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
);
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 * at the head of the list, but that's secret knowledge known only to
721 * flock_lock_file and posix_lock_file.
723 * Note that if called with an FL_EXISTS argument, the caller may determine
724 * whether or not a lock was successfully freed by testing the return
727 static int flock_lock_file(struct file
*filp
, struct file_lock
*request
)
729 struct file_lock
*new_fl
= NULL
;
730 struct file_lock
**before
;
731 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
736 if (request
->fl_flags
& FL_ACCESS
)
738 for_each_lock(inode
, before
) {
739 struct file_lock
*fl
= *before
;
744 if (filp
!= fl
->fl_file
)
746 if (request
->fl_type
== fl
->fl_type
)
749 locks_delete_lock(before
);
753 if (request
->fl_type
== F_UNLCK
) {
754 if ((request
->fl_flags
& FL_EXISTS
) && !found
)
760 new_fl
= locks_alloc_lock();
764 * If a higher-priority process was blocked on the old file lock,
765 * give it the opportunity to lock the file.
771 for_each_lock(inode
, before
) {
772 struct file_lock
*fl
= *before
;
777 if (!flock_locks_conflict(request
, fl
))
780 if (request
->fl_flags
& FL_SLEEP
)
781 locks_insert_block(fl
, request
);
784 if (request
->fl_flags
& FL_ACCESS
)
786 locks_copy_lock(new_fl
, request
);
787 locks_insert_lock(&inode
->i_flock
, new_fl
);
794 locks_free_lock(new_fl
);
798 static int __posix_lock_file(struct inode
*inode
, struct file_lock
*request
, struct file_lock
*conflock
)
800 struct file_lock
*fl
;
801 struct file_lock
*new_fl
= NULL
;
802 struct file_lock
*new_fl2
= NULL
;
803 struct file_lock
*left
= NULL
;
804 struct file_lock
*right
= NULL
;
805 struct file_lock
**before
;
806 int error
, added
= 0;
809 * We may need two file_lock structures for this operation,
810 * so we get them in advance to avoid races.
812 * In some cases we can be sure, that no new locks will be needed
814 if (!(request
->fl_flags
& FL_ACCESS
) &&
815 (request
->fl_type
!= F_UNLCK
||
816 request
->fl_start
!= 0 || request
->fl_end
!= OFFSET_MAX
)) {
817 new_fl
= locks_alloc_lock();
818 new_fl2
= locks_alloc_lock();
822 if (request
->fl_type
!= F_UNLCK
) {
823 for_each_lock(inode
, before
) {
824 struct file_lock
*fl
= *before
;
827 if (!posix_locks_conflict(request
, fl
))
830 locks_copy_lock(conflock
, fl
);
832 if (!(request
->fl_flags
& FL_SLEEP
))
835 if (posix_locks_deadlock(request
, fl
))
838 locks_insert_block(fl
, request
);
843 /* If we're just looking for a conflict, we're done. */
845 if (request
->fl_flags
& FL_ACCESS
)
849 * Find the first old lock with the same owner as the new lock.
852 before
= &inode
->i_flock
;
854 /* First skip locks owned by other processes. */
855 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
856 !posix_same_owner(request
, fl
))) {
857 before
= &fl
->fl_next
;
860 /* Process locks with this owner. */
861 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
862 /* Detect adjacent or overlapping regions (if same lock type)
864 if (request
->fl_type
== fl
->fl_type
) {
865 /* In all comparisons of start vs end, use
866 * "start - 1" rather than "end + 1". If end
867 * is OFFSET_MAX, end + 1 will become negative.
869 if (fl
->fl_end
< request
->fl_start
- 1)
871 /* If the next lock in the list has entirely bigger
872 * addresses than the new one, insert the lock here.
874 if (fl
->fl_start
- 1 > request
->fl_end
)
877 /* If we come here, the new and old lock are of the
878 * same type and adjacent or overlapping. Make one
879 * lock yielding from the lower start address of both
880 * locks to the higher end address.
