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;
28 extern int global_smbpid;
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);
105 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
107 dbuf.dptr = Realloc(dbuf.dptr, dbuf.dsize + sizeof(int));
109 DEBUG(0,("add_fd_to_close_entry: Realloc fail !\n"));
112 memcpy(dbuf.dptr + dbuf.dsize, &fsp->fd, sizeof(int));
113 dbuf.dsize += sizeof(int);
115 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
116 DEBUG(0,("add_fd_to_close_entry: tdb_store fail !\n"));
123 /****************************************************************************
124 Remove all fd entries for a specific dev/inode pair from the tdb.
125 ****************************************************************************/
127 static void delete_close_entries(files_struct *fsp)
129 TDB_DATA kbuf = locking_key_fsp(fsp);
131 if (tdb_delete(posix_pending_close_tdb, kbuf) == -1)
132 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
135 /****************************************************************************
136 Get the array of POSIX pending close records for an open fsp. Caller must
137 free. Returns number of entries.
138 ****************************************************************************/
140 static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
142 TDB_DATA kbuf = locking_key_fsp(fsp);
149 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
155 *entries = (int *)dbuf.dptr;
156 count = (size_t)(dbuf.dsize / sizeof(int));
161 /****************************************************************************
162 Get the array of POSIX locks for an fsp. Caller must free. Returns
164 ****************************************************************************/
166 static size_t get_posix_lock_entries(files_struct *fsp, struct posix_lock **entries)
168 TDB_DATA kbuf = locking_key_fsp(fsp);
176 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
182 *entries = (struct posix_lock *)dbuf.dptr;
183 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
188 /****************************************************************************
189 Deal with pending closes needed by POSIX locking support.
190 Note that posix_locking_close_file() is expected to have been called
191 to delete all locks on this fsp before this function is called.
192 ****************************************************************************/
194 int fd_close_posix(struct connection_struct *conn, files_struct *fsp)
199 struct posix_lock *entries = NULL;
200 int *fd_array = NULL;
201 BOOL locks_on_other_fds = False;
203 if (!lp_posix_locking(SNUM(conn))) {
205 * No POSIX to worry about, just close.
207 ret = conn->vfs_ops.close(fsp->fd);
213 * Get the number of outstanding POSIX locks on this dev/inode pair.
216 count = get_posix_lock_entries(fsp, &entries);
219 * Check if there are any outstanding locks belonging to
220 * other fd's. This should never be the case if posix_locking_close_file()
221 * has been called first, but it never hurts to be *sure*.
224 for (i = 0; i < count; i++) {
225 if (entries[i].fd != fsp->fd) {
226 locks_on_other_fds = True;
231 if (locks_on_other_fds) {
234 * There are outstanding locks on this dev/inode pair on other fds.
235 * Add our fd to the pending close tdb and set fsp->fd to -1.
238 if (!add_fd_to_close_entry(fsp)) {
239 free((char *)entries);
243 free((char *)entries);
249 free((char *)entries);
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(fd_array[i]) == -1) {
268 * Delete all fd's stored in the tdb
269 * for this dev/inode pair.
272 delete_close_entries(fsp);
276 free((char *)fd_array);
279 * Finally close the fd associated with this fsp.
282 ret = conn->vfs_ops.close(fsp->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(*locks)*((count-1) - entry));
333 dbuf.dsize -= sizeof(*locks);
334 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
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;
361 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
363 *pentry_num = (size_t)(dbuf.dsize / sizeof(pl));
372 pl.lock_type = lock_type;
374 dbuf.dptr = Realloc(dbuf.dptr, dbuf.dsize + sizeof(pl));
376 DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
380 memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
381 dbuf.dsize += sizeof(pl);
383 if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
384 DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
390 DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: dev=%.0f inode=%.0f\n",
391 fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size,
392 (double)fsp->dev, (double)fsp->inode ));
402 /****************************************************************************
403 Calculate if locks have any overlap at all.
404 ****************************************************************************/
406 static BOOL does_lock_overlap(SMB_OFF_T start1, SMB_OFF_T size1, SMB_OFF_T start2, SMB_OFF_T size2)
408 if (start1 >= start2 && start1 <= start2 + size2)
411 if (start1 < start2 && start1 + size1 > start2);
417 /****************************************************************************
418 Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
419 deleted and the number of records that are overlapped by this one, or -1 on error.
