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 kmem_cache_t
*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
, SLAB_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
, kmem_cache_t
*cache
, unsigned long flags
)
204 struct file_lock
*lock
= (struct file_lock
*) foo
;
206 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) !=
207 SLAB_CTOR_CONSTRUCTOR
)
210 locks_init_lock(lock
);
213 static void locks_copy_private(struct file_lock
*new, struct file_lock
*fl
)
216 if (fl
->fl_ops
->fl_copy_lock
)
217 fl
->fl_ops
->fl_copy_lock(new, fl
);
218 new->fl_ops
= fl
->fl_ops
;
221 if (fl
->fl_lmops
->fl_copy_lock
)
222 fl
->fl_lmops
->fl_copy_lock(new, fl
);
223 new->fl_lmops
= fl
->fl_lmops
;
228 * Initialize a new lock from an existing file_lock structure.
230 static void __locks_copy_lock(struct file_lock
*new, const struct file_lock
*fl
)
232 new->fl_owner
= fl
->fl_owner
;
233 new->fl_pid
= fl
->fl_pid
;
235 new->fl_flags
= fl
->fl_flags
;
236 new->fl_type
= fl
->fl_type
;
237 new->fl_start
= fl
->fl_start
;
238 new->fl_end
= fl
->fl_end
;
240 new->fl_lmops
= NULL
;
243 void locks_copy_lock(struct file_lock
*new, struct file_lock
*fl
)
245 locks_release_private(new);
247 __locks_copy_lock(new, fl
);
248 new->fl_file
= fl
->fl_file
;
249 new->fl_ops
= fl
->fl_ops
;
250 new->fl_lmops
= fl
->fl_lmops
;
252 locks_copy_private(new, fl
);
255 EXPORT_SYMBOL(locks_copy_lock
);
257 static inline int flock_translate_cmd(int cmd
) {
259 return cmd
& (LOCK_MAND
| LOCK_RW
);
271 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
272 static int flock_make_lock(struct file
*filp
, struct file_lock
**lock
,
275 struct file_lock
*fl
;
276 int type
= flock_translate_cmd(cmd
);
280 fl
= locks_alloc_lock();
285 fl
->fl_pid
= current
->tgid
;
286 fl
->fl_flags
= FL_FLOCK
;
288 fl
->fl_end
= OFFSET_MAX
;
294 static int assign_type(struct file_lock
*fl
, int type
)
308 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
311 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
316 switch (l
->l_whence
) {
324 start
= i_size_read(filp
->f_dentry
->d_inode
);
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
335 fl
->fl_end
= OFFSET_MAX
;
337 end
= start
+ l
->l_len
- 1;
339 } else if (l
->l_len
< 0) {
346 fl
->fl_start
= start
; /* we record the absolute position */
347 if (fl
->fl_end
< fl
->fl_start
)
350 fl
->fl_owner
= current
->files
;
351 fl
->fl_pid
= current
->tgid
;
353 fl
->fl_flags
= FL_POSIX
;
357 return assign_type(fl
, l
->l_type
);
360 #if BITS_PER_LONG == 32
361 static int flock64_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
366 switch (l
->l_whence
) {
374 start
= i_size_read(filp
->f_dentry
->d_inode
);
383 fl
->fl_end
= OFFSET_MAX
;
385 fl
->fl_end
= start
+ l
->l_len
- 1;
386 } else if (l
->l_len
< 0) {
387 fl
->fl_end
= start
- 1;
392 fl
->fl_start
= start
; /* we record the absolute position */
393 if (fl
->fl_end
< fl
->fl_start
)
396 fl
->fl_owner
= current
->files
;
397 fl
->fl_pid
= current
->tgid
;
399 fl
->fl_flags
= FL_POSIX
;
407 fl
->fl_type
= l
->l_type
;
417 /* default lease lock manager operations */
418 static void lease_break_callback(struct file_lock
*fl
)
420 kill_fasync(&fl
->fl_fasync
, SIGIO
, POLL_MSG
);
423 static void lease_release_private_callback(struct file_lock
*fl
)
428 f_delown(fl
->fl_file
);
429 fl
->fl_file
->f_owner
.signum
= 0;
432 static int lease_mylease_callback(struct file_lock
*fl
, struct file_lock
*try)
434 return fl
->fl_file
== try->fl_file
;
437 static struct lock_manager_operations lease_manager_ops
= {
438 .fl_break
= lease_break_callback
,
439 .fl_release_private
= lease_release_private_callback
,
440 .fl_mylease
= lease_mylease_callback
,
441 .fl_change
= lease_modify
,
445 * Initialize a lease, use the default lock manager operations
447 static int lease_init(struct file
*filp
, int type
, struct file_lock
*fl
)
449 if (assign_type(fl
, type
) != 0)
452 fl
->fl_owner
= current
->files
;
453 fl
->fl_pid
= current
->tgid
;
456 fl
->fl_flags
= FL_LEASE
;
458 fl
->fl_end
= OFFSET_MAX
;
460 fl
->fl_lmops
= &lease_manager_ops
;
464 /* Allocate a file_lock initialised to this type of lease */
465 static int lease_alloc(struct file
*filp
, int type
, struct file_lock
**flp
)
467 struct file_lock
*fl
= locks_alloc_lock();
473 error
= lease_init(filp
, type
, fl
);
483 /* Check if two locks overlap each other.
485 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
487 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
488 (fl2
->fl_end
>= fl1
->fl_start
));
492 * Check whether two locks have the same owner.
494 static int posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
496 if (fl1
->fl_lmops
&& fl1
->fl_lmops
->fl_compare_owner
)
497 return fl2
->fl_lmops
== fl1
->fl_lmops
&&
498 fl1
->fl_lmops
->fl_compare_owner(fl1
, fl2
);
499 return fl1
->fl_owner
== fl2
->fl_owner
;
502 /* Remove waiter from blocker's block list.
