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 LIST_HEAD(file_lock_list
);
144 EXPORT_SYMBOL(file_lock_list
);
146 static LIST_HEAD(blocked_list
);
148 static kmem_cache_t
*filelock_cache
;
150 /* Allocate an empty lock structure. */
151 static struct file_lock
*locks_alloc_lock(void)
153 return kmem_cache_alloc(filelock_cache
, SLAB_KERNEL
);
156 static void locks_release_private(struct file_lock
*fl
)
159 if (fl
->fl_ops
->fl_release_private
)
160 fl
->fl_ops
->fl_release_private(fl
);
164 if (fl
->fl_lmops
->fl_release_private
)
165 fl
->fl_lmops
->fl_release_private(fl
);
171 /* Free a lock which is not in use. */
172 static void locks_free_lock(struct file_lock
*fl
)
178 if (waitqueue_active(&fl
->fl_wait
))
179 panic("Attempting to free lock with active wait queue");
181 if (!list_empty(&fl
->fl_block
))
182 panic("Attempting to free lock with active block list");
184 if (!list_empty(&fl
->fl_link
))
185 panic("Attempting to free lock on active lock list");
187 locks_release_private(fl
);
188 kmem_cache_free(filelock_cache
, fl
);
191 void locks_init_lock(struct file_lock
*fl
)
193 INIT_LIST_HEAD(&fl
->fl_link
);
194 INIT_LIST_HEAD(&fl
->fl_block
);
195 init_waitqueue_head(&fl
->fl_wait
);
197 fl
->fl_fasync
= NULL
;
203 fl
->fl_start
= fl
->fl_end
= 0;
208 EXPORT_SYMBOL(locks_init_lock
);
211 * Initialises the fields of the file lock which are invariant for
214 static void init_once(void *foo
, kmem_cache_t
*cache
, unsigned long flags
)
216 struct file_lock
*lock
= (struct file_lock
*) foo
;
218 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) !=
219 SLAB_CTOR_CONSTRUCTOR
)
222 locks_init_lock(lock
);
225 static void locks_copy_private(struct file_lock
*new, struct file_lock
*fl
)
228 if (fl
->fl_ops
->fl_copy_lock
)
229 fl
->fl_ops
->fl_copy_lock(new, fl
);
230 new->fl_ops
= fl
->fl_ops
;
233 if (fl
->fl_lmops
->fl_copy_lock
)
234 fl
->fl_lmops
->fl_copy_lock(new, fl
);
235 new->fl_lmops
= fl
->fl_lmops
;
240 * Initialize a new lock from an existing file_lock structure.
242 void locks_copy_lock(struct file_lock
*new, struct file_lock
*fl
)
244 locks_release_private(new);
246 new->fl_owner
= fl
->fl_owner
;
247 new->fl_pid
= fl
->fl_pid
;
248 new->fl_file
= fl
->fl_file
;
249 new->fl_flags
= fl
->fl_flags
;
250 new->fl_type
= fl
->fl_type
;
251 new->fl_start
= fl
->fl_start
;
252 new->fl_end
= fl
->fl_end
;
253 new->fl_ops
= fl
->fl_ops
;
254 new->fl_lmops
= fl
->fl_lmops
;
256 locks_copy_private(new, fl
);
259 EXPORT_SYMBOL(locks_copy_lock
);
261 static inline int flock_translate_cmd(int cmd
) {
263 return cmd
& (LOCK_MAND
| LOCK_RW
);
275 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
276 static int flock_make_lock(struct file
*filp
, struct file_lock
**lock
,
279 struct file_lock
*fl
;
280 int type
= flock_translate_cmd(cmd
);
284 fl
= locks_alloc_lock();
289 fl
->fl_pid
= current
->tgid
;
290 fl
->fl_flags
= FL_FLOCK
;
292 fl
->fl_end
= OFFSET_MAX
;
298 static int assign_type(struct file_lock
*fl
, int type
)
312 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
315 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
320 switch (l
->l_whence
) {
328 start
= i_size_read(filp
->f_dentry
->d_inode
);
334 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
335 POSIX-2001 defines it. */
339 fl
->fl_end
= OFFSET_MAX
;
341 end
= start
+ l
->l_len
- 1;
343 } else if (l
->l_len
< 0) {
350 fl
->fl_start
= start
; /* we record the absolute position */
351 if (fl
->fl_end
< fl
->fl_start
)
354 fl
->fl_owner
= current
->files
;
355 fl
->fl_pid
= current
->tgid
;
357 fl
->fl_flags
= FL_POSIX
;
361 return assign_type(fl
, l
->l_type
);
364 #if BITS_PER_LONG == 32
365 static int flock64_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
370 switch (l
->l_whence
) {
378 start
= i_size_read(filp
->f_dentry
->d_inode
);
387 fl
->fl_end
= OFFSET_MAX
;
389 fl
->fl_end
= start
+ l
->l_len
- 1;
390 } else if (l
->l_len
< 0) {
391 fl
->fl_end
= start
- 1;
396 fl
->fl_start
= start
; /* we record the absolute position */
397 if (fl
->fl_end
< fl
->fl_start
)
400 fl
->fl_owner
= current
->files
;
401 fl
->fl_pid
= current
->tgid
;
403 fl
->fl_flags
= FL_POSIX
;
411 fl
->fl_type
= l
->l_type
;
421 /* default lease lock manager operations */
422 static void lease_break_callback(struct file_lock
*fl
)
424 kill_fasync(&fl
->fl_fasync
, SIGIO
, POLL_MSG
);
427 static void lease_release_private_callback(struct file_lock
*fl
)
432 f_delown(fl
->fl_file
);
433 fl
->fl_file
->f_owner
.signum
= 0;
436 static int lease_mylease_callback(struct file_lock
*fl
, struct file_lock
*try)
438 return fl
->fl_file
== try->fl_file
;
441 static struct lock_manager_operations lease_manager_ops
= {
442 .fl_break
= lease_break_callback
,
443 .fl_release_private
= lease_release_private_callback
,
444 .fl_mylease
= lease_mylease_callback
,
445 .fl_change
= lease_modify
,
449 * Initialize a lease, use the default lock manager operations
451 static int lease_init(struct file
*filp
, int type
, struct file_lock
*fl
)
453 fl
->fl_owner
= current
->files
;
454 fl
->fl_pid
= current
->tgid
;
457 fl
->fl_flags
= FL_LEASE
;
458 if (assign_type(fl
, type
) != 0) {
463 fl
->fl_end
= OFFSET_MAX
;
465 fl
->fl_lmops
= &lease_manager_ops
;
469 /* Allocate a file_lock initialised to this type of lease */
470 static int lease_alloc(struct file
*filp
, int type
, struct file_lock
**flp
)
472 struct file_lock
*fl
= locks_alloc_lock();
478 error
= lease_init(filp
, type
, fl
);
485 /* Check if two locks overlap each other.
