True on success.
350 ****************************************************************************/
352 static BOOL add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *pentry_num)
354 TDB_DATA kbuf = locking_key_fsp(fsp);
356 struct posix_lock pl;
361 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
363 *pentry_num = (size_t)(dbuf.dsize / sizeof(pl));
372 pl.lock_type = lock_type;
374 tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(pl));
376 DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
381 memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
382 dbuf.dsize += sizeof(pl);
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(*locks));
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->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(*locks)*((count-1) - i));
468 dbuf.dsize -= sizeof(*locks);
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->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"));
522 } else if((lock_type == READ_LOCK) && !fsp->can_read) {
524 * Ditto for read locks on write only files.
526 DEBUG(10,("map_posix_lock_type: Changing read lock to write due to write-only file.\n"));
531 * This return should be the most normal, as we attempt
532 * to always open files read/write.
535 return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
538 /****************************************************************************
539 Check to see if the given unsigned lock range is within the possible POSIX
540 range. Modifies the given args to be in range if possible, just returns
542 ****************************************************************************/
544 static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
545 SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
547 SMB_OFF_T offset = (SMB_OFF_T)u_offset;
548 SMB_OFF_T count = (SMB_OFF_T)u_count;
551 * For the type of system we are, attempt to
552 * find the maximum positive lock offset as an SMB_OFF_T.
555 #if defined(MAX_POSITIVE_LOCK_OFFSET) /* Some systems have arbitrary limits. */
557 SMB_OFF_T max_positive_lock_offset = (MAX_POSITIVE_LOCK_OFFSET);
559 #elif defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
562 * In this case SMB_OFF_T is 64 bits,
563 * and the underlying system can handle 64 bit signed locks.
566 SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
567 SMB_OFF_T mask = (mask2<<1);
568 SMB_OFF_T max_positive_lock_offset = ~mask;
570 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
573 * In this case either SMB_OFF_T is 32 bits,
574 * or the underlying system cannot handle 64 bit signed locks.
575 * All offsets & counts must be 2^31 or less.
578 SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
580 #endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
583 * POSIX locks of length zero mean lock to end-of-file.
584 * Win32 locks of length zero are point probes. Ignore
585 * any Win32 locks of length zero. JRA.
588 if (count == (SMB_OFF_T)0) {
589 DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n"));
594 * If the given offset was > max_positive_lock_offset then we cannot map this at all
598 if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
599 DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
600 (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
605 * We must truncate the count to less than max_positive_lock_offset.
608 if (u_count & ~((SMB_BIG_UINT)max_positive_lock_offset))
609 count = max_positive_lock_offset;
612 * Truncate count to end at max lock offset.
615 if (offset + count < 0 || offset + count > max_positive_lock_offset)
616 count = max_positive_lock_offset - offset;
619 * If we ate all the count, ignore this lock.
623 DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
624 (double)u_offset, (double)u_count ));
629 * The mapping was successful.
632 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
633 (double)offset, (double)count ));
635 *offset_out = offset;
641 /****************************************************************************
642 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
643 broken NFS implementations.
644 ****************************************************************************/
646 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
649 struct connection_struct *conn = fsp->conn;
651 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
653 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
655 if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno == EINVAL))) {
657 DEBUG(0,("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n",
658 (double)offset,(double)count));
659 DEBUG(0,("an %s error. This can happen when using 64 bit lock offsets\n", strerror(errno)));
660 DEBUG(0,("on 32 bit NFS mounted file systems.\n"));
663 * If the offset is > 0x7FFFFFFF then this will cause problems on
664 * 32 bit NFS mounted filesystems. Just ignore it.
667 if (offset & ~((SMB_OFF_T)0x7fffffff)) {
668 DEBUG(0,("Offset greater than 31 bits. Returning success.\n"));
672 if (count & ~((SMB_OFF_T)0x7fffffff)) {
673 /* 32 bit NFS file system, retry with smaller offset */
674 DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n"));
677 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
681 DEBUG(8,("posix_fcntl_lock: Lock call %s\n", ret ? "successful" : "failed"));
686 /****************************************************************************
687 POSIX function to see if a file region is locked. Returns True if the
688 region is locked, False otherwise.