503 * When blocker ends up pointing to itself then the list is empty.
505 static void __locks_delete_block(struct file_lock
*waiter
)
507 list_del_init(&waiter
->fl_block
);
508 list_del_init(&waiter
->fl_link
);
509 waiter
->fl_next
= NULL
;
514 static void locks_delete_block(struct file_lock
*waiter
)
517 __locks_delete_block(waiter
);
521 /* Insert waiter into blocker's block list.
522 * We use a circular list so that processes can be easily woken up in
523 * the order they blocked. The documentation doesn't require this but
524 * it seems like the reasonable thing to do.
526 static void locks_insert_block(struct file_lock
*blocker
,
527 struct file_lock
*waiter
)
529 BUG_ON(!list_empty(&waiter
->fl_block
));
530 list_add_tail(&waiter
->fl_block
, &blocker
->fl_block
);
531 waiter
->fl_next
= blocker
;
532 if (IS_POSIX(blocker
))
533 list_add(&waiter
->fl_link
, &blocked_list
);
536 /* Wake up processes blocked waiting for blocker.
537 * If told to wait then schedule the processes until the block list
538 * is empty, otherwise empty the block list ourselves.
540 static void locks_wake_up_blocks(struct file_lock
*blocker
)
542 while (!list_empty(&blocker
->fl_block
)) {
543 struct file_lock
*waiter
= list_entry(blocker
->fl_block
.next
,
544 struct file_lock
, fl_block
);
545 __locks_delete_block(waiter
);
546 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->fl_notify
)
547 waiter
->fl_lmops
->fl_notify(waiter
);
549 wake_up(&waiter
->fl_wait
);
553 /* Insert file lock fl into an inode's lock list at the position indicated
554 * by pos. At the same time add the lock to the global file lock list.
556 static void locks_insert_lock(struct file_lock
**pos
, struct file_lock
*fl
)
558 list_add(&fl
->fl_link
, &file_lock_list
);
560 /* insert into file's list */
564 if (fl
->fl_ops
&& fl
->fl_ops
->fl_insert
)
565 fl
->fl_ops
->fl_insert(fl
);
569 * Delete a lock and then free it.
570 * Wake up processes that are blocked waiting for this lock,
571 * notify the FS that the lock has been cleared and
572 * finally free the lock.
574 static void locks_delete_lock(struct file_lock
**thisfl_p
)
576 struct file_lock
*fl
= *thisfl_p
;
578 *thisfl_p
= fl
->fl_next
;
580 list_del_init(&fl
->fl_link
);
582 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
583 if (fl
->fl_fasync
!= NULL
) {
584 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
585 fl
->fl_fasync
= NULL
;
588 if (fl
->fl_ops
&& fl
->fl_ops
->fl_remove
)
589 fl
->fl_ops
->fl_remove(fl
);
591 locks_wake_up_blocks(fl
);
595 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
596 * checks for shared/exclusive status of overlapping locks.
598 static int locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
600 if (sys_fl
->fl_type
== F_WRLCK
)
602 if (caller_fl
->fl_type
== F_WRLCK
)
607 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
608 * checking before calling the locks_conflict().
610 static int posix_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
612 /* POSIX locks owned by the same process do not conflict with
615 if (!IS_POSIX(sys_fl
) || posix_same_owner(caller_fl
, sys_fl
))
618 /* Check whether they overlap */
619 if (!locks_overlap(caller_fl
, sys_fl
))
622 return (locks_conflict(caller_fl
, sys_fl
));
625 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
626 * checking before calling the locks_conflict().
628 static int flock_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
630 /* FLOCK locks referring to the same filp do not conflict with
633 if (!IS_FLOCK(sys_fl
) || (caller_fl
->fl_file
== sys_fl
->fl_file
))
635 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
638 return (locks_conflict(caller_fl
, sys_fl
));
641 static int interruptible_sleep_on_locked(wait_queue_head_t
*fl_wait
, int timeout
)
644 DECLARE_WAITQUEUE(wait
, current
);
646 __set_current_state(TASK_INTERRUPTIBLE
);
647 add_wait_queue(fl_wait
, &wait
);
651 result
= schedule_timeout(timeout
);
652 if (signal_pending(current
))
653 result
= -ERESTARTSYS
;
654 remove_wait_queue(fl_wait
, &wait
);
655 __set_current_state(TASK_RUNNING
);
659 static int locks_block_on_timeout(struct file_lock
*blocker
, struct file_lock
*waiter
, int time
)
662 locks_insert_block(blocker
, waiter
);
663 result
= interruptible_sleep_on_locked(&waiter
->fl_wait
, time
);
664 __locks_delete_block(waiter
);
669 posix_test_lock(struct file
*filp
, struct file_lock
*fl
,
670 struct file_lock
*conflock
)
672 struct file_lock
*cfl
;
675 for (cfl
= filp
->f_dentry
->d_inode
->i_flock
; cfl
; cfl
= cfl
->fl_next
) {
678 if (posix_locks_conflict(cfl
, fl
))
682 __locks_copy_lock(conflock
, cfl
);
690 EXPORT_SYMBOL(posix_test_lock
);
692 /* This function tests for deadlock condition before putting a process to
693 * sleep. The detection scheme is no longer recursive. Recursive was neat,
694 * but dangerous - we risked stack corruption if the lock data was bad, or
695 * if the recursion was too deep for any other reason.
697 * We rely on the fact that a task can only be on one lock's wait queue
698 * at a time. When we find blocked_task on a wait queue we can re-search
699 * with blocked_task equal to that queue's owner, until either blocked_task
700 * isn't found, or blocked_task is found on a queue owned by my_task.
