v2.6.22.24-op1
[linux-2.6.22.y-op.git] / fs / locks.c
blobe6d4c3b8f3a8f36f59570f379d67ca65a45cb51f
1 /*
2 * linux/fs/locks.c
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fs.h>
120 #include <linux/init.h>
121 #include <linux/module.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/smp_lock.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
129 #include <asm/semaphore.h>
130 #include <asm/uaccess.h>
132 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136 int leases_enable = 1;
137 int lease_break_time = 45;
139 #define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142 static LIST_HEAD(file_lock_list);
143 static LIST_HEAD(blocked_list);
145 static struct kmem_cache *filelock_cache __read_mostly;
147 /* Allocate an empty lock structure. */
148 static struct file_lock *locks_alloc_lock(void)
150 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
153 static void locks_release_private(struct file_lock *fl)
155 if (fl->fl_ops) {
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
158 fl->fl_ops = NULL;
160 if (fl->fl_lmops) {
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
163 fl->fl_lmops = NULL;
168 /* Free a lock which is not in use. */
169 static void locks_free_lock(struct file_lock *fl)
171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
175 locks_release_private(fl);
176 kmem_cache_free(filelock_cache, fl);
179 void locks_init_lock(struct file_lock *fl)
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
184 fl->fl_next = NULL;
185 fl->fl_fasync = NULL;
186 fl->fl_owner = NULL;
187 fl->fl_pid = 0;
188 fl->fl_file = NULL;
189 fl->fl_flags = 0;
190 fl->fl_type = 0;
191 fl->fl_start = fl->fl_end = 0;
192 fl->fl_ops = NULL;
193 fl->fl_lmops = NULL;
196 EXPORT_SYMBOL(locks_init_lock);
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
202 static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
204 struct file_lock *lock = (struct file_lock *) foo;
206 locks_init_lock(lock);
209 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
211 if (fl->fl_ops) {
212 if (fl->fl_ops->fl_copy_lock)
213 fl->fl_ops->fl_copy_lock(new, fl);
214 new->fl_ops = fl->fl_ops;
216 if (fl->fl_lmops) {
217 if (fl->fl_lmops->fl_copy_lock)
218 fl->fl_lmops->fl_copy_lock(new, fl);
219 new->fl_lmops = fl->fl_lmops;
224 * Initialize a new lock from an existing file_lock structure.
226 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
228 new->fl_owner = fl->fl_owner;
229 new->fl_pid = fl->fl_pid;
230 new->fl_file = NULL;
231 new->fl_flags = fl->fl_flags;
232 new->fl_type = fl->fl_type;
233 new->fl_start = fl->fl_start;
234 new->fl_end = fl->fl_end;
235 new->fl_ops = NULL;
236 new->fl_lmops = NULL;
239 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
241 locks_release_private(new);
243 __locks_copy_lock(new, fl);
244 new->fl_file = fl->fl_file;
245 new->fl_ops = fl->fl_ops;
246 new->fl_lmops = fl->fl_lmops;
248 locks_copy_private(new, fl);
251 EXPORT_SYMBOL(locks_copy_lock);
253 static inline int flock_translate_cmd(int cmd) {
254 if (cmd & LOCK_MAND)
255 return cmd & (LOCK_MAND | LOCK_RW);
256 switch (cmd) {
257 case LOCK_SH:
258 return F_RDLCK;
259 case LOCK_EX:
260 return F_WRLCK;
261 case LOCK_UN:
262 return F_UNLCK;
264 return -EINVAL;
267 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
268 static int flock_make_lock(struct file *filp, struct file_lock **lock,
269 unsigned int cmd)
271 struct file_lock *fl;
272 int type = flock_translate_cmd(cmd);
273 if (type < 0)
274 return type;
276 fl = locks_alloc_lock();
277 if (fl == NULL)
278 return -ENOMEM;
280 fl->fl_file = filp;
281 fl->fl_pid = current->tgid;
282 fl->fl_flags = FL_FLOCK;
283 fl->fl_type = type;
284 fl->fl_end = OFFSET_MAX;
286 *lock = fl;
287 return 0;
290 static int assign_type(struct file_lock *fl, int type)
292 switch (type) {
293 case F_RDLCK:
294 case F_WRLCK:
295 case F_UNLCK:
296 fl->fl_type = type;
297 break;
298 default:
299 return -EINVAL;
301 return 0;
304 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
305 * style lock.
307 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
308 struct flock *l)
310 off_t start, end;
312 switch (l->l_whence) {
313 case SEEK_SET:
314 start = 0;
315 break;
316 case SEEK_CUR:
317 start = filp->f_pos;
318 break;
319 case SEEK_END:
320 start = i_size_read(filp->f_path.dentry->d_inode);
321 break;
322 default:
323 return -EINVAL;
326 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
327 POSIX-2001 defines it. */
328 start += l->l_start;
329 if (start < 0)
330 return -EINVAL;
331 fl->fl_end = OFFSET_MAX;
332 if (l->l_len > 0) {
333 end = start + l->l_len - 1;
334 fl->fl_end = end;
335 } else if (l->l_len < 0) {
336 end = start - 1;
337 fl->fl_end = end;
338 start += l->l_len;
339 if (start < 0)
340 return -EINVAL;
342 fl->fl_start = start; /* we record the absolute position */
343 if (fl->fl_end < fl->fl_start)
344 return -EOVERFLOW;
346 fl->fl_owner = current->files;
347 fl->fl_pid = current->tgid;
348 fl->fl_file = filp;
349 fl->fl_flags = FL_POSIX;
350 fl->fl_ops = NULL;
351 fl->fl_lmops = NULL;
353 return assign_type(fl, l->l_type);
356 #if BITS_PER_LONG == 32
357 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
358 struct flock64 *l)
360 loff_t start;
362 switch (l->l_whence) {
363 case SEEK_SET:
364 start = 0;
365 break;
366 case SEEK_CUR:
367 start = filp->f_pos;
368 break;
369 case SEEK_END:
370 start = i_size_read(filp->f_path.dentry->d_inode);
371 break;
372 default:
373 return -EINVAL;
376 start += l->l_start;
377 if (start < 0)
378 return -EINVAL;
379 fl->fl_end = OFFSET_MAX;
380 if (l->l_len > 0) {
381 fl->fl_end = start + l->l_len - 1;
382 } else if (l->l_len < 0) {
383 fl->fl_end = start - 1;
384 start += l->l_len;
385 if (start < 0)
386 return -EINVAL;
388 fl->fl_start = start; /* we record the absolute position */
389 if (fl->fl_end < fl->fl_start)
390 return -EOVERFLOW;
392 fl->fl_owner = current->files;
393 fl->fl_pid = current->tgid;
394 fl->fl_file = filp;
395 fl->fl_flags = FL_POSIX;
396 fl->fl_ops = NULL;
397 fl->fl_lmops = NULL;
399 switch (l->l_type) {
400 case F_RDLCK:
401 case F_WRLCK:
402 case F_UNLCK:
403 fl->fl_type = l->l_type;
404 break;
405 default:
406 return -EINVAL;
409 return (0);
411 #endif
413 /* default lease lock manager operations */
414 static void lease_break_callback(struct file_lock *fl)
416 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
419 static void lease_release_private_callback(struct file_lock *fl)
421 if (!fl->fl_file)
422 return;
424 f_delown(fl->fl_file);
425 fl->fl_file->f_owner.signum = 0;
428 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
430 return fl->fl_file == try->fl_file;
433 static struct lock_manager_operations lease_manager_ops = {
434 .fl_break = lease_break_callback,
435 .fl_release_private = lease_release_private_callback,
436 .fl_mylease = lease_mylease_callback,
437 .fl_change = lease_modify,
441 * Initialize a lease, use the default lock manager operations
443 static int lease_init(struct file *filp, int type, struct file_lock *fl)
445 if (assign_type(fl, type) != 0)
446 return -EINVAL;
448 fl->fl_owner = current->files;
449 fl->fl_pid = current->tgid;
451 fl->fl_file = filp;
452 fl->fl_flags = FL_LEASE;
453 fl->fl_start = 0;
454 fl->fl_end = OFFSET_MAX;
455 fl->fl_ops = NULL;
456 fl->fl_lmops = &lease_manager_ops;
457 return 0;
460 /* Allocate a file_lock initialised to this type of lease */
461 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
463 struct file_lock *fl = locks_alloc_lock();
464 int error = -ENOMEM;
466 if (fl == NULL)
467 goto out;
469 error = lease_init(filp, type, fl);
470 if (error) {
471 locks_free_lock(fl);
472 fl = NULL;
474 out:
475 *flp = fl;
476 return error;
479 /* Check if two locks overlap each other.
