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. Returns the number of records that
349 completely overlap this request, or -1 on error. entry_num gets set to the
350 index number of the added lock (used in case we need to delete *exactly*
352 ****************************************************************************/
354 static int add_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, int lock_type, size_t *entry_num)
356 TDB_DATA kbuf = locking_key_fsp(fsp);
358 struct posix_lock pl;
359 struct posix_lock *entries;
361 int num_overlapping_records = 0;
366 * Windows is very strange. It allows read locks to be overlayed on
367 * a write lock, but leaves the write lock in force until the first
368 * unlock. It also reference counts the locks. This means the following sequence :
371 * ------------------------------------------------------------------------
372 * WRITE LOCK : start = 0, len = 10
373 * READ LOCK: start =0, len = 10 - FAIL
374 * READ LOCK : start = 5, len = 2
375 * READ LOCK: start =0, len = 10 - FAIL
376 * UNLOCK : start = 0, len = 10
377 * READ LOCK: start =0, len = 10 - OK
379 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
380 * would leave a single read lock over the 0-10 region. In order to
381 * re-create Windows semantics mapped to POSIX locks, we create multiple TDB
382 * entries, one for each overlayed lock request. We are guarenteed by the brlock
383 * semantics that if a write lock is added, then it will be first in the array.
388 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
390 count = (size_t)(dbuf.dsize / sizeof(pl));
391 entries = (struct posix_lock *)dbuf.dptr;
394 * Ensure we look for overlapping entries *before*
395 * we add this entry. Count the number of entries
396 * that completely overlap this request.
399 for (i = 0; i < count; i++) {
400 struct posix_lock *entry = &entries[i];
402 if (fsp->fd == entry->fd &&
403 start >= entry->start &&
404 start + size <= entry->start + entry->size)
405 num_overlapping_records++;
415 pl.lock_type = lock_type;
417 dbuf.dptr = Realloc(dbuf.dptr, dbuf.dsize + sizeof(pl));
419 DEBUG(0,("add_posix_lock_entry: Realloc fail !\n"));
423 memcpy(dbuf.dptr + dbuf.dsize, &pl, sizeof(pl));
424 dbuf.dsize += sizeof(pl);
428 if (tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
429 DEBUG(0,("add_posix_lock: Failed to add lock entry on file %s\n", fsp->fsp_name));
435 DEBUG(10,("add_posix_lock: File %s: type = %s: start=%.0f size=%.0f: num_records = %d : dev=%.0f inode=%.0f\n",
436 fsp->fsp_name, posix_lock_type_name(lock_type), (double)start, (double)size, num_overlapping_records,
437 (double)fsp->dev, (double)fsp->inode ));
439 return num_overlapping_records;
448 /****************************************************************************
449 Delete an entry from the POSIX locking tdb. Returns a copy of the entry being
450 deleted and the number of records that are completely overlapped by this one,
452 ****************************************************************************/
454 static int delete_posix_lock_entry(files_struct *fsp, SMB_OFF_T start, SMB_OFF_T size, struct posix_lock *pl)
456 TDB_DATA kbuf = locking_key_fsp(fsp);
458 struct posix_lock *locks;
461 int num_overlapping_records = 0;
465 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
468 DEBUG(10,("delete_posix_lock_entry: tdb_fetch failed !\n"));
472 /* There are existing locks - find a match. */
473 locks = (struct posix_lock *)dbuf.dptr;
474 count = (size_t)(dbuf.dsize / sizeof(*locks));
477 * Search for and delete the first record that matches the
481 for (i=0; i<count; i++) {
482 struct posix_lock *entry = &locks[i];
484 if (entry->fd == fsp->fd &&
485 entry->start == start &&
486 entry->size == size) {
488 /* Make a copy if requested. */
492 /* Found it - delete it. */
494 tdb_delete(posix_lock_tdb, kbuf);
497 memmove(&locks[i], &locks[i+1], sizeof(*locks)*((count-1) - i));
499 dbuf.dsize -= sizeof(*locks);
500 tdb_store(posix_lock_tdb, kbuf, dbuf, TDB_REPLACE);
512 * Count the number of entries that are
513 * overlapped completely by this unlock request.