420 ****************************************************************************/
422 static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
424 TDB_DATA kbuf = locking_key_fsp(fsp);
426 struct posix_lock *locks;
429 int num_overlapping_records = 0;
433 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
436 DEBUG(10,("delete_posix_lock_entry: tdb_fetch failed !\n"));
440 /* There are existing locks - find a match. */
441 locks = (struct posix_lock *)dbuf.dptr;
442 count = (size_t)(dbuf.dsize / sizeof(*locks));
445 * Search for and delete the first record that matches the
449 for (i=0; i<count; i++) {
450 struct posix_lock *entry = &locks[i];
452 if (entry->fd == fsp->fd &&
453 entry->start == start &&
454 entry->size == size) {
456 /* Make a copy if requested. */
460 /* Found it - delete it. */
462 tdb_delete(posix_lock_tdb, kbuf);
465 memmove(&locks[i], &locks[i+1], sizeof(*locks)*((count-1) - i));
467 dbuf.dsize -= sizeof(*locks);
468 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
480 * Count the number of entries that are
481 * overlapped by this unlock request.
484 for (i = 0; i < count; i++) {
485 struct posix_lock *entry = &locks[i];
487 if (fsp->fd == entry->fd &&
488 does_lock_overlap( start, size, entry->start, entry->size))
489 num_overlapping_records++;
492 DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
493 posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
494 (unsigned int)num_overlapping_records ));
499 return num_overlapping_records;
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(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
558 * In this case SMB_OFF_T is 64 bits,
559 * and the underlying system can handle 64 bit signed locks.
562 SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
563 SMB_OFF_T mask = (mask2<<1);
564 SMB_OFF_T max_positive_lock_offset = ~mask;
566 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
569 * In this case either SMB_OFF_T is 32 bits,
570 * or the underlying system cannot handle 64 bit signed locks.
571 * All offsets & counts must be 2^31 or less.
574 SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
576 #endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
579 * If the given offset was > max_positive_lock_offset then we cannot map this at all
583 if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
584 DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
585 (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
590 * We must truncate the offset and count to less than max_positive_lock_offset.
593 offset &= max_positive_lock_offset;
594 count &= max_positive_lock_offset;
598 * Deal with a very common case of count of all ones.
599 * (lock entire file).
602 if(count == (SMB_OFF_T)-1)
603 count = max_positive_lock_offset;
606 * Truncate count to end at max lock offset.
609 if (offset + count < 0 || offset + count > max_positive_lock_offset)
610 count = max_positive_lock_offset - offset;
613 * If we ate all the count, ignore this lock.
617 DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
618 (double)u_offset, (double)u_count ));
623 * The mapping was successful.
626 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
627 (double)offset, (double)count ));
629 *offset_out = offset;
635 #if defined(LARGE_SMB_OFF_T)
636 /****************************************************************************
637 Pathetically try and map a 64 bit lock offset into 31 bits. I hate Windows :-).
638 ****************************************************************************/
640 static uint32 map_lock_offset(uint32 high, uint32 low)
644 uint32 highcopy = high;
647 * Try and find out how many significant bits there are in high.
650 for(i = 0; highcopy; i++)
654 * We use 31 bits not 32 here as POSIX
655 * lock offsets may not be negative.
658 mask = (~0) << (31 - i);
661 return 0; /* Fail. */
669 /****************************************************************************
670 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
671 broken NFS implementations.
672 ****************************************************************************/
674 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
677 struct connection_struct *conn = fsp->conn;
679 #if defined(LARGE_SMB_OFF_T)
681 * In the 64 bit locking case we store the original
682 * values in case we have to map to a 32 bit lock on
683 * a filesystem that doesn't support 64 bit locks.
685 SMB_OFF_T orig_offset = offset;
686 SMB_OFF_T orig_count = count;
687 #endif /* LARGE_SMB_OFF_T */
689 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
691 ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
693 if (!ret && (errno == EFBIG)) {
695 dbgtext("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n", (double)offset,(double)count);
696 dbgtext("a 'file too large' error. This can happen when using 64 bit lock offsets\n");
697 dbgtext("on 32 bit NFS mounted file systems. Retrying with 32 bit truncated length.\n");
699 /* 32 bit NFS file system, retry with smaller offset */
702 ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
705 /* A lock query - just return. */
706 if (op == SMB_F_GETLK)
709 /* A lock set or unset. */
711 DEBUG(3,("posix_fcntl_lock: lock failed at offset %.0f count %.0f op %d type %d (%s)\n",
712 (double)offset,(double)count,op,type,strerror(errno)));
714 /* Perhaps it doesn't support this sort of locking ? */
715 if (errno == EINVAL) {
716 #if defined(LARGE_SMB_OFF_T)
719 * Ok - if we get here then we have a 64 bit lock request
720 * that has returned EINVAL. Try and map to 31 bits for offset
721 * and length and try again. This may happen if a filesystem
722 * doesn't support 64 bit offsets (efs/ufs) although the underlying
725 uint32 off_low = (orig_offset & 0xFFFFFFFF);
726 uint32 off_high = ((orig_offset >> 32) & 0xFFFFFFFF);
728 count = (orig_count & 0x7FFFFFFF);
729 offset = (SMB_OFF_T)map_lock_offset(off_high, off_low);
730 ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
732 if (errno == EINVAL) {
733 DEBUG(3,("posix_fcntl_lock: locking not supported? returning True\n"));
738 DEBUG(3,("posix_fcntl_lock: 64 -> 32 bit modified lock call successful\n"));
741 #else /* LARGE_SMB_OFF_T */
742 DEBUG(3,("locking not supported? returning True\n"));
744 #endif /* LARGE_SMB_OFF_T */
750 DEBUG(8,("posix_fcntl_lock: Lock call successful\n"));
755 /****************************************************************************
756 POSIX function to see if a file region is locked. Returns True if the
757 region is locked, False otherwise.