487 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
489 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
490 (fl2
->fl_end
>= fl1
->fl_start
));
494 * Check whether two locks have the same owner.
496 static int posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
498 if (fl1
->fl_lmops
&& fl1
->fl_lmops
->fl_compare_owner
)
499 return fl2
->fl_lmops
== fl1
->fl_lmops
&&
500 fl1
->fl_lmops
->fl_compare_owner(fl1
, fl2
);
501 return fl1
->fl_owner
== fl2
->fl_owner
;
504 /* Remove waiter from blocker's block list.
505 * When blocker ends up pointing to itself then the list is empty.
507 static void __locks_delete_block(struct file_lock
*waiter
)
509 list_del_init(&waiter
->fl_block
);
510 list_del_init(&waiter
->fl_link
);
511 waiter
->fl_next
= NULL
;
516 static void locks_delete_block(struct file_lock
*waiter
)
519 __locks_delete_block(waiter
);
523 /* Insert waiter into blocker's block list.
524 * We use a circular list so that processes can be easily woken up in
525 * the order they blocked. The documentation doesn't require this but
526 * it seems like the reasonable thing to do.
528 static void locks_insert_block(struct file_lock
*blocker
,
529 struct file_lock
*waiter
)
531 if (!list_empty(&waiter
->fl_block
)) {
532 printk(KERN_ERR
"locks_insert_block: removing duplicated lock "
533 "(pid=%d %Ld-%Ld type=%d)\n", waiter
->fl_pid
,
534 waiter
->fl_start
, waiter
->fl_end
, waiter
->fl_type
);
535 __locks_delete_block(waiter
);
537 list_add_tail(&waiter
->fl_block
, &blocker
->fl_block
);
538 waiter
->fl_next
= blocker
;
539 if (IS_POSIX(blocker
))
540 list_add(&waiter
->fl_link
, &blocked_list
);
543 /* Wake up processes blocked waiting for blocker.
544 * If told to wait then schedule the processes until the block list
545 * is empty, otherwise empty the block list ourselves.
547 static void locks_wake_up_blocks(struct file_lock
*blocker
)
549 while (!list_empty(&blocker
->fl_block
)) {
550 struct file_lock
*waiter
= list_entry(blocker
->fl_block
.next
,
551 struct file_lock
, fl_block
);
552 __locks_delete_block(waiter
);
553 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->fl_notify
)
554 waiter
->fl_lmops
->fl_notify(waiter
);
556 wake_up(&waiter
->fl_wait
);
560 /* Insert file lock fl into an inode's lock list at the position indicated
561 * by pos. At the same time add the lock to the global file lock list.
563 static void locks_insert_lock(struct file_lock
**pos
, struct file_lock
*fl
)
565 list_add(&fl
->fl_link
, &file_lock_list
);
567 /* insert into file's list */
571 if (fl
->fl_ops
&& fl
->fl_ops
->fl_insert
)
572 fl
->fl_ops
->fl_insert(fl
);
576 * Delete a lock and then free it.
577 * Wake up processes that are blocked waiting for this lock,
578 * notify the FS that the lock has been cleared and
579 * finally free the lock.
581 static void locks_delete_lock(struct file_lock
**thisfl_p
)
583 struct file_lock
*fl
= *thisfl_p
;
585 *thisfl_p
= fl
->fl_next
;
587 list_del_init(&fl
->fl_link
);
589 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
590 if (fl
->fl_fasync
!= NULL
) {
591 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
592 fl
->fl_fasync
= NULL
;
595 if (fl
->fl_ops
&& fl
->fl_ops
->fl_remove
)
596 fl
->fl_ops
->fl_remove(fl
);
598 locks_wake_up_blocks(fl
);
602 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
603 * checks for shared/exclusive status of overlapping locks.
605 static int locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
607 if (sys_fl
->fl_type
== F_WRLCK
)
609 if (caller_fl
->fl_type
== F_WRLCK
)
614 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
615 * checking before calling the locks_conflict().
617 static int posix_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
619 /* POSIX locks owned by the same process do not conflict with
622 if (!IS_POSIX(sys_fl
) || posix_same_owner(caller_fl
, sys_fl
))
625 /* Check whether they overlap */
626 if (!locks_overlap(caller_fl
, sys_fl
))
629 return (locks_conflict(caller_fl
, sys_fl
));
632 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
633 * checking before calling the locks_conflict().
635 static int flock_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
637 /* FLOCK locks referring to the same filp do not conflict with
640 if (!IS_FLOCK(sys_fl
) || (caller_fl
->fl_file
== sys_fl
->fl_file
))
642 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
645 return (locks_conflict(caller_fl
, sys_fl
));
648 static int interruptible_sleep_on_locked(wait_queue_head_t
*fl_wait
, int timeout
)
651 DECLARE_WAITQUEUE(wait
, current
);
653 __set_current_state(TASK_INTERRUPTIBLE
);
654 add_wait_queue(fl_wait
, &wait
);
658 result
= schedule_timeout(timeout
);
659 if (signal_pending(current
))
660 result
= -ERESTARTSYS
;
661 remove_wait_queue(fl_wait
, &wait
);
662 __set_current_state(TASK_RUNNING
);
666 static int locks_block_on_timeout(struct file_lock
*blocker
, struct file_lock
*waiter
, int time
)
669 locks_insert_block(blocker
, waiter
);
670 result
= interruptible_sleep_on_locked(&waiter
->fl_wait
, time
);
671 __locks_delete_block(waiter
);
676 posix_test_lock(struct file
*filp
, struct file_lock
*fl
,
677 struct file_lock
*conflock
)
679 struct file_lock
*cfl
;
682 for (cfl
= filp
->f_dentry
->d_inode
->i_flock
; cfl
; cfl
= cfl
->fl_next
) {
685 if (posix_locks_conflict(cfl
, fl
))
689 locks_copy_lock(conflock
, cfl
);
697 EXPORT_SYMBOL(posix_test_lock
);
699 /* This function tests for deadlock condition before putting a process to
700 * sleep. The detection scheme is no longer recursive. Recursive was neat,
701 * but dangerous - we risked stack corruption if the lock data was bad, or
702 * if the recursion was too deep for any other reason.