702 * Note: the above assumption may not be true when handling lock requests
703 * from a broken NFS client. But broken NFS clients have a lot more to
704 * worry about than proper deadlock detection anyway... --okir
706 int posix_locks_deadlock(struct file_lock
*caller_fl
,
707 struct file_lock
*block_fl
)
709 struct list_head
*tmp
;
712 if (posix_same_owner(caller_fl
, block_fl
))
714 list_for_each(tmp
, &blocked_list
) {
715 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
716 if (posix_same_owner(fl
, block_fl
)) {
725 EXPORT_SYMBOL(posix_locks_deadlock
);
727 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
728 * at the head of the list, but that's secret knowledge known only to
729 * flock_lock_file and posix_lock_file.
731 static int flock_lock_file(struct file
*filp
, struct file_lock
*request
)
733 struct file_lock
*new_fl
= NULL
;
734 struct file_lock
**before
;
735 struct inode
* inode
= filp
->f_dentry
->d_inode
;
740 for_each_lock(inode
, before
) {
741 struct file_lock
*fl
= *before
;
746 if (filp
!= fl
->fl_file
)
748 if (request
->fl_type
== fl
->fl_type
)
751 locks_delete_lock(before
);
755 if (request
->fl_type
== F_UNLCK
)
759 new_fl
= locks_alloc_lock();
763 * If a higher-priority process was blocked on the old file lock,
764 * give it the opportunity to lock the file.
769 for_each_lock(inode
, before
) {
770 struct file_lock
*fl
= *before
;
775 if (!flock_locks_conflict(request
, fl
))
778 if (request
->fl_flags
& FL_SLEEP
)
779 locks_insert_block(fl
, request
);
782 locks_copy_lock(new_fl
, request
);
783 locks_insert_lock(&inode
->i_flock
, new_fl
);
790 locks_free_lock(new_fl
);
794 static int __posix_lock_file_conf(struct inode
*inode
, struct file_lock
*request
, struct file_lock
*conflock
)
796 struct file_lock
*fl
;
797 struct file_lock
*new_fl
, *new_fl2
;
798 struct file_lock
*left
= NULL
;
799 struct file_lock
*right
= NULL
;
800 struct file_lock
**before
;
801 int error
, added
= 0;
804 * We may need two file_lock structures for this operation,
805 * so we get them in advance to avoid races.
807 new_fl
= locks_alloc_lock();
808 new_fl2
= locks_alloc_lock();
811 if (request
->fl_type
!= F_UNLCK
) {
812 for_each_lock(inode
, before
) {
813 struct file_lock
*fl
= *before
;
816 if (!posix_locks_conflict(request
, fl
))
819 locks_copy_lock(conflock
, fl
);
821 if (!(request
->fl_flags
& FL_SLEEP
))
824 if (posix_locks_deadlock(request
, fl
))
827 locks_insert_block(fl
, request
);
832 /* If we're just looking for a conflict, we're done. */
834 if (request
->fl_flags
& FL_ACCESS
)
838 * Find the first old lock with the same owner as the new lock.
841 before
= &inode
->i_flock
;
843 /* First skip locks owned by other processes. */
844 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
845 !posix_same_owner(request
, fl
))) {
846 before
= &fl
->fl_next
;
849 /* Process locks with this owner. */
850 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
851 /* Detect adjacent or overlapping regions (if same lock type)
853 if (request
->fl_type
== fl
->fl_type
) {
854 /* In all comparisons of start vs end, use
855 * "start - 1" rather than "end + 1". If end
856 * is OFFSET_MAX, end + 1 will become negative.
858 if (fl
->fl_end
< request
->fl_start
- 1)
860 /* If the next lock in the list has entirely bigger
861 * addresses than the new one, insert the lock here.
863 if (fl
->fl_start
- 1 > request
->fl_end
)
866 /* If we come here, the new and old lock are of the
867 * same type and adjacent or overlapping. Make one
868 * lock yielding from the lower start address of both
869 * locks to the higher end address.
871 if (fl
->fl_start
> request
->fl_start
)
872 fl
->fl_start
= request
->fl_start
;
874 request
->fl_start
= fl
->fl_start
;
875 if (fl
->fl_end
< request
->fl_end
)
876 fl
->fl_end
= request
->fl_end
;
878 request
->fl_end
= fl
->fl_end
;
880 locks_delete_lock(before
);
887 /* Processing for different lock types is a bit
890 if (fl
->fl_end
< request
->fl_start
)
892 if (fl
->fl_start
> request
->fl_end
)
894 if (request
->fl_type
== F_UNLCK
)
896 if (fl
->fl_start
< request
->fl_start
)
898 /* If the next lock in the list has a higher end
899 * address than the new one, insert the new one here.
901 if (fl
->fl_end
> request
->fl_end
) {
905 if (fl
->fl_start
>= request
->fl_start
) {
906 /* The new lock completely replaces an old
907 * one (This may happen several times).
910 locks_delete_lock(before
);
913 /* Replace the old lock with the new one.
914 * Wake up anybody waiting for the old one,
915 * as the change in lock type might satisfy
918 locks_wake_up_blocks(fl
);
919 fl
->fl_start
= request
->fl_start
;
920 fl
->fl_end
= request
->fl_end
;
921 fl
->fl_type
= request
->fl_type
;
922 locks_release_private(fl
);
923 locks_copy_private(fl
, request
);
928 /* Go on to next lock.
931 before
= &fl
->fl_next
;
935 * The above code only modifies existing locks in case of
936 * merging or replacing. If new lock(s) need to be inserted
937 * all modifications are done bellow this, so it's safe yet to
940 error
= -ENOLCK
; /* "no luck" */
941 if (right
&& left
== right
&& !new_fl2
)
946 if (request
->fl_type
== F_UNLCK
)
953 locks_copy_lock(new_fl
, request
);
954 locks_insert_lock(before
, new_fl
);
959 /* The new lock breaks the old one in two pieces,
960 * so we have to use the second new lock.