481 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
483 return ((fl1->fl_end >= fl2->fl_start) &&
484 (fl2->fl_end >= fl1->fl_start));
488 * Check whether two locks have the same owner.
490 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
492 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
493 return fl2->fl_lmops == fl1->fl_lmops &&
494 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
495 return fl1->fl_owner == fl2->fl_owner;
498 /* Remove waiter from blocker's block list.
499 * When blocker ends up pointing to itself then the list is empty.
501 static void __locks_delete_block(struct file_lock *waiter)
503 list_del_init(&waiter->fl_block);
504 list_del_init(&waiter->fl_link);
505 waiter->fl_next = NULL;
510 static void locks_delete_block(struct file_lock *waiter)
512 lock_kernel();
513 __locks_delete_block(waiter);
514 unlock_kernel();
517 /* Insert waiter into blocker's block list.
518 * We use a circular list so that processes can be easily woken up in
519 * the order they blocked. The documentation doesn't require this but
520 * it seems like the reasonable thing to do.
522 static void locks_insert_block(struct file_lock *blocker,
523 struct file_lock *waiter)
525 BUG_ON(!list_empty(&waiter->fl_block));
526 list_add_tail(&waiter->fl_block, &blocker->fl_block);
527 waiter->fl_next = blocker;
528 if (IS_POSIX(blocker))
529 list_add(&waiter->fl_link, &blocked_list);
532 /* Wake up processes blocked waiting for blocker.
533 * If told to wait then schedule the processes until the block list
534 * is empty, otherwise empty the block list ourselves.
536 static void locks_wake_up_blocks(struct file_lock *blocker)
538 while (!list_empty(&blocker->fl_block)) {
539 struct file_lock *waiter = list_entry(blocker->fl_block.next,
540 struct file_lock, fl_block);
541 __locks_delete_block(waiter);
542 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
543 waiter->fl_lmops->fl_notify(waiter);
544 else
545 wake_up(&waiter->fl_wait);
549 /* Insert file lock fl into an inode's lock list at the position indicated
550 * by pos. At the same time add the lock to the global file lock list.
552 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
554 list_add(&fl->fl_link, &file_lock_list);
556 /* insert into file's list */
557 fl->fl_next = *pos;
558 *pos = fl;
560 if (fl->fl_ops && fl->fl_ops->fl_insert)
561 fl->fl_ops->fl_insert(fl);
565 * Delete a lock and then free it.
566 * Wake up processes that are blocked waiting for this lock,
567 * notify the FS that the lock has been cleared and
568 * finally free the lock.
570 static void locks_delete_lock(struct file_lock **thisfl_p)
572 struct file_lock *fl = *thisfl_p;
574 *thisfl_p = fl->fl_next;
575 fl->fl_next = NULL;
576 list_del_init(&fl->fl_link);
578 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
579 if (fl->fl_fasync != NULL) {
580 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
581 fl->fl_fasync = NULL;
584 if (fl->fl_ops && fl->fl_ops->fl_remove)
585 fl->fl_ops->fl_remove(fl);
587 locks_wake_up_blocks(fl);
588 locks_free_lock(fl);
591 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
592 * checks for shared/exclusive status of overlapping locks.
594 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
596 if (sys_fl->fl_type == F_WRLCK)
597 return 1;
598 if (caller_fl->fl_type == F_WRLCK)
599 return 1;
600 return 0;
603 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
604 * checking before calling the locks_conflict().
606 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
608 /* POSIX locks owned by the same process do not conflict with
609 * each other.
611 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
612 return (0);
614 /* Check whether they overlap */
615 if (!locks_overlap(caller_fl, sys_fl))
616 return 0;
618 return (locks_conflict(caller_fl, sys_fl));
621 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
622 * checking before calling the locks_conflict().
624 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
626 /* FLOCK locks referring to the same filp do not conflict with
627 * each other.
629 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
630 return (0);
631 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
632 return 0;
634 return (locks_conflict(caller_fl, sys_fl));
637 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
639 int result = 0;
640 DECLARE_WAITQUEUE(wait, current);
642 __set_current_state(TASK_INTERRUPTIBLE);
643 add_wait_queue(fl_wait, &wait);
644 if (timeout == 0)
645 schedule();
646 else
647 result = schedule_timeout(timeout);
648 if (signal_pending(current))
649 result = -ERESTARTSYS;
650 remove_wait_queue(fl_wait, &wait);
651 __set_current_state(TASK_RUNNING);
652 return result;
655 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
657 int result;
658 locks_insert_block(blocker, waiter);
659 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
660 __locks_delete_block(waiter);
661 return result;
665 posix_test_lock(struct file *filp, struct file_lock *fl)
667 struct file_lock *cfl;
669 lock_kernel();
670 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
671 if (!IS_POSIX(cfl))
672 continue;
673 if (posix_locks_conflict(cfl, fl))
674 break;
676 if (cfl) {
677 __locks_copy_lock(fl, cfl);
678 unlock_kernel();
679 return 1;
680 } else
681 fl->fl_type = F_UNLCK;
682 unlock_kernel();
683 return 0;
686 EXPORT_SYMBOL(posix_test_lock);
688 /* This function tests for deadlock condition before putting a process to
689 * sleep. The detection scheme is no longer recursive. Recursive was neat,
690 * but dangerous - we risked stack corruption if the lock data was bad, or
691 * if the recursion was too deep for any other reason.
693 * We rely on the fact that a task can only be on one lock's wait queue
694 * at a time. When we find blocked_task on a wait queue we can re-search
695 * with blocked_task equal to that queue's owner, until either blocked_task
696 * isn't found, or blocked_task is found on a queue owned by my_task.
