2 Unix SMB/Netbios implementation.
5 Copyright (C) Jeremy Allison 1992-2000
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 2 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program; if not, write to the Free Software
19 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000.
27 extern int DEBUGLEVEL;
30 * The POSIX locking database handle.
33 static TDB_CONTEXT *posix_lock_tdb;
36 * The pending close database handle.
39 static TDB_CONTEXT *posix_pending_close_tdb;
42 * The data in POSIX lock records is an unsorted linear array of these
43 * records. It is unnecessary to store the count as tdb provides the
55 * The data in POSIX pending close records is an unsorted linear array of int
56 * records. It is unnecessary to store the count as tdb provides the
60 /* The key used in both the POSIX databases. */
62 struct posix_lock_key {
67 /*******************************************************************
68 Form a static locking key for a dev/inode pair.
69 ******************************************************************/
71 static TDB_DATA locking_key(SMB_DEV_T dev, SMB_INO_T inode)
73 static struct posix_lock_key key;
76 memset(&key, '\0', sizeof(key));
79 kbuf.dptr = (char *)&key;
80 kbuf.dsize = sizeof(key);
84 /*******************************************************************
85 Convenience function to get a key from an fsp.
86 ******************************************************************/
88 static TDB_DATA locking_key_fsp(files_struct *fsp)
90 return locking_key(fsp->dev, fsp->inode);
93 /****************************************************************************
94 Add an fd to the pending close tdb.
95 ****************************************************************************/
97 static BOOL add_fd_to_close_entry(files_struct *fsp)
99 TDB_DATA kbuf = locking_key_fsp(fsp);
105 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
107 tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(int));
109 DEBUG(0,("add_fd_to_close_entry: Realloc fail !\n"));
110 if (dbuf.dptr) free(dbuf.dptr);
114 memcpy(dbuf.dptr + dbuf.dsize, &fsp->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"));
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 = conn->vfs_ops.close(fsp,fsp->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->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->fd to -1.
240 if (!add_fd_to_close_entry(fsp)) {
241 free((char *)entries);
245 free((char *)entries);
251 free((char *)entries);
254 * No outstanding POSIX locks. Get the pending close fd's
255 * from the tdb and close them all.
258 count = get_posix_pending_close_entries(fsp, &fd_array);
261 DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
263 for(i = 0; i < count; i++) {
264 if (conn->vfs_ops.close(fsp,fd_array[i]) == -1) {
270 * Delete all fd's stored in the tdb
271 * for this dev/inode pair.
274 delete_close_entries(fsp);
278 free((char *)fd_array);
281 * Finally close the fd associated with this fsp.
284 ret = conn->vfs_ops.close(fsp,fsp->fd);
286 if (saved_errno != 0) {
296 /****************************************************************************
298 ****************************************************************************/
300 static const char *posix_lock_type_name(int lock_type)
302 return (lock_type == F_RDLCK) ? "READ" : "WRITE";
305 /****************************************************************************
306 Delete a POSIX lock entry by index number. Used if the tdb add succeeds, but
307 then the POSIX fcntl lock fails.
308 ****************************************************************************/
310 static BOOL delete_posix_lock_entry_by_index(files_struct *fsp, size_t entry)
312 TDB_DATA kbuf = locking_key_fsp(fsp);
314 struct posix_lock *locks;
319 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
322 DEBUG(10,("delete_posix_lock_entry_by_index: tdb_fetch failed !\n"));
326 count = (size_t)(dbuf.dsize / sizeof(struct posix_lock));
327 locks = (struct posix_lock *)dbuf.dptr;
330 tdb_delete(posix_lock_tdb, kbuf);
332 if (entry < count-1) {
333 memmove(&locks[entry], &locks[entry+1], sizeof(*locks)*((count-1) - entry));
335 dbuf.dsize -= sizeof(*locks);
336 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
349 /****************************************************************************