964 locks_copy_lock(left
, right
);
965 locks_insert_lock(before
, left
);
967 right
->fl_start
= request
->fl_end
+ 1;
968 locks_wake_up_blocks(right
);
971 left
->fl_end
= request
->fl_start
- 1;
972 locks_wake_up_blocks(left
);
977 * Free any unused locks.
980 locks_free_lock(new_fl
);
982 locks_free_lock(new_fl2
);
987 * posix_lock_file - Apply a POSIX-style lock to a file
988 * @filp: The file to apply the lock to
989 * @fl: The lock to be applied
991 * Add a POSIX style lock to a file.
992 * We merge adjacent & overlapping locks whenever possible.
993 * POSIX locks are sorted by owner task, then by starting address
995 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
)
997 return __posix_lock_file_conf(filp
->f_dentry
->d_inode
, fl
, NULL
);
999 EXPORT_SYMBOL(posix_lock_file
);
1002 * posix_lock_file_conf - Apply a POSIX-style lock to a file
1003 * @filp: The file to apply the lock to
1004 * @fl: The lock to be applied
1005 * @conflock: Place to return a copy of the conflicting lock, if found.
1007 * Except for the conflock parameter, acts just like posix_lock_file.
1009 int posix_lock_file_conf(struct file
*filp
, struct file_lock
*fl
,
1010 struct file_lock
*conflock
)
1012 return __posix_lock_file_conf(filp
->f_dentry
->d_inode
, fl
, conflock
);
1014 EXPORT_SYMBOL(posix_lock_file_conf
);
1017 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1018 * @filp: The file to apply the lock to
1019 * @fl: The lock to be applied
1021 * Add a POSIX style lock to a file.
1022 * We merge adjacent & overlapping locks whenever possible.
1023 * POSIX locks are sorted by owner task, then by starting address
1025 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1030 error
= posix_lock_file(filp
, fl
);
1031 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1033 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1037 locks_delete_block(fl
);
1042 EXPORT_SYMBOL(posix_lock_file_wait
);
1045 * locks_mandatory_locked - Check for an active lock
1046 * @inode: the file to check
1048 * Searches the inode's list of locks to find any POSIX locks which conflict.
1049 * This function is called from locks_verify_locked() only.
1051 int locks_mandatory_locked(struct inode
*inode
)
1053 fl_owner_t owner
= current
->files
;
1054 struct file_lock
*fl
;
1057 * Search the lock list for this inode for any POSIX locks.
1060 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1063 if (fl
->fl_owner
!= owner
)
1067 return fl
? -EAGAIN
: 0;
1071 * locks_mandatory_area - Check for a conflicting lock
1072 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1074 * @inode: the file to check
1075 * @filp: how the file was opened (if it was)
1076 * @offset: start of area to check
1077 * @count: length of area to check
1079 * Searches the inode's list of locks to find any POSIX locks which conflict.
1080 * This function is called from rw_verify_area() and
1081 * locks_verify_truncate().
1083 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1084 struct file
*filp
, loff_t offset
,
1087 struct file_lock fl
;
1090 locks_init_lock(&fl
);
1091 fl
.fl_owner
= current
->files
;
1092 fl
.fl_pid
= current
->tgid
;
1094 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1095 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1096 fl
.fl_flags
|= FL_SLEEP
;
1097 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1098 fl
.fl_start
= offset
;
1099 fl
.fl_end
= offset
+ count
- 1;
1102 error
= __posix_lock_file_conf(inode
, &fl
, NULL
);
1103 if (error
!= -EAGAIN
)
1105 if (!(fl
.fl_flags
& FL_SLEEP
))
1107 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1110 * If we've been sleeping someone might have
1111 * changed the permissions behind our back.
1113 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
1117 locks_delete_block(&fl
);
1124 EXPORT_SYMBOL(locks_mandatory_area
);
1126 /* We already had a lease on this file; just change its type */
1127 int lease_modify(struct file_lock
**before
, int arg
)
1129 struct file_lock
*fl
= *before
;
1130 int error
= assign_type(fl
, arg
);
1134 locks_wake_up_blocks(fl
);
1136 locks_delete_lock(before
);
1140 EXPORT_SYMBOL(lease_modify
);
1142 static void time_out_leases(struct inode
*inode
)
1144 struct file_lock
**before
;
1145 struct file_lock
*fl
;
1147 before
= &inode
->i_flock
;
1148 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1149 if ((fl
->fl_break_time
== 0)
1150 || time_before(jiffies
, fl
->fl_break_time
)) {
1151 before
= &fl
->fl_next
;
1154 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1155 if (fl
== *before
) /* lease_modify may have freed fl */
1156 before
= &fl
->fl_next
;
1161 * __break_lease - revoke all outstanding leases on file
1162 * @inode: the inode of the file to return
1163 * @mode: the open mode (read or write)
1165 * break_lease (inlined for speed) has checked there already
1166 * is a lease on this file. Leases are broken on a call to open()
1167 * or truncate(). This function can sleep unless you
1168 * specified %O_NONBLOCK to your open().
