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>
128 #include <linux/vs_base.h>
129 #include <linux/vs_limit.h>
131 #include <asm/semaphore.h>
132 #include <asm/uaccess.h>
134 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
135 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
136 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
138 int leases_enable = 1;
139 int lease_break_time = 45;
141 #define for_each_lock(inode, lockp) \
142 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
144 static LIST_HEAD(file_lock_list);
145 static LIST_HEAD(blocked_list);
147 static kmem_cache_t *filelock_cache __read_mostly;
149 /* Allocate an empty lock structure. */
150 static struct file_lock *locks_alloc_lock(void)
152 if (!vx_locks_avail(1))
154 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
157 static void locks_release_private(struct file_lock *fl)
160 if (fl->fl_ops->fl_release_private)
161 fl->fl_ops->fl_release_private(fl);
165 if (fl->fl_lmops->fl_release_private)
166 fl->fl_lmops->fl_release_private(fl);
172 /* Free a lock which is not in use. */
173 static void locks_free_lock(struct file_lock *fl)
175 BUG_ON(waitqueue_active(&fl->fl_wait));
176 BUG_ON(!list_empty(&fl->fl_block));
177 BUG_ON(!list_empty(&fl->fl_link));
180 locks_release_private(fl);
181 kmem_cache_free(filelock_cache, fl);
184 void locks_init_lock(struct file_lock *fl)
186 INIT_LIST_HEAD(&fl->fl_link);
187 INIT_LIST_HEAD(&fl->fl_block);
188 init_waitqueue_head(&fl->fl_wait);
190 fl->fl_fasync = NULL;
196 fl->fl_start = fl->fl_end = 0;
202 EXPORT_SYMBOL(locks_init_lock);
205 * Initialises the fields of the file lock which are invariant for
208 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
210 struct file_lock *lock = (struct file_lock *) foo;
212 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
213 SLAB_CTOR_CONSTRUCTOR)
216 locks_init_lock(lock);
219 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
222 if (fl->fl_ops->fl_copy_lock)
223 fl->fl_ops->fl_copy_lock(new, fl);
224 new->fl_ops = fl->fl_ops;
227 if (fl->fl_lmops->fl_copy_lock)
228 fl->fl_lmops->fl_copy_lock(new, fl);
229 new->fl_lmops = fl->fl_lmops;
234 * Initialize a new lock from an existing file_lock structure.
236 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
238 new->fl_owner = fl->fl_owner;
239 new->fl_pid = fl->fl_pid;
241 new->fl_flags = fl->fl_flags;
242 new->fl_type = fl->fl_type;
243 new->fl_start = fl->fl_start;
244 new->fl_end = fl->fl_end;
246 new->fl_lmops = NULL;
249 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
251 locks_release_private(new);
253 __locks_copy_lock(new, fl);
254 new->fl_file = fl->fl_file;
255 new->fl_ops = fl->fl_ops;
256 new->fl_lmops = fl->fl_lmops;
257 new->fl_xid = fl->fl_xid;
259 locks_copy_private(new, fl);
262 EXPORT_SYMBOL(locks_copy_lock);
264 static inline int flock_translate_cmd(int cmd) {
266 return cmd & (LOCK_MAND | LOCK_RW);
278 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
279 static int flock_make_lock(struct file *filp, struct file_lock **lock,
282 struct file_lock *fl;
283 int type = flock_translate_cmd(cmd);
287 fl = locks_alloc_lock();
292 fl->fl_pid = current->tgid;
293 fl->fl_flags = FL_FLOCK;
295 fl->fl_end = OFFSET_MAX;
297 fl->fl_xid = filp->f_xid;
304 static int assign_type(struct file_lock *fl, int type)
318 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
321 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
326 switch (l->l_whence) {
334 start = i_size_read(filp->f_dentry->d_inode);
340 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
341 POSIX-2001 defines it. */
345 fl->fl_end = OFFSET_MAX;
347 end = start + l->l_len - 1;
349 } else if (l->l_len < 0) {
356 fl->fl_start = start; /* we record the absolute position */
357 if (fl->fl_end < fl->fl_start)
360 fl->fl_owner = current->files;
361 fl->fl_pid = current->tgid;
363 fl->fl_flags = FL_POSIX;
367 return assign_type(fl, l->l_type);
370 #if BITS_PER_LONG == 32
371 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
376 switch (l->l_whence) {
384 start = i_size_read(filp->f_dentry->d_inode);
393 fl->fl_end = OFFSET_MAX;
395 fl->fl_end = start + l->l_len - 1;
396 } else if (l->l_len < 0) {
397 fl->fl_end = start - 1;
402 fl->fl_start = start; /* we record the absolute position */
403 if (fl->fl_end < fl->fl_start)
406 fl->fl_owner = current->files;
407 fl->fl_pid = current->tgid;
409 fl->fl_flags = FL_POSIX;
417 fl->fl_type = l->l_type;
427 /* default lease lock manager operations */
428 static void lease_break_callback(struct file_lock *fl)
430 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
433 static void lease_release_private_callback(struct file_lock *fl)
438 f_delown(fl->fl_file);
439 fl->fl_file->f_owner.signum = 0;
442 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
444 return fl->fl_file == try->fl_file;
447 static struct lock_manager_operations lease_manager_ops = {
448 .fl_break = lease_break_callback,
449 .fl_release_private = lease_release_private_callback,
450 .fl_mylease = lease_mylease_callback,
451 .fl_change = lease_modify,
455 * Initialize a lease, use the default lock manager operations
457 static int lease_init(struct file *filp, int type, struct file_lock *fl)
459 if (assign_type(fl, type) != 0)
462 fl->fl_owner = current->files;
463 fl->fl_pid = current->tgid;
464 fl->fl_xid = vx_current_xid();
467 fl->fl_flags = FL_LEASE;
469 fl->fl_end = OFFSET_MAX;
471 fl->fl_lmops = &lease_manager_ops;
475 /* Allocate a file_lock initialised to this type of lease */
476 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
478 struct file_lock *fl = locks_alloc_lock();
484 fl->fl_xid = vx_current_xid();
486 error = lease_init(filp, type, fl);
496 /* Check if two locks overlap each other.
