[GLUE] Rsync SAMBA_3_0 SVN r25598 in order to create the v3-0-test branch.
[Samba.git] / source / locking / posix.c
blobdc54a3419417ba424ca6a2eb9e2826f2e6129bfc
1 /*
2 Unix SMB/CIFS implementation.
3 Locking functions
4 Copyright (C) Jeremy Allison 1992-2006
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 This program is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with this program; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 Revision History:
22 POSIX locking support. Jeremy Allison (jeremy@valinux.com), Apr. 2000.
25 #include "includes.h"
27 #undef DBGC_CLASS
28 #define DBGC_CLASS DBGC_LOCKING
31 * The pending close database handle.
34 static TDB_CONTEXT *posix_pending_close_tdb;
36 /****************************************************************************
37 First - the functions that deal with the underlying system locks - these
38 functions are used no matter if we're mapping CIFS Windows locks or CIFS
39 POSIX locks onto POSIX.
40 ****************************************************************************/
42 /****************************************************************************
43 Utility function to map a lock type correctly depending on the open
44 mode of a file.
45 ****************************************************************************/
47 static int map_posix_lock_type( files_struct *fsp, enum brl_type lock_type)
49 if((lock_type == WRITE_LOCK) && !fsp->can_write) {
51 * Many UNIX's cannot get a write lock on a file opened read-only.
52 * Win32 locking semantics allow this.
53 * Do the best we can and attempt a read-only lock.
55 DEBUG(10,("map_posix_lock_type: Downgrading write lock to read due to read-only file.\n"));
56 return F_RDLCK;
60 * This return should be the most normal, as we attempt
61 * to always open files read/write.
64 return (lock_type == READ_LOCK) ? F_RDLCK : F_WRLCK;
67 /****************************************************************************
68 Debugging aid :-).
69 ****************************************************************************/
71 static const char *posix_lock_type_name(int lock_type)
73 return (lock_type == F_RDLCK) ? "READ" : "WRITE";
76 /****************************************************************************
77 Check to see if the given unsigned lock range is within the possible POSIX
78 range. Modifies the given args to be in range if possible, just returns
79 False if not.
80 ****************************************************************************/
82 static BOOL posix_lock_in_range(SMB_OFF_T *offset_out, SMB_OFF_T *count_out,
83 SMB_BIG_UINT u_offset, SMB_BIG_UINT u_count)
85 SMB_OFF_T offset = (SMB_OFF_T)u_offset;
86 SMB_OFF_T count = (SMB_OFF_T)u_count;
89 * For the type of system we are, attempt to
90 * find the maximum positive lock offset as an SMB_OFF_T.
93 #if defined(MAX_POSITIVE_LOCK_OFFSET) /* Some systems have arbitrary limits. */
95 SMB_OFF_T max_positive_lock_offset = (MAX_POSITIVE_LOCK_OFFSET);
97 #elif defined(LARGE_SMB_OFF_T) && !defined(HAVE_BROKEN_FCNTL64_LOCKS)
100 * In this case SMB_OFF_T is 64 bits,
101 * and the underlying system can handle 64 bit signed locks.
104 SMB_OFF_T mask2 = ((SMB_OFF_T)0x4) << (SMB_OFF_T_BITS-4);
105 SMB_OFF_T mask = (mask2<<1);
106 SMB_OFF_T max_positive_lock_offset = ~mask;
108 #else /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
111 * In this case either SMB_OFF_T is 32 bits,
112 * or the underlying system cannot handle 64 bit signed locks.
113 * All offsets & counts must be 2^31 or less.
116 SMB_OFF_T max_positive_lock_offset = 0x7FFFFFFF;
118 #endif /* !LARGE_SMB_OFF_T || HAVE_BROKEN_FCNTL64_LOCKS */
121 * POSIX locks of length zero mean lock to end-of-file.
122 * Win32 locks of length zero are point probes. Ignore
123 * any Win32 locks of length zero. JRA.
126 if (count == (SMB_OFF_T)0) {
127 DEBUG(10,("posix_lock_in_range: count = 0, ignoring.\n"));
128 return False;
132 * If the given offset was > max_positive_lock_offset then we cannot map this at all
133 * ignore this lock.
136 if (u_offset & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
137 DEBUG(10,("posix_lock_in_range: (offset = %.0f) offset > %.0f and we cannot handle this. Ignoring lock.\n",
138 (double)u_offset, (double)((SMB_BIG_UINT)max_positive_lock_offset) ));
139 return False;
143 * We must truncate the count to less than max_positive_lock_offset.
146 if (u_count & ~((SMB_BIG_UINT)max_positive_lock_offset)) {
147 count = max_positive_lock_offset;
151 * Truncate count to end at max lock offset.
154 if (offset + count < 0 || offset + count > max_positive_lock_offset) {
155 count = max_positive_lock_offset - offset;
159 * If we ate all the count, ignore this lock.
162 if (count == 0) {
163 DEBUG(10,("posix_lock_in_range: Count = 0. Ignoring lock u_offset = %.0f, u_count = %.0f\n",
164 (double)u_offset, (double)u_count ));
165 return False;
169 * The mapping was successful.
172 DEBUG(10,("posix_lock_in_range: offset_out = %.0f, count_out = %.0f\n",
173 (double)offset, (double)count ));
175 *offset_out = offset;
176 *count_out = count;
178 return True;
181 /****************************************************************************
182 Actual function that does POSIX locks. Copes with 64 -> 32 bit cruft and
183 broken NFS implementations.