882 if (fl
->fl_start
> request
->fl_start
)
883 fl
->fl_start
= request
->fl_start
;
885 request
->fl_start
= fl
->fl_start
;
886 if (fl
->fl_end
< request
->fl_end
)
887 fl
->fl_end
= request
->fl_end
;
889 request
->fl_end
= fl
->fl_end
;
891 locks_delete_lock(before
);
898 /* Processing for different lock types is a bit
901 if (fl
->fl_end
< request
->fl_start
)
903 if (fl
->fl_start
> request
->fl_end
)
905 if (request
->fl_type
== F_UNLCK
)
907 if (fl
->fl_start
< request
->fl_start
)
909 /* If the next lock in the list has a higher end
910 * address than the new one, insert the new one here.
912 if (fl
->fl_end
> request
->fl_end
) {
916 if (fl
->fl_start
>= request
->fl_start
) {
917 /* The new lock completely replaces an old
918 * one (This may happen several times).
921 locks_delete_lock(before
);
924 /* Replace the old lock with the new one.
925 * Wake up anybody waiting for the old one,
926 * as the change in lock type might satisfy
929 locks_wake_up_blocks(fl
);
930 fl
->fl_start
= request
->fl_start
;
931 fl
->fl_end
= request
->fl_end
;
932 fl
->fl_type
= request
->fl_type
;
933 locks_release_private(fl
);
934 locks_copy_private(fl
, request
);
939 /* Go on to next lock.
942 before
= &fl
->fl_next
;
946 * The above code only modifies existing locks in case of
947 * merging or replacing. If new lock(s) need to be inserted
948 * all modifications are done bellow this, so it's safe yet to
951 error
= -ENOLCK
; /* "no luck" */
952 if (right
&& left
== right
&& !new_fl2
)
957 if (request
->fl_type
== F_UNLCK
) {
958 if (request
->fl_flags
& FL_EXISTS
)
967 locks_copy_lock(new_fl
, request
);
968 locks_insert_lock(before
, new_fl
);
973 /* The new lock breaks the old one in two pieces,
974 * so we have to use the second new lock.
978 locks_copy_lock(left
, right
);
979 locks_insert_lock(before
, left
);
981 right
->fl_start
= request
->fl_end
+ 1;
982 locks_wake_up_blocks(right
);
985 left
->fl_end
= request
->fl_start
- 1;
986 locks_wake_up_blocks(left
);
991 * Free any unused locks.
994 locks_free_lock(new_fl
);
996 locks_free_lock(new_fl2
);
1001 * posix_lock_file - Apply a POSIX-style lock to a file
1002 * @filp: The file to apply the lock to
1003 * @fl: The lock to be applied
1004 * @conflock: Place to return a copy of the conflicting lock, if found.
1006 * Add a POSIX style lock to a file.
1007 * We merge adjacent & overlapping locks whenever possible.
1008 * POSIX locks are sorted by owner task, then by starting address
1010 * Note that if called with an FL_EXISTS argument, the caller may determine
1011 * whether or not a lock was successfully freed by testing the return
1012 * value for -ENOENT.
1014 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
,
1015 struct file_lock
*conflock
)
1017 return __posix_lock_file(filp
->f_path
.dentry
->d_inode
, fl
, conflock
);
1019 EXPORT_SYMBOL(posix_lock_file
);
1022 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1023 * @filp: The file to apply the lock to
1024 * @fl: The lock to be applied
1026 * Add a POSIX style lock to a file.
1027 * We merge adjacent & overlapping locks whenever possible.
1028 * POSIX locks are sorted by owner task, then by starting address
1030 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1035 error
= posix_lock_file(filp
, fl
, NULL
);
1036 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1038 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1042 locks_delete_block(fl
);
1047 EXPORT_SYMBOL(posix_lock_file_wait
);
1050 * locks_mandatory_locked - Check for an active lock
1051 * @inode: the file to check
1053 * Searches the inode's list of locks to find any POSIX locks which conflict.
1054 * This function is called from locks_verify_locked() only.
1056 int locks_mandatory_locked(struct inode
*inode
)
1058 fl_owner_t owner
= current
->files
;
1059 struct file_lock
*fl
;
1062 * Search the lock list for this inode for any POSIX locks.