498 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
500 return ((fl1->fl_end >= fl2->fl_start) &&
501 (fl2->fl_end >= fl1->fl_start));
505 * Check whether two locks have the same owner.
507 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
509 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
510 return fl2->fl_lmops == fl1->fl_lmops &&
511 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
512 return fl1->fl_owner == fl2->fl_owner;
515 /* Remove waiter from blocker's block list.
516 * When blocker ends up pointing to itself then the list is empty.
518 static void __locks_delete_block(struct file_lock *waiter)
520 list_del_init(&waiter->fl_block);
521 list_del_init(&waiter->fl_link);
522 waiter->fl_next = NULL;
527 static void locks_delete_block(struct file_lock *waiter)
530 __locks_delete_block(waiter);
534 /* Insert waiter into blocker's block list.
535 * We use a circular list so that processes can be easily woken up in
536 * the order they blocked. The documentation doesn't require this but
537 * it seems like the reasonable thing to do.
539 static void locks_insert_block(struct file_lock *blocker,
540 struct file_lock *waiter)
542 BUG_ON(!list_empty(&waiter->fl_block));
543 list_add_tail(&waiter->fl_block, &blocker->fl_block);
544 waiter->fl_next = blocker;
545 if (IS_POSIX(blocker))
546 list_add(&waiter->fl_link, &blocked_list);
549 /* Wake up processes blocked waiting for blocker.
550 * If told to wait then schedule the processes until the block list
551 * is empty, otherwise empty the block list ourselves.
553 static void locks_wake_up_blocks(struct file_lock *blocker)
555 while (!list_empty(&blocker->fl_block)) {
556 struct file_lock *waiter = list_entry(blocker->fl_block.next,
557 struct file_lock, fl_block);
558 __locks_delete_block(waiter);
559 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
560 waiter->fl_lmops->fl_notify(waiter);
562 wake_up(&waiter->fl_wait);
566 /* Insert file lock fl into an inode's lock list at the position indicated
567 * by pos. At the same time add the lock to the global file lock list.
569 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
571 list_add(&fl->fl_link, &file_lock_list);
573 /* insert into file's list */
577 if (fl->fl_ops && fl->fl_ops->fl_insert)
578 fl->fl_ops->fl_insert(fl);
582 * Delete a lock and then free it.
583 * Wake up processes that are blocked waiting for this lock,
584 * notify the FS that the lock has been cleared and
585 * finally free the lock.
587 static void locks_delete_lock(struct file_lock **thisfl_p)
589 struct file_lock *fl = *thisfl_p;
591 *thisfl_p = fl->fl_next;
593 list_del_init(&fl->fl_link);
595 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
596 if (fl->fl_fasync != NULL) {
597 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
598 fl->fl_fasync = NULL;
601 if (fl->fl_ops && fl->fl_ops->fl_remove)
602 fl->fl_ops->fl_remove(fl);
604 locks_wake_up_blocks(fl);
608 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
609 * checks for shared/exclusive status of overlapping locks.
611 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
613 if (sys_fl->fl_type == F_WRLCK)
615 if (caller_fl->fl_type == F_WRLCK)
620 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
621 * checking before calling the locks_conflict().
623 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
625 /* POSIX locks owned by the same process do not conflict with
628 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
631 /* Check whether they overlap */
632 if (!locks_overlap(caller_fl, sys_fl))
635 return (locks_conflict(caller_fl, sys_fl));
638 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
639 * checking before calling the locks_conflict().
641 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
643 /* FLOCK locks referring to the same filp do not conflict with
646 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
648 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
651 return (locks_conflict(caller_fl, sys_fl));
654 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
657 DECLARE_WAITQUEUE(wait, current);
659 __set_current_state(TASK_INTERRUPTIBLE);
660 add_wait_queue(fl_wait, &wait);
664 result = schedule_timeout(timeout);
665 if (signal_pending(current))
666 result = -ERESTARTSYS;
667 remove_wait_queue(fl_wait, &wait);
668 __set_current_state(TASK_RUNNING);
672 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
675 locks_insert_block(blocker, waiter);
676 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
677 __locks_delete_block(waiter);
682 posix_test_lock(struct file *filp, struct file_lock *fl,
683 struct file_lock *conflock)
685 struct file_lock *cfl;
688 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
691 if (posix_locks_conflict(cfl, fl))
695 __locks_copy_lock(conflock, cfl);
703 EXPORT_SYMBOL(posix_test_lock);
705 /* This function tests for deadlock condition before putting a process to
706 * sleep. The detection scheme is no longer recursive. Recursive was neat,
707 * but dangerous - we risked stack corruption if the lock data was bad, or
708 * if the recursion was too deep for any other reason.
710 * We rely on the fact that a task can only be on one lock's wait queue
711 * at a time. When we find blocked_task on a wait queue we can re-search
712 * with blocked_task equal to that queue's owner, until either blocked_task
713 * isn't found, or blocked_task is found on a queue owned by my_task.