350 Add an entry into the POSIX locking tdb. We return the index number of the
351 added lock (used in case we need to delete *exactly* this entry). Returns
352 False on fail, True on success.
353 ****************************************************************************/
355 static BOOL add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *pentry_num)
357 TDB_DATA kbuf = locking_key_fsp(fsp);
359 struct posix_lock pl;
364 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
366 *pentry_num = (size_t)(dbuf.dsize / sizeof(pl));
375 pl.lock_type = lock_type;
377 tp = Realloc(dbuf.dptr, dbuf.dsize + sizeof(pl));
379 DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
384 memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
385 dbuf.dsize += sizeof(pl);
387 if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
388 DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
394 DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: dev=%.0f inode=%.0f\n",
395 fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size,
396 (double)fsp->dev, (double)fsp->inode ));
406 /****************************************************************************
407 Calculate if locks have any overlap at all.
408 ****************************************************************************/
410 static BOOL does_lock_overlap(SMB_OFF_T start1, SMB_OFF_T size1, SMB_OFF_T start2, SMB_OFF_T size2)
412 if (start1 >= start2 && start1 <= start2 + size2)
415 if (start1 < start2 && start1 + size1 > start2)
421 /****************************************************************************
422 Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
423 deleted and the number of records that are overlapped by this one, or -1 on error.
424 ****************************************************************************/
426 static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
428 TDB_DATA kbuf = locking_key_fsp(fsp);
430 struct posix_lock *locks;
433 int num_overlapping_records = 0;
437 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
440 DEBUG(10,("delete_posix_lock_entry: tdb_fetch failed !\n"));
444 /* There are existing locks - find a match. */
445 locks = (struct posix_lock *)dbuf.dptr;
446 count = (size_t)(dbuf.dsize / sizeof(*locks));
449 * Search for and delete the first record that matches the
453 for (i=0; i<count; i++) {
454 struct posix_lock *entry = &locks[i];
456 if (entry->fd == fsp->fd &&
457 entry->start == start &&
458 entry->size == size) {
460 /* Make a copy if requested. */
464 /* Found it - delete it. */
466 tdb_delete(posix_lock_tdb, kbuf);
469 memmove(&locks[i], &locks[i+1], sizeof(*locks)*((count-1) - i));
471 dbuf.dsize -= sizeof(*locks);
472 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
484 * Count the number of entries that are
485 * overlapped by this unlock request.
488 for (i = 0; i < count; i++) {
489 struct posix_lock *entry = &locks[i];
491 if (fsp->fd == entry->fd &&
492 does_lock_overlap( start, size, entry->start, entry->size))
493 num_overlapping_records++;
496 DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
497 posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
498 (unsigned int)num_overlapping_records ));
503 return num_overlapping_records;
511 /****************************************************************************
512 Utility function to map a lock type correctly depending on the open
514 ****************************************************************************/
516 static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
518 if((lock_type == WRITE_LOCK) && !fsp->can_write) {
520 * Many UNIX's cannot get a write lock on a file opened read-only.
521 * Win32 locking semantics allow this.
522 * Do the best we can and attempt a read-only lock.
524 DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
526 } else if((lock_type == READ_LOCK) && !fsp->can_read) {
528 * Ditto for read locks on write only files.
530 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(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 * If the given offset was > max_positive_lock_offset then we cannot map this at all
587 if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
588 DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
589 (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
594 * We must truncate the offset and count to less than max_positive_lock_offset.
597 offset &= max_positive_lock_offset;
598 count &= max_positive_lock_offset;
602 * Deal with a very common case of count of all ones.
603 * (lock entire file).
606 if(count == (SMB_OFF_T)-1)
607 count = max_positive_lock_offset;
610 * Truncate count to end at max lock offset.
613 if (offset + count < 0 || offset + count > max_positive_lock_offset)
614 count = max_positive_lock_offset - offset;
617 * If we ate all the count, ignore this lock.
621 DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
622 (double)u_offset, (double)u_count ));
627 * The mapping was successful.
630 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
631 (double)offset, (double)count ));
633 *offset_out = offset;
639 /****************************************************************************
640 Pathetically try and map a 64 bit lock offset into 31 bits. I hate Windows :-).
641 ****************************************************************************/
643 uint32 map_lock_offset(uint32 high, uint32 low)
647 uint32 highcopy = high;
650 * Try and find out how many significant bits there are in high.