698 * Note: the above assumption may not be true when handling lock requests
699 * from a broken NFS client. But broken NFS clients have a lot more to
700 * worry about than proper deadlock detection anyway... --okir
702 static int posix_locks_deadlock(struct file_lock *caller_fl,
703 struct file_lock *block_fl)
705 struct list_head *tmp;
707 next_task:
708 if (posix_same_owner(caller_fl, block_fl))
709 return 1;
710 list_for_each(tmp, &blocked_list) {
711 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
712 if (posix_same_owner(fl, block_fl)) {
713 fl = fl->fl_next;
714 block_fl = fl;
715 goto next_task;
718 return 0;
721 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
722 * at the head of the list, but that's secret knowledge known only to
723 * flock_lock_file and posix_lock_file.
725 * Note that if called with an FL_EXISTS argument, the caller may determine
726 * whether or not a lock was successfully freed by testing the return
727 * value for -ENOENT.
729 static int flock_lock_file(struct file *filp, struct file_lock *request)
731 struct file_lock *new_fl = NULL;
732 struct file_lock **before;
733 struct inode * inode = filp->f_path.dentry->d_inode;
734 int error = 0;
735 int found = 0;
737 lock_kernel();
738 if (request->fl_flags & FL_ACCESS)
739 goto find_conflict;
740 for_each_lock(inode, before) {
741 struct file_lock *fl = *before;
742 if (IS_POSIX(fl))
743 break;
744 if (IS_LEASE(fl))
745 continue;
746 if (filp != fl->fl_file)
747 continue;
748 if (request->fl_type == fl->fl_type)
749 goto out;
750 found = 1;
751 locks_delete_lock(before);
752 break;
755 if (request->fl_type == F_UNLCK) {
756 if ((request->fl_flags & FL_EXISTS) && !found)
757 error = -ENOENT;
758 goto out;
761 error = -ENOMEM;
762 new_fl = locks_alloc_lock();
763 if (new_fl == NULL)
764 goto out;
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
769 if (found)
770 cond_resched();
772 find_conflict:
773 for_each_lock(inode, before) {
774 struct file_lock *fl = *before;
775 if (IS_POSIX(fl))
776 break;
777 if (IS_LEASE(fl))
778 continue;
779 if (!flock_locks_conflict(request, fl))
780 continue;
781 error = -EAGAIN;
782 if (request->fl_flags & FL_SLEEP)
783 locks_insert_block(fl, request);
784 goto out;
786 if (request->fl_flags & FL_ACCESS)
787 goto out;
788 locks_copy_lock(new_fl, request);
789 locks_insert_lock(before, new_fl);
790 new_fl = NULL;
791 error = 0;
793 out:
794 unlock_kernel();
795 if (new_fl)
796 locks_free_lock(new_fl);
797 return error;
800 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
802 struct file_lock *fl;
803 struct file_lock *new_fl = NULL;
804 struct file_lock *new_fl2 = NULL;
805 struct file_lock *left = NULL;
806 struct file_lock *right = NULL;
807 struct file_lock **before;
808 int error, added = 0;
811 * We may need two file_lock structures for this operation,
812 * so we get them in advance to avoid races.
814 * In some cases we can be sure, that no new locks will be needed
816 if (!(request->fl_flags & FL_ACCESS) &&
817 (request->fl_type != F_UNLCK ||
818 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
819 new_fl = locks_alloc_lock();
820 new_fl2 = locks_alloc_lock();
823 lock_kernel();
824 if (request->fl_type != F_UNLCK) {
825 for_each_lock(inode, before) {
826 struct file_lock *fl = *before;
827 if (!IS_POSIX(fl))
828 continue;
829 if (!posix_locks_conflict(request, fl))
830 continue;
831 if (conflock)
832 locks_copy_lock(conflock, fl);
833 error = -EAGAIN;
834 if (!(request->fl_flags & FL_SLEEP))
835 goto out;
836 error = -EDEADLK;
837 if (posix_locks_deadlock(request, fl))
838 goto out;
839 error = -EAGAIN;
840 locks_insert_block(fl, request);
841 goto out;
845 /* If we're just looking for a conflict, we're done. */
846 error = 0;
847 if (request->fl_flags & FL_ACCESS)
848 goto out;
851 * Find the first old lock with the same owner as the new lock.
854 before = &inode->i_flock;
856 /* First skip locks owned by other processes. */
857 while ((fl = *before) && (!IS_POSIX(fl) ||
858 !posix_same_owner(request, fl))) {
859 before = &fl->fl_next;
862 /* Process locks with this owner. */
863 while ((fl = *before) && posix_same_owner(request, fl)) {
864 /* Detect adjacent or overlapping regions (if same lock type)
866 if (request->fl_type == fl->fl_type) {
867 /* In all comparisons of start vs end, use
868 * "start - 1" rather than "end + 1". If end
869 * is OFFSET_MAX, end + 1 will become negative.
871 if (fl->fl_end < request->fl_start - 1)
872 goto next_lock;
873 /* If the next lock in the list has entirely bigger
874 * addresses than the new one, insert the lock here.
876 if (fl->fl_start - 1 > request->fl_end)
877 break;
879 /* If we come here, the new and old lock are of the
880 * same type and adjacent or overlapping. Make one
881 * lock yielding from the lower start address of both
882 * locks to the higher end address.
884 if (fl->fl_start > request->fl_start)
885 fl->fl_start = request->fl_start;
886 else
887 request->fl_start = fl->fl_start;
888 if (fl->fl_end < request->fl_end)
889 fl->fl_end = request->fl_end;
890 else
891 request->fl_end = fl->fl_end;
892 if (added) {
893 locks_delete_lock(before);
894 continue;
896 request = fl;
897 added = 1;
899 else {
900 /* Processing for different lock types is a bit
901 * more complex.
903 if (fl->fl_end < request->fl_start)
904 goto next_lock;
905 if (fl->fl_start > request->fl_end)
906 break;
907 if (request->fl_type == F_UNLCK)
908 added = 1;
909 if (fl->fl_start < request->fl_start)
910 left = fl;
911 /* If the next lock in the list has a higher end
912 * address than the new one, insert the new one here.
914 if (fl->fl_end > request->fl_end) {
915 right = fl;
916 break;
918 if (fl->fl_start >= request->fl_start) {
919 /* The new lock completely replaces an old
920 * one (This may happen several times).
922 if (added) {
923 locks_delete_lock(before);
924 continue;
926 /* Replace the old lock with the new one.
927 * Wake up anybody waiting for the old one,
928 * as the change in lock type might satisfy
929 * their needs.
931 locks_wake_up_blocks(fl);
932 fl->fl_start = request->fl_start;
933 fl->fl_end = request->fl_end;
934 fl->fl_type = request->fl_type;
935 locks_release_private(fl);
936 locks_copy_private(fl, request);
937 request = fl;
938 added = 1;
941 /* Go on to next lock.
943 next_lock:
944 before = &fl->fl_next;
948 * The above code only modifies existing locks in case of
949 * merging or replacing. If new lock(s) need to be inserted
950 * all modifications are done bellow this, so it's safe yet to
951 * bail out.
