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/fdtable.h>
120 #include <linux/fs.h>
121 #include <linux/init.h>
122 #include <linux/module.h>
123 #include <linux/security.h>
124 #include <linux/slab.h>
125 #include <linux/smp_lock.h>
126 #include <linux/syscalls.h>
127 #include <linux/time.h>
128 #include <linux/rcupdate.h>
129 #include <linux/pid_namespace.h>
131 #include <asm/uaccess.h>
133 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135 #define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
137 int leases_enable = 1;
138 int lease_break_time = 45;
140 #define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
143 static LIST_HEAD(file_lock_list);
144 static LIST_HEAD(blocked_list);
146 static struct kmem_cache *filelock_cache __read_mostly;
148 /* Allocate an empty lock structure. */
149 static struct file_lock *locks_alloc_lock(void)
151 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
154 void locks_release_private(struct file_lock *fl)
157 if (fl->fl_ops->fl_release_private)
158 fl->fl_ops->fl_release_private(fl);
162 if (fl->fl_lmops->fl_release_private)
163 fl->fl_lmops->fl_release_private(fl);
168 EXPORT_SYMBOL_GPL(locks_release_private);
170 /* Free a lock which is not in use. */
171 static void locks_free_lock(struct file_lock *fl)
173 BUG_ON(waitqueue_active(&fl->fl_wait));
174 BUG_ON(!list_empty(&fl->fl_block));
175 BUG_ON(!list_empty(&fl->fl_link));
177 locks_release_private(fl);
178 kmem_cache_free(filelock_cache, fl);
181 void locks_init_lock(struct file_lock *fl)
183 INIT_LIST_HEAD(&fl->fl_link);
184 INIT_LIST_HEAD(&fl->fl_block);
185 init_waitqueue_head(&fl->fl_wait);
187 fl->fl_fasync = NULL;
194 fl->fl_start = fl->fl_end = 0;
199 EXPORT_SYMBOL(locks_init_lock);
202 * Initialises the fields of the file lock which are invariant for
205 static void init_once(void *foo)
207 struct file_lock *lock = (struct file_lock *) foo;
209 locks_init_lock(lock);
212 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
215 if (fl->fl_ops->fl_copy_lock)
216 fl->fl_ops->fl_copy_lock(new, fl);
217 new->fl_ops = fl->fl_ops;
220 if (fl->fl_lmops->fl_copy_lock)
221 fl->fl_lmops->fl_copy_lock(new, fl);
222 new->fl_lmops = fl->fl_lmops;
227 * Initialize a new lock from an existing file_lock structure.
229 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
231 new->fl_owner = fl->fl_owner;
232 new->fl_pid = fl->fl_pid;
234 new->fl_flags = fl->fl_flags;
235 new->fl_type = fl->fl_type;
236 new->fl_start = fl->fl_start;
237 new->fl_end = fl->fl_end;
239 new->fl_lmops = NULL;
241 EXPORT_SYMBOL(__locks_copy_lock);
243 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
245 locks_release_private(new);
247 __locks_copy_lock(new, fl);
248 new->fl_file = fl->fl_file;
249 new->fl_ops = fl->fl_ops;
250 new->fl_lmops = fl->fl_lmops;
252 locks_copy_private(new, fl);
255 EXPORT_SYMBOL(locks_copy_lock);
257 static inline int flock_translate_cmd(int cmd) {
259 return cmd & (LOCK_MAND | LOCK_RW);
271 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
272 static int flock_make_lock(struct file *filp, struct file_lock **lock,
275 struct file_lock *fl;
276 int type = flock_translate_cmd(cmd);
280 fl = locks_alloc_lock();
285 fl->fl_pid = current->tgid;
286 fl->fl_flags = FL_FLOCK;
288 fl->fl_end = OFFSET_MAX;
294 static int assign_type(struct file_lock *fl, int type)
308 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
311 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
316 switch (l->l_whence) {
324 start = i_size_read(filp->f_path.dentry->d_inode);
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
335 fl->fl_end = OFFSET_MAX;
337 end = start + l->l_len - 1;
339 } else if (l->l_len < 0) {
346 fl->fl_start = start; /* we record the absolute position */
347 if (fl->fl_end < fl->fl_start)
350 fl->fl_owner = current->files;
351 fl->fl_pid = current->tgid;
353 fl->fl_flags = FL_POSIX;
357 return assign_type(fl, l->l_type);
360 #if BITS_PER_LONG == 32
361 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
366 switch (l->l_whence) {
374 start = i_size_read(filp->f_path.dentry->d_inode);
383 fl->fl_end = OFFSET_MAX;
385 fl->fl_end = start + l->l_len - 1;
386 } else if (l->l_len < 0) {
387 fl->fl_end = start - 1;
392 fl->fl_start = start; /* we record the absolute position */
393 if (fl->fl_end < fl->fl_start)
396 fl->fl_owner = current->files;
397 fl->fl_pid = current->tgid;
399 fl->fl_flags = FL_POSIX;
407 fl->fl_type = l->l_type;
417 /* default lease lock manager operations */
418 static void lease_break_callback(struct file_lock *fl)
420 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
423 static void lease_release_private_callback(struct file_lock *fl)
428 f_delown(fl->fl_file);
429 fl->fl_file->f_owner.signum = 0;
432 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
434 return fl->fl_file == try->fl_file;
437 static struct lock_manager_operations lease_manager_ops = {
438 .fl_break = lease_break_callback,
439 .fl_release_private = lease_release_private_callback,
440 .fl_mylease = lease_mylease_callback,
441 .fl_change = lease_modify,
445 * Initialize a lease, use the default lock manager operations
447 static int lease_init(struct file *filp, int type, struct file_lock *fl)
449 if (assign_type(fl, type) != 0)
452 fl->fl_owner = current->files;
453 fl->fl_pid = current->tgid;
456 fl->fl_flags = FL_LEASE;
458 fl->fl_end = OFFSET_MAX;
460 fl->fl_lmops = &lease_manager_ops;
464 /* Allocate a file_lock initialised to this type of lease */
465 static struct file_lock *lease_alloc(struct file *filp, int type)
467 struct file_lock *fl = locks_alloc_lock();
471 return ERR_PTR(error);
473 error = lease_init(filp, type, fl);
476 return ERR_PTR(error);
481 /* Check if two locks overlap each other.