758 ****************************************************************************/
760 BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
764 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
766 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
767 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
770 * If the requested lock won't fit in the POSIX range, we will
771 * never set it, so presume it is not locked.
774 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
778 * Note that most UNIX's can *test* for a write lock on
779 * a read-only fd, just not *set* a write lock on a read-only
780 * fd. So we don't need to use map_lock_type here.
783 return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
787 * Structure used when splitting a lock range
788 * into a POSIX lock range. Doubly linked list.
792 struct lock_list *next;
793 struct lock_list *prev;
798 /****************************************************************************
799 Create a list of lock ranges that don't overlap a given range. Used in calculating
800 POSIX locks and unlocks. This is a difficult function that requires ASCII art to
802 ****************************************************************************/
804 static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
806 TDB_DATA kbuf = locking_key_fsp(fsp);
808 struct posix_lock *locks;
813 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
818 locks = (struct posix_lock *)dbuf.dptr;
819 num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
822 * Check the current lock list on this dev/inode pair.
823 * Quit if the list is deleted.
826 DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
827 (double)lhead->start, (double)lhead->size ));
829 for (i=0; i<num_locks && lhead; i++) {
831 struct posix_lock *lock = &locks[i];
832 struct lock_list *l_curr;
835 * Walk the lock list, checking for overlaps. Note that
836 * the lock list can expand within this loop if the current
837 * range being examined needs to be split.
840 for (l_curr = lhead; l_curr;) {
842 DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
843 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
845 if ( (l_curr->start >= (lock->start + lock->size)) ||
846 (lock->start >= (l_curr->start + l_curr->size))) {
848 /* No overlap with this lock - leave this range alone. */
849 /*********************************************
860 **********************************************/
862 DEBUG(10,("no overlap case.\n" ));
864 l_curr = l_curr->next;
866 } else if ( (l_curr->start >= lock->start) &&
867 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
870 * This unlock is completely overlapped by this existing lock range
871 * and thus should have no effect (not be unlocked). Delete it from the list.
873 /*********************************************
877 +---------------------------+
879 +---------------------------+
880 **********************************************/
881 /* Save the next pointer */
882 struct lock_list *ul_next = l_curr->next;
884 DEBUG(10,("delete case.\n" ));
886 DLIST_REMOVE(lhead, l_curr);
888 break; /* No more list... */
892 } else if ( (l_curr->start >= lock->start) &&
893 (l_curr->start < lock->start + lock->size) &&
894 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
897 * This unlock overlaps the existing lock range at the high end.
898 * Truncate by moving start to existing range end and reducing size.
900 /*********************************************
911 **********************************************/
913 l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
914 l_curr->start = lock->start + lock->size;
916 DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
917 (double)l_curr->start, (double)l_curr->size ));
919 l_curr = l_curr->next;
921 } else if ( (l_curr->start < lock->start) &&
922 (l_curr->start + l_curr->size > lock->start) &&
923 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
926 * This unlock overlaps the existing lock range at the low end.
927 * Truncate by reducing size.
929 /*********************************************
940 **********************************************/
942 l_curr->size = lock->start - l_curr->start;
944 DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
945 (double)l_curr->start, (double)l_curr->size ));
947 l_curr = l_curr->next;
949 } else if ( (l_curr->start < lock->start) &&
950 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
952 * Worst case scenario. Unlock request completely overlaps an existing
953 * lock range. Split the request into two, push the new (upper) request
954 * into the dlink list, and continue with the entry after ul_new (as we
955 * know that ul_new will not overlap with this lock).
