4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 * Made mandatory locking a mount option. Default is not to allow mandatory
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fs.h>
120 #include <linux/init.h>
121 #include <linux/module.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/smp_lock.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
129 #include <asm/semaphore.h>
130 #include <asm/uaccess.h>
132 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136 int leases_enable = 1;
137 int lease_break_time = 45;
139 #define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142 static LIST_HEAD(file_lock_list);
143 static LIST_HEAD(blocked_list);
145 static struct kmem_cache *filelock_cache __read_mostly;
147 /* Allocate an empty lock structure. */
148 static struct file_lock *locks_alloc_lock(void)
150 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
153 static void locks_release_private(struct file_lock *fl)
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
168 /* Free a lock which is not in use. */
169 static void locks_free_lock(struct file_lock *fl)
171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
175 locks_release_private(fl);
176 kmem_cache_free(filelock_cache, fl);
179 void locks_init_lock(struct file_lock *fl)
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
185 fl->fl_fasync = NULL;
191 fl->fl_start = fl->fl_end = 0;
196 EXPORT_SYMBOL(locks_init_lock);
199 * Initialises the fields of the file lock which are invariant for
202 static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
204 struct file_lock *lock = (struct file_lock *) foo;
206 locks_init_lock(lock);
209 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
212 if (fl->fl_ops->fl_copy_lock)
213 fl->fl_ops->fl_copy_lock(new, fl);
214 new->fl_ops = fl->fl_ops;
217 if (fl->fl_lmops->fl_copy_lock)
218 fl->fl_lmops->fl_copy_lock(new, fl);
219 new->fl_lmops = fl->fl_lmops;
224 * Initialize a new lock from an existing file_lock structure.
226 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
228 new->fl_owner = fl->fl_owner;
229 new->fl_pid = fl->fl_pid;
231 new->fl_flags = fl->fl_flags;
232 new->fl_type = fl->fl_type;
233 new->fl_start = fl->fl_start;
234 new->fl_end = fl->fl_end;
236 new->fl_lmops = NULL;
239 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
241 locks_release_private(new);
243 __locks_copy_lock(new, fl);
244 new->fl_file = fl->fl_file;
245 new->fl_ops = fl->fl_ops;
246 new->fl_lmops = fl->fl_lmops;
248 locks_copy_private(new, fl);
251 EXPORT_SYMBOL(locks_copy_lock);
253 static inline int flock_translate_cmd(int cmd) {
255 return cmd & (LOCK_MAND | LOCK_RW);
267 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
268 static int flock_make_lock(struct file *filp, struct file_lock **lock,
271 struct file_lock *fl;
272 int type = flock_translate_cmd(cmd);
276 fl = locks_alloc_lock();
281 fl->fl_pid = current->tgid;
282 fl->fl_flags = FL_FLOCK;
284 fl->fl_end = OFFSET_MAX;
290 static int assign_type(struct file_lock *fl, int type)
304 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
307 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
312 switch (l->l_whence) {
320 start = i_size_read(filp->f_path.dentry->d_inode);
326 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
327 POSIX-2001 defines it. */
331 fl->fl_end = OFFSET_MAX;
333 end = start + l->l_len - 1;
335 } else if (l->l_len < 0) {
342 fl->fl_start = start; /* we record the absolute position */
343 if (fl->fl_end < fl->fl_start)
346 fl->fl_owner = current->files;
347 fl->fl_pid = current->tgid;
349 fl->fl_flags = FL_POSIX;
353 return assign_type(fl, l->l_type);
356 #if BITS_PER_LONG == 32
357 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
362 switch (l->l_whence) {
370 start = i_size_read(filp->f_path.dentry->d_inode);
379 fl->fl_end = OFFSET_MAX;
381 fl->fl_end = start + l->l_len - 1;
382 } else if (l->l_len < 0) {
383 fl->fl_end = start - 1;
388 fl->fl_start = start; /* we record the absolute position */
389 if (fl->fl_end < fl->fl_start)
392 fl->fl_owner = current->files;
393 fl->fl_pid = current->tgid;
395 fl->fl_flags = FL_POSIX;
403 fl->fl_type = l->l_type;
413 /* default lease lock manager operations */
414 static void lease_break_callback(struct file_lock *fl)
416 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
419 static void lease_release_private_callback(struct file_lock *fl)
424 f_delown(fl->fl_file);
425 fl->fl_file->f_owner.signum = 0;
428 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
430 return fl->fl_file == try->fl_file;
433 static struct lock_manager_operations lease_manager_ops = {
434 .fl_break = lease_break_callback,
435 .fl_release_private = lease_release_private_callback,
436 .fl_mylease = lease_mylease_callback,
437 .fl_change = lease_modify,
441 * Initialize a lease, use the default lock manager operations
443 static int lease_init(struct file *filp, int type, struct file_lock *fl)
445 if (assign_type(fl, type) != 0)
448 fl->fl_owner = current->files;
449 fl->fl_pid = current->tgid;
452 fl->fl_flags = FL_LEASE;
454 fl->fl_end = OFFSET_MAX;
456 fl->fl_lmops = &lease_manager_ops;
460 /* Allocate a file_lock initialised to this type of lease */
461 static struct file_lock *lease_alloc(struct file *filp, int type)
463 struct file_lock *fl = locks_alloc_lock();
467 return ERR_PTR(error);
469 error = lease_init(filp, type, fl);
472 return ERR_PTR(error);
477 /* Check if two locks overlap each other.
479 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
481 return ((fl1->fl_end >= fl2->fl_start) &&
482 (fl2->fl_end >= fl1->fl_start));
486 * Check whether two locks have the same owner.
