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);
145 static DEFINE_SPINLOCK(file_lock_lock);
148 * Protects the two list heads above, plus the inode->i_flock list
149 * FIXME: should use a spinlock, once lockd and ceph are ready.
151 void lock_flocks(void)
153 spin_lock(&file_lock_lock);
155 EXPORT_SYMBOL_GPL(lock_flocks);
157 void unlock_flocks(void)
159 spin_unlock(&file_lock_lock);
161 EXPORT_SYMBOL_GPL(unlock_flocks);
163 static struct kmem_cache *filelock_cache __read_mostly;
165 /* Allocate an empty lock structure. */
166 struct file_lock *locks_alloc_lock(void)
168 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
170 EXPORT_SYMBOL_GPL(locks_alloc_lock);
172 void locks_release_private(struct file_lock *fl)
175 if (fl->fl_ops->fl_release_private)
176 fl->fl_ops->fl_release_private(fl);
180 if (fl->fl_lmops->fl_release_private)
181 fl->fl_lmops->fl_release_private(fl);
186 EXPORT_SYMBOL_GPL(locks_release_private);
188 /* Free a lock which is not in use. */
189 void locks_free_lock(struct file_lock *fl)
191 BUG_ON(waitqueue_active(&fl->fl_wait));
192 BUG_ON(!list_empty(&fl->fl_block));
193 BUG_ON(!list_empty(&fl->fl_link));
195 locks_release_private(fl);
196 kmem_cache_free(filelock_cache, fl);
198 EXPORT_SYMBOL(locks_free_lock);
200 void locks_init_lock(struct file_lock *fl)
202 INIT_LIST_HEAD(&fl->fl_link);
203 INIT_LIST_HEAD(&fl->fl_block);
204 init_waitqueue_head(&fl->fl_wait);
206 fl->fl_fasync = NULL;
213 fl->fl_start = fl->fl_end = 0;
218 EXPORT_SYMBOL(locks_init_lock);
221 * Initialises the fields of the file lock which are invariant for
224 static void init_once(void *foo)
226 struct file_lock *lock = (struct file_lock *) foo;
228 locks_init_lock(lock);
231 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
234 if (fl->fl_ops->fl_copy_lock)
235 fl->fl_ops->fl_copy_lock(new, fl);
236 new->fl_ops = fl->fl_ops;
239 if (fl->fl_lmops->fl_copy_lock)
240 fl->fl_lmops->fl_copy_lock(new, fl);
241 new->fl_lmops = fl->fl_lmops;
246 * Initialize a new lock from an existing file_lock structure.
248 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
250 new->fl_owner = fl->fl_owner;
251 new->fl_pid = fl->fl_pid;
253 new->fl_flags = fl->fl_flags;
254 new->fl_type = fl->fl_type;
255 new->fl_start = fl->fl_start;
256 new->fl_end = fl->fl_end;
258 new->fl_lmops = NULL;
260 EXPORT_SYMBOL(__locks_copy_lock);
262 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
264 locks_release_private(new);
266 __locks_copy_lock(new, fl);
267 new->fl_file = fl->fl_file;
268 new->fl_ops = fl->fl_ops;
269 new->fl_lmops = fl->fl_lmops;
271 locks_copy_private(new, fl);
274 EXPORT_SYMBOL(locks_copy_lock);
276 static inline int flock_translate_cmd(int cmd) {
278 return cmd & (LOCK_MAND | LOCK_RW);
290 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
291 static int flock_make_lock(struct file *filp, struct file_lock **lock,
294 struct file_lock *fl;
295 int type = flock_translate_cmd(cmd);
299 fl = locks_alloc_lock();
304 fl->fl_pid = current->tgid;
305 fl->fl_flags = FL_FLOCK;
307 fl->fl_end = OFFSET_MAX;
313 static int assign_type(struct file_lock *fl, int type)
327 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
330 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
335 switch (l->l_whence) {
343 start = i_size_read(filp->f_path.dentry->d_inode);
349 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
350 POSIX-2001 defines it. */
354 fl->fl_end = OFFSET_MAX;
356 end = start + l->l_len - 1;
358 } else if (l->l_len < 0) {
365 fl->fl_start = start; /* we record the absolute position */