957 /*********************************************
958 +---------------------------+
960 +---------------------------+
965 +-------+ +---------+
967 +-------+ +---------+
968 **********************************************/
969 struct lock_list *l_new = (struct lock_list *)talloc(ctx,
970 sizeof(struct lock_list));
973 DEBUG(0,("posix_lock_list: talloc fail.\n"));
974 return NULL; /* The talloc_destroy takes care of cleanup. */
978 l_new->start = lock->start + lock->size;
979 l_new->size = l_curr->start + l_curr->size - l_new->start;
981 /* Truncate the l_curr. */
982 l_curr->size = lock->start - l_curr->start;
984 DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
985 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
986 (double)l_new->start, (double)l_new->size ));
989 * Add into the dlink list after the l_curr point - NOT at lhead.
990 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
993 l_new->prev = l_curr;
994 l_new->next = l_curr->next;
995 l_curr->next = l_new;
997 /* And move after the link we added. */
998 l_curr = l_new->next;
1003 * This logic case should never happen. Ensure this is the
1004 * case by forcing an abort.... Remove in production.
1008 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
1009 lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
1013 } /* end for ( l_curr = lhead; l_curr;) */
1014 } /* end for (i=0; i<num_locks && ul_head; i++) */
1022 /****************************************************************************
1023 POSIX function to acquire a lock. Returns True if the
1024 lock could be granted, False if not.
1025 ****************************************************************************/
1027 BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
1032 size_t entry_num = 0;
1034 TALLOC_CTX *l_ctx = NULL;
1035 struct lock_list *llist = NULL;
1036 struct lock_list *ll = NULL;
1037 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
1039 DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
1040 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
1043 * If the requested lock won't fit in the POSIX range, we will
1044 * pretend it was successful.
1047 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1051 * Windows is very strange. It allows read locks to be overlayed
1052 * (even over a write lock), but leaves the write lock in force until the first
1053 * unlock. It also reference counts the locks. This means the following sequence :
1056 * ------------------------------------------------------------------------
1057 * WRITE LOCK : start = 2, len = 10
1058 * READ LOCK: start =0, len = 10 - FAIL
1059 * READ LOCK : start = 0, len = 14
1060 * READ LOCK: start =0, len = 10 - FAIL
1061 * UNLOCK : start = 2, len = 10
1062 * READ LOCK: start =0, len = 10 - OK
1064 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
1065 * would leave a single read lock over the 0-14 region. In order to
1066 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
1067 * entries, one for each overlayed lock request. We are guarenteed by the brlock
1068 * semantics that if a write lock is added, then it will be first in the array.
1071 if ((l_ctx = talloc_init()) == NULL) {
1072 DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
1073 return True; /* Not a fatal error. */
1076 if ((ll = (struct lock_list *)talloc(l_ctx, sizeof(struct lock_list))) == NULL) {
1077 DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
1078 talloc_destroy(l_ctx);
1079 return True; /* Not a fatal error. */
1083 * Create the initial list entry containing the
1084 * lock we want to add.
1091 DLIST_ADD(llist, ll);
1094 * The following call calculates if there are any
1095 * overlapping locks held by this process on
1096 * fd's open on the same file and splits this list
1097 * into a list of lock ranges that do not overlap with existing
1101 llist = posix_lock_list(l_ctx, llist, fsp);
1104 * Now we have the list of ranges to lock it is safe to add the
1105 * entry into the POSIX lock tdb. We take note of the entry we
1106 * added here in case we have to remove it on POSIX lock fail.
1109 if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
1110 DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
1111 talloc_destroy(l_ctx);
1116 * Add the POSIX locks on the list of ranges returned.
1117 * As the lock is supposed to be added atomically, we need to
1118 * back out all the locks if any one of these calls fail.
1121 for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
1125 DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
1126 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1128 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1129 DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f\n",
1130 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1139 * Back out all the POSIX locks we have on fail.
1142 for (ll = llist; lock_count; ll = ll->next, lock_count--) {
1146 DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
1147 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1149 posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
1153 * Remove the tdb entry for this lock.
1156 delete_posix_lock_entry_by_index(fsp,entry_num);
1159 talloc_destroy(l_ctx);
1163 /****************************************************************************
1164 POSIX function to release a lock. Returns True if the
1165 lock could be released, False if not.
1166 ****************************************************************************/
1168 BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
1173 TALLOC_CTX *ul_ctx = NULL;
1174 struct lock_list *ulist = NULL;
1175 struct lock_list *ul = NULL;
1176 struct posix_lock deleted_lock;
1177 int num_overlapped_entries;
1179 DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
1180 fsp->fsp_name, (double)u_offset, (double)u_count ));
1183 * If the requested lock won't fit in the POSIX range, we will
1184 * pretend it was successful.