514 * (Note that this is the reverse of the test in
518 for (i = 0; i < count; i++) {
519 struct posix_lock *entry = &locks[i];
521 if (fsp->fd == entry->fd &&
522 entry->start >= start &&
523 entry->start + entry->size <= start + size)
524 num_overlapping_records++;
527 DEBUG(10,("delete_posix_lock_entry: type = %s: start=%.0f size=%.0f, num_records = %d\n",
528 posix_lock_type_name(pl->lock_type), (double)pl->start, (double)pl->size,
529 (unsigned int)num_overlapping_records ));
534 return num_overlapping_records;
542 /****************************************************************************
543 Utility function to map a lock type correctly depending on the open
545 ****************************************************************************/
547 static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
549 if((lock_type == WRITE_LOCK) && !fsp->can_write) {
551 * Many UNIX's cannot get a write lock on a file opened read-only.
552 * Win32 locking semantics allow this.
553 * Do the best we can and attempt a read-only lock.
555 DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
557 } else if((lock_type == READ_LOCK) && !fsp->can_read) {
559 * Ditto for read locks on write only files.
561 DEBUG(10,("map_posix_lock_type: Changing read lock to write due to write-only file.\n"));
566 * This return should be the most normal, as we attempt
567 * to always open files read/write.
570 return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
573 /****************************************************************************
574 Check to see if the given unsigned lock range is within the possible POSIX
575 range. Modifies the given args to be in range if possible, just returns
577 ****************************************************************************/
579 static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
580 SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
585 #if defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
587 SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
588 SMB_OFF_T mask = (mask2<<1);
589 SMB_OFF_T neg_mask = ~mask;
592 * In this case SMB_OFF_T is 64 bits,
593 * and the underlying system can handle 64 bit signed locks.
594 * Cast to signed type.
597 offset = (SMB_OFF_T)u_offset;
598 count = (SMB_OFF_T)u_count;
601 * Deal with a very common case of count of all ones.
602 * (lock entire file).
605 if(count == (SMB_OFF_T)-1)
609 * POSIX lock ranges cannot be negative.
610 * Fail if any combination becomes negative.
613 if(offset < 0 || count < 0 || (offset + count < 0)) {
614 DEBUG(10,("posix_lock_in_range: negative range: offset = %.0f, count = %.0f. Ignoring lock.\n",
615 (double)offset, (double)count ));
620 * In this case SMB_OFF_T is 64 bits, the offset and count
621 * fit within the positive range, and the underlying
622 * system can handle 64 bit locks. Just return as the
623 * cast values are ok.
626 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
629 * In this case either SMB_OFF_T is 32 bits,
630 * or the underlying system cannot handle 64 bit signed locks.
631 * Either way we have to try and mangle to fit within 31 bits.
635 #if defined(HAVE_BROKEN_FCNTL64_LOCKS)
638 * SMB_OFF_T is 64 bits, but we need to use 31 bits due to
639 * broken large locking.
643 * Deal with a very common case of count of all ones.
644 * (lock entire file).
647 if(u_count == (SMB_BIG_UINT)-1)
650 if(((u_offset >> 32) & 0xFFFFFFFF) || ((u_count >> 32) & 0xFFFFFFFF)) {
651 DEBUG(10,("posix_lock_in_range: top 32 bits not zero. offset = %.0f, count = %.0f. Ignoring lock.\n",
652 (double)u_offset, (double)u_count ));
653 /* Top 32 bits of offset or count were not zero. */
657 /* Cast from 64 bits unsigned to 64 bits signed. */
658 offset = (SMB_OFF_T)u_offset;
659 count = (SMB_OFF_T)u_count;
662 * Check if we are within the 2^31 range.
666 int32 low_offset = (int32)offset;
667 int32 low_count = (int32)count;
669 if(low_offset < 0 || low_count < 0 || (low_offset + low_count < 0)) {
670 DEBUG(10,("posix_lock_in_range: not within 2^31 range. low_offset = %d, low_count = %d. Ignoring lock.\n",
671 low_offset, low_count ));
677 * Ok - we can map from a 64 bit number to a 31 bit lock.