483 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
485 return ((fl1->fl_end >= fl2->fl_start) &&
486 (fl2->fl_end >= fl1->fl_start));
490 * Check whether two locks have the same owner.
492 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
494 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
495 return fl2->fl_lmops == fl1->fl_lmops &&
496 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
497 return fl1->fl_owner == fl2->fl_owner;
500 /* Remove waiter from blocker's block list.
501 * When blocker ends up pointing to itself then the list is empty.
503 static void __locks_delete_block(struct file_lock *waiter)
505 list_del_init(&waiter->fl_block);
506 list_del_init(&waiter->fl_link);
507 waiter->fl_next = NULL;
512 static void locks_delete_block(struct file_lock *waiter)
515 __locks_delete_block(waiter);
519 /* Insert waiter into blocker's block list.
520 * We use a circular list so that processes can be easily woken up in
521 * the order they blocked. The documentation doesn't require this but
522 * it seems like the reasonable thing to do.
524 static void locks_insert_block(struct file_lock *blocker,
525 struct file_lock *waiter)
527 BUG_ON(!list_empty(&waiter->fl_block));
528 list_add_tail(&waiter->fl_block, &blocker->fl_block);
529 waiter->fl_next = blocker;
530 if (IS_POSIX(blocker))
531 list_add(&waiter->fl_link, &blocked_list);
534 /* Wake up processes blocked waiting for blocker.
535 * If told to wait then schedule the processes until the block list
536 * is empty, otherwise empty the block list ourselves.
538 static void locks_wake_up_blocks(struct file_lock *blocker)
540 while (!list_empty(&blocker->fl_block)) {
541 struct file_lock *waiter;
543 waiter = list_first_entry(&blocker->fl_block,
544 struct file_lock, fl_block);
545 __locks_delete_block(waiter);
546 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
547 waiter->fl_lmops->fl_notify(waiter);
549 wake_up(&waiter->fl_wait);
553 /* Insert file lock fl into an inode's lock list at the position indicated
554 * by pos. At the same time add the lock to the global file lock list.
556 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
558 list_add(&fl->fl_link, &file_lock_list);
560 fl->fl_nspid = get_pid(task_tgid(current));
562 /* insert into file's list */
568 * Delete a lock and then free it.
569 * Wake up processes that are blocked waiting for this lock,
570 * notify the FS that the lock has been cleared and
571 * finally free the lock.
573 static void locks_delete_lock(struct file_lock **thisfl_p)
575 struct file_lock *fl = *thisfl_p;
577 *thisfl_p = fl->fl_next;
579 list_del_init(&fl->fl_link);
581 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
582 if (fl->fl_fasync != NULL) {
583 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
584 fl->fl_fasync = NULL;
588 put_pid(fl->fl_nspid);
592 locks_wake_up_blocks(fl);
596 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
597 * checks for shared/exclusive status of overlapping locks.
599 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
601 if (sys_fl->fl_type == F_WRLCK)
603 if (caller_fl->fl_type == F_WRLCK)
608 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
609 * checking before calling the locks_conflict().
611 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
613 /* POSIX locks owned by the same process do not conflict with
616 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
619 /* Check whether they overlap */
620 if (!locks_overlap(caller_fl, sys_fl))
623 return (locks_conflict(caller_fl, sys_fl));
626 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
627 * checking before calling the locks_conflict().
629 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
631 /* FLOCK locks referring to the same filp do not conflict with
634 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
636 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
639 return (locks_conflict(caller_fl, sys_fl));
643 posix_test_lock(struct file *filp, struct file_lock *fl)
645 struct file_lock *cfl;
648 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
651 if (posix_locks_conflict(fl, cfl))
655 __locks_copy_lock(fl, cfl);
657 fl->fl_pid = pid_vnr(cfl->fl_nspid);
659 fl->fl_type = F_UNLCK;
663 EXPORT_SYMBOL(posix_test_lock);
666 * Deadlock detection:
668 * We attempt to detect deadlocks that are due purely to posix file
671 * We assume that a task can be waiting for at most one lock at a time.
