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
= list_entry(blocker
->fl_block
.next
,
538 struct file_lock
, fl_block
);
539 __locks_delete_block(waiter
);
540 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->fl_notify
)
541 waiter
->fl_lmops
->fl_notify(waiter
);
543 wake_up(&waiter
->fl_wait
);
547 /* Insert file lock fl into an inode's lock list at the position indicated
548 * by pos. At the same time add the lock to the global file lock list.
550 static void locks_insert_lock(struct file_lock
**pos
, struct file_lock
*fl
)
552 list_add(&fl
->fl_link
, &file_lock_list
);
554 /* insert into file's list */
558 if (fl
->fl_ops
&& fl
->fl_ops
->fl_insert
)
559 fl
->fl_ops
->fl_insert(fl
);
563 * Delete a lock and then free it.
564 * Wake up processes that are blocked waiting for this lock,
565 * notify the FS that the lock has been cleared and
566 * finally free the lock.
568 static void locks_delete_lock(struct file_lock
**thisfl_p
)
570 struct file_lock
*fl
= *thisfl_p
;
572 *thisfl_p
= fl
->fl_next
;
574 list_del_init(&fl
->fl_link
);
576 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
577 if (fl
->fl_fasync
!= NULL
) {
578 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
579 fl
->fl_fasync
= NULL
;
582 if (fl
->fl_ops
&& fl
->fl_ops
->fl_remove
)
583 fl
->fl_ops
->fl_remove(fl
);
585 locks_wake_up_blocks(fl
);
589 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
590 * checks for shared/exclusive status of overlapping locks.
592 static int locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
594 if (sys_fl
->fl_type
== F_WRLCK
)
596 if (caller_fl
->fl_type
== F_WRLCK
)
601 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
602 * checking before calling the locks_conflict().
604 static int posix_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
606 /* POSIX locks owned by the same process do not conflict with
609 if (!IS_POSIX(sys_fl
) || posix_same_owner(caller_fl
, sys_fl
))
612 /* Check whether they overlap */
613 if (!locks_overlap(caller_fl
, sys_fl
))
616 return (locks_conflict(caller_fl
, sys_fl
));
619 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
620 * checking before calling the locks_conflict().
622 static int flock_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
624 /* FLOCK locks referring to the same filp do not conflict with
627 if (!IS_FLOCK(sys_fl
) || (caller_fl
->fl_file
== sys_fl
->fl_file
))
629 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
632 return (locks_conflict(caller_fl
, sys_fl
));
635 static int interruptible_sleep_on_locked(wait_queue_head_t
*fl_wait
, int timeout
)
638 DECLARE_WAITQUEUE(wait
, current
);
640 __set_current_state(TASK_INTERRUPTIBLE
);
641 add_wait_queue(fl_wait
, &wait
);
645 result
= schedule_timeout(timeout
);
646 if (signal_pending(current
))
647 result
= -ERESTARTSYS
;
648 remove_wait_queue(fl_wait
, &wait
);
649 __set_current_state(TASK_RUNNING
);
653 static int locks_block_on_timeout(struct file_lock
*blocker
, struct file_lock
*waiter
, int time
)
656 locks_insert_block(blocker
, waiter
);
657 result
= interruptible_sleep_on_locked(&waiter
->fl_wait
, time
);
658 __locks_delete_block(waiter
);
663 posix_test_lock(struct file
*filp
, struct file_lock
*fl
)
665 struct file_lock
*cfl
;
668 for (cfl
= filp
->f_path
.dentry
->d_inode
->i_flock
; cfl
; cfl
= cfl
->fl_next
) {
671 if (posix_locks_conflict(cfl
, fl
))
675 __locks_copy_lock(fl
, cfl
);
677 fl
->fl_type
= F_UNLCK
;
682 EXPORT_SYMBOL(posix_test_lock
);
684 /* This function tests for deadlock condition before putting a process to
685 * sleep. The detection scheme is no longer recursive. Recursive was neat,
686 * but dangerous - we risked stack corruption if the lock data was bad, or
687 * if the recursion was too deep for any other reason.
689 * We rely on the fact that a task can only be on one lock's wait queue
690 * at a time. When we find blocked_task on a wait queue we can re-search
691 * with blocked_task equal to that queue's owner, until either blocked_task
692 * isn't found, or blocked_task is found on a queue owned by my_task.
