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 /* Free a lock which is not in use. */
157 static inline void locks_free_lock(struct file_lock
*fl
)
163 if (waitqueue_active(&fl
->fl_wait
))
164 panic("Attempting to free lock with active wait queue");
166 if (!list_empty(&fl
->fl_block
))
167 panic("Attempting to free lock with active block list");
169 if (!list_empty(&fl
->fl_link
))
170 panic("Attempting to free lock on active lock list");
173 if (fl
->fl_ops
->fl_release_private
)
174 fl
->fl_ops
->fl_release_private(fl
);
179 if (fl
->fl_lmops
->fl_release_private
)
180 fl
->fl_lmops
->fl_release_private(fl
);
184 kmem_cache_free(filelock_cache
, fl
);
187 void locks_init_lock(struct file_lock
*fl
)
189 INIT_LIST_HEAD(&fl
->fl_link
);
190 INIT_LIST_HEAD(&fl
->fl_block
);
191 init_waitqueue_head(&fl
->fl_wait
);
193 fl
->fl_fasync
= NULL
;
199 fl
->fl_start
= fl
->fl_end
= 0;
204 EXPORT_SYMBOL(locks_init_lock
);
207 * Initialises the fields of the file lock which are invariant for
210 static void init_once(void *foo
, kmem_cache_t
*cache
, unsigned long flags
)
212 struct file_lock
*lock
= (struct file_lock
*) foo
;
214 if ((flags
& (SLAB_CTOR_VERIFY
|SLAB_CTOR_CONSTRUCTOR
)) !=
215 SLAB_CTOR_CONSTRUCTOR
)
218 locks_init_lock(lock
);
222 * Initialize a new lock from an existing file_lock structure.
224 void locks_copy_lock(struct file_lock
*new, struct file_lock
*fl
)
226 new->fl_owner
= fl
->fl_owner
;
227 new->fl_pid
= fl
->fl_pid
;
228 new->fl_file
= fl
->fl_file
;
229 new->fl_flags
= fl
->fl_flags
;
230 new->fl_type
= fl
->fl_type
;
231 new->fl_start
= fl
->fl_start
;
232 new->fl_end
= fl
->fl_end
;
233 new->fl_ops
= fl
->fl_ops
;
234 new->fl_lmops
= fl
->fl_lmops
;
235 if (fl
->fl_ops
&& fl
->fl_ops
->fl_copy_lock
)
236 fl
->fl_ops
->fl_copy_lock(new, fl
);
237 if (fl
->fl_lmops
&& fl
->fl_lmops
->fl_copy_lock
)
238 fl
->fl_lmops
->fl_copy_lock(new, fl
);
241 EXPORT_SYMBOL(locks_copy_lock
);
243 static inline int flock_translate_cmd(int cmd
) {
245 return cmd
& (LOCK_MAND
| LOCK_RW
);
257 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
258 static int flock_make_lock(struct file
*filp
, struct file_lock
**lock
,
261 struct file_lock
*fl
;
262 int type
= flock_translate_cmd(cmd
);
266 fl
= locks_alloc_lock();
271 fl
->fl_pid
= current
->tgid
;
272 fl
->fl_flags
= FL_FLOCK
;
274 fl
->fl_end
= OFFSET_MAX
;
280 static int assign_type(struct file_lock
*fl
, int type
)
294 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
297 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
302 switch (l
->l_whence
) {
310 start
= i_size_read(filp
->f_dentry
->d_inode
);
316 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
317 POSIX-2001 defines it. */
319 end
= start
+ l
->l_len
- 1;
327 if (l
->l_len
> 0 && end
< 0)
330 fl
->fl_start
= start
; /* we record the absolute position */
333 fl
->fl_end
= OFFSET_MAX
;
335 fl
->fl_owner
= current
->files
;
336 fl
->fl_pid
= current
->tgid
;
338 fl
->fl_flags
= FL_POSIX
;
342 return assign_type(fl
, l
->l_type
);
345 #if BITS_PER_LONG == 32
346 static int flock64_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
351 switch (l
->l_whence
) {
359 start
= i_size_read(filp
->f_dentry
->d_inode
);
365 if (((start
+= l
->l_start
) < 0) || (l
->l_len
< 0))
367 fl
->fl_end
= start
+ l
->l_len
- 1;
368 if (l
->l_len
> 0 && fl
->fl_end
< 0)
370 fl
->fl_start
= start
; /* we record the absolute position */
372 fl
->fl_end
= OFFSET_MAX
;
374 fl
->fl_owner
= current
->files
;
375 fl
->fl_pid
= current
->tgid
;
377 fl
->fl_flags
= FL_POSIX
;
385 fl
->fl_type
= l
->l_type
;
395 /* default lease lock manager operations */
396 static void lease_break_callback(struct file_lock
*fl
)
398 kill_fasync(&fl
->fl_fasync
, SIGIO
, POLL_MSG
);
401 static void lease_release_private_callback(struct file_lock
*fl
)
406 f_delown(fl
->fl_file
);
407 fl
->fl_file
->f_owner
.signum
= 0;
410 static int lease_mylease_callback(struct file_lock
*fl
, struct file_lock
*try)
412 return fl
->fl_file
== try->fl_file
;
415 static struct lock_manager_operations lease_manager_ops
= {
416 .fl_break
= lease_break_callback
,
417 .fl_release_private
= lease_release_private_callback
,
418 .fl_mylease
= lease_mylease_callback
,
419 .fl_change
= lease_modify
,
423 * Initialize a lease, use the default lock manager operations
425 static int lease_init(struct file
*filp
, int type
, struct file_lock
*fl
)
427 fl
->fl_owner
= current
->files
;
428 fl
->fl_pid
= current
->tgid
;
431 fl
->fl_flags
= FL_LEASE
;
432 if (assign_type(fl
, type
) != 0) {
437 fl
->fl_end
= OFFSET_MAX
;
439 fl
->fl_lmops
= &lease_manager_ops
;
443 /* Allocate a file_lock initialised to this type of lease */
444 static int lease_alloc(struct file
*filp
, int type
, struct file_lock
**flp
)
446 struct file_lock
*fl
= locks_alloc_lock();
452 error
= lease_init(filp
, type
, fl
);
459 /* Check if two locks overlap each other.
461 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
463 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
464 (fl2
->fl_end
>= fl1
->fl_start
));
468 * Check whether two locks have the same owner.
471 posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
473 if (fl1
->fl_lmops
&& fl1
->fl_lmops
->fl_compare_owner
)
474 return fl2
->fl_lmops
== fl1
->fl_lmops
&&
475 fl1
->fl_lmops
->fl_compare_owner(fl1
, fl2
);
476 return fl1
->fl_owner
== fl2
->fl_owner
;
479 /* Remove waiter from blocker's block list.