704 * We rely on the fact that a task can only be on one lock's wait queue
705 * at a time. When we find blocked_task on a wait queue we can re-search
706 * with blocked_task equal to that queue's owner, until either blocked_task
707 * isn't found, or blocked_task is found on a queue owned by my_task.
709 * Note: the above assumption may not be true when handling lock requests
710 * from a broken NFS client. But broken NFS clients have a lot more to
711 * worry about than proper deadlock detection anyway... --okir
713 int posix_locks_deadlock(struct file_lock
*caller_fl
,
714 struct file_lock
*block_fl
)
716 struct list_head
*tmp
;
719 if (posix_same_owner(caller_fl
, block_fl
))
721 list_for_each(tmp
, &blocked_list
) {
722 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
723 if (posix_same_owner(fl
, block_fl
)) {
732 EXPORT_SYMBOL(posix_locks_deadlock
);
734 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
735 * at the head of the list, but that's secret knowledge known only to
736 * flock_lock_file and posix_lock_file.
738 static int flock_lock_file(struct file
*filp
, struct file_lock
*new_fl
)
740 struct file_lock
**before
;
741 struct inode
* inode
= filp
->f_dentry
->d_inode
;
746 for_each_lock(inode
, before
) {
747 struct file_lock
*fl
= *before
;
752 if (filp
!= fl
->fl_file
)
754 if (new_fl
->fl_type
== fl
->fl_type
)
757 locks_delete_lock(before
);
762 if (new_fl
->fl_type
== F_UNLCK
)
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
773 for_each_lock(inode
, before
) {
774 struct file_lock
*fl
= *before
;
779 if (!flock_locks_conflict(new_fl
, fl
))
782 if (new_fl
->fl_flags
& FL_SLEEP
) {
783 locks_insert_block(fl
, new_fl
);
787 locks_insert_lock(&inode
->i_flock
, new_fl
);
795 EXPORT_SYMBOL(posix_lock_file
);
797 static int __posix_lock_file(struct inode
*inode
, struct file_lock
*request
)
799 struct file_lock
*fl
;
800 struct file_lock
*new_fl
, *new_fl2
;
801 struct file_lock
*left
= NULL
;
802 struct file_lock
*right
= NULL
;
803 struct file_lock
**before
;
804 int error
, added
= 0;
807 * We may need two file_lock structures for this operation,
808 * so we get them in advance to avoid races.
810 new_fl
= locks_alloc_lock();
811 new_fl2
= locks_alloc_lock();
814 if (request
->fl_type
!= F_UNLCK
) {
815 for_each_lock(inode
, before
) {
816 struct file_lock
*fl
= *before
;
819 if (!posix_locks_conflict(request
, fl
))
822 if (!(request
->fl_flags
& FL_SLEEP
))
825 if (posix_locks_deadlock(request
, fl
))
828 locks_insert_block(fl
, request
);
833 /* If we're just looking for a conflict, we're done. */
835 if (request
->fl_flags
& FL_ACCESS
)
838 error
= -ENOLCK
; /* "no luck" */
839 if (!(new_fl
&& new_fl2
))
843 * We've allocated the new locks in advance, so there are no
844 * errors possible (and no blocking operations) from here on.
846 * Find the first old lock with the same owner as the new lock.
849 before
= &inode
->i_flock
;
851 /* First skip locks owned by other processes. */
852 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
853 !posix_same_owner(request
, fl
))) {
854 before
= &fl
->fl_next
;
857 /* Process locks with this owner. */
858 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
859 /* Detect adjacent or overlapping regions (if same lock type)
861 if (request
->fl_type
== fl
->fl_type
) {
862 /* In all comparisons of start vs end, use
863 * "start - 1" rather than "end + 1". If end
864 * is OFFSET_MAX, end + 1 will become negative.
866 if (fl
->fl_end
< request
->fl_start
- 1)
868 /* If the next lock in the list has entirely bigger
869 * addresses than the new one, insert the lock here.
871 if (fl
->fl_start
- 1 > request
->fl_end
)
874 /* If we come here, the new and old lock are of the
875 * same type and adjacent or overlapping. Make one
876 * lock yielding from the lower start address of both
877 * locks to the higher end address.
879 if (fl
->fl_start
> request
->fl_start
)
880 fl
->fl_start
= request
->fl_start
;
882 request
->fl_start
= fl
->fl_start
;
883 if (fl
->fl_end
< request
->fl_end
)
884 fl
->fl_end
= request
->fl_end
;
886 request
->fl_end
= fl
->fl_end
;
888 locks_delete_lock(before
);
895 /* Processing for different lock types is a bit
898 if (fl
->fl_end
< request
->fl_start
)
900 if (fl
->fl_start
> request
->fl_end
)
902 if (request
->fl_type
== F_UNLCK
)
904 if (fl
->fl_start
< request
->fl_start
)
906 /* If the next lock in the list has a higher end
907 * address than the new one, insert the new one here.
909 if (fl
->fl_end
> request
->fl_end
) {
913 if (fl
->fl_start
>= request
->fl_start
) {
914 /* The new lock completely replaces an old
915 * one (This may happen several times).
918 locks_delete_lock(before
);
921 /* Replace the old lock with the new one.
922 * Wake up anybody waiting for the old one,
923 * as the change in lock type might satisfy
926 locks_wake_up_blocks(fl
);
927 fl
->fl_start
= request
->fl_start
;
928 fl
->fl_end
= request
->fl_end
;
929 fl
->fl_type
= request
->fl_type
;
930 locks_release_private(fl
);
931 locks_copy_private(fl
, request
);
936 /* Go on to next lock.