184 ****************************************************************************/
186 static BOOL posix_fcntl_lock(files_struct *fsp, int op, SMB_OFF_T offset, SMB_OFF_T count, int type)
188 BOOL ret;
190 DEBUG(8,("posix_fcntl_lock %d %d %.0f %.0f %d\n",fsp->fh->fd,op,(double)offset,(double)count,type));
192 ret = SMB_VFS_LOCK(fsp,fsp->fh->fd,op,offset,count,type);
194 if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno == EINVAL))) {
196 DEBUG(0,("posix_fcntl_lock: WARNING: lock request at offset %.0f, length %.0f returned\n",
197 (double)offset,(double)count));
198 DEBUG(0,("an %s error. This can happen when using 64 bit lock offsets\n", strerror(errno)));
199 DEBUG(0,("on 32 bit NFS mounted file systems.\n"));
202 * If the offset is > 0x7FFFFFFF then this will cause problems on
203 * 32 bit NFS mounted filesystems. Just ignore it.
206 if (offset & ~((SMB_OFF_T)0x7fffffff)) {
207 DEBUG(0,("Offset greater than 31 bits. Returning success.\n"));
208 return True;
211 if (count & ~((SMB_OFF_T)0x7fffffff)) {
212 /* 32 bit NFS file system, retry with smaller offset */
213 DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n"));
214 errno = 0;
215 count &= 0x7fffffff;
216 ret = SMB_VFS_LOCK(fsp,fsp->fh->fd,op,offset,count,type);
220 DEBUG(8,("posix_fcntl_lock: Lock call %s\n", ret ? "successful" : "failed"));
221 return ret;
224 /****************************************************************************
225 Actual function that gets POSIX locks. Copes with 64 -> 32 bit cruft and
226 broken NFS implementations.
227 ****************************************************************************/
229 static BOOL posix_fcntl_getlock(files_struct *fsp, SMB_OFF_T *poffset, SMB_OFF_T *pcount, int *ptype)
231 pid_t pid;
232 BOOL ret;
234 DEBUG(8,("posix_fcntl_getlock %d %.0f %.0f %d\n",
235 fsp->fh->fd,(double)*poffset,(double)*pcount,*ptype));
237 ret = SMB_VFS_GETLOCK(fsp,fsp->fh->fd,poffset,pcount,ptype,&pid);
239 if (!ret && ((errno == EFBIG) || (errno == ENOLCK) || (errno == EINVAL))) {
241 DEBUG(0,("posix_fcntl_getlock: WARNING: lock request at offset %.0f, length %.0f returned\n",
242 (double)*poffset,(double)*pcount));
243 DEBUG(0,("an %s error. This can happen when using 64 bit lock offsets\n", strerror(errno)));
244 DEBUG(0,("on 32 bit NFS mounted file systems.\n"));
247 * If the offset is > 0x7FFFFFFF then this will cause problems on
248 * 32 bit NFS mounted filesystems. Just ignore it.
251 if (*poffset & ~((SMB_OFF_T)0x7fffffff)) {
252 DEBUG(0,("Offset greater than 31 bits. Returning success.\n"));
253 return True;
256 if (*pcount & ~((SMB_OFF_T)0x7fffffff)) {
257 /* 32 bit NFS file system, retry with smaller offset */
258 DEBUG(0,("Count greater than 31 bits - retrying with 31 bit truncated length.\n"));
259 errno = 0;
260 *pcount &= 0x7fffffff;
261 ret = SMB_VFS_GETLOCK(fsp,fsp->fh->fd,poffset,pcount,ptype,&pid);
265 DEBUG(8,("posix_fcntl_getlock: Lock query call %s\n", ret ? "successful" : "failed"));
266 return ret;
269 /****************************************************************************
270 POSIX function to see if a file region is locked. Returns True if the
271 region is locked, False otherwise.
272 ****************************************************************************/
274 BOOL is_posix_locked(files_struct *fsp,
275 SMB_BIG_UINT *pu_offset,
276 SMB_BIG_UINT *pu_count,
277 enum brl_type *plock_type,
278 enum brl_flavour lock_flav)
280 SMB_OFF_T offset;
281 SMB_OFF_T count;
282 int posix_lock_type = map_posix_lock_type(fsp,*plock_type);
284 DEBUG(10,("is_posix_locked: File %s, offset = %.0f, count = %.0f, type = %s\n",
285 fsp->fsp_name, (double)*pu_offset, (double)*pu_count, posix_lock_type_name(*plock_type) ));
288 * If the requested lock won't fit in the POSIX range, we will
289 * never set it, so presume it is not locked.
292 if(!posix_lock_in_range(&offset, &count, *pu_offset, *pu_count)) {
293 return False;
296 if (!posix_fcntl_getlock(fsp,&offset,&count,&posix_lock_type)) {
297 return False;
300 if (posix_lock_type == F_UNLCK) {
301 return False;
304 if (lock_flav == POSIX_LOCK) {
305 /* Only POSIX lock queries need to know the details. */
306 *pu_offset = (SMB_BIG_UINT)offset;
307 *pu_count = (SMB_BIG_UINT)count;
308 *plock_type = (posix_lock_type == F_RDLCK) ? READ_LOCK : WRITE_LOCK;
310 return True;
313 /****************************************************************************
314 Next - the functions that deal with in memory database storing representations
315 of either Windows CIFS locks or POSIX CIFS locks.