480 * When blocker ends up pointing to itself then the list is empty.
482 static inline void __locks_delete_block(struct file_lock
*waiter
)
484 list_del_init(&waiter
->fl_block
);
485 list_del_init(&waiter
->fl_link
);
486 waiter
->fl_next
= NULL
;
491 static void locks_delete_block(struct file_lock
*waiter
)
494 __locks_delete_block(waiter
);
498 /* Insert waiter into blocker's block list.
499 * We use a circular list so that processes can be easily woken up in
500 * the order they blocked. The documentation doesn't require this but
501 * it seems like the reasonable thing to do.
503 static void locks_insert_block(struct file_lock
*blocker
,
504 struct file_lock
*waiter
)
506 if (!list_empty(&waiter
->fl_block
)) {
507 printk(KERN_ERR
"locks_insert_block: removing duplicated lock "
508 "(pid=%d %Ld-%Ld type=%d)\n", waiter
->fl_pid
,
509 waiter
->fl_start
, waiter
->fl_end
, waiter
->fl_type
);
510 __locks_delete_block(waiter
);
512 list_add_tail(&waiter
->fl_block
, &blocker
->fl_block
);
513 waiter
->fl_next
= blocker
;
514 if (IS_POSIX(blocker
))
515 list_add(&waiter
->fl_link
, &blocked_list
);
518 /* Wake up processes blocked waiting for blocker.
519 * If told to wait then schedule the processes until the block list
520 * is empty, otherwise empty the block list ourselves.
522 static void locks_wake_up_blocks(struct file_lock
*blocker
)
524 while (!list_empty(&blocker
->fl_block
)) {
525 struct file_lock
*waiter
= list_entry(blocker
->fl_block
.next
,
526 struct file_lock
, fl_block
);
527 __locks_delete_block(waiter
);
528 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->fl_notify
)
529 waiter
->fl_lmops
->fl_notify(waiter
);
531 wake_up(&waiter
->fl_wait
);
535 /* Insert file lock fl into an inode's lock list at the position indicated
536 * by pos. At the same time add the lock to the global file lock list.
538 static void locks_insert_lock(struct file_lock
**pos
, struct file_lock
*fl
)
540 list_add(&fl
->fl_link
, &file_lock_list
);
542 /* insert into file's list */
546 if (fl
->fl_ops
&& fl
->fl_ops
->fl_insert
)
547 fl
->fl_ops
->fl_insert(fl
);
551 * Delete a lock and then free it.
552 * Wake up processes that are blocked waiting for this lock,
553 * notify the FS that the lock has been cleared and
554 * finally free the lock.
556 static void locks_delete_lock(struct file_lock
**thisfl_p
)
558 struct file_lock
*fl
= *thisfl_p
;
560 *thisfl_p
= fl
->fl_next
;
562 list_del_init(&fl
->fl_link
);
564 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
565 if (fl
->fl_fasync
!= NULL
) {
566 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
567 fl
->fl_fasync
= NULL
;
570 if (fl
->fl_ops
&& fl
->fl_ops
->fl_remove
)
571 fl
->fl_ops
->fl_remove(fl
);
573 locks_wake_up_blocks(fl
);
577 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
578 * checks for shared/exclusive status of overlapping locks.
580 static int locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
582 if (sys_fl
->fl_type
== F_WRLCK
)
584 if (caller_fl
->fl_type
== F_WRLCK
)
589 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
590 * checking before calling the locks_conflict().
592 static int posix_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
594 /* POSIX locks owned by the same process do not conflict with
597 if (!IS_POSIX(sys_fl
) || posix_same_owner(caller_fl
, sys_fl
))
600 /* Check whether they overlap */
601 if (!locks_overlap(caller_fl
, sys_fl
))
604 return (locks_conflict(caller_fl
, sys_fl
));
607 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
608 * checking before calling the locks_conflict().
610 static int flock_locks_conflict(struct file_lock
*caller_fl
, struct file_lock
*sys_fl
)
612 /* FLOCK locks referring to the same filp do not conflict with
615 if (!IS_FLOCK(sys_fl
) || (caller_fl
->fl_file
== sys_fl
->fl_file
))
617 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
620 return (locks_conflict(caller_fl
, sys_fl
));
623 static int interruptible_sleep_on_locked(wait_queue_head_t
*fl_wait
, int timeout
)
626 DECLARE_WAITQUEUE(wait
, current
);
628 __set_current_state(TASK_INTERRUPTIBLE
);
629 add_wait_queue(fl_wait
, &wait
);
633 result
= schedule_timeout(timeout
);
634 if (signal_pending(current
))
635 result
= -ERESTARTSYS
;
636 remove_wait_queue(fl_wait
, &wait
);
637 __set_current_state(TASK_RUNNING
);
641 static int locks_block_on_timeout(struct file_lock
*blocker
, struct file_lock
*waiter
, int time
)
644 locks_insert_block(blocker
, waiter
);
645 result
= interruptible_sleep_on_locked(&waiter
->fl_wait
, time
);
646 __locks_delete_block(waiter
);
651 posix_test_lock(struct file
*filp
, struct file_lock
*fl
)
653 struct file_lock
*cfl
;
656 for (cfl
= filp
->f_dentry
->d_inode
->i_flock
; cfl
; cfl
= cfl
->fl_next
) {
659 if (posix_locks_conflict(cfl
, fl
))
667 EXPORT_SYMBOL(posix_test_lock
);
669 /* This function tests for deadlock condition before putting a process to
670 * sleep. The detection scheme is no longer recursive. Recursive was neat,
671 * but dangerous - we risked stack corruption if the lock data was bad, or
672 * if the recursion was too deep for any other reason.
674 * We rely on the fact that a task can only be on one lock's wait queue
675 * at a time. When we find blocked_task on a wait queue we can re-search
676 * with blocked_task equal to that queue's owner, until either blocked_task
677 * isn't found, or blocked_task is found on a queue owned by my_task.
679 * Note: the above assumption may not be true when handling lock requests
680 * from a broken NFS client. But broken NFS clients have a lot more to
681 * worry about than proper deadlock detection anyway... --okir
683 int posix_locks_deadlock(struct file_lock
*caller_fl
,
684 struct file_lock
*block_fl
)
686 struct list_head
*tmp
;
689 if (posix_same_owner(caller_fl
, block_fl
))
691 list_for_each(tmp
, &blocked_list
) {
692 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
693 if (posix_same_owner(fl
, block_fl
)) {
702 EXPORT_SYMBOL(posix_locks_deadlock
);
704 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
705 * at the head of the list, but that's secret knowledge known only to
706 * flock_lock_file and posix_lock_file.
