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
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
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
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
)
156 if (fl
->fl_ops
->fl_release_private
)
157 fl
->fl_ops
->fl_release_private(fl
);
161 if (fl
->fl_lmops
->fl_release_private
)
162 fl
->fl_lmops
->fl_release_private(fl
);
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
);
185 fl
->fl_fasync
= NULL
;
191 fl
->fl_start
= fl
->fl_end
= 0;
196 EXPORT_SYMBOL(locks_init_lock
);
199 * Initialises the fields of the file lock which are invariant for
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
)
212 if (fl
->fl_ops
->fl_copy_lock
)
213 fl
->fl_ops
->fl_copy_lock(new, fl
);
214 new->fl_ops
= fl
->fl_ops
;
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
;
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
;
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
) {
255 return cmd
& (LOCK_MAND
| LOCK_RW
);
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
,
271 struct file_lock
*fl
;
272 int type
= flock_translate_cmd(cmd
);
276 fl
= locks_alloc_lock();
281 fl
->fl_pid
= current
->tgid
;
282 fl
->fl_flags
= FL_FLOCK
;
284 fl
->fl_end
= OFFSET_MAX
;
290 static int assign_type(struct file_lock
*fl
, int type
)
304 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
307 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
312 switch (l
->l_whence
) {
320 start
= i_size_read(filp
->f_path
.dentry
->d_inode
);
326 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
327 POSIX-2001 defines it. */
331 fl
->fl_end
= OFFSET_MAX
;
333 end
= start
+ l
->l_len
- 1;
335 } else if (l
->l_len
< 0) {
342 fl
->fl_start
= start
; /* we record the absolute position */
343 if (fl
->fl_end
< fl
->fl_start
)
346 fl
->fl_owner
= current
->files
;
347 fl
->fl_pid
= current
->tgid
;
349 fl
->fl_flags
= FL_POSIX
;
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
,
362 switch (l
->l_whence
) {
370 start
= i_size_read(filp
->f_path
.dentry
->d_inode
);
379 fl
->fl_end
= OFFSET_MAX
;
381 fl
->fl_end
= start
+ l
->l_len
- 1;
382 } else if (l
->l_len
< 0) {
383 fl
->fl_end
= start
- 1;
388 fl
->fl_start
= start
; /* we record the absolute position */
389 if (fl
->fl_end
< fl
->fl_start
)
392 fl
->fl_owner
= current
->files
;
393 fl
->fl_pid
= current
->tgid
;
395 fl
->fl_flags
= FL_POSIX
;
403 fl
->fl_type
= l
->l_type
;
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
)
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)
448 fl
->fl_owner
= current
->files
;
449 fl
->fl_pid
= current
->tgid
;
452 fl
->fl_flags
= FL_LEASE
;
454 fl
->fl_end
= OFFSET_MAX
;
456 fl
->fl_lmops
= &lease_manager_ops
;
460 /* Allocate a file_lock initialised to this type of lease */
461 static struct file_lock
*lease_alloc(struct file
*filp
, int type
)
463 struct file_lock
*fl
= locks_alloc_lock();
467 return ERR_PTR(error
);
469 error
= lease_init(filp
, type
, fl
);
472 return ERR_PTR(error
);
477 /* Check if two locks overlap each other.
479 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
481 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
482 (fl2
->fl_end
>= fl1
->fl_start
));
486 * Check whether two locks have the same owner.
488 static int posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
490 if (fl1
->fl_lmops
&& fl1
->fl_lmops
->fl_compare_owner
)
491 return fl2
->fl_lmops
== fl1
->fl_lmops
&&
492 fl1
->fl_lmops
->fl_compare_owner(fl1
, fl2
);
493 return fl1
->fl_owner
== fl2
->fl_owner
;
496 /* Remove waiter from blocker's block list.
497 * When blocker ends up pointing to itself then the list is empty.
499 static void __locks_delete_block(struct file_lock
*waiter
)
501 list_del_init(&waiter
->fl_block
);
502 list_del_init(&waiter
->fl_link
);
503 waiter
->fl_next
= NULL
;
508 static void locks_delete_block(struct file_lock
*waiter
)
511 __locks_delete_block(waiter
);
515 /* Insert waiter into blocker's block list.
516 * We use a circular list so that processes can be easily woken up in
517 * the order they blocked. The documentation doesn't require this but
518 * it seems like the reasonable thing to do.
520 static void locks_insert_block(struct file_lock
*blocker
,
521 struct file_lock
*waiter
)
523 BUG_ON(!list_empty(&waiter
->fl_block
));
524 list_add_tail(&waiter
->fl_block
, &blocker
->fl_block
);
525 waiter
->fl_next
= blocker
;
526 if (IS_POSIX(blocker
))
527 list_add(&waiter
->fl_link
, &blocked_list
);
530 /* Wake up processes blocked waiting for blocker.
531 * If told to wait then schedule the processes until the block list
532 * is empty, otherwise empty the block list ourselves.