715 * Note: the above assumption may not be true when handling lock requests
716 * from a broken NFS client. But broken NFS clients have a lot more to
717 * worry about than proper deadlock detection anyway... --okir
719 static int posix_locks_deadlock(struct file_lock *caller_fl,
720 struct file_lock *block_fl)
722 struct list_head *tmp;
725 if (posix_same_owner(caller_fl, block_fl))
727 list_for_each(tmp, &blocked_list) {
728 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
729 if (posix_same_owner(fl, block_fl)) {
738 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
739 * at the head of the list, but that's secret knowledge known only to
740 * flock_lock_file and posix_lock_file.
742 * Note that if called with an FL_EXISTS argument, the caller may determine
743 * whether or not a lock was successfully freed by testing the return
746 static int flock_lock_file(struct file *filp, struct file_lock *request)
748 struct file_lock *new_fl = NULL;
749 struct file_lock **before;
750 struct inode * inode = filp->f_dentry->d_inode;
755 if (request->fl_flags & FL_ACCESS)
757 for_each_lock(inode, before) {
758 struct file_lock *fl = *before;
763 if (filp != fl->fl_file)
765 if (request->fl_type == fl->fl_type)
768 locks_delete_lock(before);
772 if (request->fl_type == F_UNLCK) {
773 if ((request->fl_flags & FL_EXISTS) && !found)
779 new_fl = locks_alloc_lock();
783 * If a higher-priority process was blocked on the old file lock,
784 * give it the opportunity to lock the file.
790 for_each_lock(inode, before) {
791 struct file_lock *fl = *before;
796 if (!flock_locks_conflict(request, fl))
799 if (request->fl_flags & FL_SLEEP)
800 locks_insert_block(fl, request);
803 if (request->fl_flags & FL_ACCESS)
805 locks_copy_lock(new_fl, request);
806 vx_locks_inc(new_fl);
807 locks_insert_lock(&inode->i_flock, new_fl);
814 locks_free_lock(new_fl);
818 static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request,
819 struct file_lock *conflock, xid_t xid)
821 struct file_lock *fl;
822 struct file_lock *new_fl = NULL;
823 struct file_lock *new_fl2 = NULL;
824 struct file_lock *left = NULL;
825 struct file_lock *right = NULL;
826 struct file_lock **before;
827 int error, added = 0;
830 * We may need two file_lock structures for this operation,
831 * so we get them in advance to avoid races.
833 * In some cases we can be sure, that no new locks will be needed
835 if (!(request->fl_flags & FL_ACCESS) &&
836 (request->fl_type != F_UNLCK ||
837 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
838 new_fl = locks_alloc_lock();
839 new_fl->fl_xid = xid;
840 vx_locks_inc(new_fl);
841 new_fl2 = locks_alloc_lock();
842 new_fl2->fl_xid = xid;
843 vx_locks_inc(new_fl2);
847 if (request->fl_type != F_UNLCK) {
848 for_each_lock(inode, before) {
849 struct file_lock *fl = *before;
852 if (!posix_locks_conflict(request, fl))
855 locks_copy_lock(conflock, fl);
857 if (!(request->fl_flags & FL_SLEEP))
860 if (posix_locks_deadlock(request, fl))
863 locks_insert_block(fl, request);
868 /* If we're just looking for a conflict, we're done. */
870 if (request->fl_flags & FL_ACCESS)
874 * Find the first old lock with the same owner as the new lock.
877 before = &inode->i_flock;
879 /* First skip locks owned by other processes. */
880 while ((fl = *before) && (!IS_POSIX(fl) ||
881 !posix_same_owner(request, fl))) {
882 before = &fl->fl_next;
885 /* Process locks with this owner. */
886 while ((fl = *before) && posix_same_owner(request, fl)) {
887 /* Detect adjacent or overlapping regions (if same lock type)
889 if (request->fl_type == fl->fl_type) {
890 /* In all comparisons of start vs end, use
891 * "start - 1" rather than "end + 1". If end
892 * is OFFSET_MAX, end + 1 will become negative.
894 if (fl->fl_end < request->fl_start - 1)
896 /* If the next lock in the list has entirely bigger
897 * addresses than the new one, insert the lock here.
899 if (fl->fl_start - 1 > request->fl_end)
902 /* If we come here, the new and old lock are of the
903 * same type and adjacent or overlapping. Make one
904 * lock yielding from the lower start address of both
905 * locks to the higher end address.
907 if (fl->fl_start > request->fl_start)
908 fl->fl_start = request->fl_start;
910 request->fl_start = fl->fl_start;
911 if (fl->fl_end < request->fl_end)
912 fl->fl_end = request->fl_end;
914 request->fl_end = fl->fl_end;
916 locks_delete_lock(before);
923 /* Processing for different lock types is a bit
926 if (fl->fl_end < request->fl_start)
928 if (fl->fl_start > request->fl_end)
930 if (request->fl_type == F_UNLCK)
932 if (fl->fl_start < request->fl_start)
934 /* If the next lock in the list has a higher end
935 * address than the new one, insert the new one here.
937 if (fl->fl_end > request->fl_end) {
941 if (fl->fl_start >= request->fl_start) {
942 /* The new lock completely replaces an old
943 * one (This may happen several times).
946 locks_delete_lock(before);
949 /* Replace the old lock with the new one.