653 for(i = 0; highcopy; i++)
657 * We use 31 bits not 32 here as POSIX
658 * lock offsets may not be negative.
661 mask = (~0) << (31 - i);
664 return 0; /* Fail. */
671 /****************************************************************************
672 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
673 broken NFS implementations.
674 ****************************************************************************/
676 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
679 struct connection_struct *conn = fsp->conn;
681 #if defined(LARGE_SMB_OFF_T)
683 * In the 64 bit locking case we store the original
684 * values in case we have to map to a 32 bit lock on
685 * a filesystem that doesn't support 64 bit locks.
687 SMB_OFF_T orig_offset = offset;
688 SMB_OFF_T orig_count = count;
689 #endif /* LARGE_SMB_OFF_T */
691 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
693 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
695 if (!ret && (errno == EFBIG)) {
697 dbgtext("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n", (double)offset,(double)count);
698 dbgtext("a 'file too large' error. This can happen when using 64 bit lock offsets\n");
699 dbgtext("on 32 bit NFS mounted file systems. Retrying with 32 bit truncated length.\n");
701 /* 32 bit NFS file system, retry with smaller offset */
704 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
707 /* A lock query - just return. */
708 if (op == SMB_F_GETLK)
711 /* A lock set or unset. */
713 DEBUG(3,("posix_fcntl_lock: lock failed at offset %.0f count %.0f op %d type %d (%s)\n",
714 (double)offset,(double)count,op,type,strerror(errno)));
716 /* Perhaps it doesn't support this sort of locking ? */
717 if (errno == EINVAL) {
718 #if defined(LARGE_SMB_OFF_T)
721 * Ok - if we get here then we have a 64 bit lock request
722 * that has returned EINVAL. Try and map to 31 bits for offset
723 * and length and try again. This may happen if a filesystem
724 * doesn't support 64 bit offsets (efs/ufs) although the underlying
727 uint32 off_low = (orig_offset & 0xFFFFFFFF);
728 uint32 off_high = ((orig_offset >> 32) & 0xFFFFFFFF);
730 count = (orig_count & 0x7FFFFFFF);
731 offset = (SMB_OFF_T)map_lock_offset(off_high, off_low);
732 ret = conn->vfs_ops.lock(fsp,fsp->fd,op,offset,count,type);
734 if (errno == EINVAL) {
735 DEBUG(3,("posix_fcntl_lock: locking not supported? returning True\n"));
740 DEBUG(3,("posix_fcntl_lock: 64 -> 32 bit modified lock call successful\n"));
743 #else /* LARGE_SMB_OFF_T */
744 DEBUG(3,("locking not supported? returning True\n"));
746 #endif /* LARGE_SMB_OFF_T */
752 DEBUG(8,("posix_fcntl_lock: Lock call successful\n"));
757 /****************************************************************************
758 POSIX function to see if a file region is locked. Returns True if the
759 region is locked, False otherwise.
760 ****************************************************************************/
762 BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
766 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
768 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
769 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
772 * If the requested lock won't fit in the POSIX range, we will
773 * never set it, so presume it is not locked.
776 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
780 * Note that most UNIX's can *test* for a write lock on
781 * a read-only fd, just not *set* a write lock on a read-only
782 * fd. So we don't need to use map_lock_type here.
785 return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
789 * Structure used when splitting a lock range
790 * into a POSIX lock range. Doubly linked list.
794 struct lock_list *next;
795 struct lock_list *prev;
800 /****************************************************************************
801 Create a list of lock ranges that don't overlap a given range. Used in calculating
802 POSIX locks and unlocks. This is a difficult function that requires ASCII art to
804 ****************************************************************************/
806 static struct lock_list *posix_lock_list(TALLOC_CTX *ctx, struct lock_list *lhead, files_struct *fsp)
808 TDB_DATA kbuf = locking_key_fsp(fsp);
810 struct posix_lock *locks;
815 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
820 locks = (struct posix_lock *)dbuf.dptr;
821 num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
824 * Check the current lock list on this dev/inode pair.