534 static void locks_wake_up_blocks(struct file_lock
*blocker
)
536 while (!list_empty(&blocker
->fl_block
)) {
537 struct file_lock
*waiter
;
539 waiter
= list_first_entry(&blocker
->fl_block
,
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
);
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 */
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
;
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
);
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
)
598 if (caller_fl
->fl_type
== F_WRLCK
)
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
611 if (!IS_POSIX(sys_fl
) || posix_same_owner(caller_fl
, sys_fl
))
614 /* Check whether they overlap */
615 if (!locks_overlap(caller_fl
, sys_fl
))
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
629 if (!IS_FLOCK(sys_fl
) || (caller_fl
->fl_file
== sys_fl
->fl_file
))
631 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
634 return (locks_conflict(caller_fl
, sys_fl
));
637 static int interruptible_sleep_on_locked(wait_queue_head_t
*fl_wait
, int timeout
)
640 DECLARE_WAITQUEUE(wait
, current
);
642 __set_current_state(TASK_INTERRUPTIBLE
);
643 add_wait_queue(fl_wait
, &wait
);
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
);
655 static int locks_block_on_timeout(struct file_lock
*blocker
, struct file_lock
*waiter
, int time
)
658 locks_insert_block(blocker
, waiter
);
659 result
= interruptible_sleep_on_locked(&waiter
->fl_wait
, time
);
660 __locks_delete_block(waiter
);
665 posix_test_lock(struct file
*filp
, struct file_lock
*fl
)
667 struct file_lock
*cfl
;
670 for (cfl
= filp
->f_path
.dentry
->d_inode
->i_flock
; cfl
; cfl
= cfl
->fl_next
) {
673 if (posix_locks_conflict(fl
, cfl
))
677 __locks_copy_lock(fl
, cfl
);
679 fl
->fl_type
= F_UNLCK
;
684 EXPORT_SYMBOL(posix_test_lock
);
686 /* This function tests for deadlock condition before putting a process to
687 * sleep. The detection scheme is no longer recursive. Recursive was neat,
688 * but dangerous - we risked stack corruption if the lock data was bad, or
689 * if the recursion was too deep for any other reason.
691 * We rely on the fact that a task can only be on one lock's wait queue
692 * at a time. When we find blocked_task on a wait queue we can re-search
693 * with blocked_task equal to that queue's owner, until either blocked_task
694 * isn't found, or blocked_task is found on a queue owned by my_task.
696 * Note: the above assumption may not be true when handling lock requests
697 * from a broken NFS client. But broken NFS clients have a lot more to
698 * worry about than proper deadlock detection anyway... --okir
700 static int posix_locks_deadlock(struct file_lock
*caller_fl
,
701 struct file_lock
*block_fl
)
703 struct list_head
*tmp
;
706 if (posix_same_owner(caller_fl
, block_fl
))
708 list_for_each(tmp
, &blocked_list
) {
709 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
710 if (posix_same_owner(fl
, block_fl
)) {
719 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
720 * after any leases, but before any posix locks.
722 * Note that if called with an FL_EXISTS argument, the caller may determine
723 * whether or not a lock was successfully freed by testing the return
726 static int flock_lock_file(struct file
*filp
, struct file_lock
*request
)
728 struct file_lock
*new_fl
= NULL
;
729 struct file_lock
**before
;
730 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
735 if (request
->fl_flags
& FL_ACCESS
)
738 if (request
->fl_type
!= F_UNLCK
) {
740 new_fl
= locks_alloc_lock();
746 for_each_lock(inode
, before
) {
747 struct file_lock
*fl
= *before
;
752 if (filp
!= fl
->fl_file
)
754 if (request
->fl_type
== fl
->fl_type
)
757 locks_delete_lock(before
);
761 if (request
->fl_type
== F_UNLCK
) {
762 if ((request
->fl_flags
& FL_EXISTS
) && !found
)
768 * If a higher-priority process was blocked on the old file lock,
769 * give it the opportunity to lock the file.
775 for_each_lock(inode
, before
) {
776 struct file_lock
*fl
= *before
;
781 if (!flock_locks_conflict(request
, fl
))
784 if (request
->fl_flags
& FL_SLEEP
)
785 locks_insert_block(fl
, request
);
788 if (request
->fl_flags
& FL_ACCESS
)
790 locks_copy_lock(new_fl
, request
);
791 locks_insert_lock(before
, new_fl
);
798 locks_free_lock(new_fl
);
802 static int __posix_lock_file(struct inode
*inode
, struct file_lock
*request
, struct file_lock
*conflock
)
804 struct file_lock
*fl
;
805 struct file_lock
*new_fl
= NULL
;
806 struct file_lock
*new_fl2
= NULL
;
807 struct file_lock
*left
= NULL
;
808 struct file_lock
*right
= NULL
;
809 struct file_lock
**before
;
810 int error
, added
= 0;
813 * We may need two file_lock structures for this operation,
814 * so we get them in advance to avoid races.
816 * In some cases we can be sure, that no new locks will be needed
818 if (!(request
->fl_flags
& FL_ACCESS
) &&
819 (request
->fl_type
!= F_UNLCK
||
820 request
->fl_start
!= 0 || request
->fl_end
!= OFFSET_MAX
)) {
821 new_fl
= locks_alloc_lock();
822 new_fl2
= locks_alloc_lock();
826 if (request
->fl_type
!= F_UNLCK
) {
827 for_each_lock(inode
, before
) {
831 if (!posix_locks_conflict(request
, fl
))
834 locks_copy_lock(conflock
, fl
);
836 if (!(request
->fl_flags
& FL_SLEEP
))
839 if (posix_locks_deadlock(request
, fl
))
842 locks_insert_block(fl
, request
);
847 /* If we're just looking for a conflict, we're done. */
849 if (request
->fl_flags
& FL_ACCESS
)
853 * Find the first old lock with the same owner as the new lock.
856 before
= &inode
->i_flock
;
858 /* First skip locks owned by other processes. */
859 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
860 !posix_same_owner(request
, fl
))) {
861 before
= &fl
->fl_next
;
864 /* Process locks with this owner. */
865 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
866 /* Detect adjacent or overlapping regions (if same lock type)
868 if (request
->fl_type
== fl
->fl_type
) {
869 /* In all comparisons of start vs end, use
870 * "start - 1" rather than "end + 1". If end
871 * is OFFSET_MAX, end + 1 will become negative.