316 ****************************************************************************/
318 /* The key used in the in-memory POSIX databases. */
320 struct lock_ref_count_key {
321 SMB_DEV_T device;
322 SMB_INO_T inode;
323 char r;
326 struct fd_key {
327 SMB_DEV_T device;
328 SMB_INO_T inode;
331 /*******************************************************************
332 Form a static locking key for a dev/inode pair for the fd array.
333 ******************************************************************/
335 static TDB_DATA fd_array_key(SMB_DEV_T dev, SMB_INO_T inode)
337 static struct fd_key key;
338 TDB_DATA kbuf;
340 memset(&key, '\0', sizeof(key));
341 key.device = dev;
342 key.inode = inode;
343 kbuf.dptr = (char *)&key;
344 kbuf.dsize = sizeof(key);
345 return kbuf;
348 /*******************************************************************
349 Form a static locking key for a dev/inode pair for the lock ref count
350 ******************************************************************/
352 static TDB_DATA locking_ref_count_key(SMB_DEV_T dev, SMB_INO_T inode)
354 static struct lock_ref_count_key key;
355 TDB_DATA kbuf;
357 memset(&key, '\0', sizeof(key));
358 key.device = dev;
359 key.inode = inode;
360 key.r = 'r';
361 kbuf.dptr = (char *)&key;
362 kbuf.dsize = sizeof(key);
363 return kbuf;
366 /*******************************************************************
367 Convenience function to get an fd_array key from an fsp.
368 ******************************************************************/
370 static TDB_DATA fd_array_key_fsp(files_struct *fsp)
372 return fd_array_key(fsp->dev, fsp->inode);
375 /*******************************************************************
376 Convenience function to get a lock ref count key from an fsp.
377 ******************************************************************/
379 static TDB_DATA locking_ref_count_key_fsp(files_struct *fsp)
381 return locking_ref_count_key(fsp->dev, fsp->inode);
384 /*******************************************************************
385 Create the in-memory POSIX lock databases.
386 ********************************************************************/
388 BOOL posix_locking_init(int read_only)
390 if (posix_pending_close_tdb) {
391 return True;
394 if (!posix_pending_close_tdb) {
395 posix_pending_close_tdb = tdb_open_log(NULL, 0, TDB_INTERNAL,
396 read_only?O_RDONLY:(O_RDWR|O_CREAT), 0644);
398 if (!posix_pending_close_tdb) {
399 DEBUG(0,("Failed to open POSIX pending close database.\n"));
400 return False;
403 return True;
406 /*******************************************************************
407 Delete the in-memory POSIX lock databases.
408 ********************************************************************/
410 BOOL posix_locking_end(void)
412 if (posix_pending_close_tdb && tdb_close(posix_pending_close_tdb) != 0) {
413 return False;
415 return True;
418 /****************************************************************************
419 Next - the functions that deal with storing fd's that have outstanding
420 POSIX locks when closed.
421 ****************************************************************************/
423 /****************************************************************************
424 The records in posix_pending_close_tdb are composed of an array of ints
425 keyed by dev/ino pair.
426 The first int is a reference count of the number of outstanding locks on
427 all open fd's on this dev/ino pair. Any subsequent ints are the fd's that
428 were open on this dev/ino pair that should have been closed, but can't as
429 the lock ref count is non zero.
430 ****************************************************************************/
432 /****************************************************************************
433 Keep a reference count of the number of Windows locks open on this dev/ino
434 pair. Creates entry if it doesn't exist.
435 ****************************************************************************/
437 static void increment_windows_lock_ref_count(files_struct *fsp)
439 TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
440 TDB_DATA dbuf;
441 int lock_ref_count;
443 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
444 if (dbuf.dptr == NULL) {
445 dbuf.dptr = (char *)SMB_MALLOC_P(int);
446 if (!dbuf.dptr) {
447 smb_panic("increment_windows_lock_ref_count: malloc fail.\n");
449 memset(dbuf.dptr, '\0', sizeof(int));
450 dbuf.dsize = sizeof(int);
453 memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
454 lock_ref_count++;
455 memcpy(dbuf.dptr, &lock_ref_count, sizeof(int));
457 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
458 smb_panic("increment_windows_lock_ref_count: tdb_store_fail.\n");
460 SAFE_FREE(dbuf.dptr);
462 DEBUG(10,("increment_windows_lock_ref_count for file now %s = %d\n",
463 fsp->fsp_name, lock_ref_count ));
466 static void decrement_windows_lock_ref_count(files_struct *fsp)
468 TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
469 TDB_DATA dbuf;
470 int lock_ref_count;
472 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
473 if (!dbuf.dptr) {
474 smb_panic("decrement_windows_lock_ref_count: logic error.\n");
477 memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
478 lock_ref_count--;
479 memcpy(dbuf.dptr, &lock_ref_count, sizeof(int));
481 if (lock_ref_count < 0) {
482 smb_panic("decrement_windows_lock_ref_count: lock_count logic error.\n");
485 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
486 smb_panic("decrement_windows_lock_ref_count: tdb_store_fail.\n");
488 SAFE_FREE(dbuf.dptr);
490 DEBUG(10,("decrement_windows_lock_ref_count for file now %s = %d\n",
491 fsp->fsp_name, lock_ref_count ));
494 /****************************************************************************