672 * So for any acquired lock, the process holding that lock may be
673 * waiting on at most one other lock. That lock in turns may be held by
674 * someone waiting for at most one other lock. Given a requested lock
675 * caller_fl which is about to wait for a conflicting lock block_fl, we
676 * follow this chain of waiters to ensure we are not about to create a
679 * Since we do this before we ever put a process to sleep on a lock, we
680 * are ensured that there is never a cycle; that is what guarantees that
681 * the while() loop in posix_locks_deadlock() eventually completes.
683 * Note: the above assumption may not be true when handling lock
684 * requests from a broken NFS client. It may also fail in the presence
685 * of tasks (such as posix threads) sharing the same open file table.
687 * To handle those cases, we just bail out after a few iterations.
690 #define MAX_DEADLK_ITERATIONS 10
692 /* Find a lock that the owner of the given block_fl is blocking on. */
693 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
695 struct file_lock *fl;
697 list_for_each_entry(fl, &blocked_list, fl_link) {
698 if (posix_same_owner(fl, block_fl))
704 static int posix_locks_deadlock(struct file_lock *caller_fl,
705 struct file_lock *block_fl)
709 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
710 if (i++ > MAX_DEADLK_ITERATIONS)
712 if (posix_same_owner(caller_fl, block_fl))
718 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
719 * after any leases, but before any posix locks.
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))
785 error = FILE_LOCK_DEFERRED;
786 locks_insert_block(fl, request);
789 if (request->fl_flags & FL_ACCESS)
791 locks_copy_lock(new_fl, request);
792 locks_insert_lock(before, new_fl);
799 locks_free_lock(new_fl);
803 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
805 struct file_lock *fl;
806 struct file_lock *new_fl = NULL;
807 struct file_lock *new_fl2 = NULL;
808 struct file_lock *left = NULL;
809 struct file_lock *right = NULL;
810 struct file_lock **before;
811 int error, added = 0;
814 * We may need two file_lock structures for this operation,
815 * so we get them in advance to avoid races.
817 * In some cases we can be sure, that no new locks will be needed
819 if (!(request->fl_flags & FL_ACCESS) &&
820 (request->fl_type != F_UNLCK ||
821 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
822 new_fl = locks_alloc_lock();
823 new_fl2 = locks_alloc_lock();
827 if (request->fl_type != F_UNLCK) {
828 for_each_lock(inode, before) {
832 if (!posix_locks_conflict(request, fl))
835 __locks_copy_lock(conflock, fl);
837 if (!(request->fl_flags & FL_SLEEP))
840 if (posix_locks_deadlock(request, fl))
842 error = FILE_LOCK_DEFERRED;
843 locks_insert_block(fl, request);
848 /* If we're just looking for a conflict, we're done. */
850 if (request->fl_flags & FL_ACCESS)
854 * Find the first old lock with the same owner as the new lock.
857 before = &inode->i_flock;
859 /* First skip locks owned by other processes. */
860 while ((fl = *before) && (!IS_POSIX(fl) ||
861 !posix_same_owner(request, fl))) {
862 before = &fl->fl_next;
865 /* Process locks with this owner. */
866 while ((fl = *before) && posix_same_owner(request, fl)) {
867 /* Detect adjacent or overlapping regions (if same lock type)
869 if (request->fl_type == fl->fl_type) {
870 /* In all comparisons of start vs end, use
871 * "start - 1" rather than "end + 1". If end
872 * is OFFSET_MAX, end + 1 will become negative.
874 if (fl->fl_end < request->fl_start - 1)
876 /* If the next lock in the list has entirely bigger
877 * addresses than the new one, insert the lock here.
879 if (fl->fl_start - 1 > request->fl_end)
882 /* If we come here, the new and old lock are of the
883 * same type and adjacent or overlapping. Make one
884 * lock yielding from the lower start address of both
885 * locks to the higher end address.
887 if (fl->fl_start > request->fl_start)
888 fl->fl_start = request->fl_start;
890 request->fl_start = fl->fl_start;
891 if (fl->fl_end < request->fl_end)
892 fl->fl_end = request->fl_end;
894 request->fl_end = fl->fl_end;
896 locks_delete_lock(before);
903 /* Processing for different lock types is a bit
906 if (fl->fl_end < request->fl_start)
908 if (fl->fl_start > request->fl_end)
910 if (request->fl_type == F_UNLCK)
912 if (fl->fl_start < request->fl_start)
914 /* If the next lock in the list has a higher end
915 * address than the new one, insert the new one here.
917 if (fl->fl_end > request->fl_end) {
921 if (fl->fl_start >= request->fl_start) {
922 /* The new lock completely replaces an old
923 * one (This may happen several times).
926 locks_delete_lock(before);
929 /* Replace the old lock with the new one.
930 * Wake up anybody waiting for the old one,
931 * as the change in lock type might satisfy
934 locks_wake_up_blocks(fl);
935 fl->fl_start = request->fl_start;
936 fl->fl_end = request->fl_end;
937 fl->fl_type = request->fl_type;
938 locks_release_private(fl);
939 locks_copy_private(fl, request);
944 /* Go on to next lock.
947 before = &fl->fl_next;
951 * The above code only modifies existing locks in case of
952 * merging or replacing. If new lock(s) need to be inserted
953 * all modifications are done bellow this, so it's safe yet to
956 error = -ENOLCK; /* "no luck" */
957 if (right && left == right && !new_fl2)
962 if (request->fl_type == F_UNLCK) {
963 if (request->fl_flags & FL_EXISTS)
972 locks_copy_lock(new_fl, request);
973 locks_insert_lock(before, new_fl);
978 /* The new lock breaks the old one in two pieces,
979 * so we have to use the second new lock.