873 if (fl
->fl_end
< request
->fl_start
- 1)
875 /* If the next lock in the list has entirely bigger
876 * addresses than the new one, insert the lock here.
878 if (fl
->fl_start
- 1 > request
->fl_end
)
881 /* If we come here, the new and old lock are of the
882 * same type and adjacent or overlapping. Make one
883 * lock yielding from the lower start address of both
884 * locks to the higher end address.
886 if (fl
->fl_start
> request
->fl_start
)
887 fl
->fl_start
= request
->fl_start
;
889 request
->fl_start
= fl
->fl_start
;
890 if (fl
->fl_end
< request
->fl_end
)
891 fl
->fl_end
= request
->fl_end
;
893 request
->fl_end
= fl
->fl_end
;
895 locks_delete_lock(before
);
902 /* Processing for different lock types is a bit
905 if (fl
->fl_end
< request
->fl_start
)
907 if (fl
->fl_start
> request
->fl_end
)
909 if (request
->fl_type
== F_UNLCK
)
911 if (fl
->fl_start
< request
->fl_start
)
913 /* If the next lock in the list has a higher end
914 * address than the new one, insert the new one here.
916 if (fl
->fl_end
> request
->fl_end
) {
920 if (fl
->fl_start
>= request
->fl_start
) {
921 /* The new lock completely replaces an old
922 * one (This may happen several times).
925 locks_delete_lock(before
);
928 /* Replace the old lock with the new one.
929 * Wake up anybody waiting for the old one,
930 * as the change in lock type might satisfy
933 locks_wake_up_blocks(fl
);
934 fl
->fl_start
= request
->fl_start
;
935 fl
->fl_end
= request
->fl_end
;
936 fl
->fl_type
= request
->fl_type
;
937 locks_release_private(fl
);
938 locks_copy_private(fl
, request
);
943 /* Go on to next lock.
946 before
= &fl
->fl_next
;
950 * The above code only modifies existing locks in case of
951 * merging or replacing. If new lock(s) need to be inserted
952 * all modifications are done bellow this, so it's safe yet to
955 error
= -ENOLCK
; /* "no luck" */
956 if (right
&& left
== right
&& !new_fl2
)
961 if (request
->fl_type
== F_UNLCK
) {
962 if (request
->fl_flags
& FL_EXISTS
)
971 locks_copy_lock(new_fl
, request
);
972 locks_insert_lock(before
, new_fl
);
977 /* The new lock breaks the old one in two pieces,
978 * so we have to use the second new lock.
982 locks_copy_lock(left
, right
);
983 locks_insert_lock(before
, left
);
985 right
->fl_start
= request
->fl_end
+ 1;
986 locks_wake_up_blocks(right
);
989 left
->fl_end
= request
->fl_start
- 1;
990 locks_wake_up_blocks(left
);
995 * Free any unused locks.
998 locks_free_lock(new_fl
);
1000 locks_free_lock(new_fl2
);
1005 * posix_lock_file - Apply a POSIX-style lock to a file
1006 * @filp: The file to apply the lock to
1007 * @fl: The lock to be applied
1008 * @conflock: Place to return a copy of the conflicting lock, if found.
1010 * Add a POSIX style lock to a file.
1011 * We merge adjacent & overlapping locks whenever possible.
1012 * POSIX locks are sorted by owner task, then by starting address
1014 * Note that if called with an FL_EXISTS argument, the caller may determine
1015 * whether or not a lock was successfully freed by testing the return
1016 * value for -ENOENT.
1018 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
,
1019 struct file_lock
*conflock
)
1021 return __posix_lock_file(filp
->f_path
.dentry
->d_inode
, fl
, conflock
);
1023 EXPORT_SYMBOL(posix_lock_file
);
1026 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1027 * @filp: The file to apply the lock to
1028 * @fl: The lock to be applied
1030 * Add a POSIX style lock to a file.
1031 * We merge adjacent & overlapping locks whenever possible.
1032 * POSIX locks are sorted by owner task, then by starting address
1034 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1039 error
= posix_lock_file(filp
, fl
, NULL
);
1040 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1042 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1046 locks_delete_block(fl
);
1051 EXPORT_SYMBOL(posix_lock_file_wait
);
1054 * locks_mandatory_locked - Check for an active lock
1055 * @inode: the file to check
1057 * Searches the inode's list of locks to find any POSIX locks which conflict.
1058 * This function is called from locks_verify_locked() only.
1060 int locks_mandatory_locked(struct inode
*inode
)
1062 fl_owner_t owner
= current
->files
;
1063 struct file_lock
*fl
;
1066 * Search the lock list for this inode for any POSIX locks.
1069 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1072 if (fl
->fl_owner
!= owner
)
1076 return fl
? -EAGAIN
: 0;
1080 * locks_mandatory_area - Check for a conflicting lock
1081 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1083 * @inode: the file to check
1084 * @filp: how the file was opened (if it was)
1085 * @offset: start of area to check
1086 * @count: length of area to check
1088 * Searches the inode's list of locks to find any POSIX locks which conflict.
1089 * This function is called from rw_verify_area() and
1090 * locks_verify_truncate().