694 * Note: the above assumption may not be true when handling lock requests
695 * from a broken NFS client. But broken NFS clients have a lot more to
696 * worry about than proper deadlock detection anyway... --okir
698 * However, the failure of this assumption (also possible in the case of
699 * multiple tasks sharing the same open file table) also means there's no
700 * guarantee that the loop below will terminate. As a hack, we give up
701 * after a few iterations.
704 #define MAX_DEADLK_ITERATIONS 10
706 static int posix_locks_deadlock(struct file_lock
*caller_fl
,
707 struct file_lock
*block_fl
)
709 struct list_head
*tmp
;
713 if (posix_same_owner(caller_fl
, block_fl
))
715 list_for_each(tmp
, &blocked_list
) {
716 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
717 if (posix_same_owner(fl
, block_fl
)) {
718 if (i
++ > MAX_DEADLK_ITERATIONS
)
728 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
729 * at the head of the list, but that's secret knowledge known only to
730 * flock_lock_file and posix_lock_file.
732 * Note that if called with an FL_EXISTS argument, the caller may determine
733 * whether or not a lock was successfully freed by testing the return
736 static int flock_lock_file(struct file
*filp
, struct file_lock
*request
)
738 struct file_lock
*new_fl
= NULL
;
739 struct file_lock
**before
;
740 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
745 if (request
->fl_flags
& FL_ACCESS
)
747 for_each_lock(inode
, before
) {
748 struct file_lock
*fl
= *before
;
753 if (filp
!= fl
->fl_file
)
755 if (request
->fl_type
== fl
->fl_type
)
758 locks_delete_lock(before
);
762 if (request
->fl_type
== F_UNLCK
) {
763 if ((request
->fl_flags
& FL_EXISTS
) && !found
)
769 new_fl
= locks_alloc_lock();
773 * If a higher-priority process was blocked on the old file lock,
774 * give it the opportunity to lock the file.
780 for_each_lock(inode
, before
) {
781 struct file_lock
*fl
= *before
;
786 if (!flock_locks_conflict(request
, fl
))
789 if (request
->fl_flags
& FL_SLEEP
)
790 locks_insert_block(fl
, request
);
793 if (request
->fl_flags
& FL_ACCESS
)
795 locks_copy_lock(new_fl
, request
);
796 locks_insert_lock(before
, new_fl
);
803 locks_free_lock(new_fl
);
807 static int __posix_lock_file(struct inode
*inode
, struct file_lock
*request
, struct file_lock
*conflock
)
809 struct file_lock
*fl
;
810 struct file_lock
*new_fl
= NULL
;
811 struct file_lock
*new_fl2
= NULL
;
812 struct file_lock
*left
= NULL
;
813 struct file_lock
*right
= NULL
;
814 struct file_lock
**before
;
815 int error
, added
= 0;
818 * We may need two file_lock structures for this operation,
819 * so we get them in advance to avoid races.
821 * In some cases we can be sure, that no new locks will be needed
823 if (!(request
->fl_flags
& FL_ACCESS
) &&
824 (request
->fl_type
!= F_UNLCK
||
825 request
->fl_start
!= 0 || request
->fl_end
!= OFFSET_MAX
)) {
826 new_fl
= locks_alloc_lock();
827 new_fl2
= locks_alloc_lock();
831 if (request
->fl_type
!= F_UNLCK
) {
832 for_each_lock(inode
, before
) {
833 struct file_lock
*fl
= *before
;
836 if (!posix_locks_conflict(request
, fl
))
839 locks_copy_lock(conflock
, fl
);
841 if (!(request
->fl_flags
& FL_SLEEP
))
844 if (posix_locks_deadlock(request
, fl
))
847 locks_insert_block(fl
, request
);
852 /* If we're just looking for a conflict, we're done. */
854 if (request
->fl_flags
& FL_ACCESS
)
858 * Find the first old lock with the same owner as the new lock.
861 before
= &inode
->i_flock
;
863 /* First skip locks owned by other processes. */
864 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
865 !posix_same_owner(request
, fl
))) {
866 before
= &fl
->fl_next
;
869 /* Process locks with this owner. */
870 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
871 /* Detect adjacent or overlapping regions (if same lock type)
873 if (request
->fl_type
== fl
->fl_type
) {
874 /* In all comparisons of start vs end, use
875 * "start - 1" rather than "end + 1". If end
876 * is OFFSET_MAX, end + 1 will become negative.
878 if (fl
->fl_end
< request
->fl_start
- 1)
880 /* If the next lock in the list has entirely bigger
881 * addresses than the new one, insert the lock here.
883 if (fl
->fl_start
- 1 > request
->fl_end
)
886 /* If we come here, the new and old lock are of the
887 * same type and adjacent or overlapping. Make one
888 * lock yielding from the lower start address of both
889 * locks to the higher end address.