1065 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1068 if (fl
->fl_owner
!= owner
)
1072 return fl
? -EAGAIN
: 0;
1076 * locks_mandatory_area - Check for a conflicting lock
1077 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1079 * @inode: the file to check
1080 * @filp: how the file was opened (if it was)
1081 * @offset: start of area to check
1082 * @count: length of area to check
1084 * Searches the inode's list of locks to find any POSIX locks which conflict.
1085 * This function is called from rw_verify_area() and
1086 * locks_verify_truncate().
1088 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1089 struct file
*filp
, loff_t offset
,
1092 struct file_lock fl
;
1095 locks_init_lock(&fl
);
1096 fl
.fl_owner
= current
->files
;
1097 fl
.fl_pid
= current
->tgid
;
1099 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1100 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1101 fl
.fl_flags
|= FL_SLEEP
;
1102 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1103 fl
.fl_start
= offset
;
1104 fl
.fl_end
= offset
+ count
- 1;
1107 error
= __posix_lock_file(inode
, &fl
, NULL
);
1108 if (error
!= -EAGAIN
)
1110 if (!(fl
.fl_flags
& FL_SLEEP
))
1112 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1115 * If we've been sleeping someone might have
1116 * changed the permissions behind our back.
1118 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
1122 locks_delete_block(&fl
);
1129 EXPORT_SYMBOL(locks_mandatory_area
);
1131 /* We already had a lease on this file; just change its type */
1132 int lease_modify(struct file_lock
**before
, int arg
)
1134 struct file_lock
*fl
= *before
;
1135 int error
= assign_type(fl
, arg
);
1139 locks_wake_up_blocks(fl
);
1141 locks_delete_lock(before
);
1145 EXPORT_SYMBOL(lease_modify
);
1147 static void time_out_leases(struct inode
*inode
)
1149 struct file_lock
**before
;
1150 struct file_lock
*fl
;
1152 before
= &inode
->i_flock
;
1153 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1154 if ((fl
->fl_break_time
== 0)
1155 || time_before(jiffies
, fl
->fl_break_time
)) {
1156 before
= &fl
->fl_next
;
1159 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1160 if (fl
== *before
) /* lease_modify may have freed fl */
1161 before
= &fl
->fl_next
;
1166 * __break_lease - revoke all outstanding leases on file
1167 * @inode: the inode of the file to return
1168 * @mode: the open mode (read or write)
1170 * break_lease (inlined for speed) has checked there already is at least
1171 * some kind of lock (maybe a lease) on this file. Leases are broken on
1172 * a call to open() or truncate(). This function can sleep unless you
1173 * specified %O_NONBLOCK to your open().
1175 int __break_lease(struct inode
*inode
, unsigned int mode
)
1177 int error
= 0, future
;
1178 struct file_lock
*new_fl
, *flock
;
1179 struct file_lock
*fl
;
1180 unsigned long break_time
;
1181 int i_have_this_lease
= 0;
1183 new_fl
= lease_alloc(NULL
, mode
& FMODE_WRITE
? F_WRLCK
: F_RDLCK
);
1187 time_out_leases(inode
);
1189 flock
= inode
->i_flock
;
1190 if ((flock
== NULL
) || !IS_LEASE(flock
))
1193 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
)
1194 if (fl
->fl_owner
== current
->files
)
1195 i_have_this_lease
= 1;
1197 if (mode
& FMODE_WRITE
) {
1198 /* If we want write access, we have to revoke any lease. */
1199 future
= F_UNLCK
| F_INPROGRESS
;
1200 } else if (flock
->fl_type
& F_INPROGRESS
) {
1201 /* If the lease is already being broken, we just leave it */
1202 future
= flock
->fl_type
;
1203 } else if (flock
->fl_type
& F_WRLCK
) {
1204 /* Downgrade the exclusive lease to a read-only lease. */
1205 future
= F_RDLCK
| F_INPROGRESS
;
1207 /* the existing lease was read-only, so we can read too. */