1187 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1191 * We treat this as one unlock request for POSIX accounting purposes even
1192 * if it may later be split into multiple smaller POSIX unlock ranges.
1193 * num_overlapped_entries is the number of existing locks that have any
1194 * overlap with this unlock request.
1197 num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
1199 if (num_overlapped_entries == -1) {
1200 smb_panic("release_posix_lock: unable find entry to delete !\n");
1204 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
1205 * a POSIX write lock, then before doing the unlock we need to downgrade
1206 * the POSIX lock to a read lock. This allows any overlapping read locks
1207 * to be atomically maintained.
1210 if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
1211 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1212 DEBUG(0,("release_posix_lock: downgrade of lock failed !\n"));
1217 if ((ul_ctx = talloc_init()) == NULL) {
1218 DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
1219 return True; /* Not a fatal error. */
1222 if ((ul = (struct lock_list *)talloc(ul_ctx, sizeof(struct lock_list))) == NULL) {
1223 DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
1224 talloc_destroy(ul_ctx);
1225 return True; /* Not a fatal error. */
1229 * Create the initial list entry containing the
1230 * lock we want to remove.
1237 DLIST_ADD(ulist, ul);
1240 * The following call calculates if there are any
1241 * overlapping locks held by this process on
1242 * fd's open on the same file and creates a
1243 * list of unlock ranges that will allow
1244 * POSIX lock ranges to remain on the file whilst the
1245 * unlocks are performed.
1248 ulist = posix_lock_list(ul_ctx, ulist, fsp);
1251 * Release the POSIX locks on the list of ranges returned.
1254 for(; ulist; ulist = ulist->next) {
1255 offset = ulist->start;
1256 count = ulist->size;
1258 DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
1259 (double)offset, (double)count ));
1261 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
1265 talloc_destroy(ul_ctx);
1270 /****************************************************************************
1271 Remove all lock entries for a specific dev/inode pair from the tdb.
1272 ****************************************************************************/
1274 static void delete_posix_lock_entries(files_struct *fsp)
1276 TDB_DATA kbuf = locking_key_fsp(fsp);
1278 if (tdb_delete(posix_lock_tdb, kbuf) == -1)
1279 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
1282 /****************************************************************************
1284 ****************************************************************************/
1286 static void dump_entry(struct posix_lock *pl)
1288 DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
1289 (double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
1292 /****************************************************************************
1293 Remove any locks on this fd. Called from file_close().
1294 ****************************************************************************/
1296 void posix_locking_close_file(files_struct *fsp)
1298 struct posix_lock *entries = NULL;
1302 * Optimization for the common case where we are the only
1303 * opener of a file. If all fd entries are our own, we don't
1304 * need to explicitly release all the locks via the POSIX functions,
1305 * we can just remove all the entries in the tdb and allow the
1306 * close to remove the real locks.
1309 count = get_posix_lock_entries(fsp, &entries);
1312 DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
1316 for (i = 0; i < count; i++) {
1317 if (entries[i].fd != fsp->fd )
1320 dump_entry(&entries[i]);
1324 /* All locks are ours. */
1325 DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n",
1326 fsp->fsp_name, (unsigned int)count ));
1327 free((char *)entries);
1328 delete_posix_lock_entries(fsp);
1333 * Difficult case. We need to delete all our locks, whilst leaving
1334 * all other POSIX locks in place.
1337 for (i = 0; i < count; i++) {
1338 struct posix_lock *pl = &entries[i];
1339 if (pl->fd == fsp->fd)
1340 release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
1342 free((char *)entries);
1345 /*******************************************************************
1346 Create the in-memory POSIX lock databases.
1347 ********************************************************************/
1349 BOOL posix_locking_init(int read_only)
1351 if (posix_lock_tdb && posix_pending_close_tdb)
1354 if (!posix_lock_tdb)
1355 posix_lock_tdb = tdb_open(NULL, 0, TDB_INTERNAL,
1356 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1357 if (!posix_lock_tdb) {
1358 DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
1361 if (!posix_pending_close_tdb)
1362 posix_pending_close_tdb = tdb_open(NULL, 0, TDB_INTERNAL,
1363 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1364 if (!posix_pending_close_tdb) {
1365 DEBUG(0,("Failed to open POSIX pending close database.\n"));
1372 /*******************************************************************
1373 Delete the in-memory POSIX lock databases.
1374 ********************************************************************/
1376 BOOL posix_locking_end(void)
1378 if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
1380 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)