939 before
= &fl
->fl_next
;
944 if (request
->fl_type
== F_UNLCK
)
946 locks_copy_lock(new_fl
, request
);
947 locks_insert_lock(before
, new_fl
);
952 /* The new lock breaks the old one in two pieces,
953 * so we have to use the second new lock.
957 locks_copy_lock(left
, right
);
958 locks_insert_lock(before
, left
);
960 right
->fl_start
= request
->fl_end
+ 1;
961 locks_wake_up_blocks(right
);
964 left
->fl_end
= request
->fl_start
- 1;
965 locks_wake_up_blocks(left
);
970 * Free any unused locks.
973 locks_free_lock(new_fl
);
975 locks_free_lock(new_fl2
);
980 * posix_lock_file - Apply a POSIX-style lock to a file
981 * @filp: The file to apply the lock to
982 * @fl: The lock to be applied
984 * Add a POSIX style lock to a file.
985 * We merge adjacent & overlapping locks whenever possible.
986 * POSIX locks are sorted by owner task, then by starting address
988 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
)
990 return __posix_lock_file(filp
->f_dentry
->d_inode
, fl
);
994 * posix_lock_file_wait - Apply a POSIX-style lock to a file
995 * @filp: The file to apply the lock to
996 * @fl: The lock to be applied
998 * Add a POSIX style lock to a file.
999 * We merge adjacent & overlapping locks whenever possible.
1000 * POSIX locks are sorted by owner task, then by starting address
1002 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1007 error
= __posix_lock_file(filp
->f_dentry
->d_inode
, fl
);
1008 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1010 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1014 locks_delete_block(fl
);
1019 EXPORT_SYMBOL(posix_lock_file_wait
);
1022 * locks_mandatory_locked - Check for an active lock
1023 * @inode: the file to check
1025 * Searches the inode's list of locks to find any POSIX locks which conflict.
1026 * This function is called from locks_verify_locked() only.
1028 int locks_mandatory_locked(struct inode
*inode
)
1030 fl_owner_t owner
= current
->files
;
1031 struct file_lock
*fl
;
1034 * Search the lock list for this inode for any POSIX locks.
1037 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1040 if (fl
->fl_owner
!= owner
)
1044 return fl
? -EAGAIN
: 0;
1048 * locks_mandatory_area - Check for a conflicting lock
1049 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1051 * @inode: the file to check
1052 * @filp: how the file was opened (if it was)
1053 * @offset: start of area to check
1054 * @count: length of area to check
1056 * Searches the inode's list of locks to find any POSIX locks which conflict.
1057 * This function is called from rw_verify_area() and
1058 * locks_verify_truncate().
1060 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1061 struct file
*filp
, loff_t offset
,
1064 struct file_lock fl
;
1067 locks_init_lock(&fl
);
1068 fl
.fl_owner
= current
->files
;
1069 fl
.fl_pid
= current
->tgid
;
1071 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1072 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1073 fl
.fl_flags
|= FL_SLEEP
;
1074 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1075 fl
.fl_start
= offset
;
1076 fl
.fl_end
= offset
+ count
- 1;
1079 error
= __posix_lock_file(inode
, &fl
);
1080 if (error
!= -EAGAIN
)
1082 if (!(fl
.fl_flags
& FL_SLEEP
))
1084 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1087 * If we've been sleeping someone might have
1088 * changed the permissions behind our back.
1090 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
1094 locks_delete_block(&fl
);
1101 EXPORT_SYMBOL(locks_mandatory_area
);
1103 /* We already had a lease on this file; just change its type */
1104 int lease_modify(struct file_lock
**before
, int arg
)
1106 struct file_lock
*fl
= *before
;
1107 int error
= assign_type(fl
, arg
);
1111 locks_wake_up_blocks(fl
);
1113 locks_delete_lock(before
);
1117 EXPORT_SYMBOL(lease_modify
);
1119 static void time_out_leases(struct inode
*inode
)
1121 struct file_lock
**before
;
1122 struct file_lock
*fl
;
1124 before
= &inode
->i_flock
;
1125 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1126 if ((fl
->fl_break_time
== 0)
1127 || time_before(jiffies
, fl
->fl_break_time
)) {
1128 before
= &fl
->fl_next
;
1131 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1132 if (fl
== *before
) /* lease_modify may have freed fl */
1133 before
= &fl
->fl_next
;
1138 * __break_lease - revoke all outstanding leases on file
1139 * @inode: the inode of the file to return
1140 * @mode: the open mode (read or write)
1142 * break_lease (inlined for speed) has checked there already
1143 * is a lease on this file. Leases are broken on a call to open()
1144 * or truncate(). This function can sleep unless you
1145 * specified %O_NONBLOCK to your open().