1170 int __break_lease(struct inode
*inode
, unsigned int mode
)
1172 int error
= 0, future
;
1173 struct file_lock
*new_fl
, *flock
;
1174 struct file_lock
*fl
;
1176 unsigned long break_time
;
1177 int i_have_this_lease
= 0;
1179 alloc_err
= lease_alloc(NULL
, mode
& FMODE_WRITE
? F_WRLCK
: F_RDLCK
,
1184 time_out_leases(inode
);
1186 flock
= inode
->i_flock
;
1187 if ((flock
== NULL
) || !IS_LEASE(flock
))
1190 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
)
1191 if (fl
->fl_owner
== current
->files
)
1192 i_have_this_lease
= 1;
1194 if (mode
& FMODE_WRITE
) {
1195 /* If we want write access, we have to revoke any lease. */
1196 future
= F_UNLCK
| F_INPROGRESS
;
1197 } else if (flock
->fl_type
& F_INPROGRESS
) {
1198 /* If the lease is already being broken, we just leave it */
1199 future
= flock
->fl_type
;
1200 } else if (flock
->fl_type
& F_WRLCK
) {
1201 /* Downgrade the exclusive lease to a read-only lease. */
1202 future
= F_RDLCK
| F_INPROGRESS
;
1204 /* the existing lease was read-only, so we can read too. */
1208 if (alloc_err
&& !i_have_this_lease
&& ((mode
& O_NONBLOCK
) == 0)) {
1214 if (lease_break_time
> 0) {
1215 break_time
= jiffies
+ lease_break_time
* HZ
;
1216 if (break_time
== 0)
1217 break_time
++; /* so that 0 means no break time */
1220 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1221 if (fl
->fl_type
!= future
) {
1222 fl
->fl_type
= future
;
1223 fl
->fl_break_time
= break_time
;
1224 /* lease must have lmops break callback */
1225 fl
->fl_lmops
->fl_break(fl
);
1229 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1230 error
= -EWOULDBLOCK
;
1235 break_time
= flock
->fl_break_time
;
1236 if (break_time
!= 0) {
1237 break_time
-= jiffies
;
1238 if (break_time
== 0)
1241 error
= locks_block_on_timeout(flock
, new_fl
, break_time
);
1244 time_out_leases(inode
);
1245 /* Wait for the next lease that has not been broken yet */
1246 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1247 flock
= flock
->fl_next
) {
1248 if (flock
->fl_type
& F_INPROGRESS
)
1257 locks_free_lock(new_fl
);
1261 EXPORT_SYMBOL(__break_lease
);
1266 * @time: pointer to a timespec which will contain the last modified time
1268 * This is to force NFS clients to flush their caches for files with
1269 * exclusive leases. The justification is that if someone has an
1270 * exclusive lease, then they could be modifiying it.
1272 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1274 struct file_lock
*flock
= inode
->i_flock
;
1275 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1276 *time
= current_fs_time(inode
->i_sb
);
1278 *time
= inode
->i_mtime
;
1281 EXPORT_SYMBOL(lease_get_mtime
);
1284 * fcntl_getlease - Enquire what lease is currently active
1287 * The value returned by this function will be one of
1288 * (if no lease break is pending):
1290 * %F_RDLCK to indicate a shared lease is held.
1292 * %F_WRLCK to indicate an exclusive lease is held.
1294 * %F_UNLCK to indicate no lease is held.
1296 * (if a lease break is pending):
1298 * %F_RDLCK to indicate an exclusive lease needs to be
1299 * changed to a shared lease (or removed).
1301 * %F_UNLCK to indicate the lease needs to be removed.
1303 * XXX: sfr & willy disagree over whether F_INPROGRESS
1304 * should be returned to userspace.
1306 int fcntl_getlease(struct file
*filp
)
1308 struct file_lock
*fl
;
1312 time_out_leases(filp
->f_dentry
->d_inode
);
1313 for (fl
= filp
->f_dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1315 if (fl
->fl_file
== filp
) {
1316 type
= fl
->fl_type
& ~F_INPROGRESS
;
1325 * __setlease - sets a lease on an open file
1326 * @filp: file pointer
1327 * @arg: type of lease to obtain
1328 * @flp: input - file_lock to use, output - file_lock inserted
1330 * The (input) flp->fl_lmops->fl_break function is required
1333 * Called with kernel lock held.
1335 static int __setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1337 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1338 struct dentry
*dentry
= filp
->f_dentry
;
1339 struct inode
*inode
= dentry
->d_inode
;
1340 int error
, rdlease_count
= 0, wrlease_count
= 0;
1342 time_out_leases(inode
);
1345 if (!flp
|| !(*flp
) || !(*flp
)->fl_lmops
|| !(*flp
)->fl_lmops
->fl_break
)
1351 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1353 if ((arg
== F_WRLCK
)
1354 && ((atomic_read(&dentry
->d_count
) > 1)
1355 || (atomic_read(&inode
->i_count
) > 1)))
1359 * At this point, we know that if there is an exclusive
1360 * lease on this file, then we hold it on this filp
1361 * (otherwise our open of this file would have blocked).
1362 * And if we are trying to acquire an exclusive lease,
1363 * then the file is not open by anyone (including us)
1364 * except for this filp.
1366 for (before
= &inode
->i_flock
;
1367 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1368 before
= &fl
->fl_next
) {
1369 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1371 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1373 * Someone is in the process of opening this
1374 * file for writing so we may not take an
1375 * exclusive lease on it.
1382 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1383 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1386 if (my_before
!= NULL
) {
1388 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1400 error
= lease_alloc(filp
, arg
, &fl
);
1404 locks_copy_lock(fl
, lease
);
1406 locks_insert_lock(before
, fl
);
1414 * setlease - sets a lease on an open file
1415 * @filp: file pointer
1416 * @arg: type of lease to obtain
1417 * @lease: file_lock to use
1419 * Call this to establish a lease on the file.
1420 * The fl_lmops fl_break function is required by break_lease
1423 int setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1425 struct dentry
*dentry
= filp
->f_dentry
;
1426 struct inode
*inode
= dentry
->d_inode
;
1429 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1431 if (!S_ISREG(inode
->i_mode
))
1433 error
= security_file_lock(filp
, arg
);
1438 error
= __setlease(filp
, arg
, lease
);
1444 EXPORT_SYMBOL(setlease
);
1447 * fcntl_setlease - sets a lease on an open file
1448 * @fd: open file descriptor
1449 * @filp: file pointer
1450 * @arg: type of lease to obtain
1452 * Call this fcntl to establish a lease on the file.
1453 * Note that you also need to call %F_SETSIG to
1454 * receive a signal when the lease is broken.