950 * Wake up anybody waiting for the old one,
951 * as the change in lock type might satisfy
954 locks_wake_up_blocks(fl);
955 fl->fl_start = request->fl_start;
956 fl->fl_end = request->fl_end;
957 fl->fl_type = request->fl_type;
958 locks_release_private(fl);
959 locks_copy_private(fl, request);
964 /* Go on to next lock.
967 before = &fl->fl_next;
971 * The above code only modifies existing locks in case of
972 * merging or replacing. If new lock(s) need to be inserted
973 * all modifications are done bellow this, so it's safe yet to
976 error = -ENOLCK; /* "no luck" */
977 if (right && left == right && !new_fl2)
982 if (request->fl_type == F_UNLCK) {
983 if (request->fl_flags & FL_EXISTS)
992 locks_copy_lock(new_fl, request);
993 locks_insert_lock(before, new_fl);
998 /* The new lock breaks the old one in two pieces,
999 * so we have to use the second new lock.
1003 locks_copy_lock(left, right);
1004 locks_insert_lock(before, left);
1006 right->fl_start = request->fl_end + 1;
1007 locks_wake_up_blocks(right);
1010 left->fl_end = request->fl_start - 1;
1011 locks_wake_up_blocks(left);
1016 * Free any unused locks.
1019 locks_free_lock(new_fl);
1021 locks_free_lock(new_fl2);
1026 * posix_lock_file - Apply a POSIX-style lock to a file
1027 * @filp: The file to apply the lock to
1028 * @fl: The lock to be applied
1030 * Add a POSIX style lock to a file.
1031 * We merge adjacent & overlapping locks whenever possible.
1032 * POSIX locks are sorted by owner task, then by starting address
1034 * Note that if called with an FL_EXISTS argument, the caller may determine
1035 * whether or not a lock was successfully freed by testing the return
1036 * value for -ENOENT.
1038 int posix_lock_file(struct file *filp, struct file_lock *fl)
1040 return __posix_lock_file_conf(filp->f_dentry->d_inode,
1041 fl, NULL, filp->f_xid);
1043 EXPORT_SYMBOL(posix_lock_file);
1046 * posix_lock_file_conf - Apply a POSIX-style lock to a file
1047 * @filp: The file to apply the lock to
1048 * @fl: The lock to be applied
1049 * @conflock: Place to return a copy of the conflicting lock, if found.
1051 * Except for the conflock parameter, acts just like posix_lock_file.
1053 int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1054 struct file_lock *conflock)
1056 return __posix_lock_file_conf(filp->f_dentry->d_inode,
1057 fl, conflock, filp->f_xid);
1059 EXPORT_SYMBOL(posix_lock_file_conf);
1062 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1063 * @filp: The file to apply the lock to
1064 * @fl: The lock to be applied
1066 * Add a POSIX style lock to a file.
1067 * We merge adjacent & overlapping locks whenever possible.
1068 * POSIX locks are sorted by owner task, then by starting address
1070 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1075 error = posix_lock_file(filp, fl);
1076 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1078 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1082 locks_delete_block(fl);
1087 EXPORT_SYMBOL(posix_lock_file_wait);
1090 * locks_mandatory_locked - Check for an active lock
1091 * @inode: the file to check
1093 * Searches the inode's list of locks to find any POSIX locks which conflict.
1094 * This function is called from locks_verify_locked() only.
1096 int locks_mandatory_locked(struct inode *inode)
1098 fl_owner_t owner = current->files;
1099 struct file_lock *fl;
1102 * Search the lock list for this inode for any POSIX locks.
1105 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1108 if (fl->fl_owner != owner)
1112 return fl ? -EAGAIN : 0;
1116 * locks_mandatory_area - Check for a conflicting lock
1117 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1119 * @inode: the file to check
1120 * @filp: how the file was opened (if it was)
1121 * @offset: start of area to check
1122 * @count: length of area to check
1124 * Searches the inode's list of locks to find any POSIX locks which conflict.
1125 * This function is called from rw_verify_area() and
1126 * locks_verify_truncate().
1128 int locks_mandatory_area(int read_write, struct inode *inode,
1129 struct file *filp, loff_t offset,
1132 struct file_lock fl;
1135 locks_init_lock(&fl);
1136 fl.fl_owner = current->files;
1137 fl.fl_pid = current->tgid;
1139 fl.fl_flags = FL_POSIX | FL_ACCESS;
1140 if (filp && !(filp->f_flags & O_NONBLOCK))
1141 fl.fl_flags |= FL_SLEEP;
1142 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1143 fl.fl_start = offset;
1144 fl.fl_end = offset + count - 1;
1147 error = __posix_lock_file_conf(inode, &fl, NULL, filp->f_xid);
1148 if (error != -EAGAIN)
1150 if (!(fl.fl_flags & FL_SLEEP))
1152 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1155 * If we've been sleeping someone might have
1156 * changed the permissions behind our back.
1158 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1162 locks_delete_block(&fl);
1169 EXPORT_SYMBOL(locks_mandatory_area);
1171 /* We already had a lease on this file; just change its type */
1172 int lease_modify(struct file_lock **before, int arg)
1174 struct file_lock *fl = *before;
1175 int error = assign_type(fl, arg);
1179 locks_wake_up_blocks(fl);
1181 locks_delete_lock(before);
1185 EXPORT_SYMBOL(lease_modify);
1187 static void time_out_leases(struct inode *inode)
1189 struct file_lock **before;
1190 struct file_lock *fl;
1192 before = &inode->i_flock;
1193 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1194 if ((fl->fl_break_time == 0)
1195 || time_before(jiffies, fl->fl_break_time)) {
1196 before = &fl->fl_next;
1199 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1200 if (fl == *before) /* lease_modify may have freed fl */
1201 before = &fl->fl_next;
1206 * __break_lease - revoke all outstanding leases on file
1207 * @inode: the inode of the file to return
1208 * @mode: the open mode (read or write)
1210 * break_lease (inlined for speed) has checked there already
1211 * is a lease on this file. Leases are broken on a call to open()
1212 * or truncate(). This function can sleep unless you
1213 * specified %O_NONBLOCK to your open().