708 static int flock_lock_file(struct file
*filp
, struct file_lock
*new_fl
)
710 struct file_lock
**before
;
711 struct inode
* inode
= filp
->f_dentry
->d_inode
;
716 for_each_lock(inode
, before
) {
717 struct file_lock
*fl
= *before
;
722 if (filp
!= fl
->fl_file
)
724 if (new_fl
->fl_type
== fl
->fl_type
)
727 locks_delete_lock(before
);
732 if (new_fl
->fl_type
== F_UNLCK
)
736 * If a higher-priority process was blocked on the old file lock,
737 * give it the opportunity to lock the file.
743 for_each_lock(inode
, before
) {
744 struct file_lock
*fl
= *before
;
749 if (!flock_locks_conflict(new_fl
, fl
))
752 if (new_fl
->fl_flags
& FL_SLEEP
) {
753 locks_insert_block(fl
, new_fl
);
757 locks_insert_lock(&inode
->i_flock
, new_fl
);
765 EXPORT_SYMBOL(posix_lock_file
);
767 static int __posix_lock_file(struct inode
*inode
, struct file_lock
*request
)
769 struct file_lock
*fl
;
770 struct file_lock
*new_fl
, *new_fl2
;
771 struct file_lock
*left
= NULL
;
772 struct file_lock
*right
= NULL
;
773 struct file_lock
**before
;
774 int error
, added
= 0;
777 * We may need two file_lock structures for this operation,
778 * so we get them in advance to avoid races.
780 new_fl
= locks_alloc_lock();
781 new_fl2
= locks_alloc_lock();
784 if (request
->fl_type
!= F_UNLCK
) {
785 for_each_lock(inode
, before
) {
786 struct file_lock
*fl
= *before
;
789 if (!posix_locks_conflict(request
, fl
))
792 if (!(request
->fl_flags
& FL_SLEEP
))
795 if (posix_locks_deadlock(request
, fl
))
798 locks_insert_block(fl
, request
);
803 /* If we're just looking for a conflict, we're done. */
805 if (request
->fl_flags
& FL_ACCESS
)
808 error
= -ENOLCK
; /* "no luck" */
809 if (!(new_fl
&& new_fl2
))
813 * We've allocated the new locks in advance, so there are no
814 * errors possible (and no blocking operations) from here on.
816 * Find the first old lock with the same owner as the new lock.
819 before
= &inode
->i_flock
;
821 /* First skip locks owned by other processes. */
822 while ((fl
= *before
) && (!IS_POSIX(fl
) ||
823 !posix_same_owner(request
, fl
))) {
824 before
= &fl
->fl_next
;
827 /* Process locks with this owner. */
828 while ((fl
= *before
) && posix_same_owner(request
, fl
)) {
829 /* Detect adjacent or overlapping regions (if same lock type)
831 if (request
->fl_type
== fl
->fl_type
) {
832 /* In all comparisons of start vs end, use
833 * "start - 1" rather than "end + 1". If end
834 * is OFFSET_MAX, end + 1 will become negative.
836 if (fl
->fl_end
< request
->fl_start
- 1)
838 /* If the next lock in the list has entirely bigger
839 * addresses than the new one, insert the lock here.
841 if (fl
->fl_start
- 1 > request
->fl_end
)
844 /* If we come here, the new and old lock are of the
845 * same type and adjacent or overlapping. Make one
846 * lock yielding from the lower start address of both
847 * locks to the higher end address.
849 if (fl
->fl_start
> request
->fl_start
)
850 fl
->fl_start
= request
->fl_start
;
852 request
->fl_start
= fl
->fl_start
;
853 if (fl
->fl_end
< request
->fl_end
)
854 fl
->fl_end
= request
->fl_end
;
856 request
->fl_end
= fl
->fl_end
;
858 locks_delete_lock(before
);
865 /* Processing for different lock types is a bit
868 if (fl
->fl_end
< request
->fl_start
)
870 if (fl
->fl_start
> request
->fl_end
)
872 if (request
->fl_type
== F_UNLCK
)
874 if (fl
->fl_start
< request
->fl_start
)
876 /* If the next lock in the list has a higher end
877 * address than the new one, insert the new one here.
879 if (fl
->fl_end
> request
->fl_end
) {
883 if (fl
->fl_start
>= request
->fl_start
) {
884 /* The new lock completely replaces an old
885 * one (This may happen several times).
888 locks_delete_lock(before
);
891 /* Replace the old lock with the new one.
892 * Wake up anybody waiting for the old one,
893 * as the change in lock type might satisfy
896 locks_wake_up_blocks(fl
);
897 fl
->fl_start
= request
->fl_start
;
898 fl
->fl_end
= request
->fl_end
;
899 fl
->fl_type
= request
->fl_type
;
900 fl
->fl_u
= request
->fl_u
;
905 /* Go on to next lock.
908 before
= &fl
->fl_next
;
913 if (request
->fl_type
== F_UNLCK
)
915 locks_copy_lock(new_fl
, request
);
916 locks_insert_lock(before
, new_fl
);
921 /* The new lock breaks the old one in two pieces,
922 * so we have to use the second new lock.
926 locks_copy_lock(left
, right
);
927 locks_insert_lock(before
, left
);
929 right
->fl_start
= request
->fl_end
+ 1;
930 locks_wake_up_blocks(right
);
933 left
->fl_end
= request
->fl_start
- 1;
934 locks_wake_up_blocks(left
);
939 * Free any unused locks.
942 locks_free_lock(new_fl
);
944 locks_free_lock(new_fl2
);
949 * posix_lock_file - Apply a POSIX-style lock to a file
950 * @filp: The file to apply the lock to
951 * @fl: The lock to be applied
953 * Add a POSIX style lock to a file.
954 * We merge adjacent & overlapping locks whenever possible.
955 * POSIX locks are sorted by owner task, then by starting address
957 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
)
959 return __posix_lock_file(filp
->f_dentry
->d_inode
, fl
);
963 * posix_lock_file_wait - Apply a POSIX-style lock to a file
964 * @filp: The file to apply the lock to
965 * @fl: The lock to be applied
967 * Add a POSIX style lock to a file.
968 * We merge adjacent & overlapping locks whenever possible.