495 Bulk delete - subtract as many locks as we've just deleted.
496 ****************************************************************************/
498 void reduce_windows_lock_ref_count(files_struct *fsp, unsigned int dcount)
500 TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
501 TDB_DATA dbuf;
502 int lock_ref_count;
504 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
505 if (!dbuf.dptr) {
506 return;
509 memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
510 lock_ref_count -= dcount;
512 if (lock_ref_count < 0) {
513 smb_panic("reduce_windows_lock_ref_count: lock_count logic error.\n");
515 memcpy(dbuf.dptr, &lock_ref_count, sizeof(int));
517 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
518 smb_panic("reduce_windows_lock_ref_count: tdb_store_fail.\n");
520 SAFE_FREE(dbuf.dptr);
522 DEBUG(10,("reduce_windows_lock_ref_count for file now %s = %d\n",
523 fsp->fsp_name, lock_ref_count ));
526 /****************************************************************************
527 Fetch the lock ref count.
528 ****************************************************************************/
530 static int get_windows_lock_ref_count(files_struct *fsp)
532 TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
533 TDB_DATA dbuf;
534 int lock_ref_count;
536 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
537 if (!dbuf.dptr) {
538 lock_ref_count = 0;
539 } else {
540 memcpy(&lock_ref_count, dbuf.dptr, sizeof(int));
542 SAFE_FREE(dbuf.dptr);
544 DEBUG(10,("get_windows_lock_count for file %s = %d\n",
545 fsp->fsp_name, lock_ref_count ));
546 return lock_ref_count;
549 /****************************************************************************
550 Delete a lock_ref_count entry.
551 ****************************************************************************/
553 static void delete_windows_lock_ref_count(files_struct *fsp)
555 TDB_DATA kbuf = locking_ref_count_key_fsp(fsp);
557 /* Not a bug if it doesn't exist - no locks were ever granted. */
558 tdb_delete(posix_pending_close_tdb, kbuf);
559 DEBUG(10,("delete_windows_lock_ref_count for file %s\n", fsp->fsp_name));
562 /****************************************************************************
563 Add an fd to the pending close tdb.
564 ****************************************************************************/
566 static void add_fd_to_close_entry(files_struct *fsp)
568 TDB_DATA kbuf = fd_array_key_fsp(fsp);
569 TDB_DATA dbuf;
571 dbuf.dptr = NULL;
572 dbuf.dsize = 0;
574 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
576 dbuf.dptr = (char *)SMB_REALLOC(dbuf.dptr, dbuf.dsize + sizeof(int));
577 if (!dbuf.dptr) {
578 smb_panic("add_fd_to_close_entry: Realloc fail !\n");
581 memcpy(dbuf.dptr + dbuf.dsize, &fsp->fh->fd, sizeof(int));
582 dbuf.dsize += sizeof(int);
584 if (tdb_store(posix_pending_close_tdb, kbuf, dbuf, TDB_REPLACE) == -1) {
585 smb_panic("add_fd_to_close_entry: tdb_store_fail.\n");
588 DEBUG(10,("add_fd_to_close_entry: added fd %d file %s\n",
589 fsp->fh->fd, fsp->fsp_name ));
591 SAFE_FREE(dbuf.dptr);
594 /****************************************************************************
595 Remove all fd entries for a specific dev/inode pair from the tdb.
596 ****************************************************************************/
598 static void delete_close_entries(files_struct *fsp)
600 TDB_DATA kbuf = fd_array_key_fsp(fsp);
602 if (tdb_delete(posix_pending_close_tdb, kbuf) == -1) {
603 smb_panic("delete_close_entries: tdb_delete fail !\n");
607 /****************************************************************************
608 Get the array of POSIX pending close records for an open fsp. Caller must
609 free. Returns number of entries.
610 ****************************************************************************/
612 static size_t get_posix_pending_close_entries(files_struct *fsp, int **entries)
614 TDB_DATA kbuf = fd_array_key_fsp(fsp);
615 TDB_DATA dbuf;
616 size_t count = 0;
618 *entries = NULL;
619 dbuf.dptr = NULL;
621 dbuf = tdb_fetch(posix_pending_close_tdb, kbuf);
623 if (!dbuf.dptr) {
624 return 0;
627 *entries = (int *)dbuf.dptr;
628 count = (size_t)(dbuf.dsize / sizeof(int));
630 return count;
633 /****************************************************************************
634 Deal with pending closes needed by POSIX locking support.
635 Note that posix_locking_close_file() is expected to have been called
636 to delete all locks on this fsp before this function is called.
637 ****************************************************************************/
639 NTSTATUS fd_close_posix(struct connection_struct *conn, files_struct *fsp)
641 int saved_errno = 0;
642 int ret;
643 int *fd_array = NULL;
644 size_t count, i;
646 if (!lp_locking(fsp->conn->params) || !lp_posix_locking(conn->params)) {
648 * No locking or POSIX to worry about or we want POSIX semantics
649 * which will lose all locks on all fd's open on this dev/inode,
650 * just close.