366 if (fl->fl_end < fl->fl_start)
369 fl->fl_owner = current->files;
370 fl->fl_pid = current->tgid;
372 fl->fl_flags = FL_POSIX;
376 return assign_type(fl, l->l_type);
379 #if BITS_PER_LONG == 32
380 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
385 switch (l->l_whence) {
393 start = i_size_read(filp->f_path.dentry->d_inode);
402 fl->fl_end = OFFSET_MAX;
404 fl->fl_end = start + l->l_len - 1;
405 } else if (l->l_len < 0) {
406 fl->fl_end = start - 1;
411 fl->fl_start = start; /* we record the absolute position */
412 if (fl->fl_end < fl->fl_start)
415 fl->fl_owner = current->files;
416 fl->fl_pid = current->tgid;
418 fl->fl_flags = FL_POSIX;
426 fl->fl_type = l->l_type;
436 /* default lease lock manager operations */
437 static void lease_break_callback(struct file_lock *fl)
439 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
442 static void lease_release_private_callback(struct file_lock *fl)
447 f_delown(fl->fl_file);
448 fl->fl_file->f_owner.signum = 0;
451 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
453 return fl->fl_file == try->fl_file;
456 static const struct lock_manager_operations lease_manager_ops = {
457 .fl_break = lease_break_callback,
458 .fl_release_private = lease_release_private_callback,
459 .fl_mylease = lease_mylease_callback,
460 .fl_change = lease_modify,
464 * Initialize a lease, use the default lock manager operations
466 static int lease_init(struct file *filp, int type, struct file_lock *fl)
468 if (assign_type(fl, type) != 0)
471 fl->fl_owner = current->files;
472 fl->fl_pid = current->tgid;
475 fl->fl_flags = FL_LEASE;
477 fl->fl_end = OFFSET_MAX;
479 fl->fl_lmops = &lease_manager_ops;
483 /* Allocate a file_lock initialised to this type of lease */
484 static struct file_lock *lease_alloc(struct file *filp, int type)
486 struct file_lock *fl = locks_alloc_lock();
490 return ERR_PTR(error);
492 error = lease_init(filp, type, fl);
495 return ERR_PTR(error);
500 /* Check if two locks overlap each other.
502 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
504 return ((fl1->fl_end >= fl2->fl_start) &&
505 (fl2->fl_end >= fl1->fl_start));
509 * Check whether two locks have the same owner.
511 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
513 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
514 return fl2->fl_lmops == fl1->fl_lmops &&
515 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
516 return fl1->fl_owner == fl2->fl_owner;
519 /* Remove waiter from blocker's block list.
520 * When blocker ends up pointing to itself then the list is empty.
522 static void __locks_delete_block(struct file_lock *waiter)
524 list_del_init(&waiter->fl_block);
525 list_del_init(&waiter->fl_link);
526 waiter->fl_next = NULL;
531 static void locks_delete_block(struct file_lock *waiter)
534 __locks_delete_block(waiter);
538 /* Insert waiter into blocker's block list.
539 * We use a circular list so that processes can be easily woken up in
540 * the order they blocked. The documentation doesn't require this but
541 * it seems like the reasonable thing to do.
543 static void locks_insert_block(struct file_lock *blocker,
544 struct file_lock *waiter)
546 BUG_ON(!list_empty(&waiter->fl_block));
547 list_add_tail(&waiter->fl_block, &blocker->fl_block);
548 waiter->fl_next = blocker;
549 if (IS_POSIX(blocker))
550 list_add(&waiter->fl_link, &blocked_list);
553 /* Wake up processes blocked waiting for blocker.
554 * If told to wait then schedule the processes until the block list
555 * is empty, otherwise empty the block list ourselves.