969 * POSIX locks are sorted by owner task, then by starting address
971 int posix_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
976 error
= __posix_lock_file(filp
->f_dentry
->d_inode
, fl
);
977 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
979 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
983 locks_delete_block(fl
);
988 EXPORT_SYMBOL(posix_lock_file_wait
);
991 * locks_mandatory_locked - Check for an active lock
992 * @inode: the file to check
994 * Searches the inode's list of locks to find any POSIX locks which conflict.
995 * This function is called from locks_verify_locked() only.
997 int locks_mandatory_locked(struct inode
*inode
)
999 fl_owner_t owner
= current
->files
;
1000 struct file_lock
*fl
;
1003 * Search the lock list for this inode for any POSIX locks.
1006 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
1009 if (fl
->fl_owner
!= owner
)
1013 return fl
? -EAGAIN
: 0;
1017 * locks_mandatory_area - Check for a conflicting lock
1018 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1020 * @inode: the file to check
1021 * @filp: how the file was opened (if it was)
1022 * @offset: start of area to check
1023 * @count: length of area to check
1025 * Searches the inode's list of locks to find any POSIX locks which conflict.
1026 * This function is called from rw_verify_area() and
1027 * locks_verify_truncate().
1029 int locks_mandatory_area(int read_write
, struct inode
*inode
,
1030 struct file
*filp
, loff_t offset
,
1033 struct file_lock fl
;
1036 locks_init_lock(&fl
);
1037 fl
.fl_owner
= current
->files
;
1038 fl
.fl_pid
= current
->tgid
;
1040 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1041 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1042 fl
.fl_flags
|= FL_SLEEP
;
1043 fl
.fl_type
= (read_write
== FLOCK_VERIFY_WRITE
) ? F_WRLCK
: F_RDLCK
;
1044 fl
.fl_start
= offset
;
1045 fl
.fl_end
= offset
+ count
- 1;
1048 error
= __posix_lock_file(inode
, &fl
);
1049 if (error
!= -EAGAIN
)
1051 if (!(fl
.fl_flags
& FL_SLEEP
))
1053 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_next
);
1056 * If we've been sleeping someone might have
1057 * changed the permissions behind our back.
1059 if ((inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
)
1063 locks_delete_block(&fl
);
1070 EXPORT_SYMBOL(locks_mandatory_area
);
1072 /* We already had a lease on this file; just change its type */
1073 int lease_modify(struct file_lock
**before
, int arg
)
1075 struct file_lock
*fl
= *before
;
1076 int error
= assign_type(fl
, arg
);
1080 locks_wake_up_blocks(fl
);
1082 locks_delete_lock(before
);
1086 EXPORT_SYMBOL(lease_modify
);
1088 static void time_out_leases(struct inode
*inode
)
1090 struct file_lock
**before
;
1091 struct file_lock
*fl
;
1093 before
= &inode
->i_flock
;
1094 while ((fl
= *before
) && IS_LEASE(fl
) && (fl
->fl_type
& F_INPROGRESS
)) {
1095 if ((fl
->fl_break_time
== 0)
1096 || time_before(jiffies
, fl
->fl_break_time
)) {
1097 before
= &fl
->fl_next
;
1100 printk(KERN_INFO
"lease broken - owner pid = %d\n", fl
->fl_pid
);
1101 lease_modify(before
, fl
->fl_type
& ~F_INPROGRESS
);
1102 if (fl
== *before
) /* lease_modify may have freed fl */
1103 before
= &fl
->fl_next
;
1108 * __break_lease - revoke all outstanding leases on file
1109 * @inode: the inode of the file to return
1110 * @mode: the open mode (read or write)
1112 * break_lease (inlined for speed) has checked there already
1113 * is a lease on this file. Leases are broken on a call to open()
1114 * or truncate(). This function can sleep unless you
1115 * specified %O_NONBLOCK to your open().
1117 int __break_lease(struct inode
*inode
, unsigned int mode
)
1119 int error
= 0, future
;
1120 struct file_lock
*new_fl
, *flock
;
1121 struct file_lock
*fl
;
1123 unsigned long break_time
;
1124 int i_have_this_lease
= 0;
1126 alloc_err
= lease_alloc(NULL
, mode
& FMODE_WRITE
? F_WRLCK
: F_RDLCK
,
1131 time_out_leases(inode
);
1133 flock
= inode
->i_flock
;
1134 if ((flock
== NULL
) || !IS_LEASE(flock
))
1137 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
)
1138 if (fl
->fl_owner
== current
->files
)
1139 i_have_this_lease
= 1;
1141 if (mode
& FMODE_WRITE
) {
1142 /* If we want write access, we have to revoke any lease. */
1143 future
= F_UNLCK
| F_INPROGRESS
;
1144 } else if (flock
->fl_type
& F_INPROGRESS
) {
1145 /* If the lease is already being broken, we just leave it */
1146 future
= flock
->fl_type
;
1147 } else if (flock
->fl_type
& F_WRLCK
) {
1148 /* Downgrade the exclusive lease to a read-only lease. */
1149 future
= F_RDLCK
| F_INPROGRESS
;
1151 /* the existing lease was read-only, so we can read too. */
1155 if (alloc_err
&& !i_have_this_lease
&& ((mode
& O_NONBLOCK
) == 0)) {
1161 if (lease_break_time
> 0) {
1162 break_time
= jiffies
+ lease_break_time
* HZ
;
1163 if (break_time
== 0)
1164 break_time
++; /* so that 0 means no break time */
1167 for (fl
= flock
; fl
&& IS_LEASE(fl
); fl
= fl
->fl_next
) {
1168 if (fl
->fl_type
!= future
) {
1169 fl
->fl_type
= future
;
1170 fl
->fl_break_time
= break_time
;
1171 /* lease must have lmops break callback */
1172 fl
->fl_lmops
->fl_break(fl
);
1176 if (i_have_this_lease
|| (mode
& O_NONBLOCK
)) {
1177 error
= -EWOULDBLOCK
;
1182 break_time
= flock
->fl_break_time
;
1183 if (break_time
!= 0) {
1184 break_time
-= jiffies
;
1185 if (break_time
== 0)
1188 error
= locks_block_on_timeout(flock
, new_fl
, break_time
);
1191 time_out_leases(inode
);
1192 /* Wait for the next lease that has not been broken yet */
1193 for (flock
= inode
->i_flock
; flock
&& IS_LEASE(flock
);
1194 flock
= flock
->fl_next
) {
1195 if (flock
->fl_type
& F_INPROGRESS
)
1204 locks_free_lock(new_fl
);
1208 EXPORT_SYMBOL(__break_lease
);
1213 * @time: pointer to a timespec which will contain the last modified time
1215 * This is to force NFS clients to flush their caches for files with
1216 * exclusive leases. The justification is that if someone has an
1217 * exclusive lease, then they could be modifiying it.