1215 int __break_lease(struct inode *inode, unsigned int mode)
1217 int error = 0, future;
1218 struct file_lock *new_fl, *flock;
1219 struct file_lock *fl;
1221 unsigned long break_time;
1222 int i_have_this_lease = 0;
1224 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1229 time_out_leases(inode);
1231 flock = inode->i_flock;
1232 if ((flock == NULL) || !IS_LEASE(flock))
1235 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1236 if (fl->fl_owner == current->files)
1237 i_have_this_lease = 1;
1239 if (mode & FMODE_WRITE) {
1240 /* If we want write access, we have to revoke any lease. */
1241 future = F_UNLCK | F_INPROGRESS;
1242 } else if (flock->fl_type & F_INPROGRESS) {
1243 /* If the lease is already being broken, we just leave it */
1244 future = flock->fl_type;
1245 } else if (flock->fl_type & F_WRLCK) {
1246 /* Downgrade the exclusive lease to a read-only lease. */
1247 future = F_RDLCK | F_INPROGRESS;
1249 /* the existing lease was read-only, so we can read too. */
1253 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1259 if (lease_break_time > 0) {
1260 break_time = jiffies + lease_break_time * HZ;
1261 if (break_time == 0)
1262 break_time++; /* so that 0 means no break time */
1265 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1266 if (fl->fl_type != future) {
1267 fl->fl_type = future;
1268 fl->fl_break_time = break_time;
1269 /* lease must have lmops break callback */
1270 fl->fl_lmops->fl_break(fl);
1274 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1275 error = -EWOULDBLOCK;
1280 break_time = flock->fl_break_time;
1281 if (break_time != 0) {
1282 break_time -= jiffies;
1283 if (break_time == 0)
1286 error = locks_block_on_timeout(flock, new_fl, break_time);
1289 time_out_leases(inode);
1290 /* Wait for the next lease that has not been broken yet */
1291 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1292 flock = flock->fl_next) {
1293 if (flock->fl_type & F_INPROGRESS)
1302 locks_free_lock(new_fl);
1306 EXPORT_SYMBOL(__break_lease);
1311 * @time: pointer to a timespec which will contain the last modified time
1313 * This is to force NFS clients to flush their caches for files with
1314 * exclusive leases. The justification is that if someone has an
1315 * exclusive lease, then they could be modifiying it.
1317 void lease_get_mtime(struct inode *inode, struct timespec *time)
1319 struct file_lock *flock = inode->i_flock;
1320 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1321 *time = current_fs_time(inode->i_sb);
1323 *time = inode->i_mtime;
1326 EXPORT_SYMBOL(lease_get_mtime);
1329 * fcntl_getlease - Enquire what lease is currently active
1332 * The value returned by this function will be one of
1333 * (if no lease break is pending):
1335 * %F_RDLCK to indicate a shared lease is held.
1337 * %F_WRLCK to indicate an exclusive lease is held.
1339 * %F_UNLCK to indicate no lease is held.
1341 * (if a lease break is pending):
1343 * %F_RDLCK to indicate an exclusive lease needs to be
1344 * changed to a shared lease (or removed).
1346 * %F_UNLCK to indicate the lease needs to be removed.
1348 * XXX: sfr & willy disagree over whether F_INPROGRESS
1349 * should be returned to userspace.
1351 int fcntl_getlease(struct file *filp)
1353 struct file_lock *fl;
1357 time_out_leases(filp->f_dentry->d_inode);
1358 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1360 if (fl->fl_file == filp) {
1361 type = fl->fl_type & ~F_INPROGRESS;
1370 * __setlease - sets a lease on an open file
1371 * @filp: file pointer
1372 * @arg: type of lease to obtain
1373 * @flp: input - file_lock to use, output - file_lock inserted
1375 * The (input) flp->fl_lmops->fl_break function is required
1378 * Called with kernel lock held.
1380 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1382 struct file_lock *fl, **before, **my_before = NULL, *lease;
1383 struct dentry *dentry = filp->f_dentry;
1384 struct inode *inode = dentry->d_inode;
1385 int error, rdlease_count = 0, wrlease_count = 0;
1387 time_out_leases(inode);
1390 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1396 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1398 if ((arg == F_WRLCK)
1399 && ((atomic_read(&dentry->d_count) > 1)
1400 || (atomic_read(&inode->i_count) > 1)))
1404 * At this point, we know that if there is an exclusive
1405 * lease on this file, then we hold it on this filp
1406 * (otherwise our open of this file would have blocked).
1407 * And if we are trying to acquire an exclusive lease,
1408 * then the file is not open by anyone (including us)
1409 * except for this filp.
1411 for (before = &inode->i_flock;
1412 ((fl = *before) != NULL) && IS_LEASE(fl);
1413 before = &fl->fl_next) {
1414 if (lease->fl_lmops->fl_mylease(fl, lease))
1416 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1418 * Someone is in the process of opening this
1419 * file for writing so we may not take an
1420 * exclusive lease on it.