1456 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1458 struct file_lock fl
, *flp
= &fl
;
1459 struct dentry
*dentry
= filp
->f_dentry
;
1460 struct inode
*inode
= dentry
->d_inode
;
1463 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1465 if (!S_ISREG(inode
->i_mode
))
1467 error
= security_file_lock(filp
, arg
);
1471 locks_init_lock(&fl
);
1472 error
= lease_init(filp
, arg
, &fl
);
1478 error
= __setlease(filp
, arg
, &flp
);
1479 if (error
|| arg
== F_UNLCK
)
1482 error
= fasync_helper(fd
, filp
, 1, &flp
->fl_fasync
);
1484 /* remove lease just inserted by __setlease */
1485 flp
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1486 flp
->fl_break_time
= jiffies
- 10;
1487 time_out_leases(inode
);
1491 error
= f_setown(filp
, current
->pid
, 0);
1498 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1499 * @filp: The file to apply the lock to
1500 * @fl: The lock to be applied
1502 * Add a FLOCK style lock to a file.
1504 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1509 error
= flock_lock_file(filp
, fl
);
1510 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1512 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1516 locks_delete_block(fl
);
1522 EXPORT_SYMBOL(flock_lock_file_wait
);
1525 * sys_flock: - flock() system call.
1526 * @fd: the file descriptor to lock.
1527 * @cmd: the type of lock to apply.
1529 * Apply a %FL_FLOCK style lock to an open file descriptor.
1530 * The @cmd can be one of
1532 * %LOCK_SH -- a shared lock.
1534 * %LOCK_EX -- an exclusive lock.
1536 * %LOCK_UN -- remove an existing lock.
1538 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1540 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1541 * processes read and write access respectively.
1543 asmlinkage
long sys_flock(unsigned int fd
, unsigned int cmd
)
1546 struct file_lock
*lock
;
1547 int can_sleep
, unlock
;
1555 can_sleep
= !(cmd
& LOCK_NB
);
1557 unlock
= (cmd
== LOCK_UN
);
1559 if (!unlock
&& !(cmd
& LOCK_MAND
) && !(filp
->f_mode
& 3))
1562 error
= flock_make_lock(filp
, &lock
, cmd
);
1566 lock
->fl_flags
|= FL_SLEEP
;
1568 error
= security_file_lock(filp
, cmd
);
1572 if (filp
->f_op
&& filp
->f_op
->flock
)
1573 error
= filp
->f_op
->flock(filp
,
1574 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1577 error
= flock_lock_file_wait(filp
, lock
);
1580 locks_free_lock(lock
);
1588 /* Report the first existing lock that would conflict with l.
1589 * This implements the F_GETLK command of fcntl().
1591 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1593 struct file_lock
*fl
, cfl
, file_lock
;
1598 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1601 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1604 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1608 if (filp
->f_op
&& filp
->f_op
->lock
) {
1609 error
= filp
->f_op
->lock(filp
, F_GETLK
, &file_lock
);
1610 if (file_lock
.fl_ops
&& file_lock
.fl_ops
->fl_release_private
)
1611 file_lock
.fl_ops
->fl_release_private(&file_lock
);
1615 fl
= (file_lock
.fl_type
== F_UNLCK
? NULL
: &file_lock
);
1617 fl
= (posix_test_lock(filp
, &file_lock
, &cfl
) ? &cfl
: NULL
);
1620 flock
.l_type
= F_UNLCK
;
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.
1629 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1631 if ((fl
->fl_end
!= OFFSET_MAX
)
1632 && (fl
->fl_end
> OFFT_OFFSET_MAX
))
1635 flock
.l_start
= fl
->fl_start
;
1636 flock
.l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1637 fl
->fl_end
- fl
->fl_start
+ 1;
1639 flock
.l_type
= fl
->fl_type
;
1642 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1648 /* Apply the lock described by l to an open file descriptor.
1649 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1651 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1652 struct flock __user
*l
)
1654 struct file_lock
*file_lock
= locks_alloc_lock();
1656 struct inode
*inode
;
1659 if (file_lock
== NULL
)
1663 * This might block, so we do it before checking the inode.
1666 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1669 inode
= filp
->f_dentry
->d_inode
;
1671 /* Don't allow mandatory locks on files that may be memory mapped
1674 if (IS_MANDLOCK(inode
) &&
1675 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1676 mapping_writably_mapped(filp
->f_mapping
)) {
1682 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1685 if (cmd
== F_SETLKW
) {
1686 file_lock
->fl_flags
|= FL_SLEEP
;
1690 switch (flock
.l_type
) {
1692 if (!(filp
->f_mode
& FMODE_READ
))
1696 if (!(filp
->f_mode
& FMODE_WRITE
))
1706 error
= security_file_lock(filp
, file_lock
->fl_type
);
1710 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
)
1711 error
= filp
->f_op
->lock(filp
, cmd
, file_lock
);
1714 error
= posix_lock_file(filp
, file_lock
);
1715 if ((error
!= -EAGAIN
) || (cmd
== F_SETLK
))
1717 error
= wait_event_interruptible(file_lock
->fl_wait
,
1718 !file_lock
->fl_next
);
1722 locks_delete_block(file_lock
);
1728 * Attempt to detect a close/fcntl race and recover by
1729 * releasing the lock that was just acquired.
1731 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1732 flock
.l_type
= F_UNLCK
;
1737 locks_free_lock(file_lock
);
1741 #if BITS_PER_LONG == 32
1742 /* Report the first existing lock that would conflict with l.
1743 * This implements the F_GETLK command of fcntl().