953 error = -ENOLCK; /* "no luck" */
954 if (right && left == right && !new_fl2)
955 goto out;
957 error = 0;
958 if (!added) {
959 if (request->fl_type == F_UNLCK) {
960 if (request->fl_flags & FL_EXISTS)
961 error = -ENOENT;
962 goto out;
965 if (!new_fl) {
966 error = -ENOLCK;
967 goto out;
969 locks_copy_lock(new_fl, request);
970 locks_insert_lock(before, new_fl);
971 new_fl = NULL;
973 if (right) {
974 if (left == right) {
975 /* The new lock breaks the old one in two pieces,
976 * so we have to use the second new lock.
978 left = new_fl2;
979 new_fl2 = NULL;
980 locks_copy_lock(left, right);
981 locks_insert_lock(before, left);
983 right->fl_start = request->fl_end + 1;
984 locks_wake_up_blocks(right);
986 if (left) {
987 left->fl_end = request->fl_start - 1;
988 locks_wake_up_blocks(left);
990 out:
991 unlock_kernel();
993 * Free any unused locks.
995 if (new_fl)
996 locks_free_lock(new_fl);
997 if (new_fl2)
998 locks_free_lock(new_fl2);
999 return error;
1003 * posix_lock_file - Apply a POSIX-style lock to a file
1004 * @filp: The file to apply the lock to
1005 * @fl: The lock to be applied
1006 * @conflock: Place to return a copy of the conflicting lock, if found.
1008 * Add a POSIX style lock to a file.
1009 * We merge adjacent & overlapping locks whenever possible.
1010 * POSIX locks are sorted by owner task, then by starting address
1012 * Note that if called with an FL_EXISTS argument, the caller may determine
1013 * whether or not a lock was successfully freed by testing the return
1014 * value for -ENOENT.
1016 int posix_lock_file(struct file *filp, struct file_lock *fl,
1017 struct file_lock *conflock)
1019 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1021 EXPORT_SYMBOL(posix_lock_file);
1024 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1025 * @filp: The file to apply the lock to
1026 * @fl: The lock to be applied
1028 * Add a POSIX style lock to a file.
1029 * We merge adjacent & overlapping locks whenever possible.
1030 * POSIX locks are sorted by owner task, then by starting address
1032 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1034 int error;
1035 might_sleep ();
1036 for (;;) {
1037 error = posix_lock_file(filp, fl, NULL);
1038 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1039 break;
1040 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1041 if (!error)
1042 continue;
1044 locks_delete_block(fl);
1045 break;
1047 return error;
1049 EXPORT_SYMBOL(posix_lock_file_wait);
1052 * locks_mandatory_locked - Check for an active lock
1053 * @inode: the file to check
1055 * Searches the inode's list of locks to find any POSIX locks which conflict.
1056 * This function is called from locks_verify_locked() only.
1058 int locks_mandatory_locked(struct inode *inode)
1060 fl_owner_t owner = current->files;
1061 struct file_lock *fl;
1064 * Search the lock list for this inode for any POSIX locks.
1066 lock_kernel();
1067 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1068 if (!IS_POSIX(fl))
1069 continue;
1070 if (fl->fl_owner != owner)
1071 break;
1073 unlock_kernel();
1074 return fl ? -EAGAIN : 0;
1078 * locks_mandatory_area - Check for a conflicting lock
1079 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1080 * for shared
1081 * @inode: the file to check
1082 * @filp: how the file was opened (if it was)
1083 * @offset: start of area to check
1084 * @count: length of area to check
1086 * Searches the inode's list of locks to find any POSIX locks which conflict.
1087 * This function is called from rw_verify_area() and
1088 * locks_verify_truncate().
1090 int locks_mandatory_area(int read_write, struct inode *inode,
1091 struct file *filp, loff_t offset,
1092 size_t count)
1094 struct file_lock fl;
1095 int error;
1097 locks_init_lock(&fl);
1098 fl.fl_owner = current->files;
1099 fl.fl_pid = current->tgid;
1100 fl.fl_file = filp;
1101 fl.fl_flags = FL_POSIX | FL_ACCESS;
1102 if (filp && !(filp->f_flags & O_NONBLOCK))
1103 fl.fl_flags |= FL_SLEEP;
1104 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1105 fl.fl_start = offset;
1106 fl.fl_end = offset + count - 1;
1108 for (;;) {
1109 error = __posix_lock_file(inode, &fl, NULL);
1110 if (error != -EAGAIN)
1111 break;
1112 if (!(fl.fl_flags & FL_SLEEP))
1113 break;
1114 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1115 if (!error) {
1117 * If we've been sleeping someone might have
1118 * changed the permissions behind our back.
1120 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1121 continue;
1124 locks_delete_block(&fl);
1125 break;
1128 return error;
1131 EXPORT_SYMBOL(locks_mandatory_area);
1133 /* We already had a lease on this file; just change its type */
1134 int lease_modify(struct file_lock **before, int arg)
1136 struct file_lock *fl = *before;
1137 int error = assign_type(fl, arg);
1139 if (error)
1140 return error;
1141 locks_wake_up_blocks(fl);
1142 if (arg == F_UNLCK)
1143 locks_delete_lock(before);
1144 return 0;
1147 EXPORT_SYMBOL(lease_modify);
1149 static void time_out_leases(struct inode *inode)
1151 struct file_lock **before;
1152 struct file_lock *fl;
1154 before = &inode->i_flock;
1155 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1156 if ((fl->fl_break_time == 0)
1157 || time_before(jiffies, fl->fl_break_time)) {
1158 before = &fl->fl_next;
1159 continue;
1161 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1162 if (fl == *before) /* lease_modify may have freed fl */
1163 before = &fl->fl_next;
1168 * __break_lease - revoke all outstanding leases on file
1169 * @inode: the inode of the file to return
1170 * @mode: the open mode (read or write)
1172 * break_lease (inlined for speed) has checked there already
1173 * is a lease on this file. Leases are broken on a call to open()
1174 * or truncate(). This function can sleep unless you
1175 * specified %O_NONBLOCK to your open().
1177 int __break_lease(struct inode *inode, unsigned int mode)
1179 int error = 0, future;
1180 struct file_lock *new_fl, *flock;
1181 struct file_lock *fl;
1182 int alloc_err;
1183 unsigned long break_time;
1184 int i_have_this_lease = 0;
1186 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1187 &new_fl);
1189 lock_kernel();
1191 time_out_leases(inode);
1193 flock = inode->i_flock;
1194 if ((flock == NULL) || !IS_LEASE(flock))
1195 goto out;
1197 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1198 if (fl->fl_owner == current->files)
1199 i_have_this_lease = 1;
1201 if (mode & FMODE_WRITE) {
1202 /* If we want write access, we have to revoke any lease. */
1203 future = F_UNLCK | F_INPROGRESS;
1204 } else if (flock->fl_type & F_INPROGRESS) {
1205 /* If the lease is already being broken, we just leave it */
1206 future = flock->fl_type;
1207 } else if (flock->fl_type & F_WRLCK) {
1208 /* Downgrade the exclusive lease to a read-only lease. */
1209 future = F_RDLCK | F_INPROGRESS;
1210 } else {
1211 /* the existing lease was read-only, so we can read too. */
1212 goto out;
1215 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1216 error = alloc_err;
1217 goto out;
1220 break_time = 0;
1221 if (lease_break_time > 0) {
1222 break_time = jiffies + lease_break_time * HZ;
1223 if (break_time == 0)
1224 break_time++; /* so that 0 means no break time */
1227 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1228 if (fl->fl_type != future) {
1229 fl->fl_type = future;
1230 fl->fl_break_time = break_time;
1231 /* lease must have lmops break callback */
1232 fl->fl_lmops->fl_break(fl);
1236 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1237 error = -EWOULDBLOCK;
1238 goto out;
1241 restart:
1242 break_time = flock->fl_break_time;
1243 if (break_time != 0) {
1244 break_time -= jiffies;
1245 if (break_time == 0)
1246 break_time++;
1248 error = locks_block_on_timeout(flock, new_fl, break_time);
1249 if (error >= 0) {
1250 if (error == 0)
1251 time_out_leases(inode);
1252 /* Wait for the next lease that has not been broken yet */
1253 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1254 flock = flock->fl_next) {
1255 if (flock->fl_type & F_INPROGRESS)
1256 goto restart;
1258 error = 0;
1261 out:
1262 unlock_kernel();
1263 if (!alloc_err)
1264 locks_free_lock(new_fl);
1265 return error;
1268 EXPORT_SYMBOL(__break_lease);
1271 * lease_get_mtime
1272 * @inode: the inode
1273 * @time: pointer to a timespec which will contain the last modified time
1275 * This is to force NFS clients to flush their caches for files with
1276 * exclusive leases. The justification is that if someone has an
1277 * exclusive lease, then they could be modifiying it.