680 #else /* HAVE_BROKEN_FCNTL64_LOCKS */
683 * SMB_OFF_T is 32 bits.
686 #if defined(HAVE_LONGLONG)
689 * SMB_BIG_UINT is 64 bits, we can do a 32 bit shift.
693 * Deal with a very common case of count of all ones.
694 * (lock entire file).
697 if(u_count == (SMB_BIG_UINT)-1)
700 if(((u_offset >> 32) & 0xFFFFFFFF) || ((u_count >> 32) & 0xFFFFFFFF)) {
701 DEBUG(10,("posix_lock_in_range: top 32 bits not zero. u_offset = %.0f, u_count = %.0f. Ignoring lock.\n",
702 (double)u_offset, (double)u_count ));
706 /* Cast from 64 bits unsigned to 32 bits signed. */
707 offset = (SMB_OFF_T)u_offset;
708 count = (SMB_OFF_T)u_count;
711 * Check if we are within the 2^31 range.
714 if(offset < 0 || count < 0 || (offset + count < 0)) {
715 DEBUG(10,("posix_lock_in_range: not within 2^31 range. offset = %d, count = %d. Ignoring lock.\n",
716 (int)offset, (int)count ));
720 #else /* HAVE_LONGLONG */
723 * SMB_BIG_UINT and SMB_OFF_T are both 32 bits,
728 * Deal with a very common case of count of all ones.
729 * (lock entire file).
732 if(u_count == (SMB_BIG_UINT)-1)
735 /* Cast from 32 bits unsigned to 32 bits signed. */
736 offset = (SMB_OFF_T)u_offset;
737 count = (SMB_OFF_T)u_count;
740 * Check if we are within the 2^31 range.
743 if(offset < 0 || count < 0 || (offset + count < 0)) {
744 DEBUG(10,("posix_lock_in_range: not within 2^31 range. offset = %d, count = %d. Ignoring lock.\n",
745 (int)offset, (int)count ));
749 #endif /* HAVE_LONGLONG */
750 #endif /* LARGE_SMB_OFF_T */
751 #endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
754 * The mapping was successful.
757 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
758 (double)offset, (double)count ));
760 *offset_out = offset;
766 #if defined(LARGE_SMB_OFF_T)
767 /****************************************************************************
768 Pathetically try and map a 64 bit lock offset into 31 bits. I hate Windows :-).
769 ****************************************************************************/
771 static uint32 map_lock_offset(uint32 high, uint32 low)
775 uint32 highcopy = high;
778 * Try and find out how many significant bits there are in high.
781 for(i = 0; highcopy; i++)
785 * We use 31 bits not 32 here as POSIX
786 * lock offsets may not be negative.
789 mask = (~0) << (31 - i);
792 return 0; /* Fail. */
800 /****************************************************************************
801 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
802 broken NFS implementations.
803 ****************************************************************************/
805 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
808 struct connection_struct *conn = fsp->conn;
810 #if defined(LARGE_SMB_OFF_T)
812 * In the 64 bit locking case we store the original
813 * values in case we have to map to a 32 bit lock on
814 * a filesystem that doesn't support 64 bit locks.