488 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
490 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
491 return fl2->fl_lmops == fl1->fl_lmops &&
492 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
493 return fl1->fl_owner == fl2->fl_owner;
496 /* Remove waiter from blocker's block list.
497 * When blocker ends up pointing to itself then the list is empty.
499 static void __locks_delete_block(struct file_lock *waiter)
501 list_del_init(&waiter->fl_block);
502 list_del_init(&waiter->fl_link);
503 waiter->fl_next = NULL;
508 static void locks_delete_block(struct file_lock *waiter)
511 __locks_delete_block(waiter);
515 /* Insert waiter into blocker's block list.
516 * We use a circular list so that processes can be easily woken up in
517 * the order they blocked. The documentation doesn't require this but
518 * it seems like the reasonable thing to do.
520 static void locks_insert_block(struct file_lock *blocker,
521 struct file_lock *waiter)
523 BUG_ON(!list_empty(&waiter->fl_block));
524 list_add_tail(&waiter->fl_block, &blocker->fl_block);
525 waiter->fl_next = blocker;
526 if (IS_POSIX(blocker))
527 list_add(&waiter->fl_link, &blocked_list);
530 /* Wake up processes blocked waiting for blocker.
531 * If told to wait then schedule the processes until the block list
532 * is empty, otherwise empty the block list ourselves.
534 static void locks_wake_up_blocks(struct file_lock *blocker)
536 while (!list_empty(&blocker->fl_block)) {
537 struct file_lock *waiter = list_entry(blocker->fl_block.next,
538 struct file_lock, fl_block);
539 __locks_delete_block(waiter);
540 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
541 waiter->fl_lmops->fl_notify(waiter);
543 wake_up(&waiter->fl_wait);
547 /* Insert file lock fl into an inode's lock list at the position indicated
548 * by pos. At the same time add the lock to the global file lock list.
550 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
552 list_add(&fl->fl_link, &file_lock_list);
554 /* insert into file's list */
558 if (fl->fl_ops && fl->fl_ops->fl_insert)
559 fl->fl_ops->fl_insert(fl);
563 * Delete a lock and then free it.
564 * Wake up processes that are blocked waiting for this lock,
565 * notify the FS that the lock has been cleared and
566 * finally free the lock.
568 static void locks_delete_lock(struct file_lock **thisfl_p)
570 struct file_lock *fl = *thisfl_p;
572 *thisfl_p = fl->fl_next;
574 list_del_init(&fl->fl_link);
576 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
577 if (fl->fl_fasync != NULL) {
578 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
579 fl->fl_fasync = NULL;
582 if (fl->fl_ops && fl->fl_ops->fl_remove)
583 fl->fl_ops->fl_remove(fl);
585 locks_wake_up_blocks(fl);
589 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
590 * checks for shared/exclusive status of overlapping locks.
592 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
594 if (sys_fl->fl_type == F_WRLCK)
596 if (caller_fl->fl_type == F_WRLCK)
601 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
602 * checking before calling the locks_conflict().
604 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
606 /* POSIX locks owned by the same process do not conflict with
609 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
612 /* Check whether they overlap */
613 if (!locks_overlap(caller_fl, sys_fl))
616 return (locks_conflict(caller_fl, sys_fl));
619 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
620 * checking before calling the locks_conflict().
622 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
624 /* FLOCK locks referring to the same filp do not conflict with
627 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
629 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
632 return (locks_conflict(caller_fl, sys_fl));
635 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
638 DECLARE_WAITQUEUE(wait, current);
640 __set_current_state(TASK_INTERRUPTIBLE);
641 add_wait_queue(fl_wait, &wait);
645 result = schedule_timeout(timeout);
646 if (signal_pending(current))
647 result = -ERESTARTSYS;
648 remove_wait_queue(fl_wait, &wait);
649 __set_current_state(TASK_RUNNING);
653 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
656 locks_insert_block(blocker, waiter);
657 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
658 __locks_delete_block(waiter);
663 posix_test_lock(struct file *filp, struct file_lock *fl)
665 struct file_lock *cfl;
668 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
671 if (posix_locks_conflict(fl, cfl))
675 __locks_copy_lock(fl, cfl);
677 fl->fl_type = F_UNLCK;
682 EXPORT_SYMBOL(posix_test_lock);
684 /* This function tests for deadlock condition before putting a process to
685 * sleep. The detection scheme is no longer recursive. Recursive was neat,
686 * but dangerous - we risked stack corruption if the lock data was bad, or
687 * if the recursion was too deep for any other reason.
689 * We rely on the fact that a task can only be on one lock's wait queue
690 * at a time. When we find blocked_task on a wait queue we can re-search
691 * with blocked_task equal to that queue's owner, until either blocked_task
692 * isn't found, or blocked_task is found on a queue owned by my_task.
694 * Note: the above assumption may not be true when handling lock requests
695 * from a broken NFS client. But broken NFS clients have a lot more to
696 * worry about than proper deadlock detection anyway... --okir
698 static int posix_locks_deadlock(struct file_lock *caller_fl,
699 struct file_lock *block_fl)
701 struct list_head *tmp;
704 if (posix_same_owner(caller_fl, block_fl))
706 list_for_each(tmp, &blocked_list) {
707 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
708 if (posix_same_owner(fl, block_fl)) {
717 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
718 * at the head of the list, but that's secret knowledge known only to
719 * flock_lock_file and posix_lock_file.
