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/fdtable.h>
120 #include <linux/fs.h>
121 #include <linux/init.h>
122 #include <linux/module.h>
123 #include <linux/security.h>
124 #include <linux/slab.h>
125 #include <linux/smp_lock.h>
126 #include <linux/syscalls.h>
127 #include <linux/time.h>
128 #include <linux/rcupdate.h>
129 #include <linux/pid_namespace.h>
131 #include <asm/uaccess.h>
133 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
137 int leases_enable
= 1;
138 int lease_break_time
= 45;
140 #define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
143 static LIST_HEAD(file_lock_list
);
144 static LIST_HEAD(blocked_list
);
147 * Protects the two list heads above, plus the inode->i_flock list
148 * FIXME: should use a spinlock, once lockd and ceph are ready.
150 void lock_flocks(void)
154 EXPORT_SYMBOL_GPL(lock_flocks
);
156 void unlock_flocks(void)
160 EXPORT_SYMBOL_GPL(unlock_flocks
);
162 static struct kmem_cache
*filelock_cache __read_mostly
;
164 /* Allocate an empty lock structure. */
165 struct file_lock
*locks_alloc_lock(void)
167 return kmem_cache_alloc(filelock_cache
, GFP_KERNEL
);
169 EXPORT_SYMBOL_GPL(locks_alloc_lock
);
171 void locks_release_private(struct file_lock
*fl
)
174 if (fl
->fl_ops
->fl_release_private
)
175 fl
->fl_ops
->fl_release_private(fl
);
179 if (fl
->fl_lmops
->fl_release_private
)
180 fl
->fl_lmops
->fl_release_private(fl
);
185 EXPORT_SYMBOL_GPL(locks_release_private
);
187 /* Free a lock which is not in use. */
188 static void locks_free_lock(struct file_lock
*fl
)
190 BUG_ON(waitqueue_active(&fl
->fl_wait
));
191 BUG_ON(!list_empty(&fl
->fl_block
));
192 BUG_ON(!list_empty(&fl
->fl_link
));
194 locks_release_private(fl
);
195 kmem_cache_free(filelock_cache
, fl
);
198 void locks_init_lock(struct file_lock
*fl
)
200 INIT_LIST_HEAD(&fl
->fl_link
);
201 INIT_LIST_HEAD(&fl
->fl_block
);
202 init_waitqueue_head(&fl
->fl_wait
);
204 fl
->fl_fasync
= NULL
;
211 fl
->fl_start
= fl
->fl_end
= 0;
216 EXPORT_SYMBOL(locks_init_lock
);
219 * Initialises the fields of the file lock which are invariant for
222 static void init_once(void *foo
)
224 struct file_lock
*lock
= (struct file_lock
*) foo
;
226 locks_init_lock(lock
);
229 static void locks_copy_private(struct file_lock
*new, struct file_lock
*fl
)
232 if (fl
->fl_ops
->fl_copy_lock
)
233 fl
->fl_ops
->fl_copy_lock(new, fl
);
234 new->fl_ops
= fl
->fl_ops
;
237 if (fl
->fl_lmops
->fl_copy_lock
)
238 fl
->fl_lmops
->fl_copy_lock(new, fl
);
239 new->fl_lmops
= fl
->fl_lmops
;
244 * Initialize a new lock from an existing file_lock structure.
246 void __locks_copy_lock(struct file_lock
*new, const struct file_lock
*fl
)
248 new->fl_owner
= fl
->fl_owner
;
249 new->fl_pid
= fl
->fl_pid
;
251 new->fl_flags
= fl
->fl_flags
;
252 new->fl_type
= fl
->fl_type
;
253 new->fl_start
= fl
->fl_start
;
254 new->fl_end
= fl
->fl_end
;
256 new->fl_lmops
= NULL
;
258 EXPORT_SYMBOL(__locks_copy_lock
);
260 void locks_copy_lock(struct file_lock
*new, struct file_lock
*fl
)
262 locks_release_private(new);
264 __locks_copy_lock(new, fl
);
265 new->fl_file
= fl
->fl_file
;
266 new->fl_ops
= fl
->fl_ops
;
267 new->fl_lmops
= fl
->fl_lmops
;
269 locks_copy_private(new, fl
);
272 EXPORT_SYMBOL(locks_copy_lock
);
274 static inline int flock_translate_cmd(int cmd
) {
276 return cmd
& (LOCK_MAND
| LOCK_RW
);
288 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
289 static int flock_make_lock(struct file
*filp
, struct file_lock
**lock
,
292 struct file_lock
*fl
;
293 int type
= flock_translate_cmd(cmd
);
297 fl
= locks_alloc_lock();
302 fl
->fl_pid
= current
->tgid
;
303 fl
->fl_flags
= FL_FLOCK
;
305 fl
->fl_end
= OFFSET_MAX
;
311 static int assign_type(struct file_lock
*fl
, int type
)
325 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
328 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
333 switch (l
->l_whence
) {
341 start
= i_size_read(filp
->f_path
.dentry
->d_inode
);
347 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
348 POSIX-2001 defines it. */
352 fl
->fl_end
= OFFSET_MAX
;
354 end
= start
+ l
->l_len
- 1;
356 } else if (l
->l_len
< 0) {
363 fl
->fl_start
= start
; /* we record the absolute position */
364 if (fl
->fl_end
< fl
->fl_start
)
367 fl
->fl_owner
= current
->files
;
368 fl
->fl_pid
= current
->tgid
;
370 fl
->fl_flags
= FL_POSIX
;
374 return assign_type(fl
, l
->l_type
);
377 #if BITS_PER_LONG == 32
378 static int flock64_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
383 switch (l
->l_whence
) {
391 start
= i_size_read(filp
->f_path
.dentry
->d_inode
);
400 fl
->fl_end
= OFFSET_MAX
;
402 fl
->fl_end
= start
+ l
->l_len
- 1;
403 } else if (l
->l_len
< 0) {
404 fl
->fl_end
= start
- 1;
409 fl
->fl_start
= start
; /* we record the absolute position */
410 if (fl
->fl_end
< fl
->fl_start
)
413 fl
->fl_owner
= current
->files
;
414 fl
->fl_pid
= current
->tgid
;
416 fl
->fl_flags
= FL_POSIX
;
424 fl
->fl_type
= l
->l_type
;
434 /* default lease lock manager operations */
435 static void lease_break_callback(struct file_lock
*fl
)
437 kill_fasync(&fl
->fl_fasync
, SIGIO
, POLL_MSG
);
440 static void lease_release_private_callback(struct file_lock
*fl
)
445 f_delown(fl
->fl_file
);
446 fl
->fl_file
->f_owner
.signum
= 0;
449 static int lease_mylease_callback(struct file_lock
*fl
, struct file_lock
*try)
451 return fl
->fl_file
== try->fl_file
;
454 static const struct lock_manager_operations lease_manager_ops
= {
455 .fl_break
= lease_break_callback
,
456 .fl_release_private
= lease_release_private_callback
,
457 .fl_mylease
= lease_mylease_callback
,
458 .fl_change
= lease_modify
,
462 * Initialize a lease, use the default lock manager operations
464 static int lease_init(struct file
*filp
, int type
, struct file_lock
*fl
)
466 if (assign_type(fl
, type
) != 0)
469 fl
->fl_owner
= current
->files
;
470 fl
->fl_pid
= current
->tgid
;
473 fl
->fl_flags
= FL_LEASE
;
475 fl
->fl_end
= OFFSET_MAX
;
477 fl
->fl_lmops
= &lease_manager_ops
;
481 /* Allocate a file_lock initialised to this type of lease */
482 static struct file_lock
*lease_alloc(struct file
*filp
, int type
)
484 struct file_lock
*fl
= locks_alloc_lock();
488 return ERR_PTR(error
);
490 error
= lease_init(filp
, type
, fl
);
493 return ERR_PTR(error
);
498 /* Check if two locks overlap each other.