689 ****************************************************************************/
691 BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
695 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
697 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
698 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
701 * If the requested lock won't fit in the POSIX range, we will
702 * never set it, so presume it is not locked.
705 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
709 * Note that most UNIX's can *test* for a write lock on
710 * a read-only fd, just not *set* a write lock on a read-only
711 * fd. So we don't need to use map_lock_type here.
714 return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
718 * Structure used when splitting a lock range
719 * into a POSIX lock range. Doubly linked list.
723 struct lock_list *next;
724 struct lock_list *prev;
729 /****************************************************************************
730 Create a list of lock ranges that don't overlap a given range. Used in calculating
731 POSIX locks and unlocks. This is a difficult function that requires ASCII art to
733 ****************************************************************************/
735 static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
737 TDB_DATA kbuf = locking_key_fsp(fsp);
739 struct posix_lock *locks;
744 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
749 locks = (struct posix_lock *)dbuf.dptr;
750 num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
753 * Check the current lock list on this dev/inode pair.
754 * Quit if the list is deleted.
757 DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
758 (double)lhead->start, (double)lhead->size ));
760 for (i=0; i<num_locks && lhead; i++) {
762 struct posix_lock *lock = &locks[i];
763 struct lock_list *l_curr;
766 * Walk the lock list, checking for overlaps. Note that
767 * the lock list can expand within this loop if the current
768 * range being examined needs to be split.
771 for (l_curr = lhead; l_curr;) {
773 DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
774 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
776 if ( (l_curr->start >= (lock->start + lock->size)) ||
777 (lock->start >= (l_curr->start + l_curr->size))) {
779 /* No overlap with this lock - leave this range alone. */
780 /*********************************************
791 **********************************************/
793 DEBUG(10,("no overlap case.\n" ));
795 l_curr = l_curr->next;
797 } else if ( (l_curr->start >= lock->start) &&
798 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
801 * This unlock is completely overlapped by this existing lock range
802 * and thus should have no effect (not be unlocked). Delete it from the list.
804 /*********************************************
808 +---------------------------+
810 +---------------------------+
811 **********************************************/
812 /* Save the next pointer */
813 struct lock_list *ul_next = l_curr->next;
815 DEBUG(10,("delete case.\n" ));
817 DLIST_REMOVE(lhead, l_curr);
819 break; /* No more list... */
823 } else if ( (l_curr->start >= lock->start) &&
824 (l_curr->start < lock->start + lock->size) &&
825 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
828 * This unlock overlaps the existing lock range at the high end.
829 * Truncate by moving start to existing range end and reducing size.
831 /*********************************************
842 **********************************************/
844 l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
845 l_curr->start = lock->start + lock->size;
847 DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
848 (double)l_curr->start, (double)l_curr->size ));
850 l_curr = l_curr->next;
852 } else if ( (l_curr->start < lock->start) &&
853 (l_curr->start + l_curr->size > lock->start) &&
854 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
857 * This unlock overlaps the existing lock range at the low end.
858 * Truncate by reducing size.
860 /*********************************************
871 **********************************************/
873 l_curr->size = lock->start - l_curr->start;
875 DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
876 (double)l_curr->start, (double)l_curr->size ));
878 l_curr = l_curr->next;
880 } else if ( (l_curr->start < lock->start) &&
881 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
883 * Worst case scenario. Unlock request completely overlaps an existing
884 * lock range. Split the request into two, push the new (upper) request
885 * into the dlink list, and continue with the entry after ul_new (as we
886 * know that ul_new will not overlap with this lock).
888 /*********************************************
889 +---------------------------+
891 +---------------------------+
896 +-------+ +---------+
898 +-------+ +---------+
899 **********************************************/
900 struct lock_list *l_new = (struct lock_list *)talloc(ctx,
901 sizeof(struct lock_list));
904 DEBUG(0,("posix_lock_list: talloc fail.\n"));
905 return NULL; /* The talloc_destroy takes care of cleanup. */
909 l_new->start = lock->start + lock->size;
910 l_new->size = l_curr->start + l_curr->size - l_new->start;
912 /* Truncate the l_curr. */
913 l_curr->size = lock->start - l_curr->start;
915 DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
916 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
917 (double)l_new->start, (double)l_new->size ));
920 * Add into the dlink list after the l_curr point - NOT at lhead.