1219 void lease_get_mtime(struct inode
*inode
, struct timespec
*time
)
1221 struct file_lock
*flock
= inode
->i_flock
;
1222 if (flock
&& IS_LEASE(flock
) && (flock
->fl_type
& F_WRLCK
))
1223 *time
= current_fs_time(inode
->i_sb
);
1225 *time
= inode
->i_mtime
;
1228 EXPORT_SYMBOL(lease_get_mtime
);
1231 * fcntl_getlease - Enquire what lease is currently active
1234 * The value returned by this function will be one of
1235 * (if no lease break is pending):
1237 * %F_RDLCK to indicate a shared lease is held.
1239 * %F_WRLCK to indicate an exclusive lease is held.
1241 * %F_UNLCK to indicate no lease is held.
1243 * (if a lease break is pending):
1245 * %F_RDLCK to indicate an exclusive lease needs to be
1246 * changed to a shared lease (or removed).
1248 * %F_UNLCK to indicate the lease needs to be removed.
1250 * XXX: sfr & willy disagree over whether F_INPROGRESS
1251 * should be returned to userspace.
1253 int fcntl_getlease(struct file
*filp
)
1255 struct file_lock
*fl
;
1259 time_out_leases(filp
->f_dentry
->d_inode
);
1260 for (fl
= filp
->f_dentry
->d_inode
->i_flock
; fl
&& IS_LEASE(fl
);
1262 if (fl
->fl_file
== filp
) {
1263 type
= fl
->fl_type
& ~F_INPROGRESS
;
1272 * __setlease - sets a lease on an open file
1273 * @filp: file pointer
1274 * @arg: type of lease to obtain
1275 * @flp: input - file_lock to use, output - file_lock inserted
1277 * The (input) flp->fl_lmops->fl_break function is required
1280 * Called with kernel lock held.
1282 static int __setlease(struct file
*filp
, long arg
, struct file_lock
**flp
)
1284 struct file_lock
*fl
, **before
, **my_before
= NULL
, *lease
;
1285 struct dentry
*dentry
= filp
->f_dentry
;
1286 struct inode
*inode
= dentry
->d_inode
;
1287 int error
, rdlease_count
= 0, wrlease_count
= 0;
1289 time_out_leases(inode
);
1292 if (!flp
|| !(*flp
) || !(*flp
)->fl_lmops
|| !(*flp
)->fl_lmops
->fl_break
)
1298 if ((arg
== F_RDLCK
) && (atomic_read(&inode
->i_writecount
) > 0))
1300 if ((arg
== F_WRLCK
)
1301 && ((atomic_read(&dentry
->d_count
) > 1)
1302 || (atomic_read(&inode
->i_count
) > 1)))
1306 * At this point, we know that if there is an exclusive
1307 * lease on this file, then we hold it on this filp
1308 * (otherwise our open of this file would have blocked).
1309 * And if we are trying to acquire an exclusive lease,
1310 * then the file is not open by anyone (including us)
1311 * except for this filp.
1313 for (before
= &inode
->i_flock
;
1314 ((fl
= *before
) != NULL
) && IS_LEASE(fl
);
1315 before
= &fl
->fl_next
) {
1316 if (lease
->fl_lmops
->fl_mylease(fl
, lease
))
1318 else if (fl
->fl_type
== (F_INPROGRESS
| F_UNLCK
))
1320 * Someone is in the process of opening this
1321 * file for writing so we may not take an
1322 * exclusive lease on it.
1329 if ((arg
== F_RDLCK
&& (wrlease_count
> 0)) ||
1330 (arg
== F_WRLCK
&& ((rdlease_count
+ wrlease_count
) > 0)))
1333 if (my_before
!= NULL
) {
1334 error
= lease
->fl_lmops
->fl_change(my_before
, arg
);
1346 error
= lease_alloc(filp
, arg
, &fl
);
1350 locks_copy_lock(fl
, lease
);
1352 locks_insert_lock(before
, fl
);
1360 * setlease - sets a lease on an open file
1361 * @filp: file pointer
1362 * @arg: type of lease to obtain
1363 * @lease: file_lock to use
1365 * Call this to establish a lease on the file.
1366 * The fl_lmops fl_break function is required by break_lease
1369 int setlease(struct file
*filp
, long arg
, struct file_lock
**lease
)
1371 struct dentry
*dentry
= filp
->f_dentry
;
1372 struct inode
*inode
= dentry
->d_inode
;
1375 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1377 if (!S_ISREG(inode
->i_mode
))
1379 error
= security_file_lock(filp
, arg
);
1384 error
= __setlease(filp
, arg
, lease
);
1390 EXPORT_SYMBOL(setlease
);
1393 * fcntl_setlease - sets a lease on an open file
1394 * @fd: open file descriptor
1395 * @filp: file pointer
1396 * @arg: type of lease to obtain
1398 * Call this fcntl to establish a lease on the file.
1399 * Note that you also need to call %F_SETSIG to
1400 * receive a signal when the lease is broken.
1402 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
1404 struct file_lock fl
, *flp
= &fl
;
1405 struct dentry
*dentry
= filp
->f_dentry
;
1406 struct inode
*inode
= dentry
->d_inode
;
1409 if ((current
->fsuid
!= inode
->i_uid
) && !capable(CAP_LEASE
))
1411 if (!S_ISREG(inode
->i_mode
))
1413 error
= security_file_lock(filp
, arg
);
1417 locks_init_lock(&fl
);
1418 error
= lease_init(filp
, arg
, &fl
);
1424 error
= __setlease(filp
, arg
, &flp
);
1428 error
= fasync_helper(fd
, filp
, 1, &flp
->fl_fasync
);
1430 /* remove lease just inserted by __setlease */
1431 flp
->fl_type
= F_UNLCK
| F_INPROGRESS
;
1432 flp
->fl_break_time
= jiffies
- 10;
1433 time_out_leases(inode
);
1437 error
= f_setown(filp
, current
->pid
, 0);
1444 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1445 * @filp: The file to apply the lock to
1446 * @fl: The lock to be applied
1448 * Add a FLOCK style lock to a file.
1450 int flock_lock_file_wait(struct file
*filp
, struct file_lock
*fl
)
1455 error
= flock_lock_file(filp
, fl
);
1456 if ((error
!= -EAGAIN
) || !(fl
->fl_flags
& FL_SLEEP
))
1458 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_next
);
1462 locks_delete_block(fl
);
1468 EXPORT_SYMBOL(flock_lock_file_wait
);
1471 * sys_flock: - flock() system call.