500 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
502 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
503 (fl2
->fl_end
>= fl1
->fl_start
));
507 * Check whether two locks have the same owner.
509 static int posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
511 if (fl1
->fl_lmops
&& fl1
->fl_lmops
->fl_compare_owner
)
512 return fl2
->fl_lmops
== fl1
->fl_lmops
&&
513 fl1
->fl_lmops
->fl_compare_owner(fl1
, fl2
);
514 return fl1
->fl_owner
== fl2
->fl_owner
;
517 /* Remove waiter from blocker's block list.
518 * When blocker ends up pointing to itself then the list is empty.
520 static void __locks_delete_block(struct file_lock
*waiter
)
522 list_del_init(&waiter
->fl_block
);
523 list_del_init(&waiter
->fl_link
);
524 waiter
->fl_next
= NULL
;
529 static void locks_delete_block(struct file_lock
*waiter
)
532 __locks_delete_block(waiter
);
536 /* Insert waiter into blocker's block list.
537 * We use a circular list so that processes can be easily woken up in
538 * the order they blocked. The documentation doesn't require this but
539 * it seems like the reasonable thing to do.
541 static void locks_insert_block(struct file_lock
*blocker
,
542 struct file_lock
*waiter
)
544 BUG_ON(!list_empty(&waiter
->fl_block
));
545 list_add_tail(&waiter
->fl_block
, &blocker
->fl_block
);
546 waiter
->fl_next
= blocker
;
547 if (IS_POSIX(blocker
))
548 list_add(&waiter
->fl_link
, &blocked_list
);
551 /* Wake up processes blocked waiting for blocker.
552 * If told to wait then schedule the processes until the block list
553 * is empty, otherwise empty the block list ourselves.
555 static void locks_wake_up_blocks(struct file_lock
*blocker
)
557 while (!list_empty(&blocker
->fl_block
)) {
558 struct file_lock
*waiter
;
560 waiter
= list_first_entry(&blocker
->fl_block
,
561 struct file_lock
, fl_block
);
562 __locks_delete_block(waiter
);
563 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->fl_notify
)
564 waiter
->fl_lmops
->fl_notify(waiter
);
566 wake_up(&waiter
->fl_wait
);
570 /* Insert file lock fl into an inode's lock list at the position indicated
571 * by pos. At the same time add the lock to the global file lock list.
573 static void locks_insert_lock(struct file_lock
**pos
, struct file_lock
*fl
)
575 list_add(&fl
->fl_link
, &file_lock_list
);
577 fl
->fl_nspid
= get_pid(task_tgid(current
));
579 /* insert into file's list */
585 * Delete a lock and then free it.
586 * Wake up processes that are blocked waiting for this lock,
587 * notify the FS that the lock has been cleared and
588 * finally free the lock.
590 static void locks_delete_lock(struct file_lock
**thisfl_p
)
592 struct file_lock
*fl
= *thisfl_p
;
594 *thisfl_p
= fl
->fl_next
;
596 list_del_init(&fl
->fl_link
);
598 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
599 if (fl
->fl_fasync
!= NULL
) {
600 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
601 fl
->fl_fasync
= NULL
;
605 put_pid(fl
->fl_nspid
);
609 locks_wake_up_blocks(fl
);
613 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
614 * checks for shared/exclusive status of overlapping locks.
616 static int locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
618 if (sys_fl
->fl_type
== F_WRLCK
)
620 if (caller_fl
->fl_type
== F_WRLCK
)
625 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
626 * checking before calling the locks_conflict().
628 static int posix_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
630 /* POSIX locks owned by the same process do not conflict with
633 if (!IS_POSIX(sys_fl
) || posix_same_owner(caller_fl
, sys_fl
))
636 /* Check whether they overlap */
637 if (!locks_overlap(caller_fl
, sys_fl
))
640 return (locks_conflict(caller_fl
, sys_fl
));
643 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
644 * checking before calling the locks_conflict().
646 static int flock_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
648 /* FLOCK locks referring to the same filp do not conflict with
651 if (!IS_FLOCK(sys_fl
) || (caller_fl
->fl_file
== sys_fl
->fl_file
))
653 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
656 return (locks_conflict(caller_fl
, sys_fl
));
660 posix_test_lock(struct file
*filp
, struct file_lock
*fl
)
662 struct file_lock
*cfl
;
665 for (cfl
= filp
->f_path
.dentry
->d_inode
->i_flock
; cfl
; cfl
= cfl
->fl_next
) {
668 if (posix_locks_conflict(fl
, cfl
))
672 __locks_copy_lock(fl
, cfl
);
674 fl
->fl_pid
= pid_vnr(cfl
->fl_nspid
);
676 fl
->fl_type
= F_UNLCK
;
680 EXPORT_SYMBOL(posix_test_lock
);
683 * Deadlock detection:
685 * We attempt to detect deadlocks that are due purely to posix file
688 * We assume that a task can be waiting for at most one lock at a time.
689 * So for any acquired lock, the process holding that lock may be
690 * waiting on at most one other lock. That lock in turns may be held by
691 * someone waiting for at most one other lock. Given a requested lock
692 * caller_fl which is about to wait for a conflicting lock block_fl, we
693 * follow this chain of waiters to ensure we are not about to create a
696 * Since we do this before we ever put a process to sleep on a lock, we
697 * are ensured that there is never a cycle; that is what guarantees that
698 * the while() loop in posix_locks_deadlock() eventually completes.