983 locks_copy_lock(left, right);
984 locks_insert_lock(before, left);
986 right->fl_start = request->fl_end + 1;
987 locks_wake_up_blocks(right);
990 left->fl_end = request->fl_start - 1;
991 locks_wake_up_blocks(left);
996 * Free any unused locks.
999 locks_free_lock(new_fl);
1001 locks_free_lock(new_fl2);
1006 * posix_lock_file - Apply a POSIX-style lock to a file
1007 * @filp: The file to apply the lock to
1008 * @fl: The lock to be applied
1009 * @conflock: Place to return a copy of the conflicting lock, if found.
1011 * Add a POSIX style lock to a file.
1012 * We merge adjacent & overlapping locks whenever possible.
1013 * POSIX locks are sorted by owner task, then by starting address
1015 * Note that if called with an FL_EXISTS argument, the caller may determine
1016 * whether or not a lock was successfully freed by testing the return
1017 * value for -ENOENT.
1019 int posix_lock_file(struct file *filp, struct file_lock *fl,
1020 struct file_lock *conflock)
1022 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1024 EXPORT_SYMBOL(posix_lock_file);
1027 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1028 * @filp: The file to apply the lock to
1029 * @fl: The lock to be applied
1031 * Add a POSIX style lock to a file.
1032 * We merge adjacent & overlapping locks whenever possible.
1033 * POSIX locks are sorted by owner task, then by starting address
1035 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1040 error = posix_lock_file(filp, fl, NULL);
1041 if (error != FILE_LOCK_DEFERRED)
1043 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1047 locks_delete_block(fl);
1052 EXPORT_SYMBOL(posix_lock_file_wait);
1055 * locks_mandatory_locked - Check for an active lock
1056 * @inode: the file to check
1058 * Searches the inode's list of locks to find any POSIX locks which conflict.
1059 * This function is called from locks_verify_locked() only.
1061 int locks_mandatory_locked(struct inode *inode)
1063 fl_owner_t owner = current->files;
1064 struct file_lock *fl;
1067 * Search the lock list for this inode for any POSIX locks.
1070 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1073 if (fl->fl_owner != owner)
1077 return fl ? -EAGAIN : 0;
1081 * locks_mandatory_area - Check for a conflicting lock
1082 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1084 * @inode: the file to check
1085 * @filp: how the file was opened (if it was)
1086 * @offset: start of area to check
1087 * @count: length of area to check
1089 * Searches the inode's list of locks to find any POSIX locks which conflict.
1090 * This function is called from rw_verify_area() and
1091 * locks_verify_truncate().
1093 int locks_mandatory_area(int read_write, struct inode *inode,
1094 struct file *filp, loff_t offset,
1097 struct file_lock fl;
1100 locks_init_lock(&fl);
1101 fl.fl_owner = current->files;
1102 fl.fl_pid = current->tgid;
1104 fl.fl_flags = FL_POSIX | FL_ACCESS;
1105 if (filp && !(filp->f_flags & O_NONBLOCK))
1106 fl.fl_flags |= FL_SLEEP;
1107 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1108 fl.fl_start = offset;
1109 fl.fl_end = offset + count - 1;
1112 error = __posix_lock_file(inode, &fl, NULL);
1113 if (error != FILE_LOCK_DEFERRED)
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 (__mandatory_lock(inode))
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 locks_insert_block(flock, new_fl);
1249 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1250 !new_fl->fl_next, break_time);
1251 __locks_delete_block(new_fl);
1254 time_out_leases(inode);
1255 /* Wait for the next lease that has not been broken yet */
1256 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1257 flock = flock->fl_next) {
1258 if (flock->fl_type & F_INPROGRESS)
1266 if (!IS_ERR(new_fl))
1267 locks_free_lock(new_fl);
1271 EXPORT_SYMBOL(__break_lease);
1274 * lease_get_mtime - get the last modified time of an inode
1276 * @time: pointer to a timespec which will contain the last modified time
1278 * This is to force NFS clients to flush their caches for files with
1279 * exclusive leases. The justification is that if someone has an
1280 * exclusive lease, then they could be modifying it.
1282 void lease_get_mtime(struct inode *inode, struct timespec *time)
1284 struct file_lock *flock = inode->i_flock;
1285 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1286 *time = current_fs_time(inode->i_sb);
1288 *time = inode->i_mtime;
1291 EXPORT_SYMBOL(lease_get_mtime);
1294 * fcntl_getlease - Enquire what lease is currently active
1297 * The value returned by this function will be one of
1298 * (if no lease break is pending):
1300 * %F_RDLCK to indicate a shared lease is held.
1302 * %F_WRLCK to indicate an exclusive lease is held.
1304 * %F_UNLCK to indicate no lease is held.
1306 * (if a lease break is pending):
1308 * %F_RDLCK to indicate an exclusive lease needs to be
1309 * changed to a shared lease (or removed).
1311 * %F_UNLCK to indicate the lease needs to be removed.