1427 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1428 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1431 if (my_before != NULL) {
1433 error = lease->fl_lmops->fl_change(my_before, arg);
1446 fl = locks_alloc_lock();
1450 locks_copy_lock(fl, lease);
1452 locks_insert_lock(before, fl);
1461 * setlease - sets a lease on an open file
1462 * @filp: file pointer
1463 * @arg: type of lease to obtain
1464 * @lease: file_lock to use
1466 * Call this to establish a lease on the file.
1467 * The fl_lmops fl_break function is required by break_lease
1470 int setlease(struct file *filp, long arg, struct file_lock **lease)
1472 struct dentry *dentry = filp->f_dentry;
1473 struct inode *inode = dentry->d_inode;
1476 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1478 if (!S_ISREG(inode->i_mode))
1480 error = security_file_lock(filp, arg);
1485 error = __setlease(filp, arg, lease);
1491 EXPORT_SYMBOL(setlease);
1494 * fcntl_setlease - sets a lease on an open file
1495 * @fd: open file descriptor
1496 * @filp: file pointer
1497 * @arg: type of lease to obtain
1499 * Call this fcntl to establish a lease on the file.
1500 * Note that you also need to call %F_SETSIG to
1501 * receive a signal when the lease is broken.
1503 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1505 struct file_lock fl, *flp = &fl;
1506 struct dentry *dentry = filp->f_dentry;
1507 struct inode *inode = dentry->d_inode;
1510 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1512 if (!S_ISREG(inode->i_mode))
1514 error = security_file_lock(filp, arg);
1518 locks_init_lock(&fl);
1519 error = lease_init(filp, arg, &fl);
1525 error = __setlease(filp, arg, &flp);
1526 if (error || arg == F_UNLCK)
1529 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1531 /* remove lease just inserted by __setlease */
1532 flp->fl_type = F_UNLCK | F_INPROGRESS;
1533 flp->fl_break_time = jiffies- 10;
1534 time_out_leases(inode);
1538 error = f_setown(filp, current->pid, 0);
1545 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1546 * @filp: The file to apply the lock to
1547 * @fl: The lock to be applied
1549 * Add a FLOCK style lock to a file.
1551 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1556 error = flock_lock_file(filp, fl);
1557 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1559 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1563 locks_delete_block(fl);
1569 EXPORT_SYMBOL(flock_lock_file_wait);
1572 * sys_flock: - flock() system call.
1573 * @fd: the file descriptor to lock.
1574 * @cmd: the type of lock to apply.
1576 * Apply a %FL_FLOCK style lock to an open file descriptor.
1577 * The @cmd can be one of
1579 * %LOCK_SH -- a shared lock.
1581 * %LOCK_EX -- an exclusive lock.
1583 * %LOCK_UN -- remove an existing lock.
1585 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1587 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1588 * processes read and write access respectively.
1590 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1593 struct file_lock *lock;
1594 int can_sleep, unlock;
1602 can_sleep = !(cmd & LOCK_NB);
1604 unlock = (cmd == LOCK_UN);
1606 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1609 error = flock_make_lock(filp, &lock, cmd);
1613 lock->fl_flags |= FL_SLEEP;
1615 error = security_file_lock(filp, cmd);
1619 if (filp->f_op && filp->f_op->flock)
1620 error = filp->f_op->flock(filp,
1621 (can_sleep) ? F_SETLKW : F_SETLK,
1624 error = flock_lock_file_wait(filp, lock);
1627 locks_free_lock(lock);
1635 /* Report the first existing lock that would conflict with l.
1636 * This implements the F_GETLK command of fcntl().
1638 int fcntl_getlk(struct file *filp, struct flock __user *l)
1640 struct file_lock *fl, cfl, file_lock;
1645 if (copy_from_user(&flock, l, sizeof(flock)))
1648 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1651 error = flock_to_posix_lock(filp, &file_lock, &flock);
1655 if (filp->f_op && filp->f_op->lock) {
1656 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1657 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1658 file_lock.fl_ops->fl_release_private(&file_lock);
1662 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1664 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1667 flock.l_type = F_UNLCK;
1669 flock.l_pid = fl->fl_pid;
1670 #if BITS_PER_LONG == 32
1672 * Make sure we can represent the posix lock via
1673 * legacy 32bit flock.
1676 if (fl->fl_start > OFFT_OFFSET_MAX)
1678 if ((fl->fl_end != OFFSET_MAX)
1679 && (fl->fl_end > OFFT_OFFSET_MAX))
1682 flock.l_start = fl->fl_start;
1683 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1684 fl->fl_end - fl->fl_start + 1;
1686 flock.l_type = fl->fl_type;
1689 if (!copy_to_user(l, &flock, sizeof(flock)))
1695 /* Apply the lock described by l to an open file descriptor.
1696 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1698 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1699 struct flock __user *l)
1701 struct file_lock *file_lock = locks_alloc_lock();
1703 struct inode *inode;
1706 if (file_lock == NULL)
1709 file_lock->fl_xid = filp->f_xid;
1710 vx_locks_inc(file_lock);
1713 * This might block, so we do it before checking the inode.