700 * Note: the above assumption may not be true when handling lock
701 * requests from a broken NFS client. It may also fail in the presence
702 * of tasks (such as posix threads) sharing the same open file table.
704 * To handle those cases, we just bail out after a few iterations.
707 #define MAX_DEADLK_ITERATIONS 10
709 /* Find a lock that the owner of the given block_fl is blocking on. */
710 static struct file_lock
*what_owner_is_waiting_for(struct file_lock
*block_fl
)
712 struct file_lock
*fl
;
714 list_for_each_entry(fl
, &blocked_list
, fl_link
) {
715 if (posix_same_owner(fl
, block_fl
))
721 static int posix_locks_deadlock(struct file_lock
*caller_fl
,
722 struct file_lock
*block_fl
)
726 while ((block_fl
= what_owner_is_waiting_for(block_fl
))) {
727 if (i
++ > MAX_DEADLK_ITERATIONS
)
729 if (posix_same_owner(caller_fl
, block_fl
))
735 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
736 * after any leases, but before any posix locks.
738 * Note that if called with an FL_EXISTS argument, the caller may determine
739 * whether or not a lock was successfully freed by testing the return
742 static int flock_lock_file(struct file
*filp
, struct file_lock
*request
)
744 struct file_lock
*new_fl
= NULL
;
745 struct file_lock
**before
;
746 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
750 if (!(request
->fl_flags
& FL_ACCESS
) && (request
->fl_type
!= F_UNLCK
)) {
751 new_fl
= locks_alloc_lock();
757 if (request
->fl_flags
& FL_ACCESS
)
760 for_each_lock(inode
, before
) {
761 struct file_lock
*fl
= *before
;
766 if (filp
!= fl
->fl_file
)
768 if (request
->fl_type
== fl
->fl_type
)
771 locks_delete_lock(before
);
775 if (request
->fl_type
== F_UNLCK
) {
776 if ((request
->fl_flags
& FL_EXISTS
) && !found
)
782 * If a higher-priority process was blocked on the old file lock,
783 * give it the opportunity to lock the file.
792 for_each_lock(inode
, before
) {
793 struct file_lock
*fl
= *before
;
798 if (!flock_locks_conflict(request
, fl
))
801 if (!(request
->fl_flags
& FL_SLEEP
))
803 error
= FILE_LOCK_DEFERRED
;
804 locks_insert_block(fl
, request
);
807 if (request
->fl_flags
& FL_ACCESS
)
809 locks_copy_lock(new_fl
, request
);
810 locks_insert_lock(before
, new_fl
);
817 locks_free_lock(new_fl
);
821 static int __posix_lock_file(struct inode
*inode
, struct file_lock
*request
, struct file_lock
*conflock
)
823 struct file_lock
*fl
;
824 struct file_lock
*new_fl
= NULL
;
825 struct file_lock
*new_fl2
= NULL
;
826 struct file_lock
*left
= NULL
;
827 struct file_lock
*right
= NULL
;
828 struct file_lock
**before
;
829 int error
, added
= 0;
832 * We may need two file_lock structures for this operation,
833 * so we get them in advance to avoid races.
835 * In some cases we can be sure, that no new locks will be needed
837 if (!(request
->fl_flags
& FL_ACCESS
) &&
838 (request
->fl_type
!= F_UNLCK
||
839 request
->fl_start
!= 0 || request
->fl_end
!= OFFSET_MAX
)) {
840 new_fl
= locks_alloc_lock();
841 new_fl2
= locks_alloc_lock();
845 if (request
->fl_type
!= F_UNLCK
) {
846 for_each_lock(inode
, before
) {
850 if (!posix_locks_conflict(request
, fl
))
853 __locks_copy_lock(conflock
, fl
);
855 if (!(request
->fl_flags
& FL_SLEEP
))
858 if (posix_locks_deadlock(request
, fl
))
860 error
= FILE_LOCK_DEFERRED
;
861 locks_insert_block(fl
, request
);
866 /* If we're just looking for a conflict, we're done. */
868 if (request
->fl_flags
& FL_ACCESS
)
872 * Find the first old lock with the same owner as the new lock.
875 before
= &inode
->i_flock
;
877 /* First skip locks owned by other processes. */
878 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
879 !posix_same_owner(request
, fl
))) {
880 before
= &fl
->fl_next
;
883 /* Process locks with this owner. */
884 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
885 /* Detect adjacent or overlapping regions (if same lock type)
887 if (request
->fl_type
== fl
->fl_type
) {
888 /* In all comparisons of start vs end, use
889 * "start - 1" rather than "end + 1". If end
890 * is OFFSET_MAX, end + 1 will become negative.
892 if (fl
->fl_end
< request
->fl_start
- 1)
894 /* If the next lock in the list has entirely bigger
895 * addresses than the new one, insert the lock here.
897 if (fl
->fl_start
- 1 > request
->fl_end
)
900 /* If we come here, the new and old lock are of the
901 * same type and adjacent or overlapping. Make one
902 * lock yielding from the lower start address of both
903 * locks to the higher end address.
905 if (fl
->fl_start
> request
->fl_start
)
906 fl
->fl_start
= request
->fl_start
;
908 request
->fl_start
= fl
->fl_start
;
909 if (fl
->fl_end
< request
->fl_end
)
910 fl
->fl_end
= request
->fl_end
;
912 request
->fl_end
= fl
->fl_end
;
914 locks_delete_lock(before
);
921 /* Processing for different lock types is a bit
924 if (fl
->fl_end
< request
->fl_start
)
926 if (fl
->fl_start
> request
->fl_end
)
928 if (request
->fl_type
== F_UNLCK
)
930 if (fl
->fl_start
< request
->fl_start
)
932 /* If the next lock in the list has a higher end
933 * address than the new one, insert the new one here.
935 if (fl
->fl_end
> request
->fl_end
) {
939 if (fl
->fl_start
>= request
->fl_start
) {
940 /* The new lock completely replaces an old
941 * one (This may happen several times).
944 locks_delete_lock(before
);
947 /* Replace the old lock with the new one.
948 * Wake up anybody waiting for the old one,
949 * as the change in lock type might satisfy
952 locks_wake_up_blocks(fl
);
953 fl
->fl_start
= request
->fl_start
;
954 fl
->fl_end
= request
->fl_end
;
955 fl
->fl_type
= request
->fl_type
;
956 locks_release_private(fl
);
957 locks_copy_private(fl
, request
);
962 /* Go on to next lock.