1211 if (IS_ERR(new_fl
) && !i_have_this_lease
1212 && ((mode
& O_NONBLOCK
) == 0)) {
1213 error
= PTR_ERR(new_fl
);
1218 if (lease_break_time
> 0) {
1219 break_time
= jiffies
+ lease_break_time
* HZ
;
1220 if (break_time
== 0)
1221 break_time
++; /* so that 0 means no break time */
1224 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1225 if (fl
->fl_type
!= future
) {
1226 fl
->fl_type
= future
;
1227 fl
->fl_break_time
= break_time
;
1228 /* lease must have lmops break callback */
1229 fl
->fl_lmops
->fl_break(fl
);
1233 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1234 error
= -EWOULDBLOCK
;
1239 break_time
= flock
->fl_break_time
;
1240 if (break_time
!= 0) {
1241 break_time
-= jiffies
;
1242 if (break_time
== 0)
1245 error
= locks_block_on_timeout(flock
, new_fl
, break_time
);
1248 time_out_leases(inode
);
1249 /* Wait for the next lease that has not been broken yet */
1250 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1251 flock
= flock
->fl_next
) {
1252 if (flock
->fl_type
& F_INPROGRESS
)
1260 if (!IS_ERR(new_fl
))
1261 locks_free_lock(new_fl
);
1265 EXPORT_SYMBOL(__break_lease
);
1270 * @time: pointer to a timespec which will contain the last modified time
1272 * This is to force NFS clients to flush their caches for files with
1273 * exclusive leases. The justification is that if someone has an
1274 * exclusive lease, then they could be modifiying it.
1276 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1278 struct file_lock
*flock
= inode
->i_flock
;
1279 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1280 *time
= current_fs_time(inode
->i_sb
);
1282 *time
= inode
->i_mtime
;
1285 EXPORT_SYMBOL(lease_get_mtime
);
1288 * fcntl_getlease - Enquire what lease is currently active
1291 * The value returned by this function will be one of
1292 * (if no lease break is pending):
1294 * %F_RDLCK to indicate a shared lease is held.
1296 * %F_WRLCK to indicate an exclusive lease is held.
1298 * %F_UNLCK to indicate no lease is held.
1300 * (if a lease break is pending):
1302 * %F_RDLCK to indicate an exclusive lease needs to be
1303 * changed to a shared lease (or removed).
1305 * %F_UNLCK to indicate the lease needs to be removed.
1307 * XXX: sfr & willy disagree over whether F_INPROGRESS
1308 * should be returned to userspace.
1310 int fcntl_getlease(struct file
*filp
)
1312 struct file_lock
*fl
;
1316 time_out_leases(filp
->f_path
.dentry
->d_inode
);
1317 for (fl
= filp
->f_path
.dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1319 if (fl
->fl_file
== filp
) {
1320 type
= fl
->fl_type
& ~F_INPROGRESS
;
1329 * setlease - sets a lease on an open file
1330 * @filp: file pointer
1331 * @arg: type of lease to obtain
1332 * @flp: input - file_lock to use, output - file_lock inserted
1334 * The (input) flp->fl_lmops->fl_break function is required
1337 * Called with kernel lock held.
1339 int setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1341 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1342 struct dentry
*dentry
= filp
->f_path
.dentry
;
1343 struct inode
*inode
= dentry
->d_inode
;
1344 int error
, rdlease_count
= 0, wrlease_count
= 0;
1346 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1348 if (!S_ISREG(inode
->i_mode
))
1350 error
= security_file_lock(filp
, arg
);
1354 time_out_leases(inode
);
1356 BUG_ON(!(*flp
)->fl_lmops
->fl_break
);
1361 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1363 if ((arg
== F_WRLCK
)
1364 && ((atomic_read(&dentry
->d_count
) > 1)
1365 || (atomic_read(&inode
->i_count
) > 1)))
1369 * At this point, we know that if there is an exclusive
1370 * lease on this file, then we hold it on this filp
1371 * (otherwise our open of this file would have blocked).