721 * Note that if called with an FL_EXISTS argument, the caller may determine
722 * whether or not a lock was successfully freed by testing the return
725 static int flock_lock_file(struct file *filp, struct file_lock *request)
727 struct file_lock *new_fl = NULL;
728 struct file_lock **before;
729 struct inode * inode = filp->f_path.dentry->d_inode;
734 if (request->fl_flags & FL_ACCESS)
737 if (request->fl_type != F_UNLCK) {
739 new_fl = locks_alloc_lock();
745 for_each_lock(inode, before) {
746 struct file_lock *fl = *before;
751 if (filp != fl->fl_file)
753 if (request->fl_type == fl->fl_type)
756 locks_delete_lock(before);
760 if (request->fl_type == F_UNLCK) {
761 if ((request->fl_flags & FL_EXISTS) && !found)
767 * If a higher-priority process was blocked on the old file lock,
768 * give it the opportunity to lock the file.
774 for_each_lock(inode, before) {
775 struct file_lock *fl = *before;
780 if (!flock_locks_conflict(request, fl))
783 if (request->fl_flags & FL_SLEEP)
784 locks_insert_block(fl, request);
787 if (request->fl_flags & FL_ACCESS)
789 locks_copy_lock(new_fl, request);
790 locks_insert_lock(before, new_fl);
797 locks_free_lock(new_fl);
801 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
803 struct file_lock *fl;
804 struct file_lock *new_fl = NULL;
805 struct file_lock *new_fl2 = NULL;
806 struct file_lock *left = NULL;
807 struct file_lock *right = NULL;
808 struct file_lock **before;
809 int error, added = 0;
812 * We may need two file_lock structures for this operation,
813 * so we get them in advance to avoid races.
815 * In some cases we can be sure, that no new locks will be needed
817 if (!(request->fl_flags & FL_ACCESS) &&
818 (request->fl_type != F_UNLCK ||
819 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
820 new_fl = locks_alloc_lock();
821 new_fl2 = locks_alloc_lock();
825 if (request->fl_type != F_UNLCK) {
826 for_each_lock(inode, before) {
830 if (!posix_locks_conflict(request, fl))
833 locks_copy_lock(conflock, fl);
835 if (!(request->fl_flags & FL_SLEEP))
838 if (posix_locks_deadlock(request, fl))
841 locks_insert_block(fl, request);
846 /* If we're just looking for a conflict, we're done. */
848 if (request->fl_flags & FL_ACCESS)
852 * Find the first old lock with the same owner as the new lock.
855 before = &inode->i_flock;
857 /* First skip locks owned by other processes. */
858 while ((fl = *before) && (!IS_POSIX(fl) ||
859 !posix_same_owner(request, fl))) {
860 before = &fl->fl_next;
863 /* Process locks with this owner. */
864 while ((fl = *before) && posix_same_owner(request, fl)) {
865 /* Detect adjacent or overlapping regions (if same lock type)
867 if (request->fl_type == fl->fl_type) {
868 /* In all comparisons of start vs end, use
869 * "start - 1" rather than "end + 1". If end
870 * is OFFSET_MAX, end + 1 will become negative.
872 if (fl->fl_end < request->fl_start - 1)
874 /* If the next lock in the list has entirely bigger
875 * addresses than the new one, insert the lock here.
877 if (fl->fl_start - 1 > request->fl_end)
880 /* If we come here, the new and old lock are of the
881 * same type and adjacent or overlapping. Make one
882 * lock yielding from the lower start address of both
883 * locks to the higher end address.
885 if (fl->fl_start > request->fl_start)
886 fl->fl_start = request->fl_start;
888 request->fl_start = fl->fl_start;
889 if (fl->fl_end < request->fl_end)
890 fl->fl_end = request->fl_end;
892 request->fl_end = fl->fl_end;
894 locks_delete_lock(before);
901 /* Processing for different lock types is a bit
904 if (fl->fl_end < request->fl_start)
906 if (fl->fl_start > request->fl_end)
908 if (request->fl_type == F_UNLCK)
910 if (fl->fl_start < request->fl_start)
912 /* If the next lock in the list has a higher end
913 * address than the new one, insert the new one here.
915 if (fl->fl_end > request->fl_end) {
919 if (fl->fl_start >= request->fl_start) {
920 /* The new lock completely replaces an old
921 * one (This may happen several times).
924 locks_delete_lock(before);
927 /* Replace the old lock with the new one.
928 * Wake up anybody waiting for the old one,
929 * as the change in lock type might satisfy
932 locks_wake_up_blocks(fl);
933 fl->fl_start = request->fl_start;
934 fl->fl_end = request->fl_end;
935 fl->fl_type = request->fl_type;
936 locks_release_private(fl);
937 locks_copy_private(fl, request);
942 /* Go on to next lock.
945 before = &fl->fl_next;
949 * The above code only modifies existing locks in case of
950 * merging or replacing. If new lock(s) need to be inserted
951 * all modifications are done bellow this, so it's safe yet to
954 error = -ENOLCK; /* "no luck" */
955 if (right && left == right && !new_fl2)
960 if (request->fl_type == F_UNLCK) {
961 if (request->fl_flags & FL_EXISTS)
970 locks_copy_lock(new_fl, request);
971 locks_insert_lock(before, new_fl);
976 /* The new lock breaks the old one in two pieces,
977 * so we have to use the second new lock.
981 locks_copy_lock(left, right);
982 locks_insert_lock(before, left);
984 right->fl_start = request->fl_end + 1;
985 locks_wake_up_blocks(right);
988 left->fl_end = request->fl_start - 1;
989 locks_wake_up_blocks(left);
994 * Free any unused locks.
997 locks_free_lock(new_fl);
999 locks_free_lock(new_fl2);
1004 * posix_lock_file - Apply a POSIX-style lock to a file
1005 * @filp: The file to apply the lock to
1006 * @fl: The lock to be applied
1007 * @conflock: Place to return a copy of the conflicting lock, if found.