965 before
= &fl
->fl_next
;
969 * The above code only modifies existing locks in case of
970 * merging or replacing. If new lock(s) need to be inserted
971 * all modifications are done bellow this, so it's safe yet to
974 error
= -ENOLCK
; /* "no luck" */
975 if (right
&& left
== right
&& !new_fl2
)
980 if (request
->fl_type
== F_UNLCK
) {
981 if (request
->fl_flags
& FL_EXISTS
)
990 locks_copy_lock(new_fl
, request
);
991 locks_insert_lock(before
, new_fl
);
996 /* The new lock breaks the old one in two pieces,
997 * so we have to use the second new lock.
1001 locks_copy_lock(left
, right
);
1002 locks_insert_lock(before
, left
);
1004 right
->fl_start
= request
->fl_end
+ 1;
1005 locks_wake_up_blocks(right
);
1008 left
->fl_end
= request
->fl_start
- 1;
1009 locks_wake_up_blocks(left
);
1014 * Free any unused locks.
1017 locks_free_lock(new_fl
);
1019 locks_free_lock(new_fl2
);
1024 * posix_lock_file - Apply a POSIX-style lock to a file
1025 * @filp: The file to apply the lock to
1026 * @fl: The lock to be applied
1027 * @conflock: Place to return a copy of the conflicting lock, if found.
1029 * Add a POSIX style lock to a file.
1030 * We merge adjacent & overlapping locks whenever possible.
1031 * POSIX locks are sorted by owner task, then by starting address
1033 * Note that if called with an FL_EXISTS argument, the caller may determine
1034 * whether or not a lock was successfully freed by testing the return
1035 * value for -ENOENT.
1037 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
,
1038 struct file_lock
*conflock
)
1040 return __posix_lock_file(filp
->f_path
.dentry
->d_inode
, fl
, conflock
);
1042 EXPORT_SYMBOL(posix_lock_file
);
1045 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1046 * @filp: The file to apply the lock to
1047 * @fl: The lock to be applied
1049 * Add a POSIX style lock to a file.
1050 * We merge adjacent & overlapping locks whenever possible.
1051 * POSIX locks are sorted by owner task, then by starting address
1053 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1058 error
= posix_lock_file(filp
, fl
, NULL
);
1059 if (error
!= FILE_LOCK_DEFERRED
)
1061 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1065 locks_delete_block(fl
);
1070 EXPORT_SYMBOL(posix_lock_file_wait
);
1073 * locks_mandatory_locked - Check for an active lock
1074 * @inode: the file to check
1076 * Searches the inode's list of locks to find any POSIX locks which conflict.
1077 * This function is called from locks_verify_locked() only.
1079 int locks_mandatory_locked(struct inode
*inode
)
1081 fl_owner_t owner
= current
->files
;
1082 struct file_lock
*fl
;
1085 * Search the lock list for this inode for any POSIX locks.
1088 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1091 if (fl
->fl_owner
!= owner
)
1095 return fl
? -EAGAIN
: 0;
1099 * locks_mandatory_area - Check for a conflicting lock
1100 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1102 * @inode: the file to check
1103 * @filp: how the file was opened (if it was)
1104 * @offset: start of area to check
1105 * @count: length of area to check
1107 * Searches the inode's list of locks to find any POSIX locks which conflict.
1108 * This function is called from rw_verify_area() and
1109 * locks_verify_truncate().
1111 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1112 struct file
*filp
, loff_t offset
,
1115 struct file_lock fl
;
1118 locks_init_lock(&fl
);
1119 fl
.fl_owner
= current
->files
;
1120 fl
.fl_pid
= current
->tgid
;
1122 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1123 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1124 fl
.fl_flags
|= FL_SLEEP
;
1125 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1126 fl
.fl_start
= offset
;
1127 fl
.fl_end
= offset
+ count
- 1;
1130 error
= __posix_lock_file(inode
, &fl
, NULL
);
1131 if (error
!= FILE_LOCK_DEFERRED
)
1133 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1136 * If we've been sleeping someone might have
1137 * changed the permissions behind our back.
1139 if (__mandatory_lock(inode
))
1143 locks_delete_block(&fl
);
1150 EXPORT_SYMBOL(locks_mandatory_area
);
1152 /* We already had a lease on this file; just change its type */
1153 int lease_modify(struct file_lock
**before
, int arg
)
1155 struct file_lock
*fl
= *before
;
1156 int error
= assign_type(fl
, arg
);
1160 locks_wake_up_blocks(fl
);
1162 locks_delete_lock(before
);
1166 EXPORT_SYMBOL(lease_modify
);
1168 static void time_out_leases(struct inode
*inode
)
1170 struct file_lock
**before
;
1171 struct file_lock
*fl
;
1173 before
= &inode
->i_flock
;
1174 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1175 if ((fl
->fl_break_time
== 0)
1176 || time_before(jiffies
, fl
->fl_break_time
)) {
1177 before
= &fl
->fl_next
;
1180 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1181 if (fl
== *before
) /* lease_modify may have freed fl */
1182 before
= &fl
->fl_next
;
1187 * __break_lease - revoke all outstanding leases on file
1188 * @inode: the inode of the file to return
1189 * @mode: the open mode (read or write)
1191 * break_lease (inlined for speed) has checked there already is at least
1192 * some kind of lock (maybe a lease) on this file. Leases are broken on
1193 * a call to open() or truncate(). This function can sleep unless you
1194 * specified %O_NONBLOCK to your open().