652 ret = SMB_VFS_CLOSE(fsp,fsp->fh->fd);
653 fsp->fh->fd = -1;
654 if (ret == -1) {
655 return map_nt_error_from_unix(errno);
657 return NT_STATUS_OK;
660 if (get_windows_lock_ref_count(fsp)) {
663 * There are outstanding locks on this dev/inode pair on other fds.
664 * Add our fd to the pending close tdb and set fsp->fh->fd to -1.
667 add_fd_to_close_entry(fsp);
668 fsp->fh->fd = -1;
669 return NT_STATUS_OK;
673 * No outstanding locks. Get the pending close fd's
674 * from the tdb and close them all.
677 count = get_posix_pending_close_entries(fsp, &fd_array);
679 if (count) {
680 DEBUG(10,("fd_close_posix: doing close on %u fd's.\n", (unsigned int)count ));
682 for(i = 0; i < count; i++) {
683 if (SMB_VFS_CLOSE(fsp,fd_array[i]) == -1) {
684 saved_errno = errno;
689 * Delete all fd's stored in the tdb
690 * for this dev/inode pair.
693 delete_close_entries(fsp);
696 SAFE_FREE(fd_array);
698 /* Don't need a lock ref count on this dev/ino anymore. */
699 delete_windows_lock_ref_count(fsp);
702 * Finally close the fd associated with this fsp.
705 ret = SMB_VFS_CLOSE(fsp,fsp->fh->fd);
707 if (ret == 0 && saved_errno != 0) {
708 errno = saved_errno;
709 ret = -1;
712 fsp->fh->fd = -1;
714 if (ret == -1) {
715 return map_nt_error_from_unix(errno);
718 return NT_STATUS_OK;
721 /****************************************************************************
722 Next - the functions that deal with the mapping CIFS Windows locks onto
723 the underlying system POSIX locks.
724 ****************************************************************************/
727 * Structure used when splitting a lock range
728 * into a POSIX lock range. Doubly linked list.
731 struct lock_list {
732 struct lock_list *next;
733 struct lock_list *prev;
734 SMB_OFF_T start;
735 SMB_OFF_T size;
738 /****************************************************************************
739 Create a list of lock ranges that don't overlap a given range. Used in calculating
740 POSIX locks and unlocks. This is a difficult function that requires ASCII art to
741 understand it :-).
742 ****************************************************************************/
744 static struct lock_list *posix_lock_list(TALLOC_CTX *ctx,
745 struct lock_list *lhead,
746 const struct lock_context *lock_ctx, /* Lock context lhead belongs to. */
747 files_struct *fsp,
748 const struct lock_struct *plocks,
749 int num_locks)
751 int i;
754 * Check the current lock list on this dev/inode pair.
755 * Quit if the list is deleted.
758 DEBUG(10,("posix_lock_list: curr: start=%.0f,size=%.0f\n",
759 (double)lhead->start, (double)lhead->size ));
761 for (i=0; i<num_locks && lhead; i++) {
762 const struct lock_struct *lock = &plocks[i];
763 struct lock_list *l_curr;
765 /* Ignore all but read/write locks. */
766 if (lock->lock_type != READ_LOCK && lock->lock_type != WRITE_LOCK) {
767 continue;
770 /* Ignore locks not owned by this process. */
771 if (!procid_equal(&lock->context.pid, &lock_ctx->pid)) {
772 continue;
776 * Walk the lock list, checking for overlaps. Note that
777 * the lock list can expand within this loop if the current
778 * range being examined needs to be split.
781 for (l_curr = lhead; l_curr;) {
783 DEBUG(10,("posix_lock_list: lock: fnum=%d: start=%.0f,size=%.0f:type=%s", lock->fnum,
784 (double)lock->start, (double)lock->size, posix_lock_type_name(lock->lock_type) ));
786 if ( (l_curr->start >= (lock->start + lock->size)) ||
787 (lock->start >= (l_curr->start + l_curr->size))) {
789 /* No overlap with existing lock - leave this range alone. */
790 /*********************************************
791 +---------+
792 | l_curr |
793 +---------+
794 +-------+
795 | lock |
796 +-------+
797 OR....
798 +---------+
799 | l_curr |
800 +---------+
801 **********************************************/
803 DEBUG(10,(" no overlap case.\n" ));
805 l_curr = l_curr->next;
807 } else if ( (l_curr->start >= lock->start) &&
808 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
811 * This range is completely overlapped by this existing lock range
812 * and thus should have no effect. Delete it from the list.
814 /*********************************************
815 +---------+
816 | l_curr |
817 +---------+
818 +---------------------------+
819 | lock |
820 +---------------------------+
821 **********************************************/
822 /* Save the next pointer */
823 struct lock_list *ul_next = l_curr->next;
825 DEBUG(10,(" delete case.\n" ));
827 DLIST_REMOVE(lhead, l_curr);
828 if(lhead == NULL) {
829 break; /* No more list... */
832 l_curr = ul_next;
834 } else if ( (l_curr->start >= lock->start) &&
835 (l_curr->start < lock->start + lock->size) &&
836 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
839 * This range overlaps the existing lock range at the high end.
840 * Truncate by moving start to existing range end and reducing size.