816 SMB_OFF_T orig_offset = offset;
817 SMB_OFF_T orig_count = count;
818 #endif /* LARGE_SMB_OFF_T */
820 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fd,op,(double)offset,(double)count,type));
822 ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
824 if (!ret && (errno == EFBIG)) {
826 dbgtext("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n", (double)offset,(double)count);
827 dbgtext("a 'file too large' error. This can happen when using 64 bit lock offsets\n");
828 dbgtext("on 32 bit NFS mounted file systems. Retrying with 32 bit truncated length.\n");
830 /* 32 bit NFS file system, retry with smaller offset */
833 ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
836 /* A lock query - just return. */
837 if (op == SMB_F_GETLK)
840 /* A lock set or unset. */
842 DEBUG(3,("posix_fcntl_lock: lock failed at offset %.0f count %.0f op %d type %d (%s)\n",
843 (double)offset,(double)count,op,type,strerror(errno)));
845 /* Perhaps it doesn't support this sort of locking ? */
846 if (errno == EINVAL) {
847 #if defined(LARGE_SMB_OFF_T)
850 * Ok - if we get here then we have a 64 bit lock request
851 * that has returned EINVAL. Try and map to 31 bits for offset
852 * and length and try again. This may happen if a filesystem
853 * doesn't support 64 bit offsets (efs/ufs) although the underlying
856 uint32 off_low = (orig_offset & 0xFFFFFFFF);
857 uint32 off_high = ((orig_offset >> 32) & 0xFFFFFFFF);
859 count = (orig_count & 0x7FFFFFFF);
860 offset = (SMB_OFF_T)map_lock_offset(off_high, off_low);
861 ret = conn->vfs_ops.lock(fsp->fd,op,offset,count,type);
863 if (errno == EINVAL) {
864 DEBUG(3,("posix_fcntl_lock: locking not supported? returning True\n"));
869 DEBUG(3,("posix_fcntl_lock: 64 -> 32 bit modified lock call successful\n"));
872 #else /* LARGE_SMB_OFF_T */
873 DEBUG(3,("locking not supported? returning True\n"));
875 #endif /* LARGE_SMB_OFF_T */
881 DEBUG(8,("posix_fcntl_lock: Lock call successful\n"));
886 /****************************************************************************
887 POSIX function to see if a file region is locked. Returns True if the
888 region is locked, False otherwise.
889 ****************************************************************************/
891 BOOL is_posix_locked(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
895 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
897 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
898 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
901 * If the requested lock won't fit in the POSIX range, we will
902 * never set it, so presume it is not locked.
905 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
909 * Note that most UNIX's can *test* for a write lock on
910 * a read-only fd, just not *set* a write lock on a read-only
911 * fd. So we don't need to use map_lock_type here.
914 return posix_fcntl_lock(fsp,SMB_F_GETLK,offset,count,posix_lock_type);
917 /****************************************************************************
918 POSIX function to acquire a lock. Returns True if the
919 lock could be granted, False if not.
920 ****************************************************************************/
922 BOOL set_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count, enum brl_type lock_type)
927 size_t entry_num = 0;
928 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
929 int num_overlapping_records;
931 DEBUG(5,("set_posix_lock: File %s, offset = %.0f, count = %.0f, type = %s\n",
932 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
935 * If the requested lock won't fit in the POSIX range, we will
936 * pretend it was successful.
939 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
943 * Note that setting multiple overlapping locks on different
944 * file descriptors will not be held separately by the kernel (POSIX
945 * braindamage), but will be merged into one continuous lock
946 * range. We cope with this case in the release_posix_lock code
947 * below. We need to add the posix lock entry into the tdb before
948 * doing the real posix lock call to deal with the locking overlay
949 * case described above in add_posix_lock_entry().
952 num_overlapping_records = add_posix_lock_entry(fsp,offset,count,posix_lock_type,&entry_num);
954 if (num_overlapping_records == -1) {
955 DEBUG(0,("set_posix_lock: Unable to create posix lock entry !\n"));
960 * num_overlapping_records is the count of lock records that
961 * completely contain this request on the same fsp. Only bother
962 * with the real lock request if there are none, otherwise ignore.
965 if (num_overlapping_records == 0) {
967 * First lock entry for this range created. Do a real POSIX lock.
969 ret = posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type);
972 * Oops, POSIX lock failed, delete the tdb entry we just added.
975 delete_posix_lock_entry_by_index(fsp,entry_num);
982 * Structure used when splitting a lock range
983 * into a POSIX lock range. Doubly linked list.
987 struct unlock_list *next;
988 struct unlock_list *prev;
993 /****************************************************************************
994 Create a list of lock ranges that don't overlap a given range. Used in calculating
995 POSIX lock unlocks. This is a difficult function that requires ASCII art to
997 ****************************************************************************/
999 static struct unlock_list *posix_unlock_list(TALLOC_CTX *ctx, struct unlock_list *ulhead, files_struct *fsp)
1001 TDB_DATA kbuf = locking_key_fsp(fsp);
1003 struct posix_lock *locks;
1004 size_t num_locks, i;
1008 dbuf = tdb_fetch(posix_lock_tdb, kbuf);
1014 locks = (struct posix_lock *)dbuf.dptr;
1015 num_locks = (size_t)(dbuf.dsize / sizeof(*locks));
1018 * Check the current lock list on this dev/inode pair.