1196 int __break_lease(struct inode
*inode
, unsigned int mode
)
1198 int error
= 0, future
;
1199 struct file_lock
*new_fl
, *flock
;
1200 struct file_lock
*fl
;
1201 unsigned long break_time
;
1202 int i_have_this_lease
= 0;
1203 int want_write
= (mode
& O_ACCMODE
) != O_RDONLY
;
1205 new_fl
= lease_alloc(NULL
, want_write
? F_WRLCK
: F_RDLCK
);
1209 time_out_leases(inode
);
1211 flock
= inode
->i_flock
;
1212 if ((flock
== NULL
) || !IS_LEASE(flock
))
1215 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
)
1216 if (fl
->fl_owner
== current
->files
)
1217 i_have_this_lease
= 1;
1220 /* If we want write access, we have to revoke any lease. */
1221 future
= F_UNLCK
| F_INPROGRESS
;
1222 } else if (flock
->fl_type
& F_INPROGRESS
) {
1223 /* If the lease is already being broken, we just leave it */
1224 future
= flock
->fl_type
;
1225 } else if (flock
->fl_type
& F_WRLCK
) {
1226 /* Downgrade the exclusive lease to a read-only lease. */
1227 future
= F_RDLCK
| F_INPROGRESS
;
1229 /* the existing lease was read-only, so we can read too. */
1233 if (IS_ERR(new_fl
) && !i_have_this_lease
1234 && ((mode
& O_NONBLOCK
) == 0)) {
1235 error
= PTR_ERR(new_fl
);
1240 if (lease_break_time
> 0) {
1241 break_time
= jiffies
+ lease_break_time
* HZ
;
1242 if (break_time
== 0)
1243 break_time
++; /* so that 0 means no break time */
1246 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1247 if (fl
->fl_type
!= future
) {
1248 fl
->fl_type
= future
;
1249 fl
->fl_break_time
= break_time
;
1250 /* lease must have lmops break callback */
1251 fl
->fl_lmops
->fl_break(fl
);
1255 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1256 error
= -EWOULDBLOCK
;
1261 break_time
= flock
->fl_break_time
;
1262 if (break_time
!= 0) {
1263 break_time
-= jiffies
;
1264 if (break_time
== 0)
1267 locks_insert_block(flock
, new_fl
);
1269 error
= wait_event_interruptible_timeout(new_fl
->fl_wait
,
1270 !new_fl
->fl_next
, break_time
);
1272 __locks_delete_block(new_fl
);
1275 time_out_leases(inode
);
1276 /* Wait for the next lease that has not been broken yet */
1277 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1278 flock
= flock
->fl_next
) {
1279 if (flock
->fl_type
& F_INPROGRESS
)
1287 if (!IS_ERR(new_fl
))
1288 locks_free_lock(new_fl
);
1292 EXPORT_SYMBOL(__break_lease
);
1295 * lease_get_mtime - get the last modified time of an inode
1297 * @time: pointer to a timespec which will contain the last modified time
1299 * This is to force NFS clients to flush their caches for files with
1300 * exclusive leases. The justification is that if someone has an
1301 * exclusive lease, then they could be modifying it.
1303 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1305 struct file_lock
*flock
= inode
->i_flock
;
1306 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1307 *time
= current_fs_time(inode
->i_sb
);
1309 *time
= inode
->i_mtime
;
1312 EXPORT_SYMBOL(lease_get_mtime
);
1315 * fcntl_getlease - Enquire what lease is currently active
1318 * The value returned by this function will be one of
1319 * (if no lease break is pending):
1321 * %F_RDLCK to indicate a shared lease is held.
1323 * %F_WRLCK to indicate an exclusive lease is held.
1325 * %F_UNLCK to indicate no lease is held.
1327 * (if a lease break is pending):
1329 * %F_RDLCK to indicate an exclusive lease needs to be
1330 * changed to a shared lease (or removed).
1332 * %F_UNLCK to indicate the lease needs to be removed.
1334 * XXX: sfr & willy disagree over whether F_INPROGRESS
1335 * should be returned to userspace.
1337 int fcntl_getlease(struct file
*filp
)
1339 struct file_lock
*fl
;
1343 time_out_leases(filp
->f_path
.dentry
->d_inode
);
1344 for (fl
= filp
->f_path
.dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1346 if (fl
->fl_file
== filp
) {
1347 type
= fl
->fl_type
& ~F_INPROGRESS
;
1356 * generic_setlease - sets a lease on an open file
1357 * @filp: file pointer
1358 * @arg: type of lease to obtain
1359 * @flp: input - file_lock to use, output - file_lock inserted
1361 * The (input) flp->fl_lmops->fl_break function is required
1364 * Called with file_lock_lock held.
1366 int generic_setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1368 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1369 struct dentry
*dentry
= filp
->f_path
.dentry
;
1370 struct inode
*inode
= dentry
->d_inode
;
1371 int error
, rdlease_count
= 0, wrlease_count
= 0;
1373 if ((current_fsuid() != inode
->i_uid
) && !capable(CAP_LEASE
))
1375 if (!S_ISREG(inode
->i_mode
))
1377 error
= security_file_lock(filp
, arg
);
1381 time_out_leases(inode
);
1383 BUG_ON(!(*flp
)->fl_lmops
->fl_break
);
1387 if (arg
!= F_UNLCK
) {
1389 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1391 if ((arg
== F_WRLCK
)
1392 && ((atomic_read(&dentry
->d_count
) > 1)
1393 || (atomic_read(&inode
->i_count
) > 1)))
1398 * At this point, we know that if there is an exclusive
1399 * lease on this file, then we hold it on this filp
1400 * (otherwise our open of this file would have blocked).
1401 * And if we are trying to acquire an exclusive lease,
1402 * then the file is not open by anyone (including us)
1403 * except for this filp.
1405 for (before
= &inode
->i_flock
;
1406 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1407 before
= &fl
->fl_next
) {
1408 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1410 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1412 * Someone is in the process of opening this
1413 * file for writing so we may not take an
1414 * exclusive lease on it.
1422 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1423 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1426 if (my_before
!= NULL
) {
1428 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1439 locks_insert_lock(before
, lease
);
1443 locks_free_lock(lease
);
1446 EXPORT_SYMBOL(generic_setlease
);
1448 static int __vfs_setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1450 if (filp
->f_op
&& filp
->f_op
->setlease
)
1451 return filp
->f_op
->setlease(filp
, arg
, lease
);
1453 return generic_setlease(filp
, arg
, lease
);
1457 * vfs_setlease - sets a lease on an open file
1458 * @filp: file pointer
1459 * @arg: type of lease to obtain
1460 * @lease: file_lock to use
1462 * Call this to establish a lease on the file.
1463 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1464 * break_lease will oops!
1466 * This will call the filesystem's setlease file method, if
1467 * defined. Note that there is no getlease method; instead, the
1468 * filesystem setlease method should call back to setlease() to
1469 * add a lease to the inode's lease list, where fcntl_getlease() can
1470 * find it. Since fcntl_getlease() only reports whether the current
1471 * task holds a lease, a cluster filesystem need only do this for
1472 * leases held by processes on this node.
1474 * There is also no break_lease method; filesystems that
1475 * handle their own leases should break leases themselves from the
1476 * filesystem's open, create, and (on truncate) setattr methods.
1478 * Warning: the only current setlease methods exist only to disable
1479 * leases in certain cases. More vfs changes may be required to
1480 * allow a full filesystem lease implementation.
1483 int vfs_setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1488 error
= __vfs_setlease(filp
, arg
, lease
);
1493 EXPORT_SYMBOL_GPL(vfs_setlease
);
1496 * fcntl_setlease - sets a lease on an open file
1497 * @fd: open file descriptor
1498 * @filp: file pointer
1499 * @arg: type of lease to obtain
1501 * Call this fcntl to establish a lease on the file.
1502 * Note that you also need to call %F_SETSIG to
1503 * receive a signal when the lease is broken.