891 if (fl
->fl_start
> request
->fl_start
)
892 fl
->fl_start
= request
->fl_start
;
894 request
->fl_start
= fl
->fl_start
;
895 if (fl
->fl_end
< request
->fl_end
)
896 fl
->fl_end
= request
->fl_end
;
898 request
->fl_end
= fl
->fl_end
;
900 locks_delete_lock(before
);
907 /* Processing for different lock types is a bit
910 if (fl
->fl_end
< request
->fl_start
)
912 if (fl
->fl_start
> request
->fl_end
)
914 if (request
->fl_type
== F_UNLCK
)
916 if (fl
->fl_start
< request
->fl_start
)
918 /* If the next lock in the list has a higher end
919 * address than the new one, insert the new one here.
921 if (fl
->fl_end
> request
->fl_end
) {
925 if (fl
->fl_start
>= request
->fl_start
) {
926 /* The new lock completely replaces an old
927 * one (This may happen several times).
930 locks_delete_lock(before
);
933 /* Replace the old lock with the new one.
934 * Wake up anybody waiting for the old one,
935 * as the change in lock type might satisfy
938 locks_wake_up_blocks(fl
);
939 fl
->fl_start
= request
->fl_start
;
940 fl
->fl_end
= request
->fl_end
;
941 fl
->fl_type
= request
->fl_type
;
942 locks_release_private(fl
);
943 locks_copy_private(fl
, request
);
948 /* Go on to next lock.
951 before
= &fl
->fl_next
;
955 * The above code only modifies existing locks in case of
956 * merging or replacing. If new lock(s) need to be inserted
957 * all modifications are done bellow this, so it's safe yet to
960 error
= -ENOLCK
; /* "no luck" */
961 if (right
&& left
== right
&& !new_fl2
)
966 if (request
->fl_type
== F_UNLCK
) {
967 if (request
->fl_flags
& FL_EXISTS
)
976 locks_copy_lock(new_fl
, request
);
977 locks_insert_lock(before
, new_fl
);
982 /* The new lock breaks the old one in two pieces,
983 * so we have to use the second new lock.
987 locks_copy_lock(left
, right
);
988 locks_insert_lock(before
, left
);
990 right
->fl_start
= request
->fl_end
+ 1;
991 locks_wake_up_blocks(right
);
994 left
->fl_end
= request
->fl_start
- 1;
995 locks_wake_up_blocks(left
);
1000 * Free any unused locks.
1003 locks_free_lock(new_fl
);
1005 locks_free_lock(new_fl2
);
1010 * posix_lock_file - Apply a POSIX-style lock to a file
1011 * @filp: The file to apply the lock to
1012 * @fl: The lock to be applied
1013 * @conflock: Place to return a copy of the conflicting lock, if found.
1015 * Add a POSIX style lock to a file.
1016 * We merge adjacent & overlapping locks whenever possible.
1017 * POSIX locks are sorted by owner task, then by starting address
1019 * Note that if called with an FL_EXISTS argument, the caller may determine
1020 * whether or not a lock was successfully freed by testing the return
1021 * value for -ENOENT.
1023 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
,
1024 struct file_lock
*conflock
)
1026 return __posix_lock_file(filp
->f_path
.dentry
->d_inode
, fl
, conflock
);
1028 EXPORT_SYMBOL(posix_lock_file
);
1031 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1032 * @filp: The file to apply the lock to
1033 * @fl: The lock to be applied
1035 * Add a POSIX style lock to a file.
1036 * We merge adjacent & overlapping locks whenever possible.
1037 * POSIX locks are sorted by owner task, then by starting address
1039 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1044 error
= posix_lock_file(filp
, fl
, NULL
);
1045 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1047 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1051 locks_delete_block(fl
);
1056 EXPORT_SYMBOL(posix_lock_file_wait
);
1059 * locks_mandatory_locked - Check for an active lock
1060 * @inode: the file to check
1062 * Searches the inode's list of locks to find any POSIX locks which conflict.
1063 * This function is called from locks_verify_locked() only.
1065 int locks_mandatory_locked(struct inode
*inode
)
1067 fl_owner_t owner
= current
->files
;
1068 struct file_lock
*fl
;
1071 * Search the lock list for this inode for any POSIX locks.