1019 * Quit if the list is deleted.
1022 DEBUG(10,("posix_unlock_list: curr: start=%.0f,size=%.0f\n",
1023 (double)ulhead->start, (double)ulhead->size ));
1025 for (i=0; i<num_locks && ulhead; i++) {
1027 struct posix_lock *lock = &locks[i];
1028 struct unlock_list *ul_curr;
1031 * Walk the unlock list, checking for overlaps. Note that
1032 * the unlock list can expand within this loop if the current
1033 * range being examined needs to be split.
1036 for (ul_curr = ulhead; ul_curr;) {
1038 DEBUG(10,("posix_unlock_list: lock: fd=%d: start=%.0f,size=%.0f:type=%s", lock->fd,
1039 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
1041 if ( (ul_curr->start >= (lock->start + lock->size)) ||
1042 (lock->start >= (ul_curr->start + ul_curr->size))) {
1044 /* No overlap with this lock - leave this range alone. */
1045 /*********************************************
1056 **********************************************/
1058 DEBUG(10,("no overlap case.\n" ));
1060 ul_curr = ul_curr->next;
1062 } else if ( (ul_curr->start >= lock->start) &&
1063 (ul_curr->start + ul_curr->size <= lock->start + lock->size) ) {
1066 * This unlock is completely overlapped by this existing lock range
1067 * and thus should have no effect (not be unlocked). Delete it from the list.
1069 /*********************************************
1073 +---------------------------+
1075 +---------------------------+
1076 **********************************************/
1077 /* Save the next pointer */
1078 struct unlock_list *ul_next = ul_curr->next;
1080 DEBUG(10,("delete case.\n" ));
1082 DLIST_REMOVE(ulhead, ul_curr);
1084 break; /* No more list... */
1088 } else if ( (ul_curr->start >= lock->start) &&
1089 (ul_curr->start < lock->start + lock->size) &&
1090 (ul_curr->start + ul_curr->size > lock->start + lock->size) ) {
1093 * This unlock overlaps the existing lock range at the high end.
1094 * Truncate by moving start to existing range end and reducing size.
1096 /*********************************************
1107 **********************************************/
1109 ul_curr->size = (ul_curr->start + ul_curr->size) - (lock->start + lock->size);
1110 ul_curr->start = lock->start + lock->size;
1112 DEBUG(10,("truncate high case: start=%.0f,size=%.0f\n",
1113 (double)ul_curr->start, (double)ul_curr->size ));
1115 ul_curr = ul_curr->next;
1117 } else if ( (ul_curr->start < lock->start) &&
1118 (ul_curr->start + ul_curr->size > lock->start) &&
1119 (ul_curr->start + ul_curr->size <= lock->start + lock->size) ) {
1122 * This unlock overlaps the existing lock range at the low end.
1123 * Truncate by reducing size.
1125 /*********************************************
1136 **********************************************/
1138 ul_curr->size = lock->start - ul_curr->start;
1140 DEBUG(10,("truncate low case: start=%.0f,size=%.0f\n",
1141 (double)ul_curr->start, (double)ul_curr->size ));
1143 ul_curr = ul_curr->next;
1145 } else if ( (ul_curr->start < lock->start) &&
1146 (ul_curr->start + ul_curr->size > lock->start + lock->size) ) {
1148 * Worst case scenario. Unlock request completely overlaps an existing
1149 * lock range. Split the request into two, push the new (upper) request
1150 * into the dlink list, and continue with the entry after ul_new (as we
1151 * know that ul_new will not overlap with this lock).