1505 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1507 struct file_lock
*fl
;
1508 struct fasync_struct
*new;
1509 struct inode
*inode
= filp
->f_path
.dentry
->d_inode
;
1512 fl
= lease_alloc(filp
, arg
);
1516 new = fasync_alloc();
1518 locks_free_lock(fl
);
1522 error
= __vfs_setlease(filp
, arg
, &fl
);
1523 if (error
|| arg
== F_UNLCK
)
1527 * fasync_insert_entry() returns the old entry if any.
1528 * If there was no old entry, then it used 'new' and
1529 * inserted it into the fasync list. Clear new so that
1530 * we don't release it here.
1532 if (!fasync_insert_entry(fd
, filp
, &fl
->fl_fasync
, new))
1536 /* remove lease just inserted by setlease */
1537 fl
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1538 fl
->fl_break_time
= jiffies
- 10;
1539 time_out_leases(inode
);
1543 error
= __f_setown(filp
, task_pid(current
), PIDTYPE_PID
, 0);
1552 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1553 * @filp: The file to apply the lock to
1554 * @fl: The lock to be applied
1556 * Add a FLOCK style lock to a file.
1558 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1563 error
= flock_lock_file(filp
, fl
);
1564 if (error
!= FILE_LOCK_DEFERRED
)
1566 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1570 locks_delete_block(fl
);
1576 EXPORT_SYMBOL(flock_lock_file_wait
);
1579 * sys_flock: - flock() system call.
1580 * @fd: the file descriptor to lock.
1581 * @cmd: the type of lock to apply.
1583 * Apply a %FL_FLOCK style lock to an open file descriptor.
1584 * The @cmd can be one of
1586 * %LOCK_SH -- a shared lock.
1588 * %LOCK_EX -- an exclusive lock.
1590 * %LOCK_UN -- remove an existing lock.
1592 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1594 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1595 * processes read and write access respectively.
1597 SYSCALL_DEFINE2(flock
, unsigned int, fd
, unsigned int, cmd
)
1600 struct file_lock
*lock
;
1601 int can_sleep
, unlock
;
1609 can_sleep
= !(cmd
& LOCK_NB
);
1611 unlock
= (cmd
== LOCK_UN
);
1613 if (!unlock
&& !(cmd
& LOCK_MAND
) &&
1614 !(filp
->f_mode
& (FMODE_READ
|FMODE_WRITE
)))
1617 error
= flock_make_lock(filp
, &lock
, cmd
);
1621 lock
->fl_flags
|= FL_SLEEP
;
1623 error
= security_file_lock(filp
, lock
->fl_type
);
1627 if (filp
->f_op
&& filp
->f_op
->flock
)
1628 error
= filp
->f_op
->flock(filp
,
1629 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1632 error
= flock_lock_file_wait(filp
, lock
);
1635 locks_free_lock(lock
);
1644 * vfs_test_lock - test file byte range lock
1645 * @filp: The file to test lock for
1646 * @fl: The lock to test; also used to hold result
1648 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1649 * setting conf->fl_type to something other than F_UNLCK.
1651 int vfs_test_lock(struct file
*filp
, struct file_lock
*fl
)
1653 if (filp
->f_op
&& filp
->f_op
->lock
)
1654 return filp
->f_op
->lock(filp
, F_GETLK
, fl
);
1655 posix_test_lock(filp
, fl
);
1658 EXPORT_SYMBOL_GPL(vfs_test_lock
);
1660 static int posix_lock_to_flock(struct flock
*flock
, struct file_lock
*fl
)
1662 flock
->l_pid
= fl
->fl_pid
;
1663 #if BITS_PER_LONG == 32
1665 * Make sure we can represent the posix lock via
1666 * legacy 32bit flock.
1668 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1670 if (fl
->fl_end
!= OFFSET_MAX
&& fl
->fl_end
> OFFT_OFFSET_MAX
)
1673 flock
->l_start
= fl
->fl_start
;
1674 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1675 fl
->fl_end
- fl
->fl_start
+ 1;
1676 flock
->l_whence
= 0;
1677 flock
->l_type
= fl
->fl_type
;
1681 #if BITS_PER_LONG == 32
1682 static void posix_lock_to_flock64(struct flock64
*flock
, struct file_lock
*fl
)
1684 flock
->l_pid
= fl
->fl_pid
;
1685 flock
->l_start
= fl
->fl_start
;
1686 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1687 fl
->fl_end
- fl
->fl_start
+ 1;
1688 flock
->l_whence
= 0;
1689 flock
->l_type
= fl
->fl_type
;
1693 /* Report the first existing lock that would conflict with l.
1694 * This implements the F_GETLK command of fcntl().
1696 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1698 struct file_lock file_lock
;
1703 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1706 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1709 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1713 error
= vfs_test_lock(filp
, &file_lock
);
1717 flock
.l_type
= file_lock
.fl_type
;
1718 if (file_lock
.fl_type
!= F_UNLCK
) {
1719 error
= posix_lock_to_flock(&flock
, &file_lock
);
1724 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1731 * vfs_lock_file - file byte range lock
1732 * @filp: The file to apply the lock to
1733 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1734 * @fl: The lock to be applied
1735 * @conf: Place to return a copy of the conflicting lock, if found.
1737 * A caller that doesn't care about the conflicting lock may pass NULL
1738 * as the final argument.
1740 * If the filesystem defines a private ->lock() method, then @conf will
1741 * be left unchanged; so a caller that cares should initialize it to
1742 * some acceptable default.
1744 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1745 * locks, the ->lock() interface may return asynchronously, before the lock has
1746 * been granted or denied by the underlying filesystem, if (and only if)
1747 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1748 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1749 * the request is for a blocking lock. When ->lock() does return asynchronously,
1750 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
1751 * request completes.
1752 * If the request is for non-blocking lock the file system should return
1753 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1754 * with the result. If the request timed out the callback routine will return a
1755 * nonzero return code and the file system should release the lock. The file
1756 * system is also responsible to keep a corresponding posix lock when it
1757 * grants a lock so the VFS can find out which locks are locally held and do
1758 * the correct lock cleanup when required.
1759 * The underlying filesystem must not drop the kernel lock or call
1760 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1763 int vfs_lock_file(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
, struct file_lock
*conf
)
1765 if (filp
->f_op
&& filp
->f_op
->lock
)
1766 return filp
->f_op
->lock(filp
, cmd
, fl
);
1768 return posix_lock_file(filp
, fl
, conf
);
1770 EXPORT_SYMBOL_GPL(vfs_lock_file
);
1772 static int do_lock_file_wait(struct file
*filp
, unsigned int cmd
,
1773 struct file_lock
*fl
)
1777 error
= security_file_lock(filp
, fl
->fl_type
);
1782 error
= vfs_lock_file(filp
, cmd
, fl
, NULL
);
1783 if (error
!= FILE_LOCK_DEFERRED
)
1785 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1789 locks_delete_block(fl
);
1796 /* Apply the lock described by l to an open file descriptor.