1472 * @fd: the file descriptor to lock.
1473 * @cmd: the type of lock to apply.
1475 * Apply a %FL_FLOCK style lock to an open file descriptor.
1476 * The @cmd can be one of
1478 * %LOCK_SH -- a shared lock.
1480 * %LOCK_EX -- an exclusive lock.
1482 * %LOCK_UN -- remove an existing lock.
1484 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1486 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1487 * processes read and write access respectively.
1489 asmlinkage
long sys_flock(unsigned int fd
, unsigned int cmd
)
1492 struct file_lock
*lock
;
1493 int can_sleep
, unlock
;
1501 can_sleep
= !(cmd
& LOCK_NB
);
1503 unlock
= (cmd
== LOCK_UN
);
1505 if (!unlock
&& !(cmd
& LOCK_MAND
) && !(filp
->f_mode
& 3))
1508 error
= flock_make_lock(filp
, &lock
, cmd
);
1512 lock
->fl_flags
|= FL_SLEEP
;
1514 error
= security_file_lock(filp
, cmd
);
1518 if (filp
->f_op
&& filp
->f_op
->flock
)
1519 error
= filp
->f_op
->flock(filp
,
1520 (can_sleep
) ? F_SETLKW
: F_SETLK
,
1523 error
= flock_lock_file_wait(filp
, lock
);
1526 if (list_empty(&lock
->fl_link
)) {
1527 locks_free_lock(lock
);
1536 /* Report the first existing lock that would conflict with l.
1537 * This implements the F_GETLK command of fcntl().
1539 int fcntl_getlk(struct file
*filp
, struct flock __user
*l
)
1541 struct file_lock
*fl
, file_lock
;
1546 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1549 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1552 error
= flock_to_posix_lock(filp
, &file_lock
, &flock
);
1556 if (filp
->f_op
&& filp
->f_op
->lock
) {
1557 error
= filp
->f_op
->lock(filp
, F_GETLK
, &file_lock
);
1558 if (file_lock
.fl_ops
&& file_lock
.fl_ops
->fl_release_private
)
1559 file_lock
.fl_ops
->fl_release_private(&file_lock
);
1563 fl
= (file_lock
.fl_type
== F_UNLCK
? NULL
: &file_lock
);
1565 fl
= posix_test_lock(filp
, &file_lock
);
1568 flock
.l_type
= F_UNLCK
;
1570 flock
.l_pid
= fl
->fl_pid
;
1571 #if BITS_PER_LONG == 32
1573 * Make sure we can represent the posix lock via
1574 * legacy 32bit flock.
1577 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
1579 if ((fl
->fl_end
!= OFFSET_MAX
)
1580 && (fl
->fl_end
> OFFT_OFFSET_MAX
))
1583 flock
.l_start
= fl
->fl_start
;
1584 flock
.l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1585 fl
->fl_end
- fl
->fl_start
+ 1;
1587 flock
.l_type
= fl
->fl_type
;
1590 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1596 /* Apply the lock described by l to an open file descriptor.
1597 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1599 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1600 struct flock __user
*l
)
1602 struct file_lock
*file_lock
= locks_alloc_lock();
1604 struct inode
*inode
;
1607 if (file_lock
== NULL
)
1611 * This might block, so we do it before checking the inode.
1614 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1617 inode
= filp
->f_dentry
->d_inode
;
1619 /* Don't allow mandatory locks on files that may be memory mapped
1622 if (IS_MANDLOCK(inode
) &&
1623 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1624 mapping_writably_mapped(filp
->f_mapping
)) {
1630 error
= flock_to_posix_lock(filp
, file_lock
, &flock
);
1633 if (cmd
== F_SETLKW
) {
1634 file_lock
->fl_flags
|= FL_SLEEP
;
1638 switch (flock
.l_type
) {
1640 if (!(filp
->f_mode
& FMODE_READ
))
1644 if (!(filp
->f_mode
& FMODE_WRITE
))
1654 error
= security_file_lock(filp
, file_lock
->fl_type
);
1658 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
)
1659 error
= filp
->f_op
->lock(filp
, cmd
, file_lock
);
1662 error
= __posix_lock_file(inode
, file_lock
);
1663 if ((error
!= -EAGAIN
) || (cmd
== F_SETLK
))
1665 error
= wait_event_interruptible(file_lock
->fl_wait
,
1666 !file_lock
->fl_next
);
1670 locks_delete_block(file_lock
);
1676 * Attempt to detect a close/fcntl race and recover by
1677 * releasing the lock that was just acquired.
1679 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1680 flock
.l_type
= F_UNLCK
;
1685 locks_free_lock(file_lock
);
1689 #if BITS_PER_LONG == 32
1690 /* Report the first existing lock that would conflict with l.
1691 * This implements the F_GETLK command of fcntl().
1693 int fcntl_getlk64(struct file
*filp
, struct flock64 __user
*l
)
1695 struct file_lock
*fl
, file_lock
;
1696 struct flock64 flock
;
1700 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1703 if ((flock
.l_type
!= F_RDLCK
) && (flock
.l_type
!= F_WRLCK
))
1706 error
= flock64_to_posix_lock(filp
, &file_lock
, &flock
);
1710 if (filp
->f_op
&& filp
->f_op
->lock
) {
1711 error
= filp
->f_op
->lock(filp
, F_GETLK
, &file_lock
);
1712 if (file_lock
.fl_ops
&& file_lock
.fl_ops
->fl_release_private
)
1713 file_lock
.fl_ops
->fl_release_private(&file_lock
);
1717 fl
= (file_lock
.fl_type
== F_UNLCK
? NULL
: &file_lock
);
1719 fl
= posix_test_lock(filp
, &file_lock
);
1722 flock
.l_type
= F_UNLCK
;
1724 flock
.l_pid
= fl
->fl_pid
;
1725 flock
.l_start
= fl
->fl_start
;
1726 flock
.l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
1727 fl
->fl_end
- fl
->fl_start
+ 1;
1729 flock
.l_type
= fl
->fl_type
;
1732 if (!copy_to_user(l
, &flock
, sizeof(flock
)))
1739 /* Apply the lock described by l to an open file descriptor.
1740 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1742 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
1743 struct flock64 __user
*l
)
1745 struct file_lock
*file_lock
= locks_alloc_lock();
1746 struct flock64 flock
;
1747 struct inode
*inode
;
1750 if (file_lock
== NULL
)
1754 * This might block, so we do it before checking the inode.