1279 void lease_get_mtime(struct inode *inode, struct timespec *time)
1281 struct file_lock *flock = inode->i_flock;
1282 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1283 *time = current_fs_time(inode->i_sb);
1284 else
1285 *time = inode->i_mtime;
1288 EXPORT_SYMBOL(lease_get_mtime);
1291 * fcntl_getlease - Enquire what lease is currently active
1292 * @filp: the file
1294 * The value returned by this function will be one of
1295 * (if no lease break is pending):
1297 * %F_RDLCK to indicate a shared lease is held.
1299 * %F_WRLCK to indicate an exclusive lease is held.
1301 * %F_UNLCK to indicate no lease is held.
1303 * (if a lease break is pending):
1305 * %F_RDLCK to indicate an exclusive lease needs to be
1306 * changed to a shared lease (or removed).
1308 * %F_UNLCK to indicate the lease needs to be removed.
1310 * XXX: sfr & willy disagree over whether F_INPROGRESS
1311 * should be returned to userspace.
1313 int fcntl_getlease(struct file *filp)
1315 struct file_lock *fl;
1316 int type = F_UNLCK;
1318 lock_kernel();
1319 time_out_leases(filp->f_path.dentry->d_inode);
1320 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1321 fl = fl->fl_next) {
1322 if (fl->fl_file == filp) {
1323 type = fl->fl_type & ~F_INPROGRESS;
1324 break;
1327 unlock_kernel();
1328 return type;
1332 * __setlease - sets a lease on an open file
1333 * @filp: file pointer
1334 * @arg: type of lease to obtain
1335 * @flp: input - file_lock to use, output - file_lock inserted
1337 * The (input) flp->fl_lmops->fl_break function is required
1338 * by break_lease().
1340 * Called with kernel lock held.
1342 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1344 struct file_lock *fl, **before, **my_before = NULL, *lease;
1345 struct dentry *dentry = filp->f_path.dentry;
1346 struct inode *inode = dentry->d_inode;
1347 int error, rdlease_count = 0, wrlease_count = 0;
1349 time_out_leases(inode);
1351 error = -EINVAL;
1352 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1353 goto out;
1355 lease = *flp;
1357 error = -EAGAIN;
1358 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1359 goto out;
1360 if ((arg == F_WRLCK)
1361 && ((atomic_read(&dentry->d_count) > 1)
1362 || (atomic_read(&inode->i_count) > 1)))
1363 goto out;
1366 * At this point, we know that if there is an exclusive
1367 * lease on this file, then we hold it on this filp
1368 * (otherwise our open of this file would have blocked).
1369 * And if we are trying to acquire an exclusive lease,
1370 * then the file is not open by anyone (including us)
1371 * except for this filp.
1373 for (before = &inode->i_flock;
1374 ((fl = *before) != NULL) && IS_LEASE(fl);
1375 before = &fl->fl_next) {
1376 if (lease->fl_lmops->fl_mylease(fl, lease))
1377 my_before = before;
1378 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1380 * Someone is in the process of opening this
1381 * file for writing so we may not take an
1382 * exclusive lease on it.
1384 wrlease_count++;
1385 else
1386 rdlease_count++;
1389 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1390 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1391 goto out;
1393 if (my_before != NULL) {
1394 *flp = *my_before;
1395 error = lease->fl_lmops->fl_change(my_before, arg);
1396 goto out;
1399 error = 0;
1400 if (arg == F_UNLCK)
1401 goto out;
1403 error = -EINVAL;
1404 if (!leases_enable)
1405 goto out;
1407 error = -ENOMEM;
1408 fl = locks_alloc_lock();
1409 if (fl == NULL)
1410 goto out;
1412 locks_copy_lock(fl, lease);
1414 locks_insert_lock(before, fl);
1416 *flp = fl;
1417 error = 0;
1418 out:
1419 return error;
1423 * setlease - sets a lease on an open file
1424 * @filp: file pointer
1425 * @arg: type of lease to obtain
1426 * @lease: file_lock to use
1428 * Call this to establish a lease on the file.
1429 * The fl_lmops fl_break function is required by break_lease
1432 int setlease(struct file *filp, long arg, struct file_lock **lease)
1434 struct dentry *dentry = filp->f_path.dentry;
1435 struct inode *inode = dentry->d_inode;
1436 int error;
1438 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1439 return -EACCES;
1440 if (!S_ISREG(inode->i_mode))
1441 return -EINVAL;
1442 error = security_file_lock(filp, arg);
1443 if (error)
1444 return error;
1446 lock_kernel();
1447 error = __setlease(filp, arg, lease);
1448 unlock_kernel();
1450 return error;
1453 EXPORT_SYMBOL(setlease);
1456 * fcntl_setlease - sets a lease on an open file
1457 * @fd: open file descriptor
1458 * @filp: file pointer
1459 * @arg: type of lease to obtain
1461 * Call this fcntl to establish a lease on the file.
1462 * Note that you also need to call %F_SETSIG to
1463 * receive a signal when the lease is broken.
1465 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1467 struct file_lock fl, *flp = &fl;
1468 struct dentry *dentry = filp->f_path.dentry;
1469 struct inode *inode = dentry->d_inode;
1470 int error;
1472 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1473 return -EACCES;
1474 if (!S_ISREG(inode->i_mode))
1475 return -EINVAL;
1476 error = security_file_lock(filp, arg);
1477 if (error)
1478 return error;
1480 locks_init_lock(&fl);
1481 error = lease_init(filp, arg, &fl);
1482 if (error)
1483 return error;
1485 lock_kernel();
1487 error = __setlease(filp, arg, &flp);
1488 if (error || arg == F_UNLCK)
1489 goto out_unlock;
1491 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1492 if (error < 0) {
1493 /* remove lease just inserted by __setlease */
1494 flp->fl_type = F_UNLCK | F_INPROGRESS;
1495 flp->fl_break_time = jiffies- 10;
1496 time_out_leases(inode);
1497 goto out_unlock;
1500 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1501 out_unlock:
1502 unlock_kernel();
1503 return error;
1507 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1508 * @filp: The file to apply the lock to
1509 * @fl: The lock to be applied
1511 * Add a FLOCK style lock to a file.