557 static void locks_wake_up_blocks(struct file_lock *blocker)
559 while (!list_empty(&blocker->fl_block)) {
560 struct file_lock *waiter;
562 waiter = list_first_entry(&blocker->fl_block,
563 struct file_lock, fl_block);
564 __locks_delete_block(waiter);
565 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
566 waiter->fl_lmops->fl_notify(waiter);
568 wake_up(&waiter->fl_wait);
572 /* Insert file lock fl into an inode's lock list at the position indicated
573 * by pos. At the same time add the lock to the global file lock list.
575 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
577 list_add(&fl->fl_link, &file_lock_list);
579 fl->fl_nspid = get_pid(task_tgid(current));
581 /* insert into file's list */
587 * Delete a lock and then free it.
588 * Wake up processes that are blocked waiting for this lock,
589 * notify the FS that the lock has been cleared and
590 * finally free the lock.
592 static void locks_delete_lock(struct file_lock **thisfl_p)
594 struct file_lock *fl = *thisfl_p;
596 *thisfl_p = fl->fl_next;
598 list_del_init(&fl->fl_link);
600 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
601 if (fl->fl_fasync != NULL) {
602 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
603 fl->fl_fasync = NULL;
607 put_pid(fl->fl_nspid);
611 locks_wake_up_blocks(fl);
615 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
616 * checks for shared/exclusive status of overlapping locks.
618 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
620 if (sys_fl->fl_type == F_WRLCK)
622 if (caller_fl->fl_type == F_WRLCK)
627 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
628 * checking before calling the locks_conflict().
630 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
632 /* POSIX locks owned by the same process do not conflict with
635 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
638 /* Check whether they overlap */
639 if (!locks_overlap(caller_fl, sys_fl))
642 return (locks_conflict(caller_fl, sys_fl));
645 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
646 * checking before calling the locks_conflict().
648 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
650 /* FLOCK locks referring to the same filp do not conflict with
653 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
655 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
658 return (locks_conflict(caller_fl, sys_fl));
662 posix_test_lock(struct file *filp, struct file_lock *fl)
664 struct file_lock *cfl;
667 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
670 if (posix_locks_conflict(fl, cfl))
674 __locks_copy_lock(fl, cfl);
676 fl->fl_pid = pid_vnr(cfl->fl_nspid);
678 fl->fl_type = F_UNLCK;
682 EXPORT_SYMBOL(posix_test_lock);
685 * Deadlock detection:
687 * We attempt to detect deadlocks that are due purely to posix file
690 * We assume that a task can be waiting for at most one lock at a time.
691 * So for any acquired lock, the process holding that lock may be
692 * waiting on at most one other lock. That lock in turns may be held by
693 * someone waiting for at most one other lock. Given a requested lock
694 * caller_fl which is about to wait for a conflicting lock block_fl, we
695 * follow this chain of waiters to ensure we are not about to create a
698 * Since we do this before we ever put a process to sleep on a lock, we
699 * are ensured that there is never a cycle; that is what guarantees that
700 * the while() loop in posix_locks_deadlock() eventually completes.
702 * Note: the above assumption may not be true when handling lock
703 * requests from a broken NFS client. It may also fail in the presence
704 * of tasks (such as posix threads) sharing the same open file table.
706 * To handle those cases, we just bail out after a few iterations.
709 #define MAX_DEADLK_ITERATIONS 10
711 /* Find a lock that the owner of the given block_fl is blocking on. */
712 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
714 struct file_lock *fl;
716 list_for_each_entry(fl, &blocked_list, fl_link) {
717 if (posix_same_owner(fl, block_fl))
723 static int posix_locks_deadlock(struct file_lock *caller_fl,
724 struct file_lock *block_fl)
728 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
729 if (i++ > MAX_DEADLK_ITERATIONS)
731 if (posix_same_owner(caller_fl, block_fl))
737 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
738 * after any leases, but before any posix locks.