1716 if (copy_from_user(&flock, l, sizeof(flock)))
1719 inode = filp->f_dentry->d_inode;
1721 /* Don't allow mandatory locks on files that may be memory mapped
1724 if (IS_MANDLOCK(inode) &&
1725 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1726 mapping_writably_mapped(filp->f_mapping)) {
1732 error = flock_to_posix_lock(filp, file_lock, &flock);
1735 if (cmd == F_SETLKW) {
1736 file_lock->fl_flags |= FL_SLEEP;
1740 switch (flock.l_type) {
1742 if (!(filp->f_mode & FMODE_READ))
1746 if (!(filp->f_mode & FMODE_WRITE))
1756 error = security_file_lock(filp, file_lock->fl_type);
1760 if (filp->f_op && filp->f_op->lock != NULL)
1761 error = filp->f_op->lock(filp, cmd, file_lock);
1764 error = posix_lock_file(filp, file_lock);
1765 if ((error != -EAGAIN) || (cmd == F_SETLK))
1767 error = wait_event_interruptible(file_lock->fl_wait,
1768 !file_lock->fl_next);
1772 locks_delete_block(file_lock);
1778 * Attempt to detect a close/fcntl race and recover by
1779 * releasing the lock that was just acquired.
1781 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1782 flock.l_type = F_UNLCK;
1787 locks_free_lock(file_lock);
1791 #if BITS_PER_LONG == 32
1792 /* Report the first existing lock that would conflict with l.
1793 * This implements the F_GETLK command of fcntl().
1795 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1797 struct file_lock *fl, cfl, file_lock;
1798 struct flock64 flock;
1802 if (copy_from_user(&flock, l, sizeof(flock)))
1805 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1808 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1812 if (filp->f_op && filp->f_op->lock) {
1813 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1814 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1815 file_lock.fl_ops->fl_release_private(&file_lock);
1819 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1821 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
1824 flock.l_type = F_UNLCK;
1826 flock.l_pid = fl->fl_pid;
1827 flock.l_start = fl->fl_start;
1828 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1829 fl->fl_end - fl->fl_start + 1;
1831 flock.l_type = fl->fl_type;
1834 if (!copy_to_user(l, &flock, sizeof(flock)))
1841 /* Apply the lock described by l to an open file descriptor.
1842 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1844 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1845 struct flock64 __user *l)
1847 struct file_lock *file_lock = locks_alloc_lock();
1848 struct flock64 flock;
1849 struct inode *inode;
1852 if (file_lock == NULL)
1855 file_lock->fl_xid = filp->f_xid;
1856 vx_locks_inc(file_lock);
1859 * This might block, so we do it before checking the inode.
1862 if (copy_from_user(&flock, l, sizeof(flock)))
1865 inode = filp->f_dentry->d_inode;
1867 /* Don't allow mandatory locks on files that may be memory mapped
1870 if (IS_MANDLOCK(inode) &&
1871 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1872 mapping_writably_mapped(filp->f_mapping)) {
1878 error = flock64_to_posix_lock(filp, file_lock, &flock);
1881 if (cmd == F_SETLKW64) {
1882 file_lock->fl_flags |= FL_SLEEP;
1886 switch (flock.l_type) {
1888 if (!(filp->f_mode & FMODE_READ))
1892 if (!(filp->f_mode & FMODE_WRITE))
1902 error = security_file_lock(filp, file_lock->fl_type);
1906 if (filp->f_op && filp->f_op->lock != NULL)
1907 error = filp->f_op->lock(filp, cmd, file_lock);
1910 error = posix_lock_file(filp, file_lock);
1911 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1913 error = wait_event_interruptible(file_lock->fl_wait,
1914 !file_lock->fl_next);
1918 locks_delete_block(file_lock);
1924 * Attempt to detect a close/fcntl race and recover by
1925 * releasing the lock that was just acquired.
1927 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1928 flock.l_type = F_UNLCK;
1933 locks_free_lock(file_lock);
1936 #endif /* BITS_PER_LONG == 32 */
1939 * This function is called when the file is being removed
1940 * from the task's fd array. POSIX locks belonging to this task
1941 * are deleted at this time.
1943 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1945 struct file_lock lock;
1948 * If there are no locks held on this file, we don't need to call
1949 * posix_lock_file(). Another process could be setting a lock on this
1950 * file at the same time, but we wouldn't remove that lock anyway.
1952 if (!filp->f_dentry->d_inode->i_flock)
1955 lock.fl_type = F_UNLCK;
1956 lock.fl_flags = FL_POSIX | FL_CLOSE;
1958 lock.fl_end = OFFSET_MAX;
1959 lock.fl_owner = owner;
1960 lock.fl_pid = current->tgid;
1961 lock.fl_file = filp;
1963 lock.fl_lmops = NULL;
1965 if (filp->f_op && filp->f_op->lock != NULL)
1966 filp->f_op->lock(filp, F_SETLK, &lock);
1968 posix_lock_file(filp, &lock);
1970 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1971 lock.fl_ops->fl_release_private(&lock);
1974 EXPORT_SYMBOL(locks_remove_posix);
1977 * This function is called on the last close of an open file.
1979 void locks_remove_flock(struct file *filp)
1981 struct inode * inode = filp->f_dentry->d_inode;
1982 struct file_lock *fl;
1983 struct file_lock **before;
1985 if (!inode->i_flock)
1988 if (filp->f_op && filp->f_op->flock) {
1989 struct file_lock fl = {
1990 .fl_pid = current->tgid,
1992 .fl_flags = FL_FLOCK,
1994 .fl_end = OFFSET_MAX,
1996 filp->f_op->flock(filp, F_SETLKW, &fl);
1997 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1998 fl.fl_ops->fl_release_private(&fl);
2002 before = &inode->i_flock;
2004 while ((fl = *before) != NULL) {
2005 if (fl->fl_file == filp) {
2007 locks_delete_lock(before);
2011 lease_modify(before, F_UNLCK);
2017 before = &fl->fl_next;
2023 * posix_unblock_lock - stop waiting for a file lock
2024 * @filp: how the file was opened
2025 * @waiter: the lock which was waiting
2027 * lockd needs to block waiting for locks.