1513 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1515 int error;
1516 might_sleep();
1517 for (;;) {
1518 error = flock_lock_file(filp, fl);
1519 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1520 break;
1521 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1522 if (!error)
1523 continue;
1525 locks_delete_block(fl);
1526 break;
1528 return error;
1531 EXPORT_SYMBOL(flock_lock_file_wait);
1534 * sys_flock: - flock() system call.
1535 * @fd: the file descriptor to lock.
1536 * @cmd: the type of lock to apply.
1538 * Apply a %FL_FLOCK style lock to an open file descriptor.
1539 * The @cmd can be one of
1541 * %LOCK_SH -- a shared lock.
1543 * %LOCK_EX -- an exclusive lock.
1545 * %LOCK_UN -- remove an existing lock.
1547 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1549 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1550 * processes read and write access respectively.
1552 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1554 struct file *filp;
1555 struct file_lock *lock;
1556 int can_sleep, unlock;
1557 int error;
1559 error = -EBADF;
1560 filp = fget(fd);
1561 if (!filp)
1562 goto out;
1564 can_sleep = !(cmd & LOCK_NB);
1565 cmd &= ~LOCK_NB;
1566 unlock = (cmd == LOCK_UN);
1568 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1569 goto out_putf;
1571 error = flock_make_lock(filp, &lock, cmd);
1572 if (error)
1573 goto out_putf;
1574 if (can_sleep)
1575 lock->fl_flags |= FL_SLEEP;
1577 error = security_file_lock(filp, cmd);
1578 if (error)
1579 goto out_free;
1581 if (filp->f_op && filp->f_op->flock)
1582 error = filp->f_op->flock(filp,
1583 (can_sleep) ? F_SETLKW : F_SETLK,
1584 lock);
1585 else
1586 error = flock_lock_file_wait(filp, lock);
1588 out_free:
1589 locks_free_lock(lock);
1591 out_putf:
1592 fput(filp);
1593 out:
1594 return error;
1598 * vfs_test_lock - test file byte range lock
1599 * @filp: The file to test lock for
1600 * @fl: The lock to test
1601 * @conf: Place to return a copy of the conflicting lock, if found
1603 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1604 * setting conf->fl_type to something other than F_UNLCK.
1606 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1608 if (filp->f_op && filp->f_op->lock)
1609 return filp->f_op->lock(filp, F_GETLK, fl);
1610 posix_test_lock(filp, fl);
1611 return 0;
1613 EXPORT_SYMBOL_GPL(vfs_test_lock);
1615 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1617 flock->l_pid = fl->fl_pid;
1618 #if BITS_PER_LONG == 32
1620 * Make sure we can represent the posix lock via
1621 * legacy 32bit flock.
1623 if (fl->fl_start > OFFT_OFFSET_MAX)
1624 return -EOVERFLOW;
1625 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1626 return -EOVERFLOW;
1627 #endif
1628 flock->l_start = fl->fl_start;
1629 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1630 fl->fl_end - fl->fl_start + 1;
1631 flock->l_whence = 0;
1632 flock->l_type = fl->fl_type;
1633 return 0;
1636 #if BITS_PER_LONG == 32
1637 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1639 flock->l_pid = fl->fl_pid;
1640 flock->l_start = fl->fl_start;
1641 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1642 fl->fl_end - fl->fl_start + 1;
1643 flock->l_whence = 0;
1644 flock->l_type = fl->fl_type;
1646 #endif
1648 /* Report the first existing lock that would conflict with l.
1649 * This implements the F_GETLK command of fcntl().
1651 int fcntl_getlk(struct file *filp, struct flock __user *l)
1653 struct file_lock file_lock;
1654 struct flock flock;
1655 int error;
1657 error = -EFAULT;
1658 if (copy_from_user(&flock, l, sizeof(flock)))
1659 goto out;
1660 error = -EINVAL;
1661 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1662 goto out;
1664 error = flock_to_posix_lock(filp, &file_lock, &flock);
1665 if (error)
1666 goto out;
1668 error = vfs_test_lock(filp, &file_lock);
1669 if (error)
1670 goto out;
1672 flock.l_type = file_lock.fl_type;
1673 if (file_lock.fl_type != F_UNLCK) {
1674 error = posix_lock_to_flock(&flock, &file_lock);
1675 if (error)
1676 goto out;
1678 error = -EFAULT;
1679 if (!copy_to_user(l, &flock, sizeof(flock)))
1680 error = 0;
1681 out:
1682 return error;
1686 * vfs_lock_file - file byte range lock
1687 * @filp: The file to apply the lock to
1688 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1689 * @fl: The lock to be applied
1690 * @conf: Place to return a copy of the conflicting lock, if found.
1692 * A caller that doesn't care about the conflicting lock may pass NULL
1693 * as the final argument.
1695 * If the filesystem defines a private ->lock() method, then @conf will
1696 * be left unchanged; so a caller that cares should initialize it to
1697 * some acceptable default.
1699 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1700 * locks, the ->lock() interface may return asynchronously, before the lock has
1701 * been granted or denied by the underlying filesystem, if (and only if)
1702 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1703 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1704 * the request is for a blocking lock. When ->lock() does return asynchronously,
1705 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1706 * request completes.
1707 * If the request is for non-blocking lock the file system should return
1708 * -EINPROGRESS then try to get the lock and call the callback routine with
1709 * the result. If the request timed out the callback routine will return a
1710 * nonzero return code and the file system should release the lock. The file
1711 * system is also responsible to keep a corresponding posix lock when it
1712 * grants a lock so the VFS can find out which locks are locally held and do
1713 * the correct lock cleanup when required.
1714 * The underlying filesystem must not drop the kernel lock or call
1715 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1716 * return code.
1718 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1720 if (filp->f_op && filp->f_op->lock)
1721 return filp->f_op->lock(filp, cmd, fl);
1722 else
1723 return posix_lock_file(filp, fl, conf);
1725 EXPORT_SYMBOL_GPL(vfs_lock_file);
1727 /* Apply the lock described by l to an open file descriptor.
1728 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1730 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1731 struct flock __user *l)
1733 struct file_lock *file_lock = locks_alloc_lock();
1734 struct flock flock;
1735 struct inode *inode;
1736 struct file *f;
1737 int error;
1739 if (file_lock == NULL)
1740 return -ENOLCK;
1743 * This might block, so we do it before checking the inode.
1745 error = -EFAULT;
1746 if (copy_from_user(&flock, l, sizeof(flock)))
1747 goto out;
1749 inode = filp->f_path.dentry->d_inode;
1751 /* Don't allow mandatory locks on files that may be memory mapped
1752 * and shared.