1147 int __break_lease(struct inode
*inode
, unsigned int mode
)
1149 int error
= 0, future
;
1150 struct file_lock
*new_fl
, *flock
;
1151 struct file_lock
*fl
;
1153 unsigned long break_time
;
1154 int i_have_this_lease
= 0;
1156 alloc_err
= lease_alloc(NULL
, mode
& FMODE_WRITE
? F_WRLCK
: F_RDLCK
,
1161 time_out_leases(inode
);
1163 flock
= inode
->i_flock
;
1164 if ((flock
== NULL
) || !IS_LEASE(flock
))
1167 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
)
1168 if (fl
->fl_owner
== current
->files
)
1169 i_have_this_lease
= 1;
1171 if (mode
& FMODE_WRITE
) {
1172 /* If we want write access, we have to revoke any lease. */
1173 future
= F_UNLCK
| F_INPROGRESS
;
1174 } else if (flock
->fl_type
& F_INPROGRESS
) {
1175 /* If the lease is already being broken, we just leave it */
1176 future
= flock
->fl_type
;
1177 } else if (flock
->fl_type
& F_WRLCK
) {
1178 /* Downgrade the exclusive lease to a read-only lease. */
1179 future
= F_RDLCK
| F_INPROGRESS
;
1181 /* the existing lease was read-only, so we can read too. */
1185 if (alloc_err
&& !i_have_this_lease
&& ((mode
& O_NONBLOCK
) == 0)) {
1191 if (lease_break_time
> 0) {
1192 break_time
= jiffies
+ lease_break_time
* HZ
;
1193 if (break_time
== 0)
1194 break_time
++; /* so that 0 means no break time */
1197 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1198 if (fl
->fl_type
!= future
) {
1199 fl
->fl_type
= future
;
1200 fl
->fl_break_time
= break_time
;
1201 /* lease must have lmops break callback */
1202 fl
->fl_lmops
->fl_break(fl
);
1206 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1207 error
= -EWOULDBLOCK
;
1212 break_time
= flock
->fl_break_time
;
1213 if (break_time
!= 0) {
1214 break_time
-= jiffies
;
1215 if (break_time
== 0)
1218 error
= locks_block_on_timeout(flock
, new_fl
, break_time
);
1221 time_out_leases(inode
);
1222 /* Wait for the next lease that has not been broken yet */
1223 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1224 flock
= flock
->fl_next
) {
1225 if (flock
->fl_type
& F_INPROGRESS
)
1234 locks_free_lock(new_fl
);
1238 EXPORT_SYMBOL(__break_lease
);
1243 * @time: pointer to a timespec which will contain the last modified time
1245 * This is to force NFS clients to flush their caches for files with
1246 * exclusive leases. The justification is that if someone has an
1247 * exclusive lease, then they could be modifiying it.
1249 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1251 struct file_lock
*flock
= inode
->i_flock
;
1252 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1253 *time
= current_fs_time(inode
->i_sb
);
1255 *time
= inode
->i_mtime
;
1258 EXPORT_SYMBOL(lease_get_mtime
);
1261 * fcntl_getlease - Enquire what lease is currently active
1264 * The value returned by this function will be one of
1265 * (if no lease break is pending):
1267 * %F_RDLCK to indicate a shared lease is held.
1269 * %F_WRLCK to indicate an exclusive lease is held.
1271 * %F_UNLCK to indicate no lease is held.
1273 * (if a lease break is pending):
1275 * %F_RDLCK to indicate an exclusive lease needs to be
1276 * changed to a shared lease (or removed).
1278 * %F_UNLCK to indicate the lease needs to be removed.
1280 * XXX: sfr & willy disagree over whether F_INPROGRESS
1281 * should be returned to userspace.
1283 int fcntl_getlease(struct file
*filp
)
1285 struct file_lock
*fl
;
1289 time_out_leases(filp
->f_dentry
->d_inode
);
1290 for (fl
= filp
->f_dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1292 if (fl
->fl_file
== filp
) {
1293 type
= fl
->fl_type
& ~F_INPROGRESS
;
1302 * __setlease - sets a lease on an open file
1303 * @filp: file pointer
1304 * @arg: type of lease to obtain
1305 * @flp: input - file_lock to use, output - file_lock inserted
1307 * The (input) flp->fl_lmops->fl_break function is required
1310 * Called with kernel lock held.
1312 static int __setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1314 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1315 struct dentry
*dentry
= filp
->f_dentry
;
1316 struct inode
*inode
= dentry
->d_inode
;
1317 int error
, rdlease_count
= 0, wrlease_count
= 0;
1319 time_out_leases(inode
);
1322 if (!flp
|| !(*flp
) || !(*flp
)->fl_lmops
|| !(*flp
)->fl_lmops
->fl_break
)
1328 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1330 if ((arg
== F_WRLCK
)
1331 && ((atomic_read(&dentry
->d_count
) > 1)
1332 || (atomic_read(&inode
->i_count
) > 1)))
1336 * At this point, we know that if there is an exclusive
1337 * lease on this file, then we hold it on this filp
1338 * (otherwise our open of this file would have blocked).
1339 * And if we are trying to acquire an exclusive lease,
1340 * then the file is not open by anyone (including us)
1341 * except for this filp.
1343 for (before
= &inode
->i_flock
;
1344 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1345 before
= &fl
->fl_next
) {
1346 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1348 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1350 * Someone is in the process of opening this
1351 * file for writing so we may not take an
1352 * exclusive lease on it.
1359 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1360 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1363 if (my_before
!= NULL
) {
1364 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1376 error
= lease_alloc(filp
, arg
, &fl
);
1380 locks_copy_lock(fl
, lease
);
1382 locks_insert_lock(before
, fl
);
1390 * setlease - sets a lease on an open file
1391 * @filp: file pointer
1392 * @arg: type of lease to obtain
1393 * @lease: file_lock to use
1395 * Call this to establish a lease on the file.
1396 * The fl_lmops fl_break function is required by break_lease
1399 int setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1401 struct dentry
*dentry
= filp
->f_dentry
;
1402 struct inode
*inode
= dentry
->d_inode
;
1405 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1407 if (!S_ISREG(inode
->i_mode
))
1409 error
= security_file_lock(filp
, arg
);
1414 error
= __setlease(filp
, arg
, lease
);
1420 EXPORT_SYMBOL(setlease
);
1423 * fcntl_setlease - sets a lease on an open file
1424 * @fd: open file descriptor
1425 * @filp: file pointer
1426 * @arg: type of lease to obtain
1428 * Call this fcntl to establish a lease on the file.
1429 * Note that you also need to call %F_SETSIG to
1430 * receive a signal when the lease is broken.
1432 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1434 struct file_lock fl
, *flp
= &fl
;
1435 struct dentry
*dentry
= filp
->f_dentry
;
1436 struct inode
*inode
= dentry
->d_inode
;
1439 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1441 if (!S_ISREG(inode
->i_mode
))
1443 error
= security_file_lock(filp
, arg
);
1447 locks_init_lock(&fl
);
1448 error
= lease_init(filp
, arg
, &fl
);
1454 error
= __setlease(filp
, arg
, &flp
);
1455 if (error
|| arg
== F_UNLCK
)
1458 error
= fasync_helper(fd
, filp
, 1, &flp
->fl_fasync
);
1460 /* remove lease just inserted by __setlease */
1461 flp
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1462 flp
->fl_break_time
= jiffies
- 10;
1463 time_out_leases(inode
);
1467 error
= f_setown(filp
, current
->pid
, 0);
1474 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1475 * @filp: The file to apply the lock to
1476 * @fl: The lock to be applied
1478 * Add a FLOCK style lock to a file.