1092 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1093 struct file
*filp
, loff_t offset
,
1096 struct file_lock fl
;
1099 locks_init_lock(&fl
);
1100 fl
.fl_owner
= current
->files
;
1101 fl
.fl_pid
= current
->tgid
;
1103 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1104 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1105 fl
.fl_flags
|= FL_SLEEP
;
1106 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1107 fl
.fl_start
= offset
;
1108 fl
.fl_end
= offset
+ count
- 1;
1111 error
= __posix_lock_file(inode
, &fl
, NULL
);
1112 if (error
!= -EAGAIN
)
1114 if (!(fl
.fl_flags
& FL_SLEEP
))
1116 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1119 * If we've been sleeping someone might have
1120 * changed the permissions behind our back.
1122 if (__mandatory_lock(inode
))
1126 locks_delete_block(&fl
);
1133 EXPORT_SYMBOL(locks_mandatory_area
);
1135 /* We already had a lease on this file; just change its type */
1136 int lease_modify(struct file_lock
**before
, int arg
)
1138 struct file_lock
*fl
= *before
;
1139 int error
= assign_type(fl
, arg
);
1143 locks_wake_up_blocks(fl
);
1145 locks_delete_lock(before
);
1149 EXPORT_SYMBOL(lease_modify
);
1151 static void time_out_leases(struct inode
*inode
)
1153 struct file_lock
**before
;
1154 struct file_lock
*fl
;
1156 before
= &inode
->i_flock
;
1157 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1158 if ((fl
->fl_break_time
== 0)
1159 || time_before(jiffies
, fl
->fl_break_time
)) {
1160 before
= &fl
->fl_next
;
1163 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1164 if (fl
== *before
) /* lease_modify may have freed fl */
1165 before
= &fl
->fl_next
;
1170 * __break_lease - revoke all outstanding leases on file
1171 * @inode: the inode of the file to return
1172 * @mode: the open mode (read or write)
1174 * break_lease (inlined for speed) has checked there already is at least
1175 * some kind of lock (maybe a lease) on this file. Leases are broken on
1176 * a call to open() or truncate(). This function can sleep unless you
1177 * specified %O_NONBLOCK to your open().
1179 int __break_lease(struct inode
*inode
, unsigned int mode
)
1181 int error
= 0, future
;
1182 struct file_lock
*new_fl
, *flock
;
1183 struct file_lock
*fl
;
1184 unsigned long break_time
;
1185 int i_have_this_lease
= 0;
1187 new_fl
= lease_alloc(NULL
, mode
& FMODE_WRITE
? F_WRLCK
: F_RDLCK
);
1191 time_out_leases(inode
);
1193 flock
= inode
->i_flock
;
1194 if ((flock
== NULL
) || !IS_LEASE(flock
))
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
;
1211 /* the existing lease was read-only, so we can read too. */
1215 if (IS_ERR(new_fl
) && !i_have_this_lease
1216 && ((mode
& O_NONBLOCK
) == 0)) {
1217 error
= PTR_ERR(new_fl
);
1222 if (lease_break_time
> 0) {
1223 break_time
= jiffies
+ lease_break_time
* HZ
;
1224 if (break_time
== 0)
1225 break_time
++; /* so that 0 means no break time */
1228 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1229 if (fl
->fl_type
!= future
) {
1230 fl
->fl_type
= future
;
1231 fl
->fl_break_time
= break_time
;
1232 /* lease must have lmops break callback */
1233 fl
->fl_lmops
->fl_break(fl
);
1237 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1238 error
= -EWOULDBLOCK
;
1243 break_time
= flock
->fl_break_time
;
1244 if (break_time
!= 0) {
1245 break_time
-= jiffies
;
1246 if (break_time
== 0)
1249 error
= locks_block_on_timeout(flock
, new_fl
, break_time
);
1252 time_out_leases(inode
);
1253 /* Wait for the next lease that has not been broken yet */
1254 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1255 flock
= flock
->fl_next
) {
1256 if (flock
->fl_type
& F_INPROGRESS
)
1264 if (!IS_ERR(new_fl
))
1265 locks_free_lock(new_fl
);
1269 EXPORT_SYMBOL(__break_lease
);
1274 * @time: pointer to a timespec which will contain the last modified time
1276 * This is to force NFS clients to flush their caches for files with
1277 * exclusive leases. The justification is that if someone has an
1278 * exclusive lease, then they could be modifiying it.
1280 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1282 struct file_lock
*flock
= inode
->i_flock
;
1283 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1284 *time
= current_fs_time(inode
->i_sb
);
1286 *time
= inode
->i_mtime
;
1289 EXPORT_SYMBOL(lease_get_mtime
);
1292 * fcntl_getlease - Enquire what lease is currently active
1295 * The value returned by this function will be one of
1296 * (if no lease break is pending):
1298 * %F_RDLCK to indicate a shared lease is held.
1300 * %F_WRLCK to indicate an exclusive lease is held.
1302 * %F_UNLCK to indicate no lease is held.
1304 * (if a lease break is pending):
1306 * %F_RDLCK to indicate an exclusive lease needs to be
1307 * changed to a shared lease (or removed).
1309 * %F_UNLCK to indicate the lease needs to be removed.
1311 * XXX: sfr & willy disagree over whether F_INPROGRESS
1312 * should be returned to userspace.
1314 int fcntl_getlease(struct file
*filp
)
1316 struct file_lock
*fl
;
1320 time_out_leases(filp
->f_path
.dentry
->d_inode
);
1321 for (fl
= filp
->f_path
.dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1323 if (fl
->fl_file
== filp
) {
1324 type
= fl
->fl_type
& ~F_INPROGRESS
;
1333 * generic_setlease - sets a lease on an open file
1334 * @filp: file pointer
1335 * @arg: type of lease to obtain
1336 * @flp: input - file_lock to use, output - file_lock inserted
1338 * The (input) flp->fl_lmops->fl_break function is required
1341 * Called with kernel lock held.