1754 if (IS_MANDLOCK(inode) &&
1755 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1756 mapping_writably_mapped(filp->f_mapping)) {
1757 error = -EAGAIN;
1758 goto out;
1761 again:
1762 error = flock_to_posix_lock(filp, file_lock, &flock);
1763 if (error)
1764 goto out;
1765 if (cmd == F_SETLKW) {
1766 file_lock->fl_flags |= FL_SLEEP;
1769 error = -EBADF;
1770 switch (flock.l_type) {
1771 case F_RDLCK:
1772 if (!(filp->f_mode & FMODE_READ))
1773 goto out;
1774 break;
1775 case F_WRLCK:
1776 if (!(filp->f_mode & FMODE_WRITE))
1777 goto out;
1778 break;
1779 case F_UNLCK:
1780 break;
1781 default:
1782 error = -EINVAL;
1783 goto out;
1786 error = security_file_lock(filp, file_lock->fl_type);
1787 if (error)
1788 goto out;
1790 for (;;) {
1791 error = vfs_lock_file(filp, cmd, file_lock, NULL);
1792 if (error != -EAGAIN || cmd == F_SETLK)
1793 break;
1794 error = wait_event_interruptible(file_lock->fl_wait,
1795 !file_lock->fl_next);
1796 if (!error)
1797 continue;
1799 locks_delete_block(file_lock);
1800 break;
1804 * Attempt to detect a close/fcntl race and recover by
1805 * releasing the lock that was just acquired.
1808 * we need that spin_lock here - it prevents reordering between
1809 * update of inode->i_flock and check for it done in close().
1810 * rcu_read_lock() wouldn't do.
1812 spin_lock(&current->files->file_lock);
1813 f = fcheck(fd);
1814 spin_unlock(&current->files->file_lock);
1815 if (!error && f != filp && flock.l_type != F_UNLCK) {
1816 flock.l_type = F_UNLCK;
1817 goto again;
1820 out:
1821 locks_free_lock(file_lock);
1822 return error;
1825 #if BITS_PER_LONG == 32
1826 /* Report the first existing lock that would conflict with l.
1827 * This implements the F_GETLK command of fcntl().
1829 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1831 struct file_lock file_lock;
1832 struct flock64 flock;
1833 int error;
1835 error = -EFAULT;
1836 if (copy_from_user(&flock, l, sizeof(flock)))
1837 goto out;
1838 error = -EINVAL;
1839 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1840 goto out;
1842 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1843 if (error)
1844 goto out;
1846 error = vfs_test_lock(filp, &file_lock);
1847 if (error)
1848 goto out;
1850 flock.l_type = file_lock.fl_type;
1851 if (file_lock.fl_type != F_UNLCK)
1852 posix_lock_to_flock64(&flock, &file_lock);
1854 error = -EFAULT;
1855 if (!copy_to_user(l, &flock, sizeof(flock)))
1856 error = 0;
1858 out:
1859 return error;
1862 /* Apply the lock described by l to an open file descriptor.
1863 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1865 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1866 struct flock64 __user *l)
1868 struct file_lock *file_lock = locks_alloc_lock();
1869 struct flock64 flock;
1870 struct inode *inode;
1871 struct file *f;
1872 int error;
1874 if (file_lock == NULL)
1875 return -ENOLCK;
1878 * This might block, so we do it before checking the inode.
1880 error = -EFAULT;
1881 if (copy_from_user(&flock, l, sizeof(flock)))
1882 goto out;
1884 inode = filp->f_path.dentry->d_inode;
1886 /* Don't allow mandatory locks on files that may be memory mapped
1887 * and shared.
1889 if (IS_MANDLOCK(inode) &&
1890 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1891 mapping_writably_mapped(filp->f_mapping)) {
1892 error = -EAGAIN;
1893 goto out;
1896 again:
1897 error = flock64_to_posix_lock(filp, file_lock, &flock);
1898 if (error)
1899 goto out;
1900 if (cmd == F_SETLKW64) {
1901 file_lock->fl_flags |= FL_SLEEP;
1904 error = -EBADF;
1905 switch (flock.l_type) {
1906 case F_RDLCK:
1907 if (!(filp->f_mode & FMODE_READ))
1908 goto out;
1909 break;
1910 case F_WRLCK:
1911 if (!(filp->f_mode & FMODE_WRITE))
1912 goto out;
1913 break;
1914 case F_UNLCK:
1915 break;
1916 default:
1917 error = -EINVAL;
1918 goto out;
1921 error = security_file_lock(filp, file_lock->fl_type);
1922 if (error)
1923 goto out;
1925 for (;;) {
1926 error = vfs_lock_file(filp, cmd, file_lock, NULL);
1927 if (error != -EAGAIN || cmd == F_SETLK64)
1928 break;
1929 error = wait_event_interruptible(file_lock->fl_wait,
1930 !file_lock->fl_next);
1931 if (!error)
1932 continue;
1934 locks_delete_block(file_lock);
1935 break;
1939 * Attempt to detect a close/fcntl race and recover by
1940 * releasing the lock that was just acquired.
1942 spin_lock(&current->files->file_lock);
1943 f = fcheck(fd);
1944 spin_unlock(&current->files->file_lock);
1945 if (!error && f != filp && flock.l_type != F_UNLCK) {
1946 flock.l_type = F_UNLCK;
1947 goto again;
1950 out:
1951 locks_free_lock(file_lock);
1952 return error;
1954 #endif /* BITS_PER_LONG == 32 */
1957 * This function is called when the file is being removed
1958 * from the task's fd array. POSIX locks belonging to this task
1959 * are deleted at this time.
1961 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1963 struct file_lock lock;
1966 * If there are no locks held on this file, we don't need to call
1967 * posix_lock_file(). Another process could be setting a lock on this
1968 * file at the same time, but we wouldn't remove that lock anyway.
1970 if (!filp->f_path.dentry->d_inode->i_flock)
1971 return;
1973 lock.fl_type = F_UNLCK;
1974 lock.fl_flags = FL_POSIX | FL_CLOSE;
1975 lock.fl_start = 0;
1976 lock.fl_end = OFFSET_MAX;
1977 lock.fl_owner = owner;
1978 lock.fl_pid = current->tgid;
1979 lock.fl_file = filp;
1980 lock.fl_ops = NULL;
1981 lock.fl_lmops = NULL;
1983 vfs_lock_file(filp, F_SETLK, &lock, NULL);
1985 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1986 lock.fl_ops->fl_release_private(&lock);
1989 EXPORT_SYMBOL(locks_remove_posix);
1992 * This function is called on the last close of an open file.
1994 void locks_remove_flock(struct file *filp)
1996 struct inode * inode = filp->f_path.dentry->d_inode;
1997 struct file_lock *fl;
1998 struct file_lock **before;
2000 if (!inode->i_flock)
2001 return;
2003 if (filp->f_op && filp->f_op->flock) {
2004 struct file_lock fl = {
2005 .fl_pid = current->tgid,
2006 .fl_file = filp,
2007 .fl_flags = FL_FLOCK,
2008 .fl_type = F_UNLCK,
2009 .fl_end = OFFSET_MAX,
2011 filp->f_op->flock(filp, F_SETLKW, &fl);
2012 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2013 fl.fl_ops->fl_release_private(&fl);
2016 lock_kernel();
2017 before = &inode->i_flock;
2019 while ((fl = *before) != NULL) {
2020 if (fl->fl_file == filp) {
2021 if (IS_FLOCK(fl)) {
2022 locks_delete_lock(before);
2023 continue;
2025 if (IS_LEASE(fl)) {
2026 lease_modify(before, F_UNLCK);
2027 continue;
2029 /* What? */
2030 BUG();
2032 before = &fl->fl_next;
2034 unlock_kernel();
2038 * posix_unblock_lock - stop waiting for a file lock
2039 * @filp: how the file was opened
2040 * @waiter: the lock which was waiting
2042 * lockd needs to block waiting for locks.