1480 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1485 error
= flock_lock_file(filp
, fl
);
1486 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1488 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1492 locks_delete_block(fl
);
1498 EXPORT_SYMBOL(flock_lock_file_wait
);
1501 * sys_flock: - flock() system call.
1502 * @fd: the file descriptor to lock.
1503 * @cmd: the type of lock to apply.
1505 * Apply a %FL_FLOCK style lock to an open file descriptor.
1506 * The @cmd can be one of
1508 * %LOCK_SH -- a shared lock.
1510 * %LOCK_EX -- an exclusive lock.
1512 * %LOCK_UN -- remove an existing lock.
1514 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1516 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1517 * processes read and write access respectively.
1519 asmlinkage
long sys_flock(unsigned int fd
, unsigned int cmd
)
1522 struct file_lock
*lock
;
1523 int can_sleep
, unlock
;
1531 can_sleep
= !(cmd
& LOCK_NB
);
1533 unlock
= (cmd
== LOCK_UN
);
1535 if (!unlock
&& !(cmd
& LOCK_MAND
) && !(filp
->f_mode
& 3))
1538 error
= flock_make_lock(filp
, &lock
, cmd
);
1542 lock
->fl_flags
|= FL_SLEEP
;
1544 error
= security_file_lock(filp
, cmd
);
1548 if (filp
->f_op
&& filp
->f_op
->flock
)
1549 error
= filp
->f_op
->flock(filp
,
1550 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1553 error
= flock_lock_file_wait(filp
, lock
);
1556 if (list_empty(&lock
->fl_link
)) {
1557 locks_free_lock(lock
);
1566 /* Report the first existing lock that would conflict with l.
1567 * This implements the F_GETLK command of fcntl().
1569 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1571 struct file_lock
*fl
, cfl
, file_lock
;
1576 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1579 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1582 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1586 if (filp
->f_op
&& filp
->f_op
->lock
) {
1587 error
= filp
->f_op
->lock(filp
, F_GETLK
, &file_lock
);
1588 if (file_lock
.fl_ops
&& file_lock
.fl_ops
->fl_release_private
)
1589 file_lock
.fl_ops
->fl_release_private(&file_lock
);
1593 fl
= (file_lock
.fl_type
== F_UNLCK
? NULL
: &file_lock
);
1595 fl
= (posix_test_lock(filp
, &file_lock
, &cfl
) ? &cfl
: NULL
);
1598 flock
.l_type
= F_UNLCK
;
1600 flock
.l_pid
= fl
->fl_pid
;
1601 #if BITS_PER_LONG == 32
1603 * Make sure we can represent the posix lock via
1604 * legacy 32bit flock.
1607 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1609 if ((fl
->fl_end
!= OFFSET_MAX
)
1610 && (fl
->fl_end
> OFFT_OFFSET_MAX
))
1613 flock
.l_start
= fl
->fl_start
;
1614 flock
.l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1615 fl
->fl_end
- fl
->fl_start
+ 1;
1617 flock
.l_type
= fl
->fl_type
;
1620 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1626 /* Apply the lock described by l to an open file descriptor.
1627 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1629 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1630 struct flock __user
*l
)
1632 struct file_lock
*file_lock
= locks_alloc_lock();
1634 struct inode
*inode
;
1637 if (file_lock
== NULL
)
1641 * This might block, so we do it before checking the inode.
1644 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1647 inode
= filp
->f_dentry
->d_inode
;
1649 /* Don't allow mandatory locks on files that may be memory mapped
1652 if (IS_MANDLOCK(inode
) &&
1653 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1654 mapping_writably_mapped(filp
->f_mapping
)) {
1660 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1663 if (cmd
== F_SETLKW
) {
1664 file_lock
->fl_flags
|= FL_SLEEP
;
1668 switch (flock
.l_type
) {
1670 if (!(filp
->f_mode
& FMODE_READ
))
1674 if (!(filp
->f_mode
& FMODE_WRITE
))
1684 error
= security_file_lock(filp
, file_lock
->fl_type
);
1688 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
)
1689 error
= filp
->f_op
->lock(filp
, cmd
, file_lock
);
1692 error
= __posix_lock_file(inode
, file_lock
);
1693 if ((error
!= -EAGAIN
) || (cmd
== F_SETLK
))
1695 error
= wait_event_interruptible(file_lock
->fl_wait
,
1696 !file_lock
->fl_next
);
1700 locks_delete_block(file_lock
);
1706 * Attempt to detect a close/fcntl race and recover by
1707 * releasing the lock that was just acquired.
1709 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1710 flock
.l_type
= F_UNLCK
;
1715 locks_free_lock(file_lock
);
1719 #if BITS_PER_LONG == 32
1720 /* Report the first existing lock that would conflict with l.
1721 * This implements the F_GETLK command of fcntl().
1723 int fcntl_getlk64(struct file
*filp
, struct flock64 __user
*l
)
1725 struct file_lock
*fl
, cfl
, file_lock
;
1726 struct flock64 flock
;
1730 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1733 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1736 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1740 if (filp
->f_op
&& filp
->f_op
->lock
) {
1741 error
= filp
->f_op
->lock(filp
, F_GETLK
, &file_lock
);
1742 if (file_lock
.fl_ops
&& file_lock
.fl_ops
->fl_release_private
)
1743 file_lock
.fl_ops
->fl_release_private(&file_lock
);
1747 fl
= (file_lock
.fl_type
== F_UNLCK
? NULL
: &file_lock
);
1749 fl
= (posix_test_lock(filp
, &file_lock
, &cfl
) ? &cfl
: NULL
);
1752 flock
.l_type
= F_UNLCK
;
1754 flock
.l_pid
= fl
->fl_pid
;
1755 flock
.l_start
= fl
->fl_start
;
1756 flock
.l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1757 fl
->fl_end
- fl
->fl_start
+ 1;
1759 flock
.l_type
= fl
->fl_type
;
1762 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1769 /* Apply the lock described by l to an open file descriptor.