1313 * XXX: sfr & willy disagree over whether F_INPROGRESS
1314 * should be returned to userspace.
1316 int fcntl_getlease(struct file *filp)
1318 struct file_lock *fl;
1322 time_out_leases(filp->f_path.dentry->d_inode);
1323 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1325 if (fl->fl_file == filp) {
1326 type = fl->fl_type & ~F_INPROGRESS;
1335 * generic_setlease - sets a lease on an open file
1336 * @filp: file pointer
1337 * @arg: type of lease to obtain
1338 * @flp: input - file_lock to use, output - file_lock inserted
1340 * The (input) flp->fl_lmops->fl_break function is required
1343 * Called with kernel lock held.
1345 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1347 struct file_lock *fl, **before, **my_before = NULL, *lease;
1348 struct file_lock *new_fl = NULL;
1349 struct dentry *dentry = filp->f_path.dentry;
1350 struct inode *inode = dentry->d_inode;
1351 int error, rdlease_count = 0, wrlease_count = 0;
1353 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
1355 if (!S_ISREG(inode->i_mode))
1357 error = security_file_lock(filp, arg);
1361 time_out_leases(inode);
1363 BUG_ON(!(*flp)->fl_lmops->fl_break);
1367 if (arg != F_UNLCK) {
1369 new_fl = locks_alloc_lock();
1374 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1376 if ((arg == F_WRLCK)
1377 && ((atomic_read(&dentry->d_count) > 1)
1378 || (atomic_read(&inode->i_count) > 1)))
1383 * At this point, we know that if there is an exclusive
1384 * lease on this file, then we hold it on this filp
1385 * (otherwise our open of this file would have blocked).
1386 * And if we are trying to acquire an exclusive lease,
1387 * then the file is not open by anyone (including us)
1388 * except for this filp.
1390 for (before = &inode->i_flock;
1391 ((fl = *before) != NULL) && IS_LEASE(fl);
1392 before = &fl->fl_next) {
1393 if (lease->fl_lmops->fl_mylease(fl, lease))
1395 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1397 * Someone is in the process of opening this
1398 * file for writing so we may not take an
1399 * exclusive lease on it.
1407 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1408 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1411 if (my_before != NULL) {
1413 error = lease->fl_lmops->fl_change(my_before, arg);
1425 locks_copy_lock(new_fl, lease);
1426 locks_insert_lock(before, new_fl);
1433 locks_free_lock(new_fl);
1436 EXPORT_SYMBOL(generic_setlease);
1439 * vfs_setlease - sets a lease on an open file
1440 * @filp: file pointer
1441 * @arg: type of lease to obtain
1442 * @lease: file_lock to use
1444 * Call this to establish a lease on the file.
1445 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1446 * break_lease will oops!
1448 * This will call the filesystem's setlease file method, if
1449 * defined. Note that there is no getlease method; instead, the
1450 * filesystem setlease method should call back to setlease() to
1451 * add a lease to the inode's lease list, where fcntl_getlease() can
1452 * find it. Since fcntl_getlease() only reports whether the current
1453 * task holds a lease, a cluster filesystem need only do this for
1454 * leases held by processes on this node.
1456 * There is also no break_lease method; filesystems that
1457 * handle their own leases shoud break leases themselves from the
1458 * filesystem's open, create, and (on truncate) setattr methods.
1460 * Warning: the only current setlease methods exist only to disable
1461 * leases in certain cases. More vfs changes may be required to
1462 * allow a full filesystem lease implementation.
1465 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1470 if (filp->f_op && filp->f_op->setlease)
1471 error = filp->f_op->setlease(filp, arg, lease);
1473 error = generic_setlease(filp, arg, lease);
1478 EXPORT_SYMBOL_GPL(vfs_setlease);
1481 * fcntl_setlease - sets a lease on an open file
1482 * @fd: open file descriptor
1483 * @filp: file pointer
1484 * @arg: type of lease to obtain
1486 * Call this fcntl to establish a lease on the file.
1487 * Note that you also need to call %F_SETSIG to
1488 * receive a signal when the lease is broken.
1490 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1492 struct file_lock fl, *flp = &fl;
1493 struct inode *inode = filp->f_path.dentry->d_inode;
1496 locks_init_lock(&fl);
1497 error = lease_init(filp, arg, &fl);
1503 error = vfs_setlease(filp, arg, &flp);
1504 if (error || arg == F_UNLCK)
1507 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1509 /* remove lease just inserted by setlease */
1510 flp->fl_type = F_UNLCK | F_INPROGRESS;
1511 flp->fl_break_time = jiffies - 10;
1512 time_out_leases(inode);
1516 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1523 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1524 * @filp: The file to apply the lock to
1525 * @fl: The lock to be applied
1527 * Add a FLOCK style lock to a file.
1529 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1534 error = flock_lock_file(filp, fl);
1535 if (error != FILE_LOCK_DEFERRED)
1537 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1541 locks_delete_block(fl);
1547 EXPORT_SYMBOL(flock_lock_file_wait);
1550 * sys_flock: - flock() system call.
1551 * @fd: the file descriptor to lock.
1552 * @cmd: the type of lock to apply.
1554 * Apply a %FL_FLOCK style lock to an open file descriptor.
1555 * The @cmd can be one of
1557 * %LOCK_SH -- a shared lock.