1797 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1799 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1800 struct flock __user
*l
)
1802 struct file_lock
*file_lock
= locks_alloc_lock();
1804 struct inode
*inode
;
1808 if (file_lock
== NULL
)
1812 * This might block, so we do it before checking the inode.
1815 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1818 inode
= filp
->f_path
.dentry
->d_inode
;
1820 /* Don't allow mandatory locks on files that may be memory mapped
1823 if (mandatory_lock(inode
) && mapping_writably_mapped(filp
->f_mapping
)) {
1829 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1832 if (cmd
== F_SETLKW
) {
1833 file_lock
->fl_flags
|= FL_SLEEP
;
1837 switch (flock
.l_type
) {
1839 if (!(filp
->f_mode
& FMODE_READ
))
1843 if (!(filp
->f_mode
& FMODE_WRITE
))
1853 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
1856 * Attempt to detect a close/fcntl race and recover by
1857 * releasing the lock that was just acquired.
1860 * we need that spin_lock here - it prevents reordering between
1861 * update of inode->i_flock and check for it done in close().
1862 * rcu_read_lock() wouldn't do.
1864 spin_lock(¤t
->files
->file_lock
);
1866 spin_unlock(¤t
->files
->file_lock
);
1867 if (!error
&& f
!= filp
&& flock
.l_type
!= F_UNLCK
) {
1868 flock
.l_type
= F_UNLCK
;
1873 locks_free_lock(file_lock
);
1877 #if BITS_PER_LONG == 32
1878 /* Report the first existing lock that would conflict with l.
1879 * This implements the F_GETLK command of fcntl().
1881 int fcntl_getlk64(struct file
*filp
, struct flock64 __user
*l
)
1883 struct file_lock file_lock
;
1884 struct flock64 flock
;
1888 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1891 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1894 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1898 error
= vfs_test_lock(filp
, &file_lock
);
1902 flock
.l_type
= file_lock
.fl_type
;
1903 if (file_lock
.fl_type
!= F_UNLCK
)
1904 posix_lock_to_flock64(&flock
, &file_lock
);
1907 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1914 /* Apply the lock described by l to an open file descriptor.
1915 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1917 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1918 struct flock64 __user
*l
)
1920 struct file_lock
*file_lock
= locks_alloc_lock();
1921 struct flock64 flock
;
1922 struct inode
*inode
;
1926 if (file_lock
== NULL
)
1930 * This might block, so we do it before checking the inode.
1933 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1936 inode
= filp
->f_path
.dentry
->d_inode
;
1938 /* Don't allow mandatory locks on files that may be memory mapped
1941 if (mandatory_lock(inode
) && mapping_writably_mapped(filp
->f_mapping
)) {
1947 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1950 if (cmd
== F_SETLKW64
) {
1951 file_lock
->fl_flags
|= FL_SLEEP
;
1955 switch (flock
.l_type
) {
1957 if (!(filp
->f_mode
& FMODE_READ
))
1961 if (!(filp
->f_mode
& FMODE_WRITE
))
1971 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
1974 * Attempt to detect a close/fcntl race and recover by
1975 * releasing the lock that was just acquired.
1977 spin_lock(¤t
->files
->file_lock
);
1979 spin_unlock(¤t
->files
->file_lock
);
1980 if (!error
&& f
!= filp
&& flock
.l_type
!= F_UNLCK
) {
1981 flock
.l_type
= F_UNLCK
;
1986 locks_free_lock(file_lock
);
1989 #endif /* BITS_PER_LONG == 32 */
1992 * This function is called when the file is being removed
1993 * from the task's fd array. POSIX locks belonging to this task
1994 * are deleted at this time.
1996 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
1998 struct file_lock lock
;
2001 * If there are no locks held on this file, we don't need to call
2002 * posix_lock_file(). Another process could be setting a lock on this
2003 * file at the same time, but we wouldn't remove that lock anyway.
2005 if (!filp
->f_path
.dentry
->d_inode
->i_flock
)
2008 lock
.fl_type
= F_UNLCK
;
2009 lock
.fl_flags
= FL_POSIX
| FL_CLOSE
;
2011 lock
.fl_end
= OFFSET_MAX
;
2012 lock
.fl_owner
= owner
;
2013 lock
.fl_pid
= current
->tgid
;
2014 lock
.fl_file
= filp
;
2016 lock
.fl_lmops
= NULL
;
2018 vfs_lock_file(filp
, F_SETLK
, &lock
, NULL
);
2020 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
2021 lock
.fl_ops
->fl_release_private(&lock
);
2024 EXPORT_SYMBOL(locks_remove_posix
);
2027 * This function is called on the last close of an open file.
2029 void locks_remove_flock(struct file
*filp
)
2031 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
2032 struct file_lock
*fl
;
2033 struct file_lock
**before
;
2035 if (!inode
->i_flock
)
2038 if (filp
->f_op
&& filp
->f_op
->flock
) {
2039 struct file_lock fl
= {
2040 .fl_pid
= current
->tgid
,
2042 .fl_flags
= FL_FLOCK
,
2044 .fl_end
= OFFSET_MAX
,
2046 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
2047 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
2048 fl
.fl_ops
->fl_release_private(&fl
);
2052 before
= &inode
->i_flock
;
2054 while ((fl
= *before
) != NULL
) {
2055 if (fl
->fl_file
== filp
) {
2057 locks_delete_lock(before
);
2061 lease_modify(before
, F_UNLCK
);
2067 before
= &fl
->fl_next
;
2073 * posix_unblock_lock - stop waiting for a file lock
2074 * @filp: how the file was opened
2075 * @waiter: the lock which was waiting
2077 * lockd needs to block waiting for locks.