842 /*********************************************
843 +---------------+
844 | l_curr |
845 +---------------+
846 +---------------+
847 | lock |
848 +---------------+
849 BECOMES....
850 +-------+
851 | l_curr|
852 +-------+
853 **********************************************/
855 l_curr->size = (l_curr->start + l_curr->size) - (lock->start + lock->size);
856 l_curr->start = lock->start + lock->size;
858 DEBUG(10,(" truncate high case: start=%.0f,size=%.0f\n",
859 (double)l_curr->start, (double)l_curr->size ));
861 l_curr = l_curr->next;
863 } else if ( (l_curr->start < lock->start) &&
864 (l_curr->start + l_curr->size > lock->start) &&
865 (l_curr->start + l_curr->size <= lock->start + lock->size) ) {
868 * This range overlaps the existing lock range at the low end.
869 * Truncate by reducing size.
871 /*********************************************
872 +---------------+
873 | l_curr |
874 +---------------+
875 +---------------+
876 | lock |
877 +---------------+
878 BECOMES....
879 +-------+
880 | l_curr|
881 +-------+
882 **********************************************/
884 l_curr->size = lock->start - l_curr->start;
886 DEBUG(10,(" truncate low case: start=%.0f,size=%.0f\n",
887 (double)l_curr->start, (double)l_curr->size ));
889 l_curr = l_curr->next;
891 } else if ( (l_curr->start < lock->start) &&
892 (l_curr->start + l_curr->size > lock->start + lock->size) ) {
894 * Worst case scenario. Range completely overlaps an existing
895 * lock range. Split the request into two, push the new (upper) request
896 * into the dlink list, and continue with the entry after l_new (as we
897 * know that l_new will not overlap with this lock).
899 /*********************************************
900 +---------------------------+
901 | l_curr |
902 +---------------------------+
903 +---------+
904 | lock |
905 +---------+
906 BECOMES.....
907 +-------+ +---------+
908 | l_curr| | l_new |
909 +-------+ +---------+
910 **********************************************/
911 struct lock_list *l_new = TALLOC_P(ctx, struct lock_list);
913 if(l_new == NULL) {
914 DEBUG(0,("posix_lock_list: talloc fail.\n"));
915 return NULL; /* The talloc_destroy takes care of cleanup. */
918 ZERO_STRUCTP(l_new);
919 l_new->start = lock->start + lock->size;
920 l_new->size = l_curr->start + l_curr->size - l_new->start;
922 /* Truncate the l_curr. */
923 l_curr->size = lock->start - l_curr->start;
925 DEBUG(10,(" split case: curr: start=%.0f,size=%.0f \
926 new: start=%.0f,size=%.0f\n", (double)l_curr->start, (double)l_curr->size,
927 (double)l_new->start, (double)l_new->size ));
930 * Add into the dlink list after the l_curr point - NOT at lhead.
931 * Note we can't use DLINK_ADD here as this inserts at the head of the given list.
934 l_new->prev = l_curr;
935 l_new->next = l_curr->next;
936 l_curr->next = l_new;
938 /* And move after the link we added. */
939 l_curr = l_new->next;
941 } else {
944 * This logic case should never happen. Ensure this is the
945 * case by forcing an abort.... Remove in production.
947 pstring msg;
949 slprintf(msg, sizeof(msg)-1, "logic flaw in cases: l_curr: start = %.0f, size = %.0f : \
950 lock: start = %.0f, size = %.0f\n", (double)l_curr->start, (double)l_curr->size, (double)lock->start, (double)lock->size );
952 smb_panic(msg);
954 } /* end for ( l_curr = lhead; l_curr;) */
955 } /* end for (i=0; i<num_locks && ul_head; i++) */
957 return lhead;
960 /****************************************************************************
961 POSIX function to acquire a lock. Returns True if the
962 lock could be granted, False if not.
963 ****************************************************************************/
965 BOOL set_posix_lock_windows_flavour(files_struct *fsp,
966 SMB_BIG_UINT u_offset,
967 SMB_BIG_UINT u_count,
968 enum brl_type lock_type,
969 const struct lock_context *lock_ctx,
970 const struct lock_struct *plocks,
971 int num_locks,
972 int *errno_ret)
974 SMB_OFF_T offset;
975 SMB_OFF_T count;
976 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
977 BOOL ret = True;
978 size_t lock_count;
979 TALLOC_CTX *l_ctx = NULL;
980 struct lock_list *llist = NULL;
981 struct lock_list *ll = NULL;
983 DEBUG(5,("set_posix_lock_windows_flavour: File %s, offset = %.0f, count = %.0f, type = %s\n",
984 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
987 * If the requested lock won't fit in the POSIX range, we will
988 * pretend it was successful.
991 if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
992 increment_windows_lock_ref_count(fsp);
993 return True;
997 * Windows is very strange. It allows read locks to be overlayed
998 * (even over a write lock), but leaves the write lock in force until the first
999 * unlock. It also reference counts the locks. This means the following sequence :
1001 * process1 process2
1002 * ------------------------------------------------------------------------
1003 * WRITE LOCK : start = 2, len = 10
1004 * READ LOCK: start =0, len = 10 - FAIL
1005 * READ LOCK : start = 0, len = 14
1006 * READ LOCK: start =0, len = 10 - FAIL
1007 * UNLOCK : start = 2, len = 10
1008 * READ LOCK: start =0, len = 10 - OK
1010 * Under POSIX, the same sequence in steps 1 and 2 would not be reference counted, but
1011 * would leave a single read lock over the 0-14 region.