1074 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1077 if (fl
->fl_owner
!= owner
)
1081 return fl
? -EAGAIN
: 0;
1085 * locks_mandatory_area - Check for a conflicting lock
1086 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1088 * @inode: the file to check
1089 * @filp: how the file was opened (if it was)
1090 * @offset: start of area to check
1091 * @count: length of area to check
1093 * Searches the inode's list of locks to find any POSIX locks which conflict.
1094 * This function is called from rw_verify_area() and
1095 * locks_verify_truncate().
1097 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1098 struct file
*filp
, loff_t offset
,
1101 struct file_lock fl
;
1104 locks_init_lock(&fl
);
1105 fl
.fl_owner
= current
->files
;
1106 fl
.fl_pid
= current
->tgid
;
1108 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1109 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1110 fl
.fl_flags
|= FL_SLEEP
;
1111 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1112 fl
.fl_start
= offset
;
1113 fl
.fl_end
= offset
+ count
- 1;
1116 error
= __posix_lock_file(inode
, &fl
, NULL
);
1117 if (error
!= -EAGAIN
)
1119 if (!(fl
.fl_flags
& FL_SLEEP
))
1121 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1124 * If we've been sleeping someone might have
1125 * changed the permissions behind our back.
1127 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
1131 locks_delete_block(&fl
);
1138 EXPORT_SYMBOL(locks_mandatory_area
);
1140 /* We already had a lease on this file; just change its type */
1141 int lease_modify(struct file_lock
**before
, int arg
)
1143 struct file_lock
*fl
= *before
;
1144 int error
= assign_type(fl
, arg
);
1148 locks_wake_up_blocks(fl
);
1150 locks_delete_lock(before
);
1154 EXPORT_SYMBOL(lease_modify
);
1156 static void time_out_leases(struct inode
*inode
)
1158 struct file_lock
**before
;
1159 struct file_lock
*fl
;
1161 before
= &inode
->i_flock
;
1162 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1163 if ((fl
->fl_break_time
== 0)
1164 || time_before(jiffies
, fl
->fl_break_time
)) {
1165 before
= &fl
->fl_next
;
1168 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1169 if (fl
== *before
) /* lease_modify may have freed fl */
1170 before
= &fl
->fl_next
;
1175 * __break_lease - revoke all outstanding leases on file
1176 * @inode: the inode of the file to return
1177 * @mode: the open mode (read or write)
1179 * break_lease (inlined for speed) has checked there already is at least
1180 * some kind of lock (maybe a lease) on this file. Leases are broken on
1181 * a call to open() or truncate(). This function can sleep unless you
1182 * specified %O_NONBLOCK to your open().
1184 int __break_lease(struct inode
*inode
, unsigned int mode
)
1186 int error
= 0, future
;
1187 struct file_lock
*new_fl
, *flock
;
1188 struct file_lock
*fl
;
1189 unsigned long break_time
;
1190 int i_have_this_lease
= 0;
1192 new_fl
= lease_alloc(NULL
, mode
& FMODE_WRITE
? F_WRLCK
: F_RDLCK
);
1196 time_out_leases(inode
);
1198 flock
= inode
->i_flock
;
1199 if ((flock
== NULL
) || !IS_LEASE(flock
))
1202 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
)
1203 if (fl
->fl_owner
== current
->files
)
1204 i_have_this_lease
= 1;
1206 if (mode
& FMODE_WRITE
) {
1207 /* If we want write access, we have to revoke any lease. */
1208 future
= F_UNLCK
| F_INPROGRESS
;
1209 } else if (flock
->fl_type
& F_INPROGRESS
) {
1210 /* If the lease is already being broken, we just leave it */
1211 future
= flock
->fl_type
;
1212 } else if (flock
->fl_type
& F_WRLCK
) {
1213 /* Downgrade the exclusive lease to a read-only lease. */
1214 future
= F_RDLCK
| F_INPROGRESS
;
1216 /* the existing lease was read-only, so we can read too. */
1220 if (IS_ERR(new_fl
) && !i_have_this_lease
1221 && ((mode
& O_NONBLOCK
) == 0)) {
1222 error
= PTR_ERR(new_fl
);
1227 if (lease_break_time
> 0) {
1228 break_time
= jiffies
+ lease_break_time
* HZ
;
1229 if (break_time
== 0)
1230 break_time
++; /* so that 0 means no break time */
1233 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1234 if (fl
->fl_type
!= future
) {
1235 fl
->fl_type
= future
;
1236 fl
->fl_break_time
= break_time
;
1237 /* lease must have lmops break callback */
1238 fl
->fl_lmops
->fl_break(fl
);
1242 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1243 error
= -EWOULDBLOCK
;
1248 break_time
= flock
->fl_break_time
;
1249 if (break_time
!= 0) {
1250 break_time
-= jiffies
;
1251 if (break_time
== 0)
1254 error
= locks_block_on_timeout(flock
, new_fl
, break_time
);
1257 time_out_leases(inode
);
1258 /* Wait for the next lease that has not been broken yet */
1259 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1260 flock
= flock
->fl_next
) {
1261 if (flock
->fl_type
& F_INPROGRESS
)
1269 if (!IS_ERR(new_fl
))
1270 locks_free_lock(new_fl
);
1274 EXPORT_SYMBOL(__break_lease
);
1279 * @time: pointer to a timespec which will contain the last modified time
1281 * This is to force NFS clients to flush their caches for files with
1282 * exclusive leases. The justification is that if someone has an
1283 * exclusive lease, then they could be modifiying it.