1153 /*********************************************
1154 +---------------------------+
1156 +---------------------------+
1161 +-------+ +---------+
1163 +-------+ +---------+
1164 **********************************************/
1165 struct unlock_list *ul_new = (struct unlock_list *)talloc(ctx,
1166 sizeof(struct unlock_list));
1168 if(ul_new == NULL) {
1169 DEBUG(0,("posix_unlock_list: talloc fail.\n"));
1170 return NULL; /* The talloc_destroy takes care of cleanup. */
1173 ZERO_STRUCTP(ul_new);
1174 ul_new->start = lock->start + lock->size;
1175 ul_new->size = ul_curr->start + ul_curr->size - ul_new->start;
1177 /* Truncate the ul_curr. */
1178 ul_curr->size = lock->start - ul_curr->start;
1180 DEBUG(10,("split case: curr: start=%.0f,size=%.0f \
1181 new: start=%.0f,size=%.0f\n", (double)ul_curr->start, (double)ul_curr->size,
1182 (double)ul_new->start, (double)ul_new->size ));
1185 * Add into the dlink list after the ul_curr point - NOT at ulhead.
1186 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
1189 ul_new->prev = ul_curr;
1190 ul_new->next = ul_curr->next;
1191 ul_curr->next = ul_new;
1193 /* And move after the link we added. */
1194 ul_curr = ul_new->next;
1199 * This logic case should never happen. Ensure this is the
1200 * case by forcing an abort.... Remove in production.
1204 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: ul_curr: start = %.0f, size = %.0f : \
1205 lock: start = %.0f, size = %.0f\n", (double)ul_curr->start, (double)ul_curr->size, (double)lock->start, (double)lock->size );
1209 } /* end for ( ul_curr = ulhead; ul_curr;) */
1210 } /* end for (i=0; i<num_locks && ul_head; i++) */
1218 /****************************************************************************
1219 POSIX function to release a lock. Returns True if the
1220 lock could be released, False if not.
1221 ****************************************************************************/
1223 BOOL release_posix_lock(files_struct *fsp, SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
1228 TALLOC_CTX *ul_ctx = NULL;
1229 struct unlock_list *ulist = NULL;
1230 struct unlock_list *ul = NULL;
1231 struct posix_lock deleted_lock;
1232 int num_overlapped_entries;
1234 DEBUG(5,("release_posix_lock: File %s, offset = %.0f, count = %.0f\n",
1235 fsp->fsp_name, (double)u_offset, (double)u_count ));
1238 * If the requested lock won't fit in the POSIX range, we will
1239 * pretend it was successful.
1242 if(!posix_lock_in_range(&offset, &count, u_offset, u_count))
1246 * We treat this as one unlock request for POSIX accounting purposes even
1247 * if it may later be split into multiple smaller POSIX unlock ranges.
1250 num_overlapped_entries = delete_posix_lock_entry(fsp, offset, count, &deleted_lock);
1252 if (num_overlapped_entries == -1) {
1253 smb_panic("release_posix_lock: unable find entry to delete !\n");
1257 * If num_overlapped_entries is > 0, and the lock_type we just deleted from the tdb was
1258 * a POSIX write lock, then before doing the unlock we need to downgrade
1259 * the POSIX lock to a read lock.
1262 if (num_overlapped_entries > 0 && deleted_lock.lock_type == F_WRLCK) {
1263 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1264 DEBUG(0,("release_posix_lock: downgrade of lock failed !\n"));
1269 if ((ul_ctx = talloc_init()) == NULL) {
1270 DEBUG(0,("release_posix_lock: unable to init talloc context.\n"));
1271 return True; /* Not a fatal error. */
1274 if ((ul = (struct unlock_list *)talloc(ul_ctx, sizeof(struct unlock_list))) == NULL) {
1275 DEBUG(0,("release_posix_lock: unable to talloc unlock list.\n"));
1276 talloc_destroy(ul_ctx);
1277 return True; /* Not a fatal error. */
1281 * Create the initial list entry containing the
1282 * lock we want to remove.
1289 DLIST_ADD(ulist, ul);
1292 * The following call calculates if there are any
1293 * overlapping locks held by this process on
1294 * fd's open on the same file and creates a
1295 * list of unlock ranges that will allow
1296 * POSIX lock ranges to remain on the file whilst the
1297 * unlocks are performed.