825 * Quit if the list is deleted.
828 DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
829 (double)lhead->start, (double)lhead->size ));
831 for (i=0; i<num_locks && lhead; i++) {
833 struct posix_lock *lock = &locks[i];
834 struct lock_list *l_curr;
837 * Walk the lock list, checking for overlaps. Note that
838 * the lock list can expand within this loop if the current
839 * range being examined needs to be split.
842 for (l_curr = lhead; l_curr;) {
844 DEBUG(10,("posix_lock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
845 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
847 if ( (l_curr->start >= (lock->start + lock->size)) ||
848 (lock->start >= (l_curr->start + l_curr->size))) {
850 /* No overlap with this lock - leave this range alone. */
851 /*********************************************
862 **********************************************/
864 DEBUG(10,("no overlap case.\n" ));
866 l_curr = l_curr->next;
868 } else if ( (l_curr->start >= lock->start) &&
869 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
872 * This unlock is completely overlapped by this existing lock range
873 * and thus should have no effect (not be unlocked). Delete it from the list.
875 /*********************************************
879 +---------------------------+
881 +---------------------------+
882 **********************************************/
883 /* Save the next pointer */
884 struct lock_list *ul_next = l_curr->next;
886 DEBUG(10,("delete case.\n" ));
888 DLIST_REMOVE(lhead, l_curr);
890 break; /* No more list... */
894 } else if ( (l_curr->start >= lock->start) &&
895 (l_curr->start < lock->start + lock->size) &&
896 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
899 * This unlock overlaps the existing lock range at the high end.
900 * Truncate by moving start to existing range end and reducing size.
902 /*********************************************
913 **********************************************/
915 l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
916 l_curr->start = lock->start + lock->size;
918 DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
919 (double)l_curr->start, (double)l_curr->size ));
921 l_curr = l_curr->next;
923 } else if ( (l_curr->start < lock->start) &&
924 (l_curr->start + l_curr->size > lock->start) &&
925 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
928 * This unlock overlaps the existing lock range at the low end.
929 * Truncate by reducing size.
931 /*********************************************
942 **********************************************/
944 l_curr->size = lock->start - l_curr->start;
946 DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
947 (double)l_curr->start, (double)l_curr->size ));
949 l_curr = l_curr->next;
951 } else if ( (l_curr->start < lock->start) &&
952 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
954 * Worst case scenario. Unlock request completely overlaps an existing
955 * lock range. Split the request into two, push the new (upper) request
956 * into the dlink list, and continue with the entry after ul_new (as we
957 * know that ul_new will not overlap with this lock).
959 /*********************************************
960 +---------------------------+
962 +---------------------------+
967 +-------+ +---------+
969 +-------+ +---------+
970 **********************************************/
971 struct lock_list *l_new = (struct lock_list *)talloc(ctx,
972 sizeof(struct lock_list));
975 DEBUG(0,("posix_lock_list: talloc fail.\n"));
976 return NULL; /* The talloc_destroy takes care of cleanup. */
980 l_new->start = lock->start + lock->size;
981 l_new->size = l_curr->start + l_curr->size - l_new->start;
983 /* Truncate the l_curr. */
984 l_curr->size = lock->start - l_curr->start;
986 DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
987 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
988 (double)l_new->start, (double)l_new->size ));
991 * Add into the dlink list after the l_curr point - NOT at lhead.
992 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
995 l_new->prev = l_curr;
996 l_new->next = l_curr->next;
997 l_curr->next = l_new;
999 /* And move after the link we added. */
1000 l_curr = l_new->next;
1005 * This logic case should never happen. Ensure this is the
1006 * case by forcing an abort.... Remove in production.
1010 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
1011 lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
1015 } /* end for ( l_curr = lhead; l_curr;) */
1016 } /* end for (i=0; i<num_locks && ul_head; i++) */
1024 /****************************************************************************
1025 POSIX function to acquire a lock. Returns True if the
1026 lock could be granted, False if not.