1009 * Add a POSIX style lock to a file.
1010 * We merge adjacent & overlapping locks whenever possible.
1011 * POSIX locks are sorted by owner task, then by starting address
1013 * Note that if called with an FL_EXISTS argument, the caller may determine
1014 * whether or not a lock was successfully freed by testing the return
1015 * value for -ENOENT.
1017 int posix_lock_file(struct file *filp, struct file_lock *fl,
1018 struct file_lock *conflock)
1020 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1022 EXPORT_SYMBOL(posix_lock_file);
1025 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1026 * @filp: The file to apply the lock to
1027 * @fl: The lock to be applied
1029 * Add a POSIX style lock to a file.
1030 * We merge adjacent & overlapping locks whenever possible.
1031 * POSIX locks are sorted by owner task, then by starting address
1033 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1038 error = posix_lock_file(filp, fl, NULL);
1039 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1041 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1045 locks_delete_block(fl);
1050 EXPORT_SYMBOL(posix_lock_file_wait);
1053 * locks_mandatory_locked - Check for an active lock
1054 * @inode: the file to check
1056 * Searches the inode's list of locks to find any POSIX locks which conflict.
1057 * This function is called from locks_verify_locked() only.
1059 int locks_mandatory_locked(struct inode *inode)
1061 fl_owner_t owner = current->files;
1062 struct file_lock *fl;
1065 * Search the lock list for this inode for any POSIX locks.
1068 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1071 if (fl->fl_owner != owner)
1075 return fl ? -EAGAIN : 0;
1079 * locks_mandatory_area - Check for a conflicting lock
1080 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1082 * @inode: the file to check
1083 * @filp: how the file was opened (if it was)
1084 * @offset: start of area to check
1085 * @count: length of area to check
1087 * Searches the inode's list of locks to find any POSIX locks which conflict.
1088 * This function is called from rw_verify_area() and
1089 * locks_verify_truncate().
1091 int locks_mandatory_area(int read_write, struct inode *inode,
1092 struct file *filp, loff_t offset,
1095 struct file_lock fl;
1098 locks_init_lock(&fl);
1099 fl.fl_owner = current->files;
1100 fl.fl_pid = current->tgid;
1102 fl.fl_flags = FL_POSIX | FL_ACCESS;
1103 if (filp && !(filp->f_flags & O_NONBLOCK))
1104 fl.fl_flags |= FL_SLEEP;
1105 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1106 fl.fl_start = offset;
1107 fl.fl_end = offset + count - 1;
1110 error = __posix_lock_file(inode, &fl, NULL);
1111 if (error != -EAGAIN)
1113 if (!(fl.fl_flags & FL_SLEEP))
1115 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1118 * If we've been sleeping someone might have
1119 * changed the permissions behind our back.
1121 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1125 locks_delete_block(&fl);
1132 EXPORT_SYMBOL(locks_mandatory_area);
1134 /* We already had a lease on this file; just change its type */
1135 int lease_modify(struct file_lock **before, int arg)
1137 struct file_lock *fl = *before;
1138 int error = assign_type(fl, arg);
1142 locks_wake_up_blocks(fl);
1144 locks_delete_lock(before);
1148 EXPORT_SYMBOL(lease_modify);
1150 static void time_out_leases(struct inode *inode)
1152 struct file_lock **before;
1153 struct file_lock *fl;
1155 before = &inode->i_flock;
1156 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1157 if ((fl->fl_break_time == 0)
1158 || time_before(jiffies, fl->fl_break_time)) {
1159 before = &fl->fl_next;
1162 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1163 if (fl == *before) /* lease_modify may have freed fl */
1164 before = &fl->fl_next;
1169 * __break_lease - revoke all outstanding leases on file
1170 * @inode: the inode of the file to return
1171 * @mode: the open mode (read or write)
1173 * break_lease (inlined for speed) has checked there already is at least
1174 * some kind of lock (maybe a lease) on this file. Leases are broken on
1175 * a call to open() or truncate(). This function can sleep unless you
1176 * specified %O_NONBLOCK to your open().
1178 int __break_lease(struct inode *inode, unsigned int mode)
1180 int error = 0, future;
1181 struct file_lock *new_fl, *flock;
1182 struct file_lock *fl;
1183 unsigned long break_time;
1184 int i_have_this_lease = 0;
1186 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
1190 time_out_leases(inode);
1192 flock = inode->i_flock;
1193 if ((flock == NULL) || !IS_LEASE(flock))
1196 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1197 if (fl->fl_owner == current->files)
1198 i_have_this_lease = 1;
1200 if (mode & FMODE_WRITE) {
1201 /* If we want write access, we have to revoke any lease. */
1202 future = F_UNLCK | F_INPROGRESS;
1203 } else if (flock->fl_type & F_INPROGRESS) {
1204 /* If the lease is already being broken, we just leave it */
1205 future = flock->fl_type;
1206 } else if (flock->fl_type & F_WRLCK) {
1207 /* Downgrade the exclusive lease to a read-only lease. */
1208 future = F_RDLCK | F_INPROGRESS;
1210 /* the existing lease was read-only, so we can read too. */
1214 if (IS_ERR(new_fl) && !i_have_this_lease
1215 && ((mode & O_NONBLOCK) == 0)) {
1216 error = PTR_ERR(new_fl);
1221 if (lease_break_time > 0) {
1222 break_time = jiffies + lease_break_time * HZ;
1223 if (break_time == 0)
1224 break_time++; /* so that 0 means no break time */
1227 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1228 if (fl->fl_type != future) {
1229 fl->fl_type = future;
1230 fl->fl_break_time = break_time;
1231 /* lease must have lmops break callback */
1232 fl->fl_lmops->fl_break(fl);
1236 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1237 error = -EWOULDBLOCK;
1242 break_time = flock->fl_break_time;
1243 if (break_time != 0) {
1244 break_time -= jiffies;
1245 if (break_time == 0)
1248 error = locks_block_on_timeout(flock, new_fl, break_time);
1251 time_out_leases(inode);
1252 /* Wait for the next lease that has not been broken yet */
1253 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1254 flock = flock->fl_next) {
1255 if (flock->fl_type & F_INPROGRESS)
1263 if (!IS_ERR(new_fl))
1264 locks_free_lock(new_fl);
1268 EXPORT_SYMBOL(__break_lease);
1273 * @time: pointer to a timespec which will contain the last modified time
1275 * This is to force NFS clients to flush their caches for files with
1276 * exclusive leases. The justification is that if someone has an
1277 * exclusive lease, then they could be modifiying it.