1300 ulist = posix_unlock_list(ul_ctx, ulist, fsp);
1303 * Release the POSIX locks on the list of ranges returned.
1306 for(; ulist; ulist = ulist->next) {
1307 offset = ulist->start;
1308 count = ulist->size;
1313 * This lock must overlap with an existing lock.
1314 * Don't do any POSIX call.
1320 DEBUG(5,("release_posix_lock: Real unlock: offset = %.0f, count = %.0f\n",
1321 (double)offset, (double)count ));
1323 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK))
1327 talloc_destroy(ul_ctx);
1332 /****************************************************************************
1333 Remove all lock entries for a specific dev/inode pair from the tdb.
1334 ****************************************************************************/
1336 static void delete_posix_lock_entries(files_struct *fsp)
1338 TDB_DATA kbuf = locking_key_fsp(fsp);
1340 if (tdb_delete(posix_lock_tdb, kbuf) == -1)
1341 DEBUG(0,("delete_close_entries: tdb_delete fail !\n"));
1344 /****************************************************************************
1346 ****************************************************************************/
1348 static void dump_entry(struct posix_lock *pl)
1350 DEBUG(10,("entry: start=%.0f, size=%.0f, type=%d, fd=%i\n",
1351 (double)pl->start, (double)pl->size, (int)pl->lock_type, pl->fd ));
1354 /****************************************************************************
1355 Remove any locks on this fd. Called from file_close().
1356 ****************************************************************************/
1358 void posix_locking_close_file(files_struct *fsp)
1360 struct posix_lock *entries = NULL;
1364 * Optimization for the common case where we are the only
1365 * opener of a file. If all fd entries are our own, we don't
1366 * need to explicitly release all the locks via the POSIX functions,
1367 * we can just remove all the entries in the tdb and allow the
1368 * close to remove the real locks.
1371 count = get_posix_lock_entries(fsp, &entries);
1374 DEBUG(10,("posix_locking_close_file: file %s has no outstanding locks.\n", fsp->fsp_name ));
1378 for (i = 0; i < count; i++) {
1379 if (entries[i].fd != fsp->fd )
1382 dump_entry(&entries[i]);
1386 /* All locks are ours. */
1387 DEBUG(10,("posix_locking_close_file: file %s has %u outstanding locks, but all on one fd.\n",
1388 fsp->fsp_name, (unsigned int)count ));
1389 free((char *)entries);
1390 delete_posix_lock_entries(fsp);
1395 * Difficult case. We need to delete all our locks, whilst leaving
1396 * all other POSIX locks in place.
1399 for (i = 0; i < count; i++) {
1400 struct posix_lock *pl = &entries[i];
1401 if (pl->fd == fsp->fd)
1402 release_posix_lock(fsp, (SMB_BIG_UINT)pl->start, (SMB_BIG_UINT)pl->size );
1404 free((char *)entries);
1407 /*******************************************************************
1408 Create the in-memory POSIX lock databases.
1409 ********************************************************************/
1411 BOOL posix_locking_init(void)
1413 if (posix_lock_tdb && posix_pending_close_tdb)
1416 if (!posix_lock_tdb)
1417 posix_lock_tdb = tdb_open(NULL, 0, TDB_INTERNAL,
1418 O_RDWR|O_CREAT, 0644);
1419 if (!posix_lock_tdb) {
1420 DEBUG(0,("Failed to open POSIX byte range locking database.\n"));
1423 if (!posix_pending_close_tdb)
1424 posix_pending_close_tdb = tdb_open(NULL, 0, TDB_INTERNAL,
1425 O_RDWR|O_CREAT, 0644);
1426 if (!posix_pending_close_tdb) {
1427 DEBUG(0,("Failed to open POSIX pending close database.\n"));
1434 /*******************************************************************
1435 Delete the in-memory POSIX lock databases.
1436 ********************************************************************/
1438 BOOL posix_locking_end(void)
1440 if (posix_lock_tdb && tdb_close(posix_lock_tdb) != 0)
1442 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0)