1014 if ((l_ctx = talloc_init("set_posix_lock")) == NULL) {
1015 DEBUG(0,("set_posix_lock_windows_flavour: unable to init talloc context.\n"));
1016 return False;
1019 if ((ll = TALLOC_P(l_ctx, struct lock_list)) == NULL) {
1020 DEBUG(0,("set_posix_lock_windows_flavour: unable to talloc unlock list.\n"));
1021 talloc_destroy(l_ctx);
1022 return False;
1026 * Create the initial list entry containing the
1027 * lock we want to add.
1030 ZERO_STRUCTP(ll);
1031 ll->start = offset;
1032 ll->size = count;
1034 DLIST_ADD(llist, ll);
1037 * The following call calculates if there are any
1038 * overlapping locks held by this process on
1039 * fd's open on the same file and splits this list
1040 * into a list of lock ranges that do not overlap with existing
1041 * POSIX locks.
1044 llist = posix_lock_list(l_ctx,
1045 llist,
1046 lock_ctx, /* Lock context llist belongs to. */
1047 fsp,
1048 plocks,
1049 num_locks);
1052 * Add the POSIX locks on the list of ranges returned.
1053 * As the lock is supposed to be added atomically, we need to
1054 * back out all the locks if any one of these calls fail.
1057 for (lock_count = 0, ll = llist; ll; ll = ll->next, lock_count++) {
1058 offset = ll->start;
1059 count = ll->size;
1061 DEBUG(5,("set_posix_lock_windows_flavour: Real lock: Type = %s: offset = %.0f, count = %.0f\n",
1062 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1064 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1065 *errno_ret = errno;
1066 DEBUG(5,("set_posix_lock_windows_flavour: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
1067 posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
1068 ret = False;
1069 break;
1073 if (!ret) {
1076 * Back out all the POSIX locks we have on fail.
1079 for (ll = llist; lock_count; ll = ll->next, lock_count--) {
1080 offset = ll->start;
1081 count = ll->size;
1083 DEBUG(5,("set_posix_lock_windows_flavour: Backing out locks: Type = %s: offset = %.0f, count = %.0f\n",
1084 posix_lock_type_name(posix_lock_type), (double)offset, (double)count ));
1086 posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK);
1088 } else {
1089 /* Remember the number of Windows locks we have on this dev/ino pair. */
1090 increment_windows_lock_ref_count(fsp);
1093 talloc_destroy(l_ctx);
1094 return ret;
1097 /****************************************************************************
1098 POSIX function to release a lock. Returns True if the
1099 lock could be released, False if not.
1100 ****************************************************************************/
1102 BOOL release_posix_lock_windows_flavour(files_struct *fsp,
1103 SMB_BIG_UINT u_offset,
1104 SMB_BIG_UINT u_count,
1105 enum brl_type deleted_lock_type,
1106 const struct lock_context *lock_ctx,
1107 const struct lock_struct *plocks,
1108 int num_locks)
1110 SMB_OFF_T offset;
1111 SMB_OFF_T count;
1112 BOOL ret = True;
1113 TALLOC_CTX *ul_ctx = NULL;
1114 struct lock_list *ulist = NULL;
1115 struct lock_list *ul = NULL;
1117 DEBUG(5,("release_posix_lock_windows_flavour: File %s, offset = %.0f, count = %.0f\n",
1118 fsp->fsp_name, (double)u_offset, (double)u_count ));
1120 /* Remember the number of Windows locks we have on this dev/ino pair. */
1121 decrement_windows_lock_ref_count(fsp);
1124 * If the requested lock won't fit in the POSIX range, we will
1125 * pretend it was successful.
1128 if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
1129 return True;
1132 if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
1133 DEBUG(0,("release_posix_lock_windows_flavour: unable to init talloc context.\n"));
1134 return False;
1137 if ((ul = TALLOC_P(ul_ctx, struct lock_list)) == NULL) {
1138 DEBUG(0,("release_posix_lock_windows_flavour: unable to talloc unlock list.\n"));
1139 talloc_destroy(ul_ctx);
1140 return False;
1144 * Create the initial list entry containing the
1145 * lock we want to remove.
1148 ZERO_STRUCTP(ul);
1149 ul->start = offset;
1150 ul->size = count;
1152 DLIST_ADD(ulist, ul);
1155 * The following call calculates if there are any
1156 * overlapping locks held by this process on
1157 * fd's open on the same file and creates a
1158 * list of unlock ranges that will allow
1159 * POSIX lock ranges to remain on the file whilst the
1160 * unlocks are performed.
1163 ulist = posix_lock_list(ul_ctx,
1164 ulist,
1165 lock_ctx, /* Lock context ulist belongs to. */
1166 fsp,
1167 plocks,
1168 num_locks);
1171 * If there were any overlapped entries (list is > 1 or size or start have changed),
1172 * and the lock_type we just deleted from
1173 * the upper layer tdb was a write lock, then before doing the unlock we need to downgrade
1174 * the POSIX lock to a read lock. This allows any overlapping read locks
1175 * to be atomically maintained.