1279 void lease_get_mtime(struct inode *inode, struct timespec *time)
1281 struct file_lock *flock = inode->i_flock;
1282 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1283 *time = current_fs_time(inode->i_sb);
1285 *time = inode->i_mtime;
1288 EXPORT_SYMBOL(lease_get_mtime);
1291 * fcntl_getlease - Enquire what lease is currently active
1294 * The value returned by this function will be one of
1295 * (if no lease break is pending):
1297 * %F_RDLCK to indicate a shared lease is held.
1299 * %F_WRLCK to indicate an exclusive lease is held.
1301 * %F_UNLCK to indicate no lease is held.
1303 * (if a lease break is pending):
1305 * %F_RDLCK to indicate an exclusive lease needs to be
1306 * changed to a shared lease (or removed).
1308 * %F_UNLCK to indicate the lease needs to be removed.
1310 * XXX: sfr & willy disagree over whether F_INPROGRESS
1311 * should be returned to userspace.
1313 int fcntl_getlease(struct file *filp)
1315 struct file_lock *fl;
1319 time_out_leases(filp->f_path.dentry->d_inode);
1320 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1322 if (fl->fl_file == filp) {
1323 type = fl->fl_type & ~F_INPROGRESS;
1332 * generic_setlease - sets a lease on an open file
1333 * @filp: file pointer
1334 * @arg: type of lease to obtain
1335 * @flp: input - file_lock to use, output - file_lock inserted
1337 * The (input) flp->fl_lmops->fl_break function is required
1340 * Called with kernel lock held.
1342 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1344 struct file_lock *fl, **before, **my_before = NULL, *lease;
1345 struct dentry *dentry = filp->f_path.dentry;
1346 struct inode *inode = dentry->d_inode;
1347 int error, rdlease_count = 0, wrlease_count = 0;
1349 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1351 if (!S_ISREG(inode->i_mode))
1353 error = security_file_lock(filp, arg);
1357 time_out_leases(inode);
1359 BUG_ON(!(*flp)->fl_lmops->fl_break);
1364 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1366 if ((arg == F_WRLCK)
1367 && ((atomic_read(&dentry->d_count) > 1)
1368 || (atomic_read(&inode->i_count) > 1)))
1372 * At this point, we know that if there is an exclusive
1373 * lease on this file, then we hold it on this filp
1374 * (otherwise our open of this file would have blocked).
1375 * And if we are trying to acquire an exclusive lease,
1376 * then the file is not open by anyone (including us)
1377 * except for this filp.
1379 for (before = &inode->i_flock;
1380 ((fl = *before) != NULL) && IS_LEASE(fl);
1381 before = &fl->fl_next) {
1382 if (lease->fl_lmops->fl_mylease(fl, lease))
1384 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1386 * Someone is in the process of opening this
1387 * file for writing so we may not take an
1388 * exclusive lease on it.
1395 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1396 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1399 if (my_before != NULL) {
1401 error = lease->fl_lmops->fl_change(my_before, arg);
1414 fl = locks_alloc_lock();
1418 locks_copy_lock(fl, lease);
1420 locks_insert_lock(before, fl);
1427 EXPORT_SYMBOL(generic_setlease);
1430 * vfs_setlease - sets a lease on an open file
1431 * @filp: file pointer
1432 * @arg: type of lease to obtain
1433 * @lease: file_lock to use
1435 * Call this to establish a lease on the file.
1436 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1437 * break_lease will oops!
1439 * This will call the filesystem's setlease file method, if
1440 * defined. Note that there is no getlease method; instead, the
1441 * filesystem setlease method should call back to setlease() to
1442 * add a lease to the inode's lease list, where fcntl_getlease() can
1443 * find it. Since fcntl_getlease() only reports whether the current
1444 * task holds a lease, a cluster filesystem need only do this for
1445 * leases held by processes on this node.
1447 * There is also no break_lease method; filesystems that
1448 * handle their own leases shoud break leases themselves from the
1449 * filesystem's open, create, and (on truncate) setattr methods.
1451 * Warning: the only current setlease methods exist only to disable
1452 * leases in certain cases. More vfs changes may be required to
1453 * allow a full filesystem lease implementation.
1456 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1461 if (filp->f_op && filp->f_op->setlease)
1462 error = filp->f_op->setlease(filp, arg, lease);
1464 error = generic_setlease(filp, arg, lease);
1469 EXPORT_SYMBOL_GPL(vfs_setlease);
1472 * fcntl_setlease - sets a lease on an open file
1473 * @fd: open file descriptor
1474 * @filp: file pointer
1475 * @arg: type of lease to obtain
1477 * Call this fcntl to establish a lease on the file.