1343 int generic_setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1345 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1346 struct file_lock
*new_fl
= NULL
;
1347 struct dentry
*dentry
= filp
->f_path
.dentry
;
1348 struct inode
*inode
= dentry
->d_inode
;
1349 int error
, rdlease_count
= 0, wrlease_count
= 0;
1351 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1353 if (!S_ISREG(inode
->i_mode
))
1355 error
= security_file_lock(filp
, arg
);
1359 time_out_leases(inode
);
1361 BUG_ON(!(*flp
)->fl_lmops
->fl_break
);
1366 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1368 if ((arg
== F_WRLCK
)
1369 && ((atomic_read(&dentry
->d_count
) > 1)
1370 || (atomic_read(&inode
->i_count
) > 1)))
1374 new_fl
= locks_alloc_lock();
1379 * At this point, we know that if there is an exclusive
1380 * lease on this file, then we hold it on this filp
1381 * (otherwise our open of this file would have blocked).
1382 * And if we are trying to acquire an exclusive lease,
1383 * then the file is not open by anyone (including us)
1384 * except for this filp.
1386 for (before
= &inode
->i_flock
;
1387 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1388 before
= &fl
->fl_next
) {
1389 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1391 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1393 * Someone is in the process of opening this
1394 * file for writing so we may not take an
1395 * exclusive lease on it.
1402 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1403 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1406 if (my_before
!= NULL
) {
1408 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1420 locks_copy_lock(new_fl
, lease
);
1421 locks_insert_lock(before
, new_fl
);
1428 locks_free_lock(new_fl
);
1431 EXPORT_SYMBOL(generic_setlease
);
1434 * vfs_setlease - sets a lease on an open file
1435 * @filp: file pointer
1436 * @arg: type of lease to obtain
1437 * @lease: file_lock to use
1439 * Call this to establish a lease on the file.
1440 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1441 * break_lease will oops!
1443 * This will call the filesystem's setlease file method, if
1444 * defined. Note that there is no getlease method; instead, the
1445 * filesystem setlease method should call back to setlease() to
1446 * add a lease to the inode's lease list, where fcntl_getlease() can
1447 * find it. Since fcntl_getlease() only reports whether the current
1448 * task holds a lease, a cluster filesystem need only do this for
1449 * leases held by processes on this node.
1451 * There is also no break_lease method; filesystems that
1452 * handle their own leases shoud break leases themselves from the
1453 * filesystem's open, create, and (on truncate) setattr methods.
1455 * Warning: the only current setlease methods exist only to disable
1456 * leases in certain cases. More vfs changes may be required to
1457 * allow a full filesystem lease implementation.
1460 int vfs_setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1465 if (filp
->f_op
&& filp
->f_op
->setlease
)
1466 error
= filp
->f_op
->setlease(filp
, arg
, lease
);
1468 error
= generic_setlease(filp
, arg
, lease
);
1473 EXPORT_SYMBOL_GPL(vfs_setlease
);
1476 * fcntl_setlease - sets a lease on an open file
1477 * @fd: open file descriptor
1478 * @filp: file pointer
1479 * @arg: type of lease to obtain
1481 * Call this fcntl to establish a lease on the file.
1482 * Note that you also need to call %F_SETSIG to
1483 * receive a signal when the lease is broken.
1485 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1487 struct file_lock fl
, *flp
= &fl
;
1488 struct dentry
*dentry
= filp
->f_path
.dentry
;
1489 struct inode
*inode
= dentry
->d_inode
;
1492 locks_init_lock(&fl
);
1493 error
= lease_init(filp
, arg
, &fl
);
1499 error
= vfs_setlease(filp
, arg
, &flp
);
1500 if (error
|| arg
== F_UNLCK
)
1503 error
= fasync_helper(fd
, filp
, 1, &flp
->fl_fasync
);
1505 /* remove lease just inserted by setlease */
1506 flp
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1507 flp
->fl_break_time
= jiffies
- 10;
1508 time_out_leases(inode
);
1512 error
= __f_setown(filp
, task_pid(current
), PIDTYPE_PID
, 0);
1519 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1520 * @filp: The file to apply the lock to
1521 * @fl: The lock to be applied
1523 * Add a FLOCK style lock to a file.
1525 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1530 error
= flock_lock_file(filp
, fl
);
1531 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1533 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1537 locks_delete_block(fl
);
1543 EXPORT_SYMBOL(flock_lock_file_wait
);
1546 * sys_flock: - flock() system call.
1547 * @fd: the file descriptor to lock.
1548 * @cmd: the type of lock to apply.
1550 * Apply a %FL_FLOCK style lock to an open file descriptor.
1551 * The @cmd can be one of
1553 * %LOCK_SH -- a shared lock.
1555 * %LOCK_EX -- an exclusive lock.
1557 * %LOCK_UN -- remove an existing lock.
1559 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1561 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1562 * processes read and write access respectively.
1564 asmlinkage
long sys_flock(unsigned int fd
, unsigned int cmd
)
1567 struct file_lock
*lock
;
1568 int can_sleep
, unlock
;
1576 can_sleep
= !(cmd
& LOCK_NB
);
1578 unlock
= (cmd
== LOCK_UN
);
1580 if (!unlock
&& !(cmd
& LOCK_MAND
) && !(filp
->f_mode
& 3))
1583 error
= flock_make_lock(filp
, &lock
, cmd
);
1587 lock
->fl_flags
|= FL_SLEEP
;
1589 error
= security_file_lock(filp
, cmd
);
1593 if (filp
->f_op
&& filp
->f_op
->flock
)
1594 error
= filp
->f_op
->flock(filp
,
1595 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1598 error
= flock_lock_file_wait(filp
, lock
);
1601 locks_free_lock(lock
);
1610 * vfs_test_lock - test file byte range lock
1611 * @filp: The file to test lock for
1612 * @fl: The lock to test; also used to hold result
1614 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1615 * setting conf->fl_type to something other than F_UNLCK.