1027 ****************************************************************************/
1029 BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
1034 size_t entry_num = 0;
1036 TALLOC_CTX *l_ctx = NULL;
1037 struct lock_list *llist = NULL;
1038 struct lock_list *ll = NULL;
1039 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
1041 DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
1042 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
1045 * If the requested lock won't fit in the POSIX range, we will
1046 * pretend it was successful.
1049 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1053 * Windows is very strange. It allows read locks to be overlayed
1054 * (even over a write lock), but leaves the write lock in force until the first
1055 * unlock. It also reference counts the locks. This means the following sequence :
1058 * ------------------------------------------------------------------------
1059 * WRITE LOCK : start = 2, len = 10
1060 * READ LOCK: start =0, len = 10 - FAIL
1061 * READ LOCK : start = 0, len = 14
1062 * READ LOCK: start =0, len = 10 - FAIL
1063 * UNLOCK : start = 2, len = 10
1064 * READ LOCK: start =0, len = 10 - OK
1066 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
1067 * would leave a single read lock over the 0-14 region. In order to
1068 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
1069 * entries, one for each overlayed lock request. We are guarenteed by the brlock
1070 * semantics that if a write lock is added, then it will be first in the array.
1073 if ((l_ctx = talloc_init()) == NULL) {
1074 DEBUG(0,("set_posix_lock: unable to init talloc context.\n"));
1075 return True; /* Not a fatal error. */
1078 if ((ll = (struct lock_list *)talloc(l_ctx, sizeof(struct lock_list))) == NULL) {
1079 DEBUG(0,("set_posix_lock: unable to talloc unlock list.\n"));
1080 talloc_destroy(l_ctx);
1081 return True; /* Not a fatal error. */
1085 * Create the initial list entry containing the
1086 * lock we want to add.
1093 DLIST_ADD(llist, ll);
1096 * The following call calculates if there are any
1097 * overlapping locks held by this process on
1098 * fd's open on the same file and splits this list
1099 * into a list of lock ranges that do not overlap with existing
1103 llist = posix_lock_list(l_ctx, llist, fsp);
1106 * Now we have the list of ranges to lock it is safe to add the
1107 * entry into the POSIX lock tdb. We take note of the entry we
1108 * added here in case we have to remove it on POSIX lock fail.
1111 if (!add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num)) {
1112 DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
1113 talloc_destroy(l_ctx);
1118 * Add the POSIX locks on the list of ranges returned.
1119 * As the lock is supposed to be added atomically, we need to
1120 * back out all the locks if any one of these calls fail.
1123 for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
1127 DEBUG(5,("set_posix_lock: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
1128 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1130 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1131 DEBUG(5,("set_posix_lock: Lock fail !: Type = %s: offset = %.0f, count = %.0f\n",
1132 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1141 * Back out all the POSIX locks we have on fail.
1144 for (ll = llist; lock_count; ll = ll->next, lock_count--) {
1148 DEBUG(5,("set_posix_lock: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
1149 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1151 posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
1155 * Remove the tdb entry for this lock.
1158 delete_posix_lock_entry_by_index(fsp,entry_num);
1161 talloc_destroy(l_ctx);
1165 /****************************************************************************
1166 POSIX function to release a lock. Returns True if the
1167 lock could be released, False if not.
1168 ****************************************************************************/
1170 BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
1175 TALLOC_CTX *ul_ctx = NULL;
1176 struct lock_list *ulist = NULL;
1177 struct lock_list *ul = NULL;
1178 struct posix_lock deleted_lock;
1179 int num_overlapped_entries;
1181 DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
1182 fsp->fsp_name, (double)u_offset, (double)u_count ));
1185 * If the requested lock won't fit in the POSIX range, we will
1186 * pretend it was successful.
1189 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1193 * We treat this as one unlock request for POSIX accounting purposes even
1194 * if it may later be split into multiple smaller POSIX unlock ranges.
1195 * num_overlapped_entries is the number of existing locks that have any
1196 * overlap with this unlock request.