1285 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1287 struct file_lock
*flock
= inode
->i_flock
;
1288 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1289 *time
= current_fs_time(inode
->i_sb
);
1291 *time
= inode
->i_mtime
;
1294 EXPORT_SYMBOL(lease_get_mtime
);
1297 * fcntl_getlease - Enquire what lease is currently active
1300 * The value returned by this function will be one of
1301 * (if no lease break is pending):
1303 * %F_RDLCK to indicate a shared lease is held.
1305 * %F_WRLCK to indicate an exclusive lease is held.
1307 * %F_UNLCK to indicate no lease is held.
1309 * (if a lease break is pending):
1311 * %F_RDLCK to indicate an exclusive lease needs to be
1312 * changed to a shared lease (or removed).
1314 * %F_UNLCK to indicate the lease needs to be removed.
1316 * XXX: sfr & willy disagree over whether F_INPROGRESS
1317 * should be returned to userspace.
1319 int fcntl_getlease(struct file
*filp
)
1321 struct file_lock
*fl
;
1325 time_out_leases(filp
->f_path
.dentry
->d_inode
);
1326 for (fl
= filp
->f_path
.dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1328 if (fl
->fl_file
== filp
) {
1329 type
= fl
->fl_type
& ~F_INPROGRESS
;
1338 * generic_setlease - sets a lease on an open file
1339 * @filp: file pointer
1340 * @arg: type of lease to obtain
1341 * @flp: input - file_lock to use, output - file_lock inserted
1343 * The (input) flp->fl_lmops->fl_break function is required
1346 * Called with kernel lock held.
1348 int generic_setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1350 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1351 struct dentry
*dentry
= filp
->f_path
.dentry
;
1352 struct inode
*inode
= dentry
->d_inode
;
1353 int error
, rdlease_count
= 0, wrlease_count
= 0;
1355 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1357 if (!S_ISREG(inode
->i_mode
))
1359 error
= security_file_lock(filp
, arg
);
1363 time_out_leases(inode
);
1365 BUG_ON(!(*flp
)->fl_lmops
->fl_break
);
1370 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1372 if ((arg
== F_WRLCK
)
1373 && ((atomic_read(&dentry
->d_count
) > 1)
1374 || (atomic_read(&inode
->i_count
) > 1)))
1378 * At this point, we know that if there is an exclusive
1379 * lease on this file, then we hold it on this filp
1380 * (otherwise our open of this file would have blocked).
1381 * And if we are trying to acquire an exclusive lease,
1382 * then the file is not open by anyone (including us)
1383 * except for this filp.
1385 for (before
= &inode
->i_flock
;
1386 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1387 before
= &fl
->fl_next
) {
1388 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1390 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1392 * Someone is in the process of opening this
1393 * file for writing so we may not take an
1394 * exclusive lease on it.
1401 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1402 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1405 if (my_before
!= NULL
) {
1407 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1420 fl
= locks_alloc_lock();
1424 locks_copy_lock(fl
, lease
);
1426 locks_insert_lock(before
, fl
);
1433 EXPORT_SYMBOL(generic_setlease
);
1436 * vfs_setlease - sets a lease on an open file
1437 * @filp: file pointer
1438 * @arg: type of lease to obtain
1439 * @lease: file_lock to use
1441 * Call this to establish a lease on the file.
1442 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1443 * break_lease will oops!
1445 * This will call the filesystem's setlease file method, if
1446 * defined. Note that there is no getlease method; instead, the
1447 * filesystem setlease method should call back to setlease() to
1448 * add a lease to the inode's lease list, where fcntl_getlease() can
1449 * find it. Since fcntl_getlease() only reports whether the current
1450 * task holds a lease, a cluster filesystem need only do this for
1451 * leases held by processes on this node.