1617 int vfs_test_lock(struct file
*filp
, struct file_lock
*fl
)
1619 if (filp
->f_op
&& filp
->f_op
->lock
)
1620 return filp
->f_op
->lock(filp
, F_GETLK
, fl
);
1621 posix_test_lock(filp
, fl
);
1624 EXPORT_SYMBOL_GPL(vfs_test_lock
);
1626 static int posix_lock_to_flock(struct flock
*flock
, struct file_lock
*fl
)
1628 flock
->l_pid
= fl
->fl_pid
;
1629 #if BITS_PER_LONG == 32
1631 * Make sure we can represent the posix lock via
1632 * legacy 32bit flock.
1634 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1636 if (fl
->fl_end
!= OFFSET_MAX
&& fl
->fl_end
> OFFT_OFFSET_MAX
)
1639 flock
->l_start
= fl
->fl_start
;
1640 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1641 fl
->fl_end
- fl
->fl_start
+ 1;
1642 flock
->l_whence
= 0;
1643 flock
->l_type
= fl
->fl_type
;
1647 #if BITS_PER_LONG == 32
1648 static void posix_lock_to_flock64(struct flock64
*flock
, struct file_lock
*fl
)
1650 flock
->l_pid
= fl
->fl_pid
;
1651 flock
->l_start
= fl
->fl_start
;
1652 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1653 fl
->fl_end
- fl
->fl_start
+ 1;
1654 flock
->l_whence
= 0;
1655 flock
->l_type
= fl
->fl_type
;
1659 /* Report the first existing lock that would conflict with l.
1660 * This implements the F_GETLK command of fcntl().
1662 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1664 struct file_lock file_lock
;
1669 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1672 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1675 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1679 error
= vfs_test_lock(filp
, &file_lock
);
1683 flock
.l_type
= file_lock
.fl_type
;
1684 if (file_lock
.fl_type
!= F_UNLCK
) {
1685 error
= posix_lock_to_flock(&flock
, &file_lock
);
1690 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1697 * vfs_lock_file - file byte range lock
1698 * @filp: The file to apply the lock to
1699 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1700 * @fl: The lock to be applied
1701 * @conf: Place to return a copy of the conflicting lock, if found.
1703 * A caller that doesn't care about the conflicting lock may pass NULL
1704 * as the final argument.
1706 * If the filesystem defines a private ->lock() method, then @conf will
1707 * be left unchanged; so a caller that cares should initialize it to
1708 * some acceptable default.
1710 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1711 * locks, the ->lock() interface may return asynchronously, before the lock has
1712 * been granted or denied by the underlying filesystem, if (and only if)
1713 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1714 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1715 * the request is for a blocking lock. When ->lock() does return asynchronously,
1716 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1717 * request completes.
1718 * If the request is for non-blocking lock the file system should return
1719 * -EINPROGRESS then try to get the lock and call the callback routine with
1720 * the result. If the request timed out the callback routine will return a
1721 * nonzero return code and the file system should release the lock. The file
1722 * system is also responsible to keep a corresponding posix lock when it
1723 * grants a lock so the VFS can find out which locks are locally held and do
1724 * the correct lock cleanup when required.
1725 * The underlying filesystem must not drop the kernel lock or call
1726 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1729 int vfs_lock_file(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
, struct file_lock
*conf
)
1731 if (filp
->f_op
&& filp
->f_op
->lock
)
1732 return filp
->f_op
->lock(filp
, cmd
, fl
);
1734 return posix_lock_file(filp
, fl
, conf
);
1736 EXPORT_SYMBOL_GPL(vfs_lock_file
);
1738 /* Apply the lock described by l to an open file descriptor.
1739 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1741 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1742 struct flock __user
*l
)
1744 struct file_lock
*file_lock
= locks_alloc_lock();
1746 struct inode
*inode
;
1749 if (file_lock
== NULL
)
1753 * This might block, so we do it before checking the inode.
1756 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1759 inode
= filp
->f_path
.dentry
->d_inode
;
1761 /* Don't allow mandatory locks on files that may be memory mapped
1764 if (mandatory_lock(inode
) && mapping_writably_mapped(filp
->f_mapping
)) {
1770 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1773 if (cmd
== F_SETLKW
) {
1774 file_lock
->fl_flags
|= FL_SLEEP
;
1778 switch (flock
.l_type
) {
1780 if (!(filp
->f_mode
& FMODE_READ
))
1784 if (!(filp
->f_mode
& FMODE_WRITE
))
1794 error
= security_file_lock(filp
, file_lock
->fl_type
);
1799 error
= vfs_lock_file(filp
, cmd
, file_lock
, NULL
);
1800 if (error
!= -EAGAIN
|| cmd
== F_SETLK
)
1802 error
= wait_event_interruptible(file_lock
->fl_wait
,
1803 !file_lock
->fl_next
);
1807 locks_delete_block(file_lock
);
1812 * Attempt to detect a close/fcntl race and recover by
1813 * releasing the lock that was just acquired.