1372 * And if we are trying to acquire an exclusive lease,
1373 * then the file is not open by anyone (including us)
1374 * except for this filp.
1376 for (before
= &inode
->i_flock
;
1377 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1378 before
= &fl
->fl_next
) {
1379 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1381 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1383 * Someone is in the process of opening this
1384 * file for writing so we may not take an
1385 * exclusive lease on it.
1392 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1393 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1396 if (my_before
!= NULL
) {
1398 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1411 fl
= locks_alloc_lock();
1415 locks_copy_lock(fl
, lease
);
1417 locks_insert_lock(before
, fl
);
1424 EXPORT_SYMBOL(setlease
);
1427 * vfs_setlease - sets a lease on an open file
1428 * @filp: file pointer
1429 * @arg: type of lease to obtain
1430 * @lease: file_lock to use
1432 * Call this to establish a lease on the file.
1433 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1434 * break_lease will oops!
1436 * This will call the filesystem's setlease file method, if
1437 * defined. Note that there is no getlease method; instead, the
1438 * filesystem setlease method should call back to setlease() to
1439 * add a lease to the inode's lease list, where fcntl_getlease() can
1440 * find it. Since fcntl_getlease() only reports whether the current
1441 * task holds a lease, a cluster filesystem need only do this for
1442 * leases held by processes on this node.
1444 * There is also no break_lease method; filesystems that
1445 * handle their own leases shoud break leases themselves from the
1446 * filesystem's open, create, and (on truncate) setattr methods.
1448 * Warning: the only current setlease methods exist only to disable
1449 * leases in certain cases. More vfs changes may be required to
1450 * allow a full filesystem lease implementation.
1453 int vfs_setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1458 if (filp
->f_op
&& filp
->f_op
->setlease
)
1459 error
= filp
->f_op
->setlease(filp
, arg
, lease
);
1461 error
= setlease(filp
, arg
, lease
);
1466 EXPORT_SYMBOL_GPL(vfs_setlease
);
1469 * fcntl_setlease - sets a lease on an open file
1470 * @fd: open file descriptor
1471 * @filp: file pointer
1472 * @arg: type of lease to obtain
1474 * Call this fcntl to establish a lease on the file.
1475 * Note that you also need to call %F_SETSIG to
1476 * receive a signal when the lease is broken.
1478 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1480 struct file_lock fl
, *flp
= &fl
;
1481 struct dentry
*dentry
= filp
->f_path
.dentry
;
1482 struct inode
*inode
= dentry
->d_inode
;
1485 locks_init_lock(&fl
);
1486 error
= lease_init(filp
, arg
, &fl
);
1492 error
= vfs_setlease(filp
, arg
, &flp
);
1493 if (error
|| arg
== F_UNLCK
)
1496 error
= fasync_helper(fd
, filp
, 1, &flp
->fl_fasync
);
1498 /* remove lease just inserted by setlease */
1499 flp
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1500 flp
->fl_break_time
= jiffies
- 10;
1501 time_out_leases(inode
);
1505 error
= __f_setown(filp
, task_pid(current
), PIDTYPE_PID
, 0);
1512 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1513 * @filp: The file to apply the lock to
1514 * @fl: The lock to be applied
1516 * Add a FLOCK style lock to a file.
1518 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1523 error
= flock_lock_file(filp
, fl
);
1524 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1526 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1530 locks_delete_block(fl
);
1536 EXPORT_SYMBOL(flock_lock_file_wait
);
1539 * sys_flock: - flock() system call.
1540 * @fd: the file descriptor to lock.
1541 * @cmd: the type of lock to apply.
1543 * Apply a %FL_FLOCK style lock to an open file descriptor.
1544 * The @cmd can be one of
1546 * %LOCK_SH -- a shared lock.
1548 * %LOCK_EX -- an exclusive lock.
1550 * %LOCK_UN -- remove an existing lock.
1552 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1554 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1555 * processes read and write access respectively.