2045 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2047 int status = 0;
2049 lock_kernel();
2050 if (waiter->fl_next)
2051 __locks_delete_block(waiter);
2052 else
2053 status = -ENOENT;
2054 unlock_kernel();
2055 return status;
2058 EXPORT_SYMBOL(posix_unblock_lock);
2061 * vfs_cancel_lock - file byte range unblock lock
2062 * @filp: The file to apply the unblock to
2063 * @fl: The lock to be unblocked
2065 * Used by lock managers to cancel blocked requests
2067 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2069 if (filp->f_op && filp->f_op->lock)
2070 return filp->f_op->lock(filp, F_CANCELLK, fl);
2071 return 0;
2074 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2076 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2078 struct inode *inode = NULL;
2080 if (fl->fl_file != NULL)
2081 inode = fl->fl_file->f_path.dentry->d_inode;
2083 out += sprintf(out, "%d:%s ", id, pfx);
2084 if (IS_POSIX(fl)) {
2085 out += sprintf(out, "%6s %s ",
2086 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2087 (inode == NULL) ? "*NOINODE*" :
2088 (IS_MANDLOCK(inode) &&
2089 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2090 "MANDATORY" : "ADVISORY ");
2091 } else if (IS_FLOCK(fl)) {
2092 if (fl->fl_type & LOCK_MAND) {
2093 out += sprintf(out, "FLOCK MSNFS ");
2094 } else {
2095 out += sprintf(out, "FLOCK ADVISORY ");
2097 } else if (IS_LEASE(fl)) {
2098 out += sprintf(out, "LEASE ");
2099 if (fl->fl_type & F_INPROGRESS)
2100 out += sprintf(out, "BREAKING ");
2101 else if (fl->fl_file)
2102 out += sprintf(out, "ACTIVE ");
2103 else
2104 out += sprintf(out, "BREAKER ");
2105 } else {
2106 out += sprintf(out, "UNKNOWN UNKNOWN ");
2108 if (fl->fl_type & LOCK_MAND) {
2109 out += sprintf(out, "%s ",
2110 (fl->fl_type & LOCK_READ)
2111 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2112 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2113 } else {
2114 out += sprintf(out, "%s ",
2115 (fl->fl_type & F_INPROGRESS)
2116 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2117 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2119 if (inode) {
2120 #ifdef WE_CAN_BREAK_LSLK_NOW
2121 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2122 inode->i_sb->s_id, inode->i_ino);
2123 #else
2124 /* userspace relies on this representation of dev_t ;-( */
2125 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2126 MAJOR(inode->i_sb->s_dev),
2127 MINOR(inode->i_sb->s_dev), inode->i_ino);
2128 #endif
2129 } else {
2130 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2132 if (IS_POSIX(fl)) {
2133 if (fl->fl_end == OFFSET_MAX)
2134 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2135 else
2136 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2137 fl->fl_end);
2138 } else {
2139 out += sprintf(out, "0 EOF\n");
2143 static void move_lock_status(char **p, off_t* pos, off_t offset)
2145 int len;
2146 len = strlen(*p);
2147 if(*pos >= offset) {
2148 /* the complete line is valid */
2149 *p += len;
2150 *pos += len;
2151 return;
2153 if(*pos+len > offset) {
2154 /* use the second part of the line */
2155 int i = offset-*pos;
2156 memmove(*p,*p+i,len-i);
2157 *p += len-i;
2158 *pos += len;
2159 return;
2161 /* discard the complete line */
2162 *pos += len;
2166 * get_locks_status - reports lock usage in /proc/locks
2167 * @buffer: address in userspace to write into
2168 * @start: ?
2169 * @offset: how far we are through the buffer
2170 * @length: how much to read
2173 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2175 struct list_head *tmp;
2176 char *q = buffer;
2177 off_t pos = 0;
2178 int i = 0;
2180 lock_kernel();
2181 list_for_each(tmp, &file_lock_list) {
2182 struct list_head *btmp;
2183 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2184 lock_get_status(q, fl, ++i, "");
2185 move_lock_status(&q, &pos, offset);
2187 if(pos >= offset+length)
2188 goto done;
2190 list_for_each(btmp, &fl->fl_block) {
2191 struct file_lock *bfl = list_entry(btmp,
2192 struct file_lock, fl_block);
2193 lock_get_status(q, bfl, i, " ->");
2194 move_lock_status(&q, &pos, offset);
2196 if(pos >= offset+length)
2197 goto done;
2200 done:
2201 unlock_kernel();
2202 *start = buffer;
2203 if(q-buffer < length)
2204 return (q-buffer);
2205 return length;
2209 * lock_may_read - checks that the region is free of locks
2210 * @inode: the inode that is being read
2211 * @start: the first byte to read
2212 * @len: the number of bytes to read
2214 * Emulates Windows locking requirements. Whole-file
2215 * mandatory locks (share modes) can prohibit a read and
2216 * byte-range POSIX locks can prohibit a read if they overlap.
2218 * N.B. this function is only ever called
2219 * from knfsd and ownership of locks is never checked.
2221 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2223 struct file_lock *fl;
2224 int result = 1;
2225 lock_kernel();
2226 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2227 if (IS_POSIX(fl)) {
2228 if (fl->fl_type == F_RDLCK)
2229 continue;
2230 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2231 continue;
2232 } else if (IS_FLOCK(fl)) {
2233 if (!(fl->fl_type & LOCK_MAND))
2234 continue;
2235 if (fl->fl_type & LOCK_READ)
2236 continue;
2237 } else
2238 continue;
2239 result = 0;
2240 break;
2242 unlock_kernel();
2243 return result;
2246 EXPORT_SYMBOL(lock_may_read);
2249 * lock_may_write - checks that the region is free of locks
2250 * @inode: the inode that is being written
2251 * @start: the first byte to write
2252 * @len: the number of bytes to write
2254 * Emulates Windows locking requirements. Whole-file
2255 * mandatory locks (share modes) can prohibit a write and
2256 * byte-range POSIX locks can prohibit a write if they overlap.
2258 * N.B. this function is only ever called
2259 * from knfsd and ownership of locks is never checked.
2261 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2263 struct file_lock *fl;
2264 int result = 1;
2265 lock_kernel();
2266 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2267 if (IS_POSIX(fl)) {
2268 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2269 continue;
2270 } else if (IS_FLOCK(fl)) {
2271 if (!(fl->fl_type & LOCK_MAND))
2272 continue;
2273 if (fl->fl_type & LOCK_WRITE)
2274 continue;
2275 } else
2276 continue;
2277 result = 0;
2278 break;
2280 unlock_kernel();
2281 return result;
2284 EXPORT_SYMBOL(lock_may_write);
2286 static int __init filelock_init(void)
2288 filelock_cache = kmem_cache_create("file_lock_cache",
2289 sizeof(struct file_lock), 0, SLAB_PANIC,
2290 init_once, NULL);
2291 return 0;
2294 core_initcall(filelock_init);