1478 * Note that you also need to call %F_SETSIG to
1479 * receive a signal when the lease is broken.
1481 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1483 struct file_lock fl, *flp = &fl;
1484 struct dentry *dentry = filp->f_path.dentry;
1485 struct inode *inode = dentry->d_inode;
1488 locks_init_lock(&fl);
1489 error = lease_init(filp, arg, &fl);
1495 error = vfs_setlease(filp, arg, &flp);
1496 if (error || arg == F_UNLCK)
1499 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1501 /* remove lease just inserted by setlease */
1502 flp->fl_type = F_UNLCK | F_INPROGRESS;
1503 flp->fl_break_time = jiffies - 10;
1504 time_out_leases(inode);
1508 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1515 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1516 * @filp: The file to apply the lock to
1517 * @fl: The lock to be applied
1519 * Add a FLOCK style lock to a file.
1521 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1526 error = flock_lock_file(filp, fl);
1527 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1529 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1533 locks_delete_block(fl);
1539 EXPORT_SYMBOL(flock_lock_file_wait);
1542 * sys_flock: - flock() system call.
1543 * @fd: the file descriptor to lock.
1544 * @cmd: the type of lock to apply.
1546 * Apply a %FL_FLOCK style lock to an open file descriptor.
1547 * The @cmd can be one of
1549 * %LOCK_SH -- a shared lock.
1551 * %LOCK_EX -- an exclusive lock.
1553 * %LOCK_UN -- remove an existing lock.
1555 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1557 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1558 * processes read and write access respectively.
1560 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1563 struct file_lock *lock;
1564 int can_sleep, unlock;
1572 can_sleep = !(cmd & LOCK_NB);
1574 unlock = (cmd == LOCK_UN);
1576 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1579 error = flock_make_lock(filp, &lock, cmd);
1583 lock->fl_flags |= FL_SLEEP;
1585 error = security_file_lock(filp, cmd);
1589 if (filp->f_op && filp->f_op->flock)
1590 error = filp->f_op->flock(filp,
1591 (can_sleep) ? F_SETLKW : F_SETLK,
1594 error = flock_lock_file_wait(filp, lock);
1597 locks_free_lock(lock);
1606 * vfs_test_lock - test file byte range lock
1607 * @filp: The file to test lock for
1608 * @fl: The lock to test; also used to hold result
1610 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1611 * setting conf->fl_type to something other than F_UNLCK.
1613 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1615 if (filp->f_op && filp->f_op->lock)
1616 return filp->f_op->lock(filp, F_GETLK, fl);
1617 posix_test_lock(filp, fl);
1620 EXPORT_SYMBOL_GPL(vfs_test_lock);
1622 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1624 flock->l_pid = fl->fl_pid;
1625 #if BITS_PER_LONG == 32
1627 * Make sure we can represent the posix lock via
1628 * legacy 32bit flock.
1630 if (fl->fl_start > OFFT_OFFSET_MAX)
1632 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1635 flock->l_start = fl->fl_start;
1636 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1637 fl->fl_end - fl->fl_start + 1;
1638 flock->l_whence = 0;
1639 flock->l_type = fl->fl_type;
1643 #if BITS_PER_LONG == 32
1644 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1646 flock->l_pid = fl->fl_pid;
1647 flock->l_start = fl->fl_start;
1648 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1649 fl->fl_end - fl->fl_start + 1;
1650 flock->l_whence = 0;
1651 flock->l_type = fl->fl_type;
1655 /* Report the first existing lock that would conflict with l.
1656 * This implements the F_GETLK command of fcntl().
1658 int fcntl_getlk(struct file *filp, struct flock __user *l)
1660 struct file_lock file_lock;
1665 if (copy_from_user(&flock, l, sizeof(flock)))
1668 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1671 error = flock_to_posix_lock(filp, &file_lock, &flock);
1675 error = vfs_test_lock(filp, &file_lock);
1679 flock.l_type = file_lock.fl_type;
1680 if (file_lock.fl_type != F_UNLCK) {
1681 error = posix_lock_to_flock(&flock, &file_lock);
1686 if (!copy_to_user(l, &flock, sizeof(flock)))
1693 * vfs_lock_file - file byte range lock
1694 * @filp: The file to apply the lock to
1695 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1696 * @fl: The lock to be applied
1697 * @conf: Place to return a copy of the conflicting lock, if found.
1699 * A caller that doesn't care about the conflicting lock may pass NULL
1700 * as the final argument.
1702 * If the filesystem defines a private ->lock() method, then @conf will
1703 * be left unchanged; so a caller that cares should initialize it to
1704 * some acceptable default.
1706 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1707 * locks, the ->lock() interface may return asynchronously, before the lock has
1708 * been granted or denied by the underlying filesystem, if (and only if)
1709 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1710 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1711 * the request is for a blocking lock. When ->lock() does return asynchronously,
1712 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1713 * request completes.
1714 * If the request is for non-blocking lock the file system should return
1715 * -EINPROGRESS then try to get the lock and call the callback routine with
1716 * the result. If the request timed out the callback routine will return a
1717 * nonzero return code and the file system should release the lock. The file
1718 * system is also responsible to keep a corresponding posix lock when it
1719 * grants a lock so the VFS can find out which locks are locally held and do
1720 * the correct lock cleanup when required.
1721 * The underlying filesystem must not drop the kernel lock or call
1722 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1725 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1727 if (filp->f_op && filp->f_op->lock)
1728 return filp->f_op->lock(filp, cmd, fl);
1730 return posix_lock_file(filp, fl, conf);
1732 EXPORT_SYMBOL_GPL(vfs_lock_file);
1734 /* Apply the lock described by l to an open file descriptor.