1815 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1816 flock
.l_type
= F_UNLCK
;
1821 locks_free_lock(file_lock
);
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
;
1836 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1839 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1842 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1846 error
= vfs_test_lock(filp
, &file_lock
);
1850 flock
.l_type
= file_lock
.fl_type
;
1851 if (file_lock
.fl_type
!= F_UNLCK
)
1852 posix_lock_to_flock64(&flock
, &file_lock
);
1855 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
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
;
1873 if (file_lock
== NULL
)
1877 * This might block, so we do it before checking the inode.
1880 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1883 inode
= filp
->f_path
.dentry
->d_inode
;
1885 /* Don't allow mandatory locks on files that may be memory mapped
1888 if (mandatory_lock(inode
) && mapping_writably_mapped(filp
->f_mapping
)) {
1894 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1897 if (cmd
== F_SETLKW64
) {
1898 file_lock
->fl_flags
|= FL_SLEEP
;
1902 switch (flock
.l_type
) {
1904 if (!(filp
->f_mode
& FMODE_READ
))
1908 if (!(filp
->f_mode
& FMODE_WRITE
))
1918 error
= security_file_lock(filp
, file_lock
->fl_type
);
1923 error
= vfs_lock_file(filp
, cmd
, file_lock
, NULL
);
1924 if (error
!= -EAGAIN
|| cmd
== F_SETLK64
)
1926 error
= wait_event_interruptible(file_lock
->fl_wait
,
1927 !file_lock
->fl_next
);
1931 locks_delete_block(file_lock
);
1936 * Attempt to detect a close/fcntl race and recover by
1937 * releasing the lock that was just acquired.
1939 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1940 flock
.l_type
= F_UNLCK
;
1945 locks_free_lock(file_lock
);
1948 #endif /* BITS_PER_LONG == 32 */
1951 * This function is called when the file is being removed
1952 * from the task's fd array. POSIX locks belonging to this task
1953 * are deleted at this time.
1955 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
1957 struct file_lock lock
;
1960 * If there are no locks held on this file, we don't need to call
1961 * posix_lock_file(). Another process could be setting a lock on this
1962 * file at the same time, but we wouldn't remove that lock anyway.
1964 if (!filp
->f_path
.dentry
->d_inode
->i_flock
)
1967 lock
.fl_type
= F_UNLCK
;
1968 lock
.fl_flags
= FL_POSIX
| FL_CLOSE
;
1970 lock
.fl_end
= OFFSET_MAX
;
1971 lock
.fl_owner
= owner
;
1972 lock
.fl_pid
= current
->tgid
;
1973 lock
.fl_file
= filp
;
1975 lock
.fl_lmops
= NULL
;
1977 vfs_lock_file(filp
, F_SETLK
, &lock
, NULL
);
1979 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
1980 lock
.fl_ops
->fl_release_private(&lock
);
1983 EXPORT_SYMBOL(locks_remove_posix
);
1986 * This function is called on the last close of an open file.
1988 void locks_remove_flock(struct file
*filp
)
1990 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
1991 struct file_lock
*fl
;
1992 struct file_lock
**before
;
1994 if (!inode
->i_flock
)
1997 if (filp
->f_op
&& filp
->f_op
->flock
) {
1998 struct file_lock fl
= {
1999 .fl_pid
= current
->tgid
,
2001 .fl_flags
= FL_FLOCK
,
2003 .fl_end
= OFFSET_MAX
,
2005 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
2006 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
2007 fl
.fl_ops
->fl_release_private(&fl
);
2011 before
= &inode
->i_flock
;
2013 while ((fl
= *before
) != NULL
) {
2014 if (fl
->fl_file
== filp
) {
2016 locks_delete_lock(before
);
2020 lease_modify(before
, F_UNLCK
);
2026 before
= &fl
->fl_next
;
2032 * posix_unblock_lock - stop waiting for a file lock
2033 * @filp: how the file was opened
2034 * @waiter: the lock which was waiting
2036 * lockd needs to block waiting for locks.