1453 * There is also no break_lease method; filesystems that
1454 * handle their own leases shoud break leases themselves from the
1455 * filesystem's open, create, and (on truncate) setattr methods.
1457 * Warning: the only current setlease methods exist only to disable
1458 * leases in certain cases. More vfs changes may be required to
1459 * allow a full filesystem lease implementation.
1462 int vfs_setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1467 if (filp
->f_op
&& filp
->f_op
->setlease
)
1468 error
= filp
->f_op
->setlease(filp
, arg
, lease
);
1470 error
= generic_setlease(filp
, arg
, lease
);
1475 EXPORT_SYMBOL_GPL(vfs_setlease
);
1478 * fcntl_setlease - sets a lease on an open file
1479 * @fd: open file descriptor
1480 * @filp: file pointer
1481 * @arg: type of lease to obtain
1483 * Call this fcntl to establish a lease on the file.
1484 * Note that you also need to call %F_SETSIG to
1485 * receive a signal when the lease is broken.
1487 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1489 struct file_lock fl
, *flp
= &fl
;
1490 struct dentry
*dentry
= filp
->f_path
.dentry
;
1491 struct inode
*inode
= dentry
->d_inode
;
1494 locks_init_lock(&fl
);
1495 error
= lease_init(filp
, arg
, &fl
);
1501 error
= vfs_setlease(filp
, arg
, &flp
);
1502 if (error
|| arg
== F_UNLCK
)
1505 error
= fasync_helper(fd
, filp
, 1, &flp
->fl_fasync
);
1507 /* remove lease just inserted by setlease */
1508 flp
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1509 flp
->fl_break_time
= jiffies
- 10;
1510 time_out_leases(inode
);
1514 error
= __f_setown(filp
, task_pid(current
), PIDTYPE_PID
, 0);
1521 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1522 * @filp: The file to apply the lock to
1523 * @fl: The lock to be applied
1525 * Add a FLOCK style lock to a file.
1527 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1532 error
= flock_lock_file(filp
, fl
);
1533 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1535 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1539 locks_delete_block(fl
);
1545 EXPORT_SYMBOL(flock_lock_file_wait
);
1548 * sys_flock: - flock() system call.
1549 * @fd: the file descriptor to lock.
1550 * @cmd: the type of lock to apply.
1552 * Apply a %FL_FLOCK style lock to an open file descriptor.
1553 * The @cmd can be one of
1555 * %LOCK_SH -- a shared lock.
1557 * %LOCK_EX -- an exclusive lock.
1559 * %LOCK_UN -- remove an existing lock.
1561 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1563 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1564 * processes read and write access respectively.
1566 asmlinkage
long sys_flock(unsigned int fd
, unsigned int cmd
)
1569 struct file_lock
*lock
;
1570 int can_sleep
, unlock
;
1578 can_sleep
= !(cmd
& LOCK_NB
);
1580 unlock
= (cmd
== LOCK_UN
);
1582 if (!unlock
&& !(cmd
& LOCK_MAND
) && !(filp
->f_mode
& 3))
1585 error
= flock_make_lock(filp
, &lock
, cmd
);
1589 lock
->fl_flags
|= FL_SLEEP
;
1591 error
= security_file_lock(filp
, cmd
);
1595 if (filp
->f_op
&& filp
->f_op
->flock
)
1596 error
= filp
->f_op
->flock(filp
,
1597 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1600 error
= flock_lock_file_wait(filp
, lock
);
1603 locks_free_lock(lock
);
1612 * vfs_test_lock - test file byte range lock
1613 * @filp: The file to test lock for
1614 * @fl: The lock to test; also used to hold result
1616 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1617 * setting conf->fl_type to something other than F_UNLCK.
1619 int vfs_test_lock(struct file
*filp
, struct file_lock
*fl
)
1621 if (filp
->f_op
&& filp
->f_op
->lock
)
1622 return filp
->f_op
->lock(filp
, F_GETLK
, fl
);
1623 posix_test_lock(filp
, fl
);
1626 EXPORT_SYMBOL_GPL(vfs_test_lock
);
1628 static int posix_lock_to_flock(struct flock
*flock
, struct file_lock
*fl
)
1630 flock
->l_pid
= fl
->fl_pid
;
1631 #if BITS_PER_LONG == 32
1633 * Make sure we can represent the posix lock via
1634 * legacy 32bit flock.