1559 * %LOCK_EX -- an exclusive lock.
1561 * %LOCK_UN -- remove an existing lock.
1563 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1565 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1566 * processes read and write access respectively.
1568 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1571 struct file_lock *lock;
1572 int can_sleep, unlock;
1580 can_sleep = !(cmd & LOCK_NB);
1582 unlock = (cmd == LOCK_UN);
1584 if (!unlock && !(cmd & LOCK_MAND) &&
1585 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
1588 error = flock_make_lock(filp, &lock, cmd);
1592 lock->fl_flags |= FL_SLEEP;
1594 error = security_file_lock(filp, lock->fl_type);
1598 if (filp->f_op && filp->f_op->flock)
1599 error = filp->f_op->flock(filp,
1600 (can_sleep) ? F_SETLKW : F_SETLK,
1603 error = flock_lock_file_wait(filp, lock);
1606 locks_free_lock(lock);
1615 * vfs_test_lock - test file byte range lock
1616 * @filp: The file to test lock for
1617 * @fl: The lock to test; also used to hold result
1619 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1620 * setting conf->fl_type to something other than F_UNLCK.
1622 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1624 if (filp->f_op && filp->f_op->lock)
1625 return filp->f_op->lock(filp, F_GETLK, fl);
1626 posix_test_lock(filp, fl);
1629 EXPORT_SYMBOL_GPL(vfs_test_lock);
1631 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1633 flock->l_pid = fl->fl_pid;
1634 #if BITS_PER_LONG == 32
1636 * Make sure we can represent the posix lock via
1637 * legacy 32bit flock.
1639 if (fl->fl_start > OFFT_OFFSET_MAX)
1641 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1644 flock->l_start = fl->fl_start;
1645 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1646 fl->fl_end - fl->fl_start + 1;
1647 flock->l_whence = 0;
1648 flock->l_type = fl->fl_type;
1652 #if BITS_PER_LONG == 32
1653 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1655 flock->l_pid = fl->fl_pid;
1656 flock->l_start = fl->fl_start;
1657 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1658 fl->fl_end - fl->fl_start + 1;
1659 flock->l_whence = 0;
1660 flock->l_type = fl->fl_type;
1664 /* Report the first existing lock that would conflict with l.
1665 * This implements the F_GETLK command of fcntl().
1667 int fcntl_getlk(struct file *filp, struct flock __user *l)
1669 struct file_lock file_lock;
1674 if (copy_from_user(&flock, l, sizeof(flock)))
1677 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1680 error = flock_to_posix_lock(filp, &file_lock, &flock);
1684 error = vfs_test_lock(filp, &file_lock);
1688 flock.l_type = file_lock.fl_type;
1689 if (file_lock.fl_type != F_UNLCK) {
1690 error = posix_lock_to_flock(&flock, &file_lock);
1695 if (!copy_to_user(l, &flock, sizeof(flock)))
1702 * vfs_lock_file - file byte range lock
1703 * @filp: The file to apply the lock to
1704 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1705 * @fl: The lock to be applied
1706 * @conf: Place to return a copy of the conflicting lock, if found.
1708 * A caller that doesn't care about the conflicting lock may pass NULL
1709 * as the final argument.
1711 * If the filesystem defines a private ->lock() method, then @conf will
1712 * be left unchanged; so a caller that cares should initialize it to
1713 * some acceptable default.
1715 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1716 * locks, the ->lock() interface may return asynchronously, before the lock has
1717 * been granted or denied by the underlying filesystem, if (and only if)
1718 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1719 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1720 * the request is for a blocking lock. When ->lock() does return asynchronously,
1721 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
1722 * request completes.
1723 * If the request is for non-blocking lock the file system should return
1724 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1725 * with the result. If the request timed out the callback routine will return a
1726 * nonzero return code and the file system should release the lock. The file
1727 * system is also responsible to keep a corresponding posix lock when it
1728 * grants a lock so the VFS can find out which locks are locally held and do
1729 * the correct lock cleanup when required.
1730 * The underlying filesystem must not drop the kernel lock or call
1731 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1734 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1736 if (filp->f_op && filp->f_op->lock)
1737 return filp->f_op->lock(filp, cmd, fl);
1739 return posix_lock_file(filp, fl, conf);
1741 EXPORT_SYMBOL_GPL(vfs_lock_file);
1743 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1744 struct file_lock *fl)
1748 error = security_file_lock(filp, fl->fl_type);
1753 error = vfs_lock_file(filp, cmd, fl, NULL);
1754 if (error != FILE_LOCK_DEFERRED)
1756 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1760 locks_delete_block(fl);
1767 /* Apply the lock described by l to an open file descriptor.
1768 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1770 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1771 struct flock __user *l)
1773 struct file_lock *file_lock = locks_alloc_lock();
1775 struct inode *inode;
1779 if (file_lock == NULL)
1783 * This might block, so we do it before checking the inode.
1786 if (copy_from_user(&flock, l, sizeof(flock)))
1789 inode = filp->f_path.dentry->d_inode;
1791 /* Don't allow mandatory locks on files that may be memory mapped
1794 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1800 error = flock_to_posix_lock(filp, file_lock, &flock);
1803 if (cmd == F_SETLKW) {
1804 file_lock->fl_flags |= FL_SLEEP;
1808 switch (flock.l_type) {
1810 if (!(filp->f_mode & FMODE_READ))
1814 if (!(filp->f_mode & FMODE_WRITE))
1824 error = do_lock_file_wait(filp, cmd, file_lock);
1827 * Attempt to detect a close/fcntl race and recover by
1828 * releasing the lock that was just acquired.