1557 asmlinkage
long sys_flock(unsigned int fd
, unsigned int cmd
)
1560 struct file_lock
*lock
;
1561 int can_sleep
, unlock
;
1569 can_sleep
= !(cmd
& LOCK_NB
);
1571 unlock
= (cmd
== LOCK_UN
);
1573 if (!unlock
&& !(cmd
& LOCK_MAND
) && !(filp
->f_mode
& 3))
1576 error
= flock_make_lock(filp
, &lock
, cmd
);
1580 lock
->fl_flags
|= FL_SLEEP
;
1582 error
= security_file_lock(filp
, cmd
);
1586 if (filp
->f_op
&& filp
->f_op
->flock
)
1587 error
= filp
->f_op
->flock(filp
,
1588 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1591 error
= flock_lock_file_wait(filp
, lock
);
1594 locks_free_lock(lock
);
1603 * vfs_test_lock - test file byte range lock
1604 * @filp: The file to test lock for
1605 * @fl: The lock to test
1606 * @conf: Place to return a copy of the conflicting lock, if found
1608 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1609 * setting conf->fl_type to something other than F_UNLCK.
1611 int vfs_test_lock(struct file
*filp
, struct file_lock
*fl
)
1613 if (filp
->f_op
&& filp
->f_op
->lock
)
1614 return filp
->f_op
->lock(filp
, F_GETLK
, fl
);
1615 posix_test_lock(filp
, fl
);
1618 EXPORT_SYMBOL_GPL(vfs_test_lock
);
1620 static int posix_lock_to_flock(struct flock
*flock
, struct file_lock
*fl
)
1622 flock
->l_pid
= fl
->fl_pid
;
1623 #if BITS_PER_LONG == 32
1625 * Make sure we can represent the posix lock via
1626 * legacy 32bit flock.
1628 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1630 if (fl
->fl_end
!= OFFSET_MAX
&& fl
->fl_end
> OFFT_OFFSET_MAX
)
1633 flock
->l_start
= fl
->fl_start
;
1634 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1635 fl
->fl_end
- fl
->fl_start
+ 1;
1636 flock
->l_whence
= 0;
1637 flock
->l_type
= fl
->fl_type
;
1641 #if BITS_PER_LONG == 32
1642 static void posix_lock_to_flock64(struct flock64
*flock
, struct file_lock
*fl
)
1644 flock
->l_pid
= fl
->fl_pid
;
1645 flock
->l_start
= fl
->fl_start
;
1646 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1647 fl
->fl_end
- fl
->fl_start
+ 1;
1648 flock
->l_whence
= 0;
1649 flock
->l_type
= fl
->fl_type
;
1653 /* Report the first existing lock that would conflict with l.
1654 * This implements the F_GETLK command of fcntl().
1656 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1658 struct file_lock file_lock
;
1663 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1666 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1669 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1673 error
= vfs_test_lock(filp
, &file_lock
);
1677 flock
.l_type
= file_lock
.fl_type
;
1678 if (file_lock
.fl_type
!= F_UNLCK
) {
1679 error
= posix_lock_to_flock(&flock
, &file_lock
);
1684 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1691 * vfs_lock_file - file byte range lock
1692 * @filp: The file to apply the lock to
1693 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1694 * @fl: The lock to be applied
1695 * @conf: Place to return a copy of the conflicting lock, if found.
1697 * A caller that doesn't care about the conflicting lock may pass NULL
1698 * as the final argument.
1700 * If the filesystem defines a private ->lock() method, then @conf will
1701 * be left unchanged; so a caller that cares should initialize it to
1702 * some acceptable default.
1704 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1705 * locks, the ->lock() interface may return asynchronously, before the lock has
1706 * been granted or denied by the underlying filesystem, if (and only if)
1707 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1708 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1709 * the request is for a blocking lock. When ->lock() does return asynchronously,
1710 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1711 * request completes.
1712 * If the request is for non-blocking lock the file system should return
1713 * -EINPROGRESS then try to get the lock and call the callback routine with
1714 * the result. If the request timed out the callback routine will return a
1715 * nonzero return code and the file system should release the lock. The file
1716 * system is also responsible to keep a corresponding posix lock when it
1717 * grants a lock so the VFS can find out which locks are locally held and do
1718 * the correct lock cleanup when required.