1636 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1638 if (fl
->fl_end
!= OFFSET_MAX
&& fl
->fl_end
> OFFT_OFFSET_MAX
)
1641 flock
->l_start
= fl
->fl_start
;
1642 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1643 fl
->fl_end
- fl
->fl_start
+ 1;
1644 flock
->l_whence
= 0;
1645 flock
->l_type
= fl
->fl_type
;
1649 #if BITS_PER_LONG == 32
1650 static void posix_lock_to_flock64(struct flock64
*flock
, struct file_lock
*fl
)
1652 flock
->l_pid
= fl
->fl_pid
;
1653 flock
->l_start
= fl
->fl_start
;
1654 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1655 fl
->fl_end
- fl
->fl_start
+ 1;
1656 flock
->l_whence
= 0;
1657 flock
->l_type
= fl
->fl_type
;
1661 /* Report the first existing lock that would conflict with l.
1662 * This implements the F_GETLK command of fcntl().
1664 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1666 struct file_lock file_lock
;
1671 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1674 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1677 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1681 error
= vfs_test_lock(filp
, &file_lock
);
1685 flock
.l_type
= file_lock
.fl_type
;
1686 if (file_lock
.fl_type
!= F_UNLCK
) {
1687 error
= posix_lock_to_flock(&flock
, &file_lock
);
1692 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1699 * vfs_lock_file - file byte range lock
1700 * @filp: The file to apply the lock to
1701 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1702 * @fl: The lock to be applied
1703 * @conf: Place to return a copy of the conflicting lock, if found.
1705 * A caller that doesn't care about the conflicting lock may pass NULL
1706 * as the final argument.
1708 * If the filesystem defines a private ->lock() method, then @conf will
1709 * be left unchanged; so a caller that cares should initialize it to
1710 * some acceptable default.
1712 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1713 * locks, the ->lock() interface may return asynchronously, before the lock has
1714 * been granted or denied by the underlying filesystem, if (and only if)
1715 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1716 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1717 * the request is for a blocking lock. When ->lock() does return asynchronously,
1718 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1719 * request completes.
1720 * If the request is for non-blocking lock the file system should return
1721 * -EINPROGRESS then try to get the lock and call the callback routine with
1722 * the result. If the request timed out the callback routine will return a
1723 * nonzero return code and the file system should release the lock. The file
1724 * system is also responsible to keep a corresponding posix lock when it
1725 * grants a lock so the VFS can find out which locks are locally held and do
1726 * the correct lock cleanup when required.
1727 * The underlying filesystem must not drop the kernel lock or call
1728 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1731 int vfs_lock_file(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
, struct file_lock
*conf
)
1733 if (filp
->f_op
&& filp
->f_op
->lock
)
1734 return filp
->f_op
->lock(filp
, cmd
, fl
);
1736 return posix_lock_file(filp
, fl
, conf
);
1738 EXPORT_SYMBOL_GPL(vfs_lock_file
);
1740 /* Apply the lock described by l to an open file descriptor.
1741 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1743 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1744 struct flock __user
*l
)
1746 struct file_lock
*file_lock
= locks_alloc_lock();
1748 struct inode
*inode
;
1751 if (file_lock
== NULL
)
1755 * This might block, so we do it before checking the inode.
1758 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1761 inode
= filp
->f_path
.dentry
->d_inode
;
1763 /* Don't allow mandatory locks on files that may be memory mapped
1766 if (IS_MANDLOCK(inode
) &&
1767 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1768 mapping_writably_mapped(filp
->f_mapping
)) {
1774 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1777 if (cmd
== F_SETLKW
) {
1778 file_lock
->fl_flags
|= FL_SLEEP
;
1782 switch (flock
.l_type
) {
1784 if (!(filp
->f_mode
& FMODE_READ
))
1788 if (!(filp
->f_mode
& FMODE_WRITE
))
1798 error
= security_file_lock(filp
, file_lock
->fl_type
);
1803 error
= vfs_lock_file(filp
, cmd
, file_lock
, NULL
);
1804 if (error
!= -EAGAIN
|| cmd
== F_SETLK
)
1806 error
= wait_event_interruptible(file_lock
->fl_wait
,
1807 !file_lock
->fl_next
);
1811 locks_delete_block(file_lock
);
1816 * Attempt to detect a close/fcntl race and recover by
1817 * releasing the lock that was just acquired.
1819 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1820 flock
.l_type
= F_UNLCK
;
1825 locks_free_lock(file_lock
);
1829 #if BITS_PER_LONG == 32
1830 /* Report the first existing lock that would conflict with l.
1831 * This implements the F_GETLK command of fcntl().