1831 * we need that spin_lock here - it prevents reordering between
1832 * update of inode->i_flock and check for it done in close().
1833 * rcu_read_lock() wouldn't do.
1835 spin_lock(¤t->files->file_lock);
1837 spin_unlock(¤t->files->file_lock);
1838 if (!error && f != filp && flock.l_type != F_UNLCK) {
1839 flock.l_type = F_UNLCK;
1844 locks_free_lock(file_lock);
1848 #if BITS_PER_LONG == 32
1849 /* Report the first existing lock that would conflict with l.
1850 * This implements the F_GETLK command of fcntl().
1852 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1854 struct file_lock file_lock;
1855 struct flock64 flock;
1859 if (copy_from_user(&flock, l, sizeof(flock)))
1862 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1865 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1869 error = vfs_test_lock(filp, &file_lock);
1873 flock.l_type = file_lock.fl_type;
1874 if (file_lock.fl_type != F_UNLCK)
1875 posix_lock_to_flock64(&flock, &file_lock);
1878 if (!copy_to_user(l, &flock, sizeof(flock)))
1885 /* Apply the lock described by l to an open file descriptor.
1886 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1888 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1889 struct flock64 __user *l)
1891 struct file_lock *file_lock = locks_alloc_lock();
1892 struct flock64 flock;
1893 struct inode *inode;
1897 if (file_lock == NULL)
1901 * This might block, so we do it before checking the inode.
1904 if (copy_from_user(&flock, l, sizeof(flock)))
1907 inode = filp->f_path.dentry->d_inode;
1909 /* Don't allow mandatory locks on files that may be memory mapped
1912 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1918 error = flock64_to_posix_lock(filp, file_lock, &flock);
1921 if (cmd == F_SETLKW64) {
1922 file_lock->fl_flags |= FL_SLEEP;
1926 switch (flock.l_type) {
1928 if (!(filp->f_mode & FMODE_READ))
1932 if (!(filp->f_mode & FMODE_WRITE))
1942 error = do_lock_file_wait(filp, cmd, file_lock);
1945 * Attempt to detect a close/fcntl race and recover by
1946 * releasing the lock that was just acquired.
1948 spin_lock(¤t->files->file_lock);
1950 spin_unlock(¤t->files->file_lock);
1951 if (!error && f != filp && flock.l_type != F_UNLCK) {
1952 flock.l_type = F_UNLCK;
1957 locks_free_lock(file_lock);
1960 #endif /* BITS_PER_LONG == 32 */
1963 * This function is called when the file is being removed
1964 * from the task's fd array. POSIX locks belonging to this task
1965 * are deleted at this time.
1967 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1969 struct file_lock lock;
1972 * If there are no locks held on this file, we don't need to call
1973 * posix_lock_file(). Another process could be setting a lock on this
1974 * file at the same time, but we wouldn't remove that lock anyway.
1976 if (!filp->f_path.dentry->d_inode->i_flock)
1979 lock.fl_type = F_UNLCK;
1980 lock.fl_flags = FL_POSIX | FL_CLOSE;
1982 lock.fl_end = OFFSET_MAX;
1983 lock.fl_owner = owner;
1984 lock.fl_pid = current->tgid;
1985 lock.fl_file = filp;
1987 lock.fl_lmops = NULL;
1989 vfs_lock_file(filp, F_SETLK, &lock, NULL);
1991 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1992 lock.fl_ops->fl_release_private(&lock);
1995 EXPORT_SYMBOL(locks_remove_posix);
1998 * This function is called on the last close of an open file.
2000 void locks_remove_flock(struct file *filp)
2002 struct inode * inode = filp->f_path.dentry->d_inode;
2003 struct file_lock *fl;
2004 struct file_lock **before;
2006 if (!inode->i_flock)
2009 if (filp->f_op && filp->f_op->flock) {
2010 struct file_lock fl = {
2011 .fl_pid = current->tgid,
2013 .fl_flags = FL_FLOCK,
2015 .fl_end = OFFSET_MAX,
2017 filp->f_op->flock(filp, F_SETLKW, &fl);
2018 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2019 fl.fl_ops->fl_release_private(&fl);
2023 before = &inode->i_flock;
2025 while ((fl = *before) != NULL) {
2026 if (fl->fl_file == filp) {
2028 locks_delete_lock(before);
2032 lease_modify(before, F_UNLCK);
2038 before = &fl->fl_next;
2044 * posix_unblock_lock - stop waiting for a file lock
2045 * @filp: how the file was opened
2046 * @waiter: the lock which was waiting
2048 * lockd needs to block waiting for locks.