2030 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2035 if (waiter->fl_next)
2036 __locks_delete_block(waiter);
2043 EXPORT_SYMBOL(posix_unblock_lock);
2045 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2047 struct inode *inode = NULL;
2049 if (fl->fl_file != NULL)
2050 inode = fl->fl_file->f_dentry->d_inode;
2052 out += sprintf(out, "%d:%s ", id, pfx);
2054 out += sprintf(out, "%6s %s ",
2055 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2056 (inode == NULL) ? "*NOINODE*" :
2057 (IS_MANDLOCK(inode) &&
2058 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2059 "MANDATORY" : "ADVISORY ");
2060 } else if (IS_FLOCK(fl)) {
2061 if (fl->fl_type & LOCK_MAND) {
2062 out += sprintf(out, "FLOCK MSNFS ");
2064 out += sprintf(out, "FLOCK ADVISORY ");
2066 } else if (IS_LEASE(fl)) {
2067 out += sprintf(out, "LEASE ");
2068 if (fl->fl_type & F_INPROGRESS)
2069 out += sprintf(out, "BREAKING ");
2070 else if (fl->fl_file)
2071 out += sprintf(out, "ACTIVE ");
2073 out += sprintf(out, "BREAKER ");
2075 out += sprintf(out, "UNKNOWN UNKNOWN ");
2077 if (fl->fl_type & LOCK_MAND) {
2078 out += sprintf(out, "%s ",
2079 (fl->fl_type & LOCK_READ)
2080 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2081 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2083 out += sprintf(out, "%s ",
2084 (fl->fl_type & F_INPROGRESS)
2085 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2086 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2089 #ifdef WE_CAN_BREAK_LSLK_NOW
2090 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2091 inode->i_sb->s_id, inode->i_ino);
2093 /* userspace relies on this representation of dev_t ;-( */
2094 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2095 MAJOR(inode->i_sb->s_dev),
2096 MINOR(inode->i_sb->s_dev), inode->i_ino);
2099 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2102 if (fl->fl_end == OFFSET_MAX)
2103 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2105 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2108 out += sprintf(out, "0 EOF\n");
2112 static void move_lock_status(char **p, off_t* pos, off_t offset)
2116 if(*pos >= offset) {
2117 /* the complete line is valid */
2122 if(*pos+len > offset) {
2123 /* use the second part of the line */
2124 int i = offset-*pos;
2125 memmove(*p,*p+i,len-i);
2130 /* discard the complete line */
2135 * get_locks_status - reports lock usage in /proc/locks
2136 * @buffer: address in userspace to write into
2138 * @offset: how far we are through the buffer
2139 * @length: how much to read
2142 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2144 struct list_head *tmp;
2150 list_for_each(tmp, &file_lock_list) {
2151 struct list_head *btmp;
2152 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2154 if (!vx_check(fl->fl_xid, VX_IDENT|VX_WATCH))
2157 lock_get_status(q, fl, ++i, "");
2158 move_lock_status(&q, &pos, offset);
2160 if(pos >= offset+length)
2163 list_for_each(btmp, &fl->fl_block) {
2164 struct file_lock *bfl = list_entry(btmp,
2165 struct file_lock, fl_block);
2166 lock_get_status(q, bfl, i, " ->");
2167 move_lock_status(&q, &pos, offset);
2169 if(pos >= offset+length)
2176 if(q-buffer < length)
2182 * lock_may_read - checks that the region is free of locks
2183 * @inode: the inode that is being read
2184 * @start: the first byte to read
2185 * @len: the number of bytes to read
2187 * Emulates Windows locking requirements. Whole-file
2188 * mandatory locks (share modes) can prohibit a read and
2189 * byte-range POSIX locks can prohibit a read if they overlap.
2191 * N.B. this function is only ever called
2192 * from knfsd and ownership of locks is never checked.
2194 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2196 struct file_lock *fl;
2199 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2201 if (fl->fl_type == F_RDLCK)
2203 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2205 } else if (IS_FLOCK(fl)) {
2206 if (!(fl->fl_type & LOCK_MAND))
2208 if (fl->fl_type & LOCK_READ)
2219 EXPORT_SYMBOL(lock_may_read);
2222 * lock_may_write - checks that the region is free of locks
2223 * @inode: the inode that is being written
2224 * @start: the first byte to write
2225 * @len: the number of bytes to write
2227 * Emulates Windows locking requirements. Whole-file
2228 * mandatory locks (share modes) can prohibit a write and
2229 * byte-range POSIX locks can prohibit a write if they overlap.
2231 * N.B. this function is only ever called
2232 * from knfsd and ownership of locks is never checked.
2234 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2236 struct file_lock *fl;
2239 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2241 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2243 } else if (IS_FLOCK(fl)) {
2244 if (!(fl->fl_type & LOCK_MAND))
2246 if (fl->fl_type & LOCK_WRITE)
2257 EXPORT_SYMBOL(lock_may_write);
2259 static int __init filelock_init(void)
2261 filelock_cache = kmem_cache_create("file_lock_cache",
2262 sizeof(struct file_lock), 0, SLAB_PANIC,
2267 core_initcall(filelock_init);