1770 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1772 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1773 struct flock64 __user
*l
)
1775 struct file_lock
*file_lock
= locks_alloc_lock();
1776 struct flock64 flock
;
1777 struct inode
*inode
;
1780 if (file_lock
== NULL
)
1784 * This might block, so we do it before checking the inode.
1787 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1790 inode
= filp
->f_dentry
->d_inode
;
1792 /* Don't allow mandatory locks on files that may be memory mapped
1795 if (IS_MANDLOCK(inode
) &&
1796 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1797 mapping_writably_mapped(filp
->f_mapping
)) {
1803 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1806 if (cmd
== F_SETLKW64
) {
1807 file_lock
->fl_flags
|= FL_SLEEP
;
1811 switch (flock
.l_type
) {
1813 if (!(filp
->f_mode
& FMODE_READ
))
1817 if (!(filp
->f_mode
& FMODE_WRITE
))
1827 error
= security_file_lock(filp
, file_lock
->fl_type
);
1831 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
)
1832 error
= filp
->f_op
->lock(filp
, cmd
, file_lock
);
1835 error
= __posix_lock_file(inode
, file_lock
);
1836 if ((error
!= -EAGAIN
) || (cmd
== F_SETLK64
))
1838 error
= wait_event_interruptible(file_lock
->fl_wait
,
1839 !file_lock
->fl_next
);
1843 locks_delete_block(file_lock
);
1849 * Attempt to detect a close/fcntl race and recover by
1850 * releasing the lock that was just acquired.
1852 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1853 flock
.l_type
= F_UNLCK
;
1858 locks_free_lock(file_lock
);
1861 #endif /* BITS_PER_LONG == 32 */
1864 * This function is called when the file is being removed
1865 * from the task's fd array. POSIX locks belonging to this task
1866 * are deleted at this time.
1868 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
1870 struct file_lock lock
, **before
;
1873 * If there are no locks held on this file, we don't need to call
1874 * posix_lock_file(). Another process could be setting a lock on this
1875 * file at the same time, but we wouldn't remove that lock anyway.
1877 before
= &filp
->f_dentry
->d_inode
->i_flock
;
1878 if (*before
== NULL
)
1881 lock
.fl_type
= F_UNLCK
;
1882 lock
.fl_flags
= FL_POSIX
;
1884 lock
.fl_end
= OFFSET_MAX
;
1885 lock
.fl_owner
= owner
;
1886 lock
.fl_pid
= current
->tgid
;
1887 lock
.fl_file
= filp
;
1889 lock
.fl_lmops
= NULL
;
1891 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
) {
1892 filp
->f_op
->lock(filp
, F_SETLK
, &lock
);
1896 /* Can't use posix_lock_file here; we need to remove it no matter
1897 * which pid we have.
1900 while (*before
!= NULL
) {
1901 struct file_lock
*fl
= *before
;
1902 if (IS_POSIX(fl
) && posix_same_owner(fl
, &lock
)) {
1903 locks_delete_lock(before
);
1906 before
= &fl
->fl_next
;
1910 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
1911 lock
.fl_ops
->fl_release_private(&lock
);
1914 EXPORT_SYMBOL(locks_remove_posix
);
1917 * This function is called on the last close of an open file.
1919 void locks_remove_flock(struct file
*filp
)
1921 struct inode
* inode
= filp
->f_dentry
->d_inode
;
1922 struct file_lock
*fl
;
1923 struct file_lock
**before
;
1925 if (!inode
->i_flock
)
1928 if (filp
->f_op
&& filp
->f_op
->flock
) {
1929 struct file_lock fl
= {
1930 .fl_pid
= current
->tgid
,
1932 .fl_flags
= FL_FLOCK
,
1934 .fl_end
= OFFSET_MAX
,
1936 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
1937 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
1938 fl
.fl_ops
->fl_release_private(&fl
);
1942 before
= &inode
->i_flock
;
1944 while ((fl
= *before
) != NULL
) {
1945 if (fl
->fl_file
== filp
) {
1947 locks_delete_lock(before
);
1951 lease_modify(before
, F_UNLCK
);
1957 before
= &fl
->fl_next
;
1963 * posix_unblock_lock - stop waiting for a file lock
1964 * @filp: how the file was opened
1965 * @waiter: the lock which was waiting
1967 * lockd needs to block waiting for locks.
1970 posix_unblock_lock(struct file
*filp
, struct file_lock
*waiter
)
1975 if (waiter
->fl_next
)
1976 __locks_delete_block(waiter
);
1983 EXPORT_SYMBOL(posix_unblock_lock
);
1985 static void lock_get_status(char* out
, struct file_lock
*fl
, int id
, char *pfx
)
1987 struct inode
*inode
= NULL
;
1989 if (fl
->fl_file
!= NULL
)
1990 inode
= fl
->fl_file
->f_dentry
->d_inode
;
1992 out
+= sprintf(out
, "%d:%s ", id
, pfx
);
1994 out
+= sprintf(out
, "%6s %s ",
1995 (fl
->fl_flags
& FL_ACCESS
) ? "ACCESS" : "POSIX ",
1996 (inode
== NULL
) ? "*NOINODE*" :
1997 (IS_MANDLOCK(inode
) &&
1998 (inode
->i_mode
& (S_IXGRP
| S_ISGID
)) == S_ISGID
) ?