921 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
924 l_new->prev = l_curr;
925 l_new->next = l_curr->next;
926 l_curr->next = l_new;
928 /* And move after the link we added. */
929 l_curr = l_new->next;
934 * This logic case should never happen. Ensure this is the
935 * case by forcing an abort.... Remove in production.
939 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
940 lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
944 } /* end for ( l_curr = lhead; l_curr;) */
945 } /* end for (i=0; i<num_locks && ul_head; i++) */
947 SAFE_FREE(dbuf.dptr);
952 /****************************************************************************
953 POSIX function to acquire a lock. Returns True if the
954 lock could be granted, False if not.
955 ****************************************************************************/
957 BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
962 size_t entry_num = 0;
964 TALLOC_CTX *l_ctx = NULL;
965 struct lock_list *llist = NULL;
966 struct lock_list *ll = NULL;
967 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
969 DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
970 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
973 * If the requested lock won't fit in the POSIX range, we will
974 * pretend it was successful.
977 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
981 * Windows is very strange. It allows read locks to be overlayed
982 * (even over a write lock), but leaves the write lock in force until the first
983 * unlock. It also reference counts the locks. This means the following sequence :
986 * ------------------------------------------------------------------------
987 * WRITE LOCK : start = 2, len = 10
988 * READ LOCK: start =0, len = 10 - FAIL
989 * READ LOCK : start = 0, len = 14
990 * READ LOCK: start =0, len = 10 - FAIL
991 * UNLOCK : start = 2, len = 10
992 * READ LOCK: start =0, len = 10 - OK
994 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
995 * would leave a single read lock over the 0-14 region. In order to
996 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
997 * entries, one for each overlayed lock request. We are guarenteed by the brlock
998 * semantics that if a write lock is added, then it will be first in the array.
1001 if ((l_ctx = talloc_init("set_posix_lock")) == NULL) {
1002 DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
1003 return True; /* Not a fatal error. */
1006 if ((ll = (struct lock_list *)talloc(l_ctx, sizeof(struct lock_list))) == NULL) {
1007 DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
1008 talloc_destroy(l_ctx);
1009 return True; /* Not a fatal error. */
1013 * Create the initial list entry containing the
1014 * lock we want to add.
1021 DLIST_ADD(llist, ll);
1024 * The following call calculates if there are any
1025 * overlapping locks held by this process on
1026 * fd's open on the same file and splits this list
1027 * into a list of lock ranges that do not overlap with existing
1031 llist = posix_lock_list(l_ctx, llist, fsp);
1034 * Now we have the list of ranges to lock it is safe to add the
1035 * entry into the POSIX lock tdb. We take note of the entry we
1036 * added here in case we have to remove it on POSIX lock fail.
1039 if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
1040 DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
1041 talloc_destroy(l_ctx);
1046 * Add the POSIX locks on the list of ranges returned.
1047 * As the lock is supposed to be added atomically, we need to
1048 * back out all the locks if any one of these calls fail.
1051 for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
1055 DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
1056 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1058 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1059 DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
1060 posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
1069 * Back out all the POSIX locks we have on fail.
1072 for (ll = llist; lock_count; ll = ll->next, lock_count--) {
1076 DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
1077 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1079 posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
1083 * Remove the tdb entry for this lock.
1086 delete_posix_lock_entry_by_index(fsp,entry_num);
1089 talloc_destroy(l_ctx);
1093 /****************************************************************************
1094 POSIX function to release a lock. Returns True if the
1095 lock could be released, False if not.
1096 ****************************************************************************/
1098 BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
1103 TALLOC_CTX *ul_ctx = NULL;
1104 struct lock_list *ulist = NULL;
1105 struct lock_list *ul = NULL;
1106 struct posix_lock deleted_lock;
1107 int num_overlapped_entries;
1109 DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
1110 fsp->fsp_name, (double)u_offset, (double)u_count ));
1113 * If the requested lock won't fit in the POSIX range, we will
1114 * pretend it was successful.