2080 posix_unblock_lock(struct file
*filp
, struct file_lock
*waiter
)
2085 if (waiter
->fl_next
)
2086 __locks_delete_block(waiter
);
2093 EXPORT_SYMBOL(posix_unblock_lock
);
2096 * vfs_cancel_lock - file byte range unblock lock
2097 * @filp: The file to apply the unblock to
2098 * @fl: The lock to be unblocked
2100 * Used by lock managers to cancel blocked requests
2102 int vfs_cancel_lock(struct file
*filp
, struct file_lock
*fl
)
2104 if (filp
->f_op
&& filp
->f_op
->lock
)
2105 return filp
->f_op
->lock(filp
, F_CANCELLK
, fl
);
2109 EXPORT_SYMBOL_GPL(vfs_cancel_lock
);
2111 #ifdef CONFIG_PROC_FS
2112 #include <linux/proc_fs.h>
2113 #include <linux/seq_file.h>
2115 static void lock_get_status(struct seq_file
*f
, struct file_lock
*fl
,
2118 struct inode
*inode
= NULL
;
2119 unsigned int fl_pid
;
2122 fl_pid
= pid_vnr(fl
->fl_nspid
);
2124 fl_pid
= fl
->fl_pid
;
2126 if (fl
->fl_file
!= NULL
)
2127 inode
= fl
->fl_file
->f_path
.dentry
->d_inode
;
2129 seq_printf(f
, "%d:%s ", id
, pfx
);
2131 seq_printf(f
, "%6s %s ",
2132 (fl
->fl_flags
& FL_ACCESS
) ? "ACCESS" : "POSIX ",
2133 (inode
== NULL
) ? "*NOINODE*" :
2134 mandatory_lock(inode
) ? "MANDATORY" : "ADVISORY ");
2135 } else if (IS_FLOCK(fl
)) {
2136 if (fl
->fl_type
& LOCK_MAND
) {
2137 seq_printf(f
, "FLOCK MSNFS ");
2139 seq_printf(f
, "FLOCK ADVISORY ");
2141 } else if (IS_LEASE(fl
)) {
2142 seq_printf(f
, "LEASE ");
2143 if (fl
->fl_type
& F_INPROGRESS
)
2144 seq_printf(f
, "BREAKING ");
2145 else if (fl
->fl_file
)
2146 seq_printf(f
, "ACTIVE ");
2148 seq_printf(f
, "BREAKER ");
2150 seq_printf(f
, "UNKNOWN UNKNOWN ");
2152 if (fl
->fl_type
& LOCK_MAND
) {
2153 seq_printf(f
, "%s ",
2154 (fl
->fl_type
& LOCK_READ
)
2155 ? (fl
->fl_type
& LOCK_WRITE
) ? "RW " : "READ "
2156 : (fl
->fl_type
& LOCK_WRITE
) ? "WRITE" : "NONE ");
2158 seq_printf(f
, "%s ",
2159 (fl
->fl_type
& F_INPROGRESS
)
2160 ? (fl
->fl_type
& F_UNLCK
) ? "UNLCK" : "READ "
2161 : (fl
->fl_type
& F_WRLCK
) ? "WRITE" : "READ ");
2164 #ifdef WE_CAN_BREAK_LSLK_NOW
2165 seq_printf(f
, "%d %s:%ld ", fl_pid
,
2166 inode
->i_sb
->s_id
, inode
->i_ino
);
2168 /* userspace relies on this representation of dev_t ;-( */
2169 seq_printf(f
, "%d %02x:%02x:%ld ", fl_pid
,
2170 MAJOR(inode
->i_sb
->s_dev
),
2171 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2174 seq_printf(f
, "%d <none>:0 ", fl_pid
);
2177 if (fl
->fl_end
== OFFSET_MAX
)
2178 seq_printf(f
, "%Ld EOF\n", fl
->fl_start
);
2180 seq_printf(f
, "%Ld %Ld\n", fl
->fl_start
, fl
->fl_end
);
2182 seq_printf(f
, "0 EOF\n");
2186 static int locks_show(struct seq_file
*f
, void *v
)
2188 struct file_lock
*fl
, *bfl
;
2190 fl
= list_entry(v
, struct file_lock
, fl_link
);
2192 lock_get_status(f
, fl
, (long)f
->private, "");
2194 list_for_each_entry(bfl
, &fl
->fl_block
, fl_block
)
2195 lock_get_status(f
, bfl
, (long)f
->private, " ->");
2201 static void *locks_start(struct seq_file
*f
, loff_t
*pos
)
2204 f
->private = (void *)1;
2205 return seq_list_start(&file_lock_list
, *pos
);
2208 static void *locks_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2210 return seq_list_next(v
, &file_lock_list
, pos
);
2213 static void locks_stop(struct seq_file
*f
, void *v
)
2218 static const struct seq_operations locks_seq_operations
= {
2219 .start
= locks_start
,
2225 static int locks_open(struct inode
*inode
, struct file
*filp
)
2227 return seq_open(filp
, &locks_seq_operations
);
2230 static const struct file_operations proc_locks_operations
= {
2233 .llseek
= seq_lseek
,
2234 .release
= seq_release
,
2237 static int __init
proc_locks_init(void)
2239 proc_create("locks", 0, NULL
, &proc_locks_operations
);
2242 module_init(proc_locks_init
);
2246 * lock_may_read - checks that the region is free of locks
2247 * @inode: the inode that is being read
2248 * @start: the first byte to read
2249 * @len: the number of bytes to read
2251 * Emulates Windows locking requirements. Whole-file
2252 * mandatory locks (share modes) can prohibit a read and
2253 * byte-range POSIX locks can prohibit a read if they overlap.
2255 * N.B. this function is only ever called
2256 * from knfsd and ownership of locks is never checked.
2258 int lock_may_read(struct inode
*inode
, loff_t start
, unsigned long len
)
2260 struct file_lock
*fl
;
2263 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2265 if (fl
->fl_type
== F_RDLCK
)
2267 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2269 } else if (IS_FLOCK(fl
)) {
2270 if (!(fl
->fl_type
& LOCK_MAND
))
2272 if (fl
->fl_type
& LOCK_READ
)
2283 EXPORT_SYMBOL(lock_may_read
);
2286 * lock_may_write - checks that the region is free of locks
2287 * @inode: the inode that is being written
2288 * @start: the first byte to write
2289 * @len: the number of bytes to write
2291 * Emulates Windows locking requirements. Whole-file
2292 * mandatory locks (share modes) can prohibit a write and
2293 * byte-range POSIX locks can prohibit a write if they overlap.
2295 * N.B. this function is only ever called
2296 * from knfsd and ownership of locks is never checked.
2298 int lock_may_write(struct inode
*inode
, loff_t start
, unsigned long len
)
2300 struct file_lock
*fl
;
2303 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2305 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2307 } else if (IS_FLOCK(fl
)) {
2308 if (!(fl
->fl_type
& LOCK_MAND
))
2310 if (fl
->fl_type
& LOCK_WRITE
)
2321 EXPORT_SYMBOL(lock_may_write
);
2323 static int __init
filelock_init(void)
2325 filelock_cache
= kmem_cache_create("file_lock_cache",
2326 sizeof(struct file_lock
), 0, SLAB_PANIC
,
2331 core_initcall(filelock_init
);