1719 * The underlying filesystem must not drop the kernel lock or call
1720 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1723 int vfs_lock_file(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
, struct file_lock
*conf
)
1725 if (filp
->f_op
&& filp
->f_op
->lock
)
1726 return filp
->f_op
->lock(filp
, cmd
, fl
);
1728 return posix_lock_file(filp
, fl
, conf
);
1730 EXPORT_SYMBOL_GPL(vfs_lock_file
);
1732 /* Apply the lock described by l to an open file descriptor.
1733 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1735 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1736 struct flock __user
*l
)
1738 struct file_lock
*file_lock
= locks_alloc_lock();
1740 struct inode
*inode
;
1743 if (file_lock
== NULL
)
1747 * This might block, so we do it before checking the inode.
1750 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1753 inode
= filp
->f_path
.dentry
->d_inode
;
1755 /* Don't allow mandatory locks on files that may be memory mapped
1758 if (IS_MANDLOCK(inode
) &&
1759 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1760 mapping_writably_mapped(filp
->f_mapping
)) {
1766 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1769 if (cmd
== F_SETLKW
) {
1770 file_lock
->fl_flags
|= FL_SLEEP
;
1774 switch (flock
.l_type
) {
1776 if (!(filp
->f_mode
& FMODE_READ
))
1780 if (!(filp
->f_mode
& FMODE_WRITE
))
1790 error
= security_file_lock(filp
, file_lock
->fl_type
);
1795 error
= vfs_lock_file(filp
, cmd
, file_lock
, NULL
);
1796 if (error
!= -EAGAIN
|| cmd
== F_SETLK
)
1798 error
= wait_event_interruptible(file_lock
->fl_wait
,
1799 !file_lock
->fl_next
);
1803 locks_delete_block(file_lock
);
1808 * Attempt to detect a close/fcntl race and recover by
1809 * releasing the lock that was just acquired.
1811 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1812 flock
.l_type
= F_UNLCK
;
1817 locks_free_lock(file_lock
);
1821 #if BITS_PER_LONG == 32
1822 /* Report the first existing lock that would conflict with l.
1823 * This implements the F_GETLK command of fcntl().
1825 int fcntl_getlk64(struct file
*filp
, struct flock64 __user
*l
)
1827 struct file_lock file_lock
;
1828 struct flock64 flock
;
1832 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1835 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1838 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1842 error
= vfs_test_lock(filp
, &file_lock
);
1846 flock
.l_type
= file_lock
.fl_type
;
1847 if (file_lock
.fl_type
!= F_UNLCK
)
1848 posix_lock_to_flock64(&flock
, &file_lock
);
1851 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1858 /* Apply the lock described by l to an open file descriptor.
1859 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1861 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1862 struct flock64 __user
*l
)
1864 struct file_lock
*file_lock
= locks_alloc_lock();
1865 struct flock64 flock
;
1866 struct inode
*inode
;
1869 if (file_lock
== NULL
)
1873 * This might block, so we do it before checking the inode.
1876 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1879 inode
= filp
->f_path
.dentry
->d_inode
;
1881 /* Don't allow mandatory locks on files that may be memory mapped
1884 if (IS_MANDLOCK(inode
) &&
1885 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1886 mapping_writably_mapped(filp
->f_mapping
)) {
1892 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1895 if (cmd
== F_SETLKW64
) {
1896 file_lock
->fl_flags
|= FL_SLEEP
;
1900 switch (flock
.l_type
) {
1902 if (!(filp
->f_mode
& FMODE_READ
))
1906 if (!(filp
->f_mode
& FMODE_WRITE
))
1916 error
= security_file_lock(filp
, file_lock
->fl_type
);
1921 error
= vfs_lock_file(filp
, cmd
, file_lock
, NULL
);
1922 if (error
!= -EAGAIN
|| cmd
== F_SETLK64
)
1924 error
= wait_event_interruptible(file_lock
->fl_wait
,
1925 !file_lock
->fl_next
);
1929 locks_delete_block(file_lock
);
1934 * Attempt to detect a close/fcntl race and recover by
1935 * releasing the lock that was just acquired.