1178 if (deleted_lock_type == WRITE_LOCK &&
1179 (!ulist || ulist->next != NULL || ulist->start != offset || ulist->size != count)) {
1181 DEBUG(5,("release_posix_lock_windows_flavour: downgrading lock to READ: offset = %.0f, count = %.0f\n",
1182 (double)offset, (double)count ));
1184 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_RDLCK)) {
1185 DEBUG(0,("release_posix_lock_windows_flavour: downgrade of lock failed with error %s !\n", strerror(errno) ));
1186 talloc_destroy(ul_ctx);
1187 return False;
1192 * Release the POSIX locks on the list of ranges returned.
1195 for(; ulist; ulist = ulist->next) {
1196 offset = ulist->start;
1197 count = ulist->size;
1199 DEBUG(5,("release_posix_lock_windows_flavour: Real unlock: offset = %.0f, count = %.0f\n",
1200 (double)offset, (double)count ));
1202 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK)) {
1203 ret = False;
1207 talloc_destroy(ul_ctx);
1208 return ret;
1211 /****************************************************************************
1212 Next - the functions that deal with mapping CIFS POSIX locks onto
1213 the underlying system POSIX locks.
1214 ****************************************************************************/
1216 /****************************************************************************
1217 POSIX function to acquire a lock. Returns True if the
1218 lock could be granted, False if not.
1219 As POSIX locks don't stack or conflict (they just overwrite)
1220 we can map the requested lock directly onto a system one. We
1221 know it doesn't conflict with locks on other contexts as the
1222 upper layer would have refused it.
1223 ****************************************************************************/
1225 BOOL set_posix_lock_posix_flavour(files_struct *fsp,
1226 SMB_BIG_UINT u_offset,
1227 SMB_BIG_UINT u_count,
1228 enum brl_type lock_type,
1229 int *errno_ret)
1231 SMB_OFF_T offset;
1232 SMB_OFF_T count;
1233 int posix_lock_type = map_posix_lock_type(fsp,lock_type);
1235 DEBUG(5,("set_posix_lock_posix_flavour: File %s, offset = %.0f, count = %.0f, type = %s\n",
1236 fsp->fsp_name, (double)u_offset, (double)u_count, posix_lock_type_name(lock_type) ));
1239 * If the requested lock won't fit in the POSIX range, we will
1240 * pretend it was successful.
1243 if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
1244 return True;
1247 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,posix_lock_type)) {
1248 *errno_ret = errno;
1249 DEBUG(5,("set_posix_lock_posix_flavour: Lock fail !: Type = %s: offset = %.0f, count = %.0f. Errno = %s\n",
1250 posix_lock_type_name(posix_lock_type), (double)offset, (double)count, strerror(errno) ));
1251 return False;
1253 return True;
1256 /****************************************************************************
1257 POSIX function to release a lock. Returns True if the
1258 lock could be released, False if not.
1259 We are given a complete lock state from the upper layer which is what the lock
1260 state should be after the unlock has already been done, so what
1261 we do is punch out holes in the unlock range where locks owned by this process
1262 have a different lock context.
1263 ****************************************************************************/
1265 BOOL release_posix_lock_posix_flavour(files_struct *fsp,
1266 SMB_BIG_UINT u_offset,
1267 SMB_BIG_UINT u_count,
1268 const struct lock_context *lock_ctx,
1269 const struct lock_struct *plocks,
1270 int num_locks)
1272 BOOL ret = True;
1273 SMB_OFF_T offset;
1274 SMB_OFF_T count;
1275 TALLOC_CTX *ul_ctx = NULL;
1276 struct lock_list *ulist = NULL;
1277 struct lock_list *ul = NULL;
1279 DEBUG(5,("release_posix_lock_posix_flavour: File %s, offset = %.0f, count = %.0f\n",
1280 fsp->fsp_name, (double)u_offset, (double)u_count ));
1283 * If the requested lock won't fit in the POSIX range, we will
1284 * pretend it was successful.
1287 if(!posix_lock_in_range(&offset, &count, u_offset, u_count)) {
1288 return True;
1291 if ((ul_ctx = talloc_init("release_posix_lock")) == NULL) {
1292 DEBUG(0,("release_posix_lock_windows_flavour: unable to init talloc context.\n"));
1293 return False;
1296 if ((ul = TALLOC_P(ul_ctx, struct lock_list)) == NULL) {
1297 DEBUG(0,("release_posix_lock_windows_flavour: unable to talloc unlock list.\n"));
1298 talloc_destroy(ul_ctx);
1299 return False;
1303 * Create the initial list entry containing the
1304 * lock we want to remove.
1307 ZERO_STRUCTP(ul);
1308 ul->start = offset;
1309 ul->size = count;
1311 DLIST_ADD(ulist, ul);
1314 * Walk the given array creating a linked list
1315 * of unlock requests.
1318 ulist = posix_lock_list(ul_ctx,
1319 ulist,
1320 lock_ctx, /* Lock context ulist belongs to. */
1321 fsp,
1322 plocks,
1323 num_locks);
1326 * Release the POSIX locks on the list of ranges returned.
1329 for(; ulist; ulist = ulist->next) {
1330 offset = ulist->start;
1331 count = ulist->size;
1333 DEBUG(5,("release_posix_lock_posix_flavour: Real unlock: offset = %.0f, count = %.0f\n",
1334 (double)offset, (double)count ));
1336 if (!posix_fcntl_lock(fsp,SMB_F_SETLK,offset,count,F_UNLCK)) {
1337 ret = False;
1341 talloc_destroy(ul_ctx);
1342 return ret;