1735 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1737 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1738 struct flock __user *l)
1740 struct file_lock *file_lock = locks_alloc_lock();
1742 struct inode *inode;
1745 if (file_lock == NULL)
1749 * This might block, so we do it before checking the inode.
1752 if (copy_from_user(&flock, l, sizeof(flock)))
1755 inode = filp->f_path.dentry->d_inode;
1757 /* Don't allow mandatory locks on files that may be memory mapped
1760 if (IS_MANDLOCK(inode) &&
1761 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1762 mapping_writably_mapped(filp->f_mapping)) {
1768 error = flock_to_posix_lock(filp, file_lock, &flock);
1771 if (cmd == F_SETLKW) {
1772 file_lock->fl_flags |= FL_SLEEP;
1776 switch (flock.l_type) {
1778 if (!(filp->f_mode & FMODE_READ))
1782 if (!(filp->f_mode & FMODE_WRITE))
1792 error = security_file_lock(filp, file_lock->fl_type);
1797 error = vfs_lock_file(filp, cmd, file_lock, NULL);
1798 if (error != -EAGAIN || cmd == F_SETLK)
1800 error = wait_event_interruptible(file_lock->fl_wait,
1801 !file_lock->fl_next);
1805 locks_delete_block(file_lock);
1810 * Attempt to detect a close/fcntl race and recover by
1811 * releasing the lock that was just acquired.
1813 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1814 flock.l_type = F_UNLCK;
1819 locks_free_lock(file_lock);
1823 #if BITS_PER_LONG == 32
1824 /* Report the first existing lock that would conflict with l.
1825 * This implements the F_GETLK command of fcntl().
1827 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1829 struct file_lock file_lock;
1830 struct flock64 flock;
1834 if (copy_from_user(&flock, l, sizeof(flock)))
1837 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1840 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1844 error = vfs_test_lock(filp, &file_lock);
1848 flock.l_type = file_lock.fl_type;
1849 if (file_lock.fl_type != F_UNLCK)
1850 posix_lock_to_flock64(&flock, &file_lock);
1853 if (!copy_to_user(l, &flock, sizeof(flock)))
1860 /* Apply the lock described by l to an open file descriptor.
1861 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1863 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1864 struct flock64 __user *l)
1866 struct file_lock *file_lock = locks_alloc_lock();
1867 struct flock64 flock;
1868 struct inode *inode;
1871 if (file_lock == NULL)
1875 * This might block, so we do it before checking the inode.
1878 if (copy_from_user(&flock, l, sizeof(flock)))
1881 inode = filp->f_path.dentry->d_inode;
1883 /* Don't allow mandatory locks on files that may be memory mapped
1886 if (IS_MANDLOCK(inode) &&
1887 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1888 mapping_writably_mapped(filp->f_mapping)) {
1894 error = flock64_to_posix_lock(filp, file_lock, &flock);
1897 if (cmd == F_SETLKW64) {
1898 file_lock->fl_flags |= FL_SLEEP;
1902 switch (flock.l_type) {
1904 if (!(filp->f_mode & FMODE_READ))
1908 if (!(filp->f_mode & FMODE_WRITE))
1918 error = security_file_lock(filp, file_lock->fl_type);
1923 error = vfs_lock_file(filp, cmd, file_lock, NULL);
1924 if (error != -EAGAIN || cmd == F_SETLK64)
1926 error = wait_event_interruptible(file_lock->fl_wait,
1927 !file_lock->fl_next);
1931 locks_delete_block(file_lock);
1936 * Attempt to detect a close/fcntl race and recover by
1937 * releasing the lock that was just acquired.
1939 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1940 flock.l_type = F_UNLCK;
1945 locks_free_lock(file_lock);
1948 #endif /* BITS_PER_LONG == 32 */
1951 * This function is called when the file is being removed
1952 * from the task's fd array. POSIX locks belonging to this task
1953 * are deleted at this time.
1955 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1957 struct file_lock lock;
1960 * If there are no locks held on this file, we don't need to call
1961 * posix_lock_file(). Another process could be setting a lock on this
1962 * file at the same time, but we wouldn't remove that lock anyway.
1964 if (!filp->f_path.dentry->d_inode->i_flock)
1967 lock.fl_type = F_UNLCK;
1968 lock.fl_flags = FL_POSIX | FL_CLOSE;
1970 lock.fl_end = OFFSET_MAX;
1971 lock.fl_owner = owner;
1972 lock.fl_pid = current->tgid;
1973 lock.fl_file = filp;
1975 lock.fl_lmops = NULL;
1977 vfs_lock_file(filp, F_SETLK, &lock, NULL);
1979 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1980 lock.fl_ops->fl_release_private(&lock);
1983 EXPORT_SYMBOL(locks_remove_posix);
1986 * This function is called on the last close of an open file.
1988 void locks_remove_flock(struct file *filp)
1990 struct inode * inode = filp->f_path.dentry->d_inode;
1991 struct file_lock *fl;
1992 struct file_lock **before;
1994 if (!inode->i_flock)
1997 if (filp->f_op && filp->f_op->flock) {
1998 struct file_lock fl = {
1999 .fl_pid = current->tgid,
2001 .fl_flags = FL_FLOCK,
2003 .fl_end = OFFSET_MAX,
2005 filp->f_op->flock(filp, F_SETLKW, &fl);
2006 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2007 fl.fl_ops->fl_release_private(&fl);
2011 before = &inode->i_flock;
2013 while ((fl = *before) != NULL) {
2014 if (fl->fl_file == filp) {
2016 locks_delete_lock(before);
2020 lease_modify(before, F_UNLCK);
2026 before = &fl->fl_next;
2032 * posix_unblock_lock - stop waiting for a file lock
2033 * @filp: how the file was opened
2034 * @waiter: the lock which was waiting
2036 * lockd needs to block waiting for locks.