1999 "MANDATORY" : "ADVISORY ");
2000 } else if (IS_FLOCK(fl
)) {
2001 if (fl
->fl_type
& LOCK_MAND
) {
2002 out
+= sprintf(out
, "FLOCK MSNFS ");
2004 out
+= sprintf(out
, "FLOCK ADVISORY ");
2006 } else if (IS_LEASE(fl
)) {
2007 out
+= sprintf(out
, "LEASE ");
2008 if (fl
->fl_type
& F_INPROGRESS
)
2009 out
+= sprintf(out
, "BREAKING ");
2010 else if (fl
->fl_file
)
2011 out
+= sprintf(out
, "ACTIVE ");
2013 out
+= sprintf(out
, "BREAKER ");
2015 out
+= sprintf(out
, "UNKNOWN UNKNOWN ");
2017 if (fl
->fl_type
& LOCK_MAND
) {
2018 out
+= sprintf(out
, "%s ",
2019 (fl
->fl_type
& LOCK_READ
)
2020 ? (fl
->fl_type
& LOCK_WRITE
) ? "RW " : "READ "
2021 : (fl
->fl_type
& LOCK_WRITE
) ? "WRITE" : "NONE ");
2023 out
+= sprintf(out
, "%s ",
2024 (fl
->fl_type
& F_INPROGRESS
)
2025 ? (fl
->fl_type
& F_UNLCK
) ? "UNLCK" : "READ "
2026 : (fl
->fl_type
& F_WRLCK
) ? "WRITE" : "READ ");
2029 #ifdef WE_CAN_BREAK_LSLK_NOW
2030 out
+= sprintf(out
, "%d %s:%ld ", fl
->fl_pid
,
2031 inode
->i_sb
->s_id
, inode
->i_ino
);
2033 /* userspace relies on this representation of dev_t ;-( */
2034 out
+= sprintf(out
, "%d %02x:%02x:%ld ", fl
->fl_pid
,
2035 MAJOR(inode
->i_sb
->s_dev
),
2036 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2039 out
+= sprintf(out
, "%d <none>:0 ", fl
->fl_pid
);
2042 if (fl
->fl_end
== OFFSET_MAX
)
2043 out
+= sprintf(out
, "%Ld EOF\n", fl
->fl_start
);
2045 out
+= sprintf(out
, "%Ld %Ld\n", fl
->fl_start
,
2048 out
+= sprintf(out
, "0 EOF\n");
2052 static void move_lock_status(char **p
, off_t
* pos
, off_t offset
)
2056 if(*pos
>= offset
) {
2057 /* the complete line is valid */
2062 if(*pos
+len
> offset
) {
2063 /* use the second part of the line */
2064 int i
= offset
-*pos
;
2065 memmove(*p
,*p
+i
,len
-i
);
2070 /* discard the complete line */
2075 * get_locks_status - reports lock usage in /proc/locks
2076 * @buffer: address in userspace to write into
2078 * @offset: how far we are through the buffer
2079 * @length: how much to read
2082 int get_locks_status(char *buffer
, char **start
, off_t offset
, int length
)
2084 struct list_head
*tmp
;
2090 list_for_each(tmp
, &file_lock_list
) {
2091 struct list_head
*btmp
;
2092 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
2093 lock_get_status(q
, fl
, ++i
, "");
2094 move_lock_status(&q
, &pos
, offset
);
2096 if(pos
>= offset
+length
)
2099 list_for_each(btmp
, &fl
->fl_block
) {
2100 struct file_lock
*bfl
= list_entry(btmp
,
2101 struct file_lock
, fl_block
);
2102 lock_get_status(q
, bfl
, i
, " ->");
2103 move_lock_status(&q
, &pos
, offset
);
2105 if(pos
>= offset
+length
)
2112 if(q
-buffer
< length
)
2118 * lock_may_read - checks that the region is free of locks
2119 * @inode: the inode that is being read
2120 * @start: the first byte to read
2121 * @len: the number of bytes to read
2123 * Emulates Windows locking requirements. Whole-file
2124 * mandatory locks (share modes) can prohibit a read and
2125 * byte-range POSIX locks can prohibit a read if they overlap.
2127 * N.B. this function is only ever called
2128 * from knfsd and ownership of locks is never checked.
2130 int lock_may_read(struct inode
*inode
, loff_t start
, unsigned long len
)
2132 struct file_lock
*fl
;
2135 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2137 if (fl
->fl_type
== F_RDLCK
)
2139 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2141 } else if (IS_FLOCK(fl
)) {
2142 if (!(fl
->fl_type
& LOCK_MAND
))
2144 if (fl
->fl_type
& LOCK_READ
)
2155 EXPORT_SYMBOL(lock_may_read
);
2158 * lock_may_write - checks that the region is free of locks
2159 * @inode: the inode that is being written
2160 * @start: the first byte to write
2161 * @len: the number of bytes to write
2163 * Emulates Windows locking requirements. Whole-file
2164 * mandatory locks (share modes) can prohibit a write and
2165 * byte-range POSIX locks can prohibit a write if they overlap.
2167 * N.B. this function is only ever called
2168 * from knfsd and ownership of locks is never checked.
2170 int lock_may_write(struct inode
*inode
, loff_t start
, unsigned long len
)
2172 struct file_lock
*fl
;
2175 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2177 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2179 } else if (IS_FLOCK(fl
)) {
2180 if (!(fl
->fl_type
& LOCK_MAND
))
2182 if (fl
->fl_type
& LOCK_WRITE
)
2193 EXPORT_SYMBOL(lock_may_write
);
2195 static inline void __steal_locks(struct file
*file
, fl_owner_t from
)
2197 struct inode
*inode
= file
->f_dentry
->d_inode
;
2198 struct file_lock
*fl
= inode
->i_flock
;
2201 if (fl
->fl_file
== file
&& fl
->fl_owner
== from
)
2202 fl
->fl_owner
= current
->files
;
2207 /* When getting ready for executing a binary, we make sure that current
2208 * has a files_struct on its own. Before dropping the old files_struct,
2209 * we take over ownership of all locks for all file descriptors we own.
2210 * Note that we may accidentally steal a lock for a file that a sibling
2211 * has created since the unshare_files() call.
2213 void steal_locks(fl_owner_t from
)
2215 struct files_struct
*files
= current
->files
;
2217 struct fdtable
*fdt
;
2225 fdt
= files_fdtable(files
);
2229 if (i
>= fdt
->max_fdset
|| i
>= fdt
->max_fds
)
2231 set
= fdt
->open_fds
->fds_bits
[j
++];
2234 struct file
*file
= fdt
->fd
[i
];
2236 __steal_locks(file
, from
);
2245 EXPORT_SYMBOL(steal_locks
);
2247 static int __init
filelock_init(void)
2249 filelock_cache
= kmem_cache_create("file_lock_cache",
2250 sizeof(struct file_lock
), 0, SLAB_PANIC
,
2255 core_initcall(filelock_init
);