1745 int fcntl_getlk64(struct file
*filp
, struct flock64 __user
*l
)
1747 struct file_lock
*fl
, cfl
, file_lock
;
1748 struct flock64 flock
;
1752 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1755 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1758 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1762 if (filp
->f_op
&& filp
->f_op
->lock
) {
1763 error
= filp
->f_op
->lock(filp
, F_GETLK
, &file_lock
);
1764 if (file_lock
.fl_ops
&& file_lock
.fl_ops
->fl_release_private
)
1765 file_lock
.fl_ops
->fl_release_private(&file_lock
);
1769 fl
= (file_lock
.fl_type
== F_UNLCK
? NULL
: &file_lock
);
1771 fl
= (posix_test_lock(filp
, &file_lock
, &cfl
) ? &cfl
: NULL
);
1774 flock
.l_type
= F_UNLCK
;
1776 flock
.l_pid
= fl
->fl_pid
;
1777 flock
.l_start
= fl
->fl_start
;
1778 flock
.l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1779 fl
->fl_end
- fl
->fl_start
+ 1;
1781 flock
.l_type
= fl
->fl_type
;
1784 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1791 /* Apply the lock described by l to an open file descriptor.
1792 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1794 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1795 struct flock64 __user
*l
)
1797 struct file_lock
*file_lock
= locks_alloc_lock();
1798 struct flock64 flock
;
1799 struct inode
*inode
;
1802 if (file_lock
== NULL
)
1806 * This might block, so we do it before checking the inode.
1809 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1812 inode
= filp
->f_dentry
->d_inode
;
1814 /* Don't allow mandatory locks on files that may be memory mapped
1817 if (IS_MANDLOCK(inode
) &&
1818 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1819 mapping_writably_mapped(filp
->f_mapping
)) {
1825 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1828 if (cmd
== F_SETLKW64
) {
1829 file_lock
->fl_flags
|= FL_SLEEP
;
1833 switch (flock
.l_type
) {
1835 if (!(filp
->f_mode
& FMODE_READ
))
1839 if (!(filp
->f_mode
& FMODE_WRITE
))
1849 error
= security_file_lock(filp
, file_lock
->fl_type
);
1853 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
)
1854 error
= filp
->f_op
->lock(filp
, cmd
, file_lock
);
1857 error
= posix_lock_file(filp
, file_lock
);
1858 if ((error
!= -EAGAIN
) || (cmd
== F_SETLK64
))
1860 error
= wait_event_interruptible(file_lock
->fl_wait
,
1861 !file_lock
->fl_next
);
1865 locks_delete_block(file_lock
);
1871 * Attempt to detect a close/fcntl race and recover by
1872 * releasing the lock that was just acquired.
1874 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1875 flock
.l_type
= F_UNLCK
;
1880 locks_free_lock(file_lock
);
1883 #endif /* BITS_PER_LONG == 32 */
1886 * This function is called when the file is being removed
1887 * from the task's fd array. POSIX locks belonging to this task
1888 * are deleted at this time.
1890 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
1892 struct file_lock lock
, **before
;
1895 * If there are no locks held on this file, we don't need to call
1896 * posix_lock_file(). Another process could be setting a lock on this
1897 * file at the same time, but we wouldn't remove that lock anyway.
1899 before
= &filp
->f_dentry
->d_inode
->i_flock
;
1900 if (*before
== NULL
)
1903 lock
.fl_type
= F_UNLCK
;
1904 lock
.fl_flags
= FL_POSIX
;
1906 lock
.fl_end
= OFFSET_MAX
;
1907 lock
.fl_owner
= owner
;
1908 lock
.fl_pid
= current
->tgid
;
1909 lock
.fl_file
= filp
;
1911 lock
.fl_lmops
= NULL
;
1913 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
) {
1914 filp
->f_op
->lock(filp
, F_SETLK
, &lock
);
1918 /* Can't use posix_lock_file here; we need to remove it no matter
1919 * which pid we have.
1922 while (*before
!= NULL
) {
1923 struct file_lock
*fl
= *before
;
1924 if (IS_POSIX(fl
) && posix_same_owner(fl
, &lock
)) {
1925 locks_delete_lock(before
);
1928 before
= &fl
->fl_next
;
1932 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
1933 lock
.fl_ops
->fl_release_private(&lock
);
1936 EXPORT_SYMBOL(locks_remove_posix
);
1939 * This function is called on the last close of an open file.
1941 void locks_remove_flock(struct file
*filp
)
1943 struct inode
* inode
= filp
->f_dentry
->d_inode
;
1944 struct file_lock
*fl
;
1945 struct file_lock
**before
;
1947 if (!inode
->i_flock
)
1950 if (filp
->f_op
&& filp
->f_op
->flock
) {
1951 struct file_lock fl
= {
1952 .fl_pid
= current
->tgid
,
1954 .fl_flags
= FL_FLOCK
,
1956 .fl_end
= OFFSET_MAX
,
1958 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
1959 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
1960 fl
.fl_ops
->fl_release_private(&fl
);
1964 before
= &inode
->i_flock
;
1966 while ((fl
= *before
) != NULL
) {
1967 if (fl
->fl_file
== filp
) {
1969 locks_delete_lock(before
);
1973 lease_modify(before
, F_UNLCK
);
1979 before
= &fl
->fl_next
;
1985 * posix_unblock_lock - stop waiting for a file lock
1986 * @filp: how the file was opened
1987 * @waiter: the lock which was waiting
1989 * lockd needs to block waiting for locks.
1992 posix_unblock_lock(struct file
*filp
, struct file_lock
*waiter
)
1997 if (waiter
->fl_next
)
1998 __locks_delete_block(waiter
);
2005 EXPORT_SYMBOL(posix_unblock_lock
);
2007 static void lock_get_status(char* out
, struct file_lock
*fl
, int id
, char *pfx
)
2009 struct inode
*inode
= NULL
;
2011 if (fl
->fl_file
!= NULL
)
2012 inode
= fl
->fl_file
->f_dentry
->d_inode
;
2014 out
+= sprintf(out
, "%d:%s ", id
, pfx
);
2016 out
+= sprintf(out
, "%6s %s ",
2017 (fl
->fl_flags
& FL_ACCESS
) ? "ACCESS" : "POSIX ",
2018 (inode
== NULL
) ? "*NOINODE*" :
2019 (IS_MANDLOCK(inode
) &&
2020 (inode
->i_mode
& (S_IXGRP
| S_ISGID
)) == S_ISGID
) ?