1757 if (copy_from_user(&flock
, l
, sizeof(flock
)))
1760 inode
= filp
->f_dentry
->d_inode
;
1762 /* Don't allow mandatory locks on files that may be memory mapped
1765 if (IS_MANDLOCK(inode
) &&
1766 (inode
->i_mode
& (S_ISGID
| S_IXGRP
)) == S_ISGID
&&
1767 mapping_writably_mapped(filp
->f_mapping
)) {
1773 error
= flock64_to_posix_lock(filp
, file_lock
, &flock
);
1776 if (cmd
== F_SETLKW64
) {
1777 file_lock
->fl_flags
|= FL_SLEEP
;
1781 switch (flock
.l_type
) {
1783 if (!(filp
->f_mode
& FMODE_READ
))
1787 if (!(filp
->f_mode
& FMODE_WRITE
))
1797 error
= security_file_lock(filp
, file_lock
->fl_type
);
1801 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
)
1802 error
= filp
->f_op
->lock(filp
, cmd
, file_lock
);
1805 error
= __posix_lock_file(inode
, file_lock
);
1806 if ((error
!= -EAGAIN
) || (cmd
== F_SETLK64
))
1808 error
= wait_event_interruptible(file_lock
->fl_wait
,
1809 !file_lock
->fl_next
);
1813 locks_delete_block(file_lock
);
1819 * Attempt to detect a close/fcntl race and recover by
1820 * releasing the lock that was just acquired.
1822 if (!error
&& fcheck(fd
) != filp
&& flock
.l_type
!= F_UNLCK
) {
1823 flock
.l_type
= F_UNLCK
;
1828 locks_free_lock(file_lock
);
1831 #endif /* BITS_PER_LONG == 32 */
1834 * This function is called when the file is being removed
1835 * from the task's fd array. POSIX locks belonging to this task
1836 * are deleted at this time.
1838 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
1840 struct file_lock lock
, **before
;
1843 * If there are no locks held on this file, we don't need to call
1844 * posix_lock_file(). Another process could be setting a lock on this
1845 * file at the same time, but we wouldn't remove that lock anyway.
1847 before
= &filp
->f_dentry
->d_inode
->i_flock
;
1848 if (*before
== NULL
)
1851 lock
.fl_type
= F_UNLCK
;
1852 lock
.fl_flags
= FL_POSIX
;
1854 lock
.fl_end
= OFFSET_MAX
;
1855 lock
.fl_owner
= owner
;
1856 lock
.fl_pid
= current
->tgid
;
1857 lock
.fl_file
= filp
;
1859 lock
.fl_lmops
= NULL
;
1861 if (filp
->f_op
&& filp
->f_op
->lock
!= NULL
) {
1862 filp
->f_op
->lock(filp
, F_SETLK
, &lock
);
1866 /* Can't use posix_lock_file here; we need to remove it no matter
1867 * which pid we have.
1870 while (*before
!= NULL
) {
1871 struct file_lock
*fl
= *before
;
1872 if (IS_POSIX(fl
) && posix_same_owner(fl
, &lock
)) {
1873 locks_delete_lock(before
);
1876 before
= &fl
->fl_next
;
1880 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
1881 lock
.fl_ops
->fl_release_private(&lock
);
1884 EXPORT_SYMBOL(locks_remove_posix
);
1887 * This function is called on the last close of an open file.
1889 void locks_remove_flock(struct file
*filp
)
1891 struct inode
* inode
= filp
->f_dentry
->d_inode
;
1892 struct file_lock
*fl
;
1893 struct file_lock
**before
;
1895 if (!inode
->i_flock
)
1898 if (filp
->f_op
&& filp
->f_op
->flock
) {
1899 struct file_lock fl
= {
1900 .fl_pid
= current
->tgid
,
1902 .fl_flags
= FL_FLOCK
,
1904 .fl_end
= OFFSET_MAX
,
1906 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
1907 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
1908 fl
.fl_ops
->fl_release_private(&fl
);
1912 before
= &inode
->i_flock
;
1914 while ((fl
= *before
) != NULL
) {
1915 if (fl
->fl_file
== filp
) {
1917 locks_delete_lock(before
);
1921 lease_modify(before
, F_UNLCK
);
1927 before
= &fl
->fl_next
;
1933 * posix_block_lock - blocks waiting for a file lock
1934 * @blocker: the lock which is blocking
1935 * @waiter: the lock which conflicts and has to wait
1937 * lockd needs to block waiting for locks.
1940 posix_block_lock(struct file_lock
*blocker
, struct file_lock
*waiter
)
1942 locks_insert_block(blocker
, waiter
);
1945 EXPORT_SYMBOL(posix_block_lock
);
1948 * posix_unblock_lock - stop waiting for a file lock
1949 * @filp: how the file was opened
1950 * @waiter: the lock which was waiting
1952 * lockd needs to block waiting for locks.
1955 posix_unblock_lock(struct file
*filp
, struct file_lock
*waiter
)
1958 * A remote machine may cancel the lock request after it's been
1959 * granted locally. If that happens, we need to delete the lock.
1962 if (waiter
->fl_next
) {
1963 __locks_delete_block(waiter
);
1967 waiter
->fl_type
= F_UNLCK
;
1968 posix_lock_file(filp
, waiter
);
1972 EXPORT_SYMBOL(posix_unblock_lock
);
1974 static void lock_get_status(char* out
, struct file_lock
*fl
, int id
, char *pfx
)
1976 struct inode
*inode
= NULL
;
1978 if (fl
->fl_file
!= NULL
)
1979 inode
= fl
->fl_file
->f_dentry
->d_inode
;
1981 out
+= sprintf(out
, "%d:%s ", id
, pfx
);
1983 out
+= sprintf(out
, "%6s %s ",
1984 (fl
->fl_flags
& FL_ACCESS
) ? "ACCESS" : "POSIX ",
1985 (inode
== NULL
) ? "*NOINODE*" :
1986 (IS_MANDLOCK(inode
) &&
1987 (inode
->i_mode
& (S_IXGRP
| S_ISGID
)) == S_ISGID
) ?