2051 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2056 if (waiter->fl_next)
2057 __locks_delete_block(waiter);
2064 EXPORT_SYMBOL(posix_unblock_lock);
2067 * vfs_cancel_lock - file byte range unblock lock
2068 * @filp: The file to apply the unblock to
2069 * @fl: The lock to be unblocked
2071 * Used by lock managers to cancel blocked requests
2073 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2075 if (filp->f_op && filp->f_op->lock)
2076 return filp->f_op->lock(filp, F_CANCELLK, fl);
2080 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2082 #ifdef CONFIG_PROC_FS
2083 #include <linux/proc_fs.h>
2084 #include <linux/seq_file.h>
2086 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2089 struct inode *inode = NULL;
2090 unsigned int fl_pid;
2093 fl_pid = pid_vnr(fl->fl_nspid);
2095 fl_pid = fl->fl_pid;
2097 if (fl->fl_file != NULL)
2098 inode = fl->fl_file->f_path.dentry->d_inode;
2100 seq_printf(f, "%d:%s ", id, pfx);
2102 seq_printf(f, "%6s %s ",
2103 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2104 (inode == NULL) ? "*NOINODE*" :
2105 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2106 } else if (IS_FLOCK(fl)) {
2107 if (fl->fl_type & LOCK_MAND) {
2108 seq_printf(f, "FLOCK MSNFS ");
2110 seq_printf(f, "FLOCK ADVISORY ");
2112 } else if (IS_LEASE(fl)) {
2113 seq_printf(f, "LEASE ");
2114 if (fl->fl_type & F_INPROGRESS)
2115 seq_printf(f, "BREAKING ");
2116 else if (fl->fl_file)
2117 seq_printf(f, "ACTIVE ");
2119 seq_printf(f, "BREAKER ");
2121 seq_printf(f, "UNKNOWN UNKNOWN ");
2123 if (fl->fl_type & LOCK_MAND) {
2124 seq_printf(f, "%s ",
2125 (fl->fl_type & LOCK_READ)
2126 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2127 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2129 seq_printf(f, "%s ",
2130 (fl->fl_type & F_INPROGRESS)
2131 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2132 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2135 #ifdef WE_CAN_BREAK_LSLK_NOW
2136 seq_printf(f, "%d %s:%ld ", fl_pid,
2137 inode->i_sb->s_id, inode->i_ino);
2139 /* userspace relies on this representation of dev_t ;-( */
2140 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2141 MAJOR(inode->i_sb->s_dev),
2142 MINOR(inode->i_sb->s_dev), inode->i_ino);
2145 seq_printf(f, "%d <none>:0 ", fl_pid);
2148 if (fl->fl_end == OFFSET_MAX)
2149 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2151 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2153 seq_printf(f, "0 EOF\n");
2157 static int locks_show(struct seq_file *f, void *v)
2159 struct file_lock *fl, *bfl;
2161 fl = list_entry(v, struct file_lock, fl_link);
2163 lock_get_status(f, fl, (long)f->private, "");
2165 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2166 lock_get_status(f, bfl, (long)f->private, " ->");
2172 static void *locks_start(struct seq_file *f, loff_t *pos)
2175 f->private = (void *)1;
2176 return seq_list_start(&file_lock_list, *pos);
2179 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2181 return seq_list_next(v, &file_lock_list, pos);
2184 static void locks_stop(struct seq_file *f, void *v)
2189 static const struct seq_operations locks_seq_operations = {
2190 .start = locks_start,
2196 static int locks_open(struct inode *inode, struct file *filp)
2198 return seq_open(filp, &locks_seq_operations);
2201 static const struct file_operations proc_locks_operations = {
2204 .llseek = seq_lseek,
2205 .release = seq_release,
2208 static int __init proc_locks_init(void)
2210 proc_create("locks", 0, NULL, &proc_locks_operations);
2213 module_init(proc_locks_init);
2217 * lock_may_read - checks that the region is free of locks
2218 * @inode: the inode that is being read
2219 * @start: the first byte to read
2220 * @len: the number of bytes to read
2222 * Emulates Windows locking requirements. Whole-file
2223 * mandatory locks (share modes) can prohibit a read and
2224 * byte-range POSIX locks can prohibit a read if they overlap.
2226 * N.B. this function is only ever called
2227 * from knfsd and ownership of locks is never checked.
2229 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2231 struct file_lock *fl;
2234 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2236 if (fl->fl_type == F_RDLCK)
2238 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2240 } else if (IS_FLOCK(fl)) {
2241 if (!(fl->fl_type & LOCK_MAND))
2243 if (fl->fl_type & LOCK_READ)
2254 EXPORT_SYMBOL(lock_may_read);
2257 * lock_may_write - checks that the region is free of locks
2258 * @inode: the inode that is being written
2259 * @start: the first byte to write
2260 * @len: the number of bytes to write
2262 * Emulates Windows locking requirements. Whole-file
2263 * mandatory locks (share modes) can prohibit a write and
2264 * byte-range POSIX locks can prohibit a write if they overlap.
2266 * N.B. this function is only ever called
2267 * from knfsd and ownership of locks is never checked.
2269 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2271 struct file_lock *fl;
2274 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2276 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2278 } else if (IS_FLOCK(fl)) {
2279 if (!(fl->fl_type & LOCK_MAND))
2281 if (fl->fl_type & LOCK_WRITE)
2292 EXPORT_SYMBOL(lock_may_write);
2294 static int __init filelock_init(void)
2296 filelock_cache = kmem_cache_create("file_lock_cache",
2297 sizeof(struct file_lock), 0, SLAB_PANIC,
2302 core_initcall(filelock_init);