2021 "MANDATORY" : "ADVISORY ");
2022 } else if (IS_FLOCK(fl
)) {
2023 if (fl
->fl_type
& LOCK_MAND
) {
2024 out
+= sprintf(out
, "FLOCK MSNFS ");
2026 out
+= sprintf(out
, "FLOCK ADVISORY ");
2028 } else if (IS_LEASE(fl
)) {
2029 out
+= sprintf(out
, "LEASE ");
2030 if (fl
->fl_type
& F_INPROGRESS
)
2031 out
+= sprintf(out
, "BREAKING ");
2032 else if (fl
->fl_file
)
2033 out
+= sprintf(out
, "ACTIVE ");
2035 out
+= sprintf(out
, "BREAKER ");
2037 out
+= sprintf(out
, "UNKNOWN UNKNOWN ");
2039 if (fl
->fl_type
& LOCK_MAND
) {
2040 out
+= sprintf(out
, "%s ",
2041 (fl
->fl_type
& LOCK_READ
)
2042 ? (fl
->fl_type
& LOCK_WRITE
) ? "RW " : "READ "
2043 : (fl
->fl_type
& LOCK_WRITE
) ? "WRITE" : "NONE ");
2045 out
+= sprintf(out
, "%s ",
2046 (fl
->fl_type
& F_INPROGRESS
)
2047 ? (fl
->fl_type
& F_UNLCK
) ? "UNLCK" : "READ "
2048 : (fl
->fl_type
& F_WRLCK
) ? "WRITE" : "READ ");
2051 #ifdef WE_CAN_BREAK_LSLK_NOW
2052 out
+= sprintf(out
, "%d %s:%ld ", fl
->fl_pid
,
2053 inode
->i_sb
->s_id
, inode
->i_ino
);
2055 /* userspace relies on this representation of dev_t ;-( */
2056 out
+= sprintf(out
, "%d %02x:%02x:%ld ", fl
->fl_pid
,
2057 MAJOR(inode
->i_sb
->s_dev
),
2058 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2061 out
+= sprintf(out
, "%d <none>:0 ", fl
->fl_pid
);
2064 if (fl
->fl_end
== OFFSET_MAX
)
2065 out
+= sprintf(out
, "%Ld EOF\n", fl
->fl_start
);
2067 out
+= sprintf(out
, "%Ld %Ld\n", fl
->fl_start
,
2070 out
+= sprintf(out
, "0 EOF\n");
2074 static void move_lock_status(char **p
, off_t
* pos
, off_t offset
)
2078 if(*pos
>= offset
) {
2079 /* the complete line is valid */
2084 if(*pos
+len
> offset
) {
2085 /* use the second part of the line */
2086 int i
= offset
-*pos
;
2087 memmove(*p
,*p
+i
,len
-i
);
2092 /* discard the complete line */
2097 * get_locks_status - reports lock usage in /proc/locks
2098 * @buffer: address in userspace to write into
2100 * @offset: how far we are through the buffer
2101 * @length: how much to read
2104 int get_locks_status(char *buffer
, char **start
, off_t offset
, int length
)
2106 struct list_head
*tmp
;
2112 list_for_each(tmp
, &file_lock_list
) {
2113 struct list_head
*btmp
;
2114 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
2115 lock_get_status(q
, fl
, ++i
, "");
2116 move_lock_status(&q
, &pos
, offset
);
2118 if(pos
>= offset
+length
)
2121 list_for_each(btmp
, &fl
->fl_block
) {
2122 struct file_lock
*bfl
= list_entry(btmp
,
2123 struct file_lock
, fl_block
);
2124 lock_get_status(q
, bfl
, i
, " ->");
2125 move_lock_status(&q
, &pos
, offset
);
2127 if(pos
>= offset
+length
)
2134 if(q
-buffer
< length
)
2140 * lock_may_read - checks that the region is free of locks
2141 * @inode: the inode that is being read
2142 * @start: the first byte to read
2143 * @len: the number of bytes to read
2145 * Emulates Windows locking requirements. Whole-file
2146 * mandatory locks (share modes) can prohibit a read and
2147 * byte-range POSIX locks can prohibit a read if they overlap.
2149 * N.B. this function is only ever called
2150 * from knfsd and ownership of locks is never checked.
2152 int lock_may_read(struct inode
*inode
, loff_t start
, unsigned long len
)
2154 struct file_lock
*fl
;
2157 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2159 if (fl
->fl_type
== F_RDLCK
)
2161 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2163 } else if (IS_FLOCK(fl
)) {
2164 if (!(fl
->fl_type
& LOCK_MAND
))
2166 if (fl
->fl_type
& LOCK_READ
)
2177 EXPORT_SYMBOL(lock_may_read
);
2180 * lock_may_write - checks that the region is free of locks
2181 * @inode: the inode that is being written
2182 * @start: the first byte to write
2183 * @len: the number of bytes to write
2185 * Emulates Windows locking requirements. Whole-file
2186 * mandatory locks (share modes) can prohibit a write and
2187 * byte-range POSIX locks can prohibit a write if they overlap.
2189 * N.B. this function is only ever called
2190 * from knfsd and ownership of locks is never checked.
2192 int lock_may_write(struct inode
*inode
, loff_t start
, unsigned long len
)
2194 struct file_lock
*fl
;
2197 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2199 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2201 } else if (IS_FLOCK(fl
)) {
2202 if (!(fl
->fl_type
& LOCK_MAND
))
2204 if (fl
->fl_type
& LOCK_WRITE
)
2215 EXPORT_SYMBOL(lock_may_write
);
2217 static int __init
filelock_init(void)
2219 filelock_cache
= kmem_cache_create("file_lock_cache",
2220 sizeof(struct file_lock
), 0, SLAB_PANIC
,
2225 core_initcall(filelock_init
);