1833 int fcntl_getlk64(struct file
*filp
, struct flock64 __user
*l
)
1835 struct file_lock file_lock
;
1836 struct flock64 flock
;
1840 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1843 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1846 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1850 error
= vfs_test_lock(filp
, &file_lock
);
1854 flock
.l_type
= file_lock
.fl_type
;
1855 if (file_lock
.fl_type
!= F_UNLCK
)
1856 posix_lock_to_flock64(&flock
, &file_lock
);
1859 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1866 /* Apply the lock described by l to an open file descriptor.
1867 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1869 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1870 struct flock64 __user
*l
)
1872 struct file_lock
*file_lock
= locks_alloc_lock();
1873 struct flock64 flock
;
1874 struct inode
*inode
;
1877 if (file_lock
== NULL
)
1881 * This might block, so we do it before checking the inode.
1884 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1887 inode
= filp
->f_path
.dentry
->d_inode
;
1889 /* Don't allow mandatory locks on files that may be memory mapped
1892 if (IS_MANDLOCK(inode
) &&
1893 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1894 mapping_writably_mapped(filp
->f_mapping
)) {
1900 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1903 if (cmd
== F_SETLKW64
) {
1904 file_lock
->fl_flags
|= FL_SLEEP
;
1908 switch (flock
.l_type
) {
1910 if (!(filp
->f_mode
& FMODE_READ
))
1914 if (!(filp
->f_mode
& FMODE_WRITE
))
1924 error
= security_file_lock(filp
, file_lock
->fl_type
);
1929 error
= vfs_lock_file(filp
, cmd
, file_lock
, NULL
);
1930 if (error
!= -EAGAIN
|| cmd
== F_SETLK64
)
1932 error
= wait_event_interruptible(file_lock
->fl_wait
,
1933 !file_lock
->fl_next
);
1937 locks_delete_block(file_lock
);
1942 * Attempt to detect a close/fcntl race and recover by
1943 * releasing the lock that was just acquired.
1945 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1946 flock
.l_type
= F_UNLCK
;
1951 locks_free_lock(file_lock
);
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
)
1973 lock
.fl_type
= F_UNLCK
;
1974 lock
.fl_flags
= FL_POSIX
| FL_CLOSE
;
1976 lock
.fl_end
= OFFSET_MAX
;
1977 lock
.fl_owner
= owner
;
1978 lock
.fl_pid
= current
->tgid
;
1979 lock
.fl_file
= filp
;
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
)
2003 if (filp
->f_op
&& filp
->f_op
->flock
) {
2004 struct file_lock fl
= {
2005 .fl_pid
= current
->tgid
,
2007 .fl_flags
= FL_FLOCK
,
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
);
2017 before
= &inode
->i_flock
;
2019 while ((fl
= *before
) != NULL
) {
2020 if (fl
->fl_file
== filp
) {
2022 locks_delete_lock(before
);
2026 lease_modify(before
, F_UNLCK
);
2032 before
= &fl
->fl_next
;
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
)
2050 if (waiter
->fl_next
)
2051 __locks_delete_block(waiter
);
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
);
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
);
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 ");
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 ");
2104 out
+= sprintf(out
, "BREAKER ");
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 ");
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 ");
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
);
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
);
2130 out
+= sprintf(out
, "%d <none>:0 ", fl
->fl_pid
);
2133 if (fl
->fl_end
== OFFSET_MAX
)
2134 out
+= sprintf(out
, "%Ld EOF\n", fl
->fl_start
);
2136 out
+= sprintf(out
, "%Ld %Ld\n", fl
->fl_start
,
2139 out
+= sprintf(out
, "0 EOF\n");
2143 static void move_lock_status(char **p
, off_t
* pos
, off_t offset
)
2147 if(*pos
>= offset
) {
2148 /* the complete line is valid */
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
);
2161 /* discard the complete line */
2166 * get_locks_status - reports lock usage in /proc/locks
2167 * @buffer: address in userspace to write into
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
;
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
)
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
)
2203 if(q
-buffer
< 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
;
2226 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2228 if (fl
->fl_type
== F_RDLCK
)
2230 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2232 } else if (IS_FLOCK(fl
)) {
2233 if (!(fl
->fl_type
& LOCK_MAND
))
2235 if (fl
->fl_type
& LOCK_READ
)
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
;
2266 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2268 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2270 } else if (IS_FLOCK(fl
)) {
2271 if (!(fl
->fl_type
& LOCK_MAND
))
2273 if (fl
->fl_type
& LOCK_WRITE
)
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
,
2294 core_initcall(filelock_init
);