1988 "MANDATORY" : "ADVISORY ");
1989 } else if (IS_FLOCK(fl
)) {
1990 if (fl
->fl_type
& LOCK_MAND
) {
1991 out
+= sprintf(out
, "FLOCK MSNFS ");
1993 out
+= sprintf(out
, "FLOCK ADVISORY ");
1995 } else if (IS_LEASE(fl
)) {
1996 out
+= sprintf(out
, "LEASE ");
1997 if (fl
->fl_type
& F_INPROGRESS
)
1998 out
+= sprintf(out
, "BREAKING ");
1999 else if (fl
->fl_file
)
2000 out
+= sprintf(out
, "ACTIVE ");
2002 out
+= sprintf(out
, "BREAKER ");
2004 out
+= sprintf(out
, "UNKNOWN UNKNOWN ");
2006 if (fl
->fl_type
& LOCK_MAND
) {
2007 out
+= sprintf(out
, "%s ",
2008 (fl
->fl_type
& LOCK_READ
)
2009 ? (fl
->fl_type
& LOCK_WRITE
) ? "RW " : "READ "
2010 : (fl
->fl_type
& LOCK_WRITE
) ? "WRITE" : "NONE ");
2012 out
+= sprintf(out
, "%s ",
2013 (fl
->fl_type
& F_INPROGRESS
)
2014 ? (fl
->fl_type
& F_UNLCK
) ? "UNLCK" : "READ "
2015 : (fl
->fl_type
& F_WRLCK
) ? "WRITE" : "READ ");
2018 #ifdef WE_CAN_BREAK_LSLK_NOW
2019 out
+= sprintf(out
, "%d %s:%ld ", fl
->fl_pid
,
2020 inode
->i_sb
->s_id
, inode
->i_ino
);
2022 /* userspace relies on this representation of dev_t ;-( */
2023 out
+= sprintf(out
, "%d %02x:%02x:%ld ", fl
->fl_pid
,
2024 MAJOR(inode
->i_sb
->s_dev
),
2025 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2028 out
+= sprintf(out
, "%d <none>:0 ", fl
->fl_pid
);
2031 if (fl
->fl_end
== OFFSET_MAX
)
2032 out
+= sprintf(out
, "%Ld EOF\n", fl
->fl_start
);
2034 out
+= sprintf(out
, "%Ld %Ld\n", fl
->fl_start
,
2037 out
+= sprintf(out
, "0 EOF\n");
2041 static void move_lock_status(char **p
, off_t
* pos
, off_t offset
)
2045 if(*pos
>= offset
) {
2046 /* the complete line is valid */
2051 if(*pos
+len
> offset
) {
2052 /* use the second part of the line */
2053 int i
= offset
-*pos
;
2054 memmove(*p
,*p
+i
,len
-i
);
2059 /* discard the complete line */
2064 * get_locks_status - reports lock usage in /proc/locks
2065 * @buffer: address in userspace to write into
2067 * @offset: how far we are through the buffer
2068 * @length: how much to read
2071 int get_locks_status(char *buffer
, char **start
, off_t offset
, int length
)
2073 struct list_head
*tmp
;
2079 list_for_each(tmp
, &file_lock_list
) {
2080 struct list_head
*btmp
;
2081 struct file_lock
*fl
= list_entry(tmp
, struct file_lock
, fl_link
);
2082 lock_get_status(q
, fl
, ++i
, "");
2083 move_lock_status(&q
, &pos
, offset
);
2085 if(pos
>= offset
+length
)
2088 list_for_each(btmp
, &fl
->fl_block
) {
2089 struct file_lock
*bfl
= list_entry(btmp
,
2090 struct file_lock
, fl_block
);
2091 lock_get_status(q
, bfl
, i
, " ->");
2092 move_lock_status(&q
, &pos
, offset
);
2094 if(pos
>= offset
+length
)
2101 if(q
-buffer
< length
)
2107 * lock_may_read - checks that the region is free of locks
2108 * @inode: the inode that is being read
2109 * @start: the first byte to read
2110 * @len: the number of bytes to read
2112 * Emulates Windows locking requirements. Whole-file
2113 * mandatory locks (share modes) can prohibit a read and
2114 * byte-range POSIX locks can prohibit a read if they overlap.
2116 * N.B. this function is only ever called
2117 * from knfsd and ownership of locks is never checked.
2119 int lock_may_read(struct inode
*inode
, loff_t start
, unsigned long len
)
2121 struct file_lock
*fl
;
2124 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2126 if (fl
->fl_type
== F_RDLCK
)
2128 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2130 } else if (IS_FLOCK(fl
)) {
2131 if (!(fl
->fl_type
& LOCK_MAND
))
2133 if (fl
->fl_type
& LOCK_READ
)
2144 EXPORT_SYMBOL(lock_may_read
);
2147 * lock_may_write - checks that the region is free of locks
2148 * @inode: the inode that is being written
2149 * @start: the first byte to write
2150 * @len: the number of bytes to write
2152 * Emulates Windows locking requirements. Whole-file
2153 * mandatory locks (share modes) can prohibit a write and
2154 * byte-range POSIX locks can prohibit a write if they overlap.
2156 * N.B. this function is only ever called
2157 * from knfsd and ownership of locks is never checked.
2159 int lock_may_write(struct inode
*inode
, loff_t start
, unsigned long len
)
2161 struct file_lock
*fl
;
2164 for (fl
= inode
->i_flock
; fl
!= NULL
; fl
= fl
->fl_next
) {
2166 if ((fl
->fl_end
< start
) || (fl
->fl_start
> (start
+ len
)))
2168 } else if (IS_FLOCK(fl
)) {
2169 if (!(fl
->fl_type
& LOCK_MAND
))
2171 if (fl
->fl_type
& LOCK_WRITE
)
2182 EXPORT_SYMBOL(lock_may_write
);
2184 static inline void __steal_locks(struct file
*file
, fl_owner_t from
)
2186 struct inode
*inode
= file
->f_dentry
->d_inode
;
2187 struct file_lock
*fl
= inode
->i_flock
;
2190 if (fl
->fl_file
== file
&& fl
->fl_owner
== from
)
2191 fl
->fl_owner
= current
->files
;
2196 /* When getting ready for executing a binary, we make sure that current
2197 * has a files_struct on its own. Before dropping the old files_struct,
2198 * we take over ownership of all locks for all file descriptors we own.
2199 * Note that we may accidentally steal a lock for a file that a sibling
2200 * has created since the unshare_files() call.
2202 void steal_locks(fl_owner_t from
)
2204 struct files_struct
*files
= current
->files
;
2206 struct fdtable
*fdt
;
2214 fdt
= files_fdtable(files
);
2218 if (i
>= fdt
->max_fdset
|| i
>= fdt
->max_fds
)
2220 set
= fdt
->open_fds
->fds_bits
[j
++];
2223 struct file
*file
= fdt
->fd
[i
];
2225 __steal_locks(file
, from
);
2234 EXPORT_SYMBOL(steal_locks
);
2236 static int __init
filelock_init(void)
2238 filelock_cache
= kmem_cache_create("file_lock_cache",
2239 sizeof(struct file_lock
), 0, SLAB_PANIC
,
2244 core_initcall(filelock_init
);