1199 num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
1201 if (num_overlapped_entries == -1) {
1202 smb_panic("release_posix_lock: unable find entry to delete !\n");
1206 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
1207 * a POSIX write lock, then before doing the unlock we need to downgrade
1208 * the POSIX lock to a read lock. This allows any overlapping read locks
1209 * to be atomically maintained.
1212 if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
1213 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1214 DEBUG(0,("release_posix_lock: downgrade of lock failed !\n"));
1219 if ((ul_ctx = talloc_init()) == NULL) {
1220 DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
1221 return True; /* Not a fatal error. */
1224 if ((ul = (struct lock_list *)talloc(ul_ctx, sizeof(struct lock_list))) == NULL) {
1225 DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
1226 talloc_destroy(ul_ctx);
1227 return True; /* Not a fatal error. */
1231 * Create the initial list entry containing the
1232 * lock we want to remove.
1239 DLIST_ADD(ulist, ul);
1242 * The following call calculates if there are any
1243 * overlapping locks held by this process on
1244 * fd's open on the same file and creates a
1245 * list of unlock ranges that will allow
1246 * POSIX lock ranges to remain on the file whilst the
1247 * unlocks are performed.
1250 ulist = posix_lock_list(ul_ctx, ulist, fsp);
1253 * Release the POSIX locks on the list of ranges returned.
1256 for(; ulist; ulist = ulist->next) {
1257 offset = ulist->start;
1258 count = ulist->size;
1260 DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
1261 (double)offset, (double)count ));
1263 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
1267 talloc_destroy(ul_ctx);
1272 /****************************************************************************
1273 Remove all lock entries for a specific dev/inode pair from the tdb.
1274 ****************************************************************************/
1276 static void delete_posix_lock_entries(files_struct *fsp)
1278 TDB_DATA kbuf = locking_key_fsp(fsp);
1280 if (tdb_delete(posix_lock_tdb, kbuf) == -1)
1281 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
1284 /****************************************************************************
1286 ****************************************************************************/
1288 static void dump_entry(struct posix_lock *pl)
1290 DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
1291 (double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
1294 /****************************************************************************
1295 Remove any locks on this fd. Called from file_close().
1296 ****************************************************************************/
1298 void posix_locking_close_file(files_struct *fsp)
1300 struct posix_lock *entries = NULL;
1304 * Optimization for the common case where we are the only
1305 * opener of a file. If all fd entries are our own, we don't
1306 * need to explicitly release all the locks via the POSIX functions,
1307 * we can just remove all the entries in the tdb and allow the
1308 * close to remove the real locks.
1311 count = get_posix_lock_entries(fsp, &entries);
1314 DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
1318 for (i = 0; i < count; i++) {
1319 if (entries[i].fd != fsp->fd )
1322 dump_entry(&entries[i]);
1326 /* All locks are ours. */
1327 DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n",
1328 fsp->fsp_name, (unsigned int)count ));
1329 free((char *)entries);
1330 delete_posix_lock_entries(fsp);
1335 * Difficult case. We need to delete all our locks, whilst leaving
1336 * all other POSIX locks in place.
1339 for (i = 0; i < count; i++) {
1340 struct posix_lock *pl = &entries[i];
1341 if (pl->fd == fsp->fd)
1342 release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
1344 free((char *)entries);
1347 /*******************************************************************
1348 Create the in-memory POSIX lock databases.
1349 ********************************************************************/
1351 BOOL posix_locking_init(int read_only)
1353 if (posix_lock_tdb && posix_pending_close_tdb)
1356 if (!posix_lock_tdb)
1357 posix_lock_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1358 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1359 if (!posix_lock_tdb) {
1360 DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
1363 if (!posix_pending_close_tdb)
1364 posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
1365 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
1366 if (!posix_pending_close_tdb) {
1367 DEBUG(0,("Failed to open POSIX pending close database.\n"));
1374 /*******************************************************************
1375 Delete the in-memory POSIX lock databases.
1376 ********************************************************************/
1378 BOOL posix_locking_end(void)
1380 if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
1382 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)