2039 posix_unblock_lock(struct file
*filp
, struct file_lock
*waiter
)
2044 if (waiter
->fl_next
)
2045 __locks_delete_block(waiter
);
2052 EXPORT_SYMBOL(posix_unblock_lock
);
2055 * vfs_cancel_lock - file byte range unblock lock
2056 * @filp: The file to apply the unblock to
2057 * @fl: The lock to be unblocked
2059 * Used by lock managers to cancel blocked requests
2061 int vfs_cancel_lock(struct file
*filp
, struct file_lock
*fl
)
2063 if (filp
->f_op
&& filp
->f_op
->lock
)
2064 return filp
->f_op
->lock(filp
, F_CANCELLK
, fl
);
2068 EXPORT_SYMBOL_GPL(vfs_cancel_lock
);
2070 static void lock_get_status(char* out
, struct file_lock
*fl
, int id
, char *pfx
)
2072 struct inode
*inode
= NULL
;
2074 if (fl
->fl_file
!= NULL
)
2075 inode
= fl
->fl_file
->f_path
.dentry
->d_inode
;
2077 out
+= sprintf(out
, "%d:%s ", id
, pfx
);
2079 out
+= sprintf(out
, "%6s %s ",
2080 (fl
->fl_flags
& FL_ACCESS
) ? "ACCESS" : "POSIX ",
2081 (inode
== NULL
) ? "*NOINODE*" :
2082 mandatory_lock(inode
) ? "MANDATORY" : "ADVISORY ");
2083 } else if (IS_FLOCK(fl
)) {
2084 if (fl
->fl_type
& LOCK_MAND
) {
2085 out
+= sprintf(out
, "FLOCK MSNFS ");
2087 out
+= sprintf(out
, "FLOCK ADVISORY ");
2089 } else if (IS_LEASE(fl
)) {
2090 out
+= sprintf(out
, "LEASE ");
2091 if (fl
->fl_type
& F_INPROGRESS
)
2092 out
+= sprintf(out
, "BREAKING ");
2093 else if (fl
->fl_file
)
2094 out
+= sprintf(out
, "ACTIVE ");
2096 out
+= sprintf(out
, "BREAKER ");
2098 out
+= sprintf(out
, "UNKNOWN UNKNOWN ");
2100 if (fl
->fl_type
& LOCK_MAND
) {
2101 out
+= sprintf(out
, "%s ",
2102 (fl
->fl_type
& LOCK_READ
)
2103 ? (fl
->fl_type
& LOCK_WRITE
) ? "RW " : "READ "
2104 : (fl
->fl_type
& LOCK_WRITE
) ? "WRITE" : "NONE ");
2106 out
+= sprintf(out
, "%s ",
2107 (fl
->fl_type
& F_INPROGRESS
)
2108 ? (fl
->fl_type
& F_UNLCK
) ? "UNLCK" : "READ "
2109 : (fl
->fl_type
& F_WRLCK
) ? "WRITE" : "READ ");
2112 #ifdef WE_CAN_BREAK_LSLK_NOW
2113 out
+= sprintf(out
, "%d %s:%ld ", fl
->fl_pid
,
2114 inode
->i_sb
->s_id
, inode
->i_ino
);
2116 /* userspace relies on this representation of dev_t ;-( */
2117 out
+= sprintf(out
, "%d %02x:%02x:%ld ", fl
->fl_pid
,
2118 MAJOR(inode
->i_sb
->s_dev
),
2119 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2122 out
+= sprintf(out
, "%d <none>:0 ", fl
->fl_pid
);
2125 if (fl
->fl_end
== OFFSET_MAX
)
2126 out
+= sprintf(out
, "%Ld EOF\n", fl
->fl_start
);
2128 out
+= sprintf(out
, "%Ld %Ld\n", fl
->fl_start
,
2131 out
+= sprintf(out
, "0 EOF\n");
2135 static void move_lock_status(char **p
, off_t
* pos
, off_t offset
)
2139 if(*pos
>= offset
) {
2140 /* the complete line is valid */
2145 if(*pos
+len
> offset
) {
2146 /* use the second part of the line */
2147 int i
= offset
-*pos
;
2148 memmove(*p
,*p
+i
,len
-i
);
2153 /* discard the complete line */
2158 * get_locks_status - reports lock usage in /proc/locks
2159 * @buffer: address in userspace to write into
2161 * @offset: how far we are through the buffer
2162 * @length: how much to read
2165 int get_locks_status(char *buffer
, char **start
, off_t offset
, int length
)
2167 struct list_head
*tmp
;
2173 list_for_each(tmp
, &file_lock_list
) {
2174 struct list_head
*btmp
;
2175 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
2176 lock_get_status(q
, fl
, ++i
, "");
2177 move_lock_status(&q
, &pos
, offset
);
2179 if(pos
>= offset
+length
)
2182 list_for_each(btmp
, &fl
->fl_block
) {
2183 struct file_lock
*bfl
= list_entry(btmp
,
2184 struct file_lock
, fl_block
);
2185 lock_get_status(q
, bfl
, i
, " ->");
2186 move_lock_status(&q
, &pos
, offset
);
2188 if(pos
>= offset
+length
)
2195 if(q
-buffer
< length
)
2201 * lock_may_read - checks that the region is free of locks
2202 * @inode: the inode that is being read
2203 * @start: the first byte to read
2204 * @len: the number of bytes to read
2206 * Emulates Windows locking requirements. Whole-file
2207 * mandatory locks (share modes) can prohibit a read and
2208 * byte-range POSIX locks can prohibit a read if they overlap.
2210 * N.B. this function is only ever called
2211 * from knfsd and ownership of locks is never checked.
2213 int lock_may_read(struct inode
*inode
, loff_t start
, unsigned long len
)
2215 struct file_lock
*fl
;
2218 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2220 if (fl
->fl_type
== F_RDLCK
)
2222 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2224 } else if (IS_FLOCK(fl
)) {
2225 if (!(fl
->fl_type
& LOCK_MAND
))
2227 if (fl
->fl_type
& LOCK_READ
)
2238 EXPORT_SYMBOL(lock_may_read
);
2241 * lock_may_write - checks that the region is free of locks
2242 * @inode: the inode that is being written
2243 * @start: the first byte to write
2244 * @len: the number of bytes to write
2246 * Emulates Windows locking requirements. Whole-file
2247 * mandatory locks (share modes) can prohibit a write and
2248 * byte-range POSIX locks can prohibit a write if they overlap.
2250 * N.B. this function is only ever called
2251 * from knfsd and ownership of locks is never checked.
2253 int lock_may_write(struct inode
*inode
, loff_t start
, unsigned long len
)
2255 struct file_lock
*fl
;
2258 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2260 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2262 } else if (IS_FLOCK(fl
)) {
2263 if (!(fl
->fl_type
& LOCK_MAND
))
2265 if (fl
->fl_type
& LOCK_WRITE
)
2276 EXPORT_SYMBOL(lock_may_write
);
2278 static int __init
filelock_init(void)
2280 filelock_cache
= kmem_cache_create("file_lock_cache",
2281 sizeof(struct file_lock
), 0, SLAB_PANIC
,
2286 core_initcall(filelock_init
);