1117 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1121 * We treat this as one unlock request for POSIX accounting purposes even
1122 * if it may later be split into multiple smaller POSIX unlock ranges.
1123 * num_overlapped_entries is the number of existing locks that have any
1124 * overlap with this unlock request.
1127 num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
1129 if (num_overlapped_entries == -1) {
1130 smb_panic("release_posix_lock: unable find entry to delete !\n");
1134 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
1135 * a POSIX write lock, then before doing the unlock we need to downgrade
1136 * the POSIX lock to a read lock. This allows any overlapping read locks
1137 * to be atomically maintained.
1140 if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
1141 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1142 DEBUG(0,("release_posix_lock: downgrade of lock failed with error %s !\n", strerror(errno) ));
1147 if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
1148 DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
1149 return True; /* Not a fatal error. */
1152 if ((ul = (struct lock_list *)talloc(ul_ctx, sizeof(struct lock_list))) == NULL) {
1153 DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
1154 talloc_destroy(ul_ctx);
1155 return True; /* Not a fatal error. */
1159 * Create the initial list entry containing the
1160 * lock we want to remove.
1167 DLIST_ADD(ulist, ul);
1170 * The following call calculates if there are any
1171 * overlapping locks held by this process on
1172 * fd's open on the same file and creates a
1173 * list of unlock ranges that will allow
1174 * POSIX lock ranges to remain on the file whilst the
1175 * unlocks are performed.
1178 ulist = posix_lock_list(ul_ctx, ulist, fsp);
1181 * Release the POSIX locks on the list of ranges returned.
1184 for(; ulist; ulist = ulist->next) {
1185 offset = ulist->start;
1186 count = ulist->size;
1188 DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
1189 (double)offset, (double)count ));
1191 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
1195 talloc_destroy(ul_ctx);
1200 /****************************************************************************
1201 Remove all lock entries for a specific dev/inode pair from the tdb.
1202 ****************************************************************************/
1204 static void delete_posix_lock_entries(files_struct *fsp)
1206 TDB_DATA kbuf = locking_key_fsp(fsp);
1208 if (tdb_delete(posix_lock_tdb, kbuf) == -1)
1209 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
1212 /****************************************************************************
1214 ****************************************************************************/
1216 static void dump_entry(struct posix_lock *pl)
1218 DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
1219 (double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
1222 /****************************************************************************
1223 Remove any locks on this fd. Called from file_close().
1224 ****************************************************************************/
1226 void posix_locking_close_file(files_struct *fsp)
1228 struct posix_lock *entries = NULL;
1232 * Optimization for the common case where we are the only
1233 * opener of a file. If all fd entries are our own, we don't
1234 * need to explicitly release all the locks via the POSIX functions,
1235 * we can just remove all the entries in the tdb and allow the
1236 * close to remove the real locks.
1239 count = get_posix_lock_entries(fsp, &entries);
1242 DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
1246 for (i = 0; i < count; i++) {
1247 if (entries[i].fd != fsp->fd )
1250 dump_entry(&entries[i]);
1254 /* All locks are ours. */
1255 DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n",
1256 fsp->fsp_name, (unsigned int)count ));
1258 delete_posix_lock_entries(fsp);
1263 * Difficult case. We need to delete all our locks, whilst leaving
1264 * all other POSIX locks in place.
1267 for (i = 0; i < count; i++) {
1268 struct posix_lock *pl = &entries[i];
1269 if (pl->fd == fsp->fd)
1270 release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
1275 /*******************************************************************
1276 Create the in-memory POSIX lock databases.
1277 ********************************************************************/
1279 BOOL posix_locking_init(int read_only)
1281 if (posix_lock_tdb && posix_pending_close_tdb)
1284 if (!posix_lock_tdb)
1285 posix_lock_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1286 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1287 if (!posix_lock_tdb) {
1288 DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
1291 if (!posix_pending_close_tdb)
1292 posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1293 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1294 if (!posix_pending_close_tdb) {
1295 DEBUG(0,("Failed to open POSIX pending close database.\n"));
1302 /*******************************************************************
1303 Delete the in-memory POSIX lock databases.
1304 ********************************************************************/
1306 BOOL posix_locking_end(void)
1308 if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
1310 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)