1937 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1938 flock
.l_type
= F_UNLCK
;
1943 locks_free_lock(file_lock
);
1946 #endif /* BITS_PER_LONG == 32 */
1949 * This function is called when the file is being removed
1950 * from the task's fd array. POSIX locks belonging to this task
1951 * are deleted at this time.
1953 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
1955 struct file_lock lock
;
1958 * If there are no locks held on this file, we don't need to call
1959 * posix_lock_file(). Another process could be setting a lock on this
1960 * file at the same time, but we wouldn't remove that lock anyway.
1962 if (!filp
->f_path
.dentry
->d_inode
->i_flock
)
1965 lock
.fl_type
= F_UNLCK
;
1966 lock
.fl_flags
= FL_POSIX
| FL_CLOSE
;
1968 lock
.fl_end
= OFFSET_MAX
;
1969 lock
.fl_owner
= owner
;
1970 lock
.fl_pid
= current
->tgid
;
1971 lock
.fl_file
= filp
;
1973 lock
.fl_lmops
= NULL
;
1975 vfs_lock_file(filp
, F_SETLK
, &lock
, NULL
);
1977 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
1978 lock
.fl_ops
->fl_release_private(&lock
);
1981 EXPORT_SYMBOL(locks_remove_posix
);
1984 * This function is called on the last close of an open file.
1986 void locks_remove_flock(struct file
*filp
)
1988 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
1989 struct file_lock
*fl
;
1990 struct file_lock
**before
;
1992 if (!inode
->i_flock
)
1995 if (filp
->f_op
&& filp
->f_op
->flock
) {
1996 struct file_lock fl
= {
1997 .fl_pid
= current
->tgid
,
1999 .fl_flags
= FL_FLOCK
,
2001 .fl_end
= OFFSET_MAX
,
2003 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
2004 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
2005 fl
.fl_ops
->fl_release_private(&fl
);
2009 before
= &inode
->i_flock
;
2011 while ((fl
= *before
) != NULL
) {
2012 if (fl
->fl_file
== filp
) {
2014 locks_delete_lock(before
);
2018 lease_modify(before
, F_UNLCK
);
2024 before
= &fl
->fl_next
;
2030 * posix_unblock_lock - stop waiting for a file lock
2031 * @filp: how the file was opened
2032 * @waiter: the lock which was waiting
2034 * lockd needs to block waiting for locks.
2037 posix_unblock_lock(struct file
*filp
, struct file_lock
*waiter
)
2042 if (waiter
->fl_next
)
2043 __locks_delete_block(waiter
);
2050 EXPORT_SYMBOL(posix_unblock_lock
);
2053 * vfs_cancel_lock - file byte range unblock lock
2054 * @filp: The file to apply the unblock to
2055 * @fl: The lock to be unblocked
2057 * Used by lock managers to cancel blocked requests
2059 int vfs_cancel_lock(struct file
*filp
, struct file_lock
*fl
)
2061 if (filp
->f_op
&& filp
->f_op
->lock
)
2062 return filp
->f_op
->lock(filp
, F_CANCELLK
, fl
);
2066 EXPORT_SYMBOL_GPL(vfs_cancel_lock
);
2068 static void lock_get_status(char* out
, struct file_lock
*fl
, int id
, char *pfx
)
2070 struct inode
*inode
= NULL
;
2072 if (fl
->fl_file
!= NULL
)
2073 inode
= fl
->fl_file
->f_path
.dentry
->d_inode
;
2075 out
+= sprintf(out
, "%d:%s ", id
, pfx
);
2077 out
+= sprintf(out
, "%6s %s ",
2078 (fl
->fl_flags
& FL_ACCESS
) ? "ACCESS" : "POSIX ",
2079 (inode
== NULL
) ? "*NOINODE*" :
2080 (IS_MANDLOCK(inode
) &&
2081 (inode
->i_mode
& (S_IXGRP
| S_ISGID
)) == S_ISGID
) ?
2082 "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
);