2039 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2044 if (waiter->fl_next)
2045 __locks_delete_block(waiter);
2052 EXPORT_SYMBOL(posix_unblock_lock);
2055 * vfs_cancel_lock - file byte range unblock lock
2056 * @filp: The file to apply the unblock to
2057 * @fl: The lock to be unblocked
2059 * Used by lock managers to cancel blocked requests
2061 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2063 if (filp->f_op && filp->f_op->lock)
2064 return filp->f_op->lock(filp, F_CANCELLK, fl);
2068 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2070 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2072 struct inode *inode = NULL;
2074 if (fl->fl_file != NULL)
2075 inode = fl->fl_file->f_path.dentry->d_inode;
2077 out += sprintf(out, "%d:%s ", id, pfx);
2079 out += sprintf(out, "%6s %s ",
2080 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2081 (inode == NULL) ? "*NOINODE*" :
2082 (IS_MANDLOCK(inode) &&
2083 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2084 "MANDATORY" : "ADVISORY ");
2085 } else if (IS_FLOCK(fl)) {
2086 if (fl->fl_type & LOCK_MAND) {
2087 out += sprintf(out, "FLOCK MSNFS ");
2089 out += sprintf(out, "FLOCK ADVISORY ");
2091 } else if (IS_LEASE(fl)) {
2092 out += sprintf(out, "LEASE ");
2093 if (fl->fl_type & F_INPROGRESS)
2094 out += sprintf(out, "BREAKING ");
2095 else if (fl->fl_file)
2096 out += sprintf(out, "ACTIVE ");
2098 out += sprintf(out, "BREAKER ");
2100 out += sprintf(out, "UNKNOWN UNKNOWN ");
2102 if (fl->fl_type & LOCK_MAND) {
2103 out += sprintf(out, "%s ",
2104 (fl->fl_type & LOCK_READ)
2105 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2106 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2108 out += sprintf(out, "%s ",
2109 (fl->fl_type & F_INPROGRESS)
2110 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2111 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2114 #ifdef WE_CAN_BREAK_LSLK_NOW
2115 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2116 inode->i_sb->s_id, inode->i_ino);
2118 /* userspace relies on this representation of dev_t ;-( */
2119 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2120 MAJOR(inode->i_sb->s_dev),
2121 MINOR(inode->i_sb->s_dev), inode->i_ino);
2124 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2127 if (fl->fl_end == OFFSET_MAX)
2128 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2130 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2133 out += sprintf(out, "0 EOF\n");
2137 static void move_lock_status(char **p, off_t* pos, off_t offset)
2141 if(*pos >= offset) {
2142 /* the complete line is valid */
2147 if(*pos+len > offset) {
2148 /* use the second part of the line */
2149 int i = offset-*pos;
2150 memmove(*p,*p+i,len-i);
2155 /* discard the complete line */
2160 * get_locks_status - reports lock usage in /proc/locks
2161 * @buffer: address in userspace to write into
2163 * @offset: how far we are through the buffer
2164 * @length: how much to read
2167 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2169 struct list_head *tmp;
2175 list_for_each(tmp, &file_lock_list) {
2176 struct list_head *btmp;
2177 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2178 lock_get_status(q, fl, ++i, "");
2179 move_lock_status(&q, &pos, offset);
2181 if(pos >= offset+length)
2184 list_for_each(btmp, &fl->fl_block) {
2185 struct file_lock *bfl = list_entry(btmp,
2186 struct file_lock, fl_block);
2187 lock_get_status(q, bfl, i, " ->");
2188 move_lock_status(&q, &pos, offset);
2190 if(pos >= offset+length)
2197 if(q-buffer < length)
2203 * lock_may_read - checks that the region is free of locks
2204 * @inode: the inode that is being read
2205 * @start: the first byte to read
2206 * @len: the number of bytes to read
2208 * Emulates Windows locking requirements. Whole-file
2209 * mandatory locks (share modes) can prohibit a read and
2210 * byte-range POSIX locks can prohibit a read if they overlap.
2212 * N.B. this function is only ever called
2213 * from knfsd and ownership of locks is never checked.
2215 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2217 struct file_lock *fl;
2220 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2222 if (fl->fl_type == F_RDLCK)
2224 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2226 } else if (IS_FLOCK(fl)) {
2227 if (!(fl->fl_type & LOCK_MAND))
2229 if (fl->fl_type & LOCK_READ)
2240 EXPORT_SYMBOL(lock_may_read);
2243 * lock_may_write - checks that the region is free of locks
2244 * @inode: the inode that is being written
2245 * @start: the first byte to write
2246 * @len: the number of bytes to write
2248 * Emulates Windows locking requirements. Whole-file
2249 * mandatory locks (share modes) can prohibit a write and
2250 * byte-range POSIX locks can prohibit a write if they overlap.
2252 * N.B. this function is only ever called
2253 * from knfsd and ownership of locks is never checked.
2255 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2257 struct file_lock *fl;
2260 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2262 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2264 } else if (IS_FLOCK(fl)) {
2265 if (!(fl->fl_type & LOCK_MAND))
2267 if (fl->fl_type & LOCK_WRITE)
2278 EXPORT_SYMBOL(lock_may_write);
2280 static int __init filelock_init(void)
2282 filelock_cache = kmem_cache_create("file_lock_cache",
2283 sizeof(struct file_lock), 0, SLAB_PANIC,
2288 core_initcall(filelock_init);