740 * Note that if called with an FL_EXISTS argument, the caller may determine
741 * whether or not a lock was successfully freed by testing the return
744 static int flock_lock_file(struct file *filp, struct file_lock *request)
746 struct file_lock *new_fl = NULL;
747 struct file_lock **before;
748 struct inode * inode = filp->f_path.dentry->d_inode;
752 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
753 new_fl = locks_alloc_lock();
759 if (request->fl_flags & FL_ACCESS)
762 for_each_lock(inode, before) {
763 struct file_lock *fl = *before;
768 if (filp != fl->fl_file)
770 if (request->fl_type == fl->fl_type)
773 locks_delete_lock(before);
777 if (request->fl_type == F_UNLCK) {
778 if ((request->fl_flags & FL_EXISTS) && !found)
784 * If a higher-priority process was blocked on the old file lock,
785 * give it the opportunity to lock the file.
794 for_each_lock(inode, before) {
795 struct file_lock *fl = *before;
800 if (!flock_locks_conflict(request, fl))
803 if (!(request->fl_flags & FL_SLEEP))
805 error = FILE_LOCK_DEFERRED;
806 locks_insert_block(fl, request);
809 if (request->fl_flags & FL_ACCESS)
811 locks_copy_lock(new_fl, request);
812 locks_insert_lock(before, new_fl);
819 locks_free_lock(new_fl);
823 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
825 struct file_lock *fl;
826 struct file_lock *new_fl = NULL;
827 struct file_lock *new_fl2 = NULL;
828 struct file_lock *left = NULL;
829 struct file_lock *right = NULL;
830 struct file_lock **before;
831 int error, added = 0;
834 * We may need two file_lock structures for this operation,
835 * so we get them in advance to avoid races.
837 * In some cases we can be sure, that no new locks will be needed
839 if (!(request->fl_flags & FL_ACCESS) &&
840 (request->fl_type != F_UNLCK ||
841 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
842 new_fl = locks_alloc_lock();
843 new_fl2 = locks_alloc_lock();
847 if (request->fl_type != F_UNLCK) {
848 for_each_lock(inode, 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))
862 error = FILE_LOCK_DEFERRED;
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
1029 * @conflock: Place to return a copy of the conflicting lock, if found.
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 * Note that if called with an FL_EXISTS argument, the caller may determine
1036 * whether or not a lock was successfully freed by testing the return
1037 * value for -ENOENT.
1039 int posix_lock_file(struct file *filp, struct file_lock *fl,
1040 struct file_lock *conflock)
1042 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1044 EXPORT_SYMBOL(posix_lock_file);
1047 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1048 * @filp: The file to apply the lock to
1049 * @fl: The lock to be applied
1051 * Add a POSIX style lock to a file.
1052 * We merge adjacent & overlapping locks whenever possible.
1053 * POSIX locks are sorted by owner task, then by starting address
1055 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1060 error = posix_lock_file(filp, fl, NULL);
1061 if (error != FILE_LOCK_DEFERRED)
1063 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1067 locks_delete_block(fl);
1072 EXPORT_SYMBOL(posix_lock_file_wait);
1075 * locks_mandatory_locked - Check for an active lock
1076 * @inode: the file to check
1078 * Searches the inode's list of locks to find any POSIX locks which conflict.
1079 * This function is called from locks_verify_locked() only.
1081 int locks_mandatory_locked(struct inode *inode)
1083 fl_owner_t owner = current->files;
1084 struct file_lock *fl;
1087 * Search the lock list for this inode for any POSIX locks.
1090 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1093 if (fl->fl_owner != owner)
1097 return fl ? -EAGAIN : 0;
1101 * locks_mandatory_area - Check for a conflicting lock
1102 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1104 * @inode: the file to check
1105 * @filp: how the file was opened (if it was)
1106 * @offset: start of area to check
1107 * @count: length of area to check
1109 * Searches the inode's list of locks to find any POSIX locks which conflict.
1110 * This function is called from rw_verify_area() and
1111 * locks_verify_truncate().
1113 int locks_mandatory_area(int read_write, struct inode *inode,
1114 struct file *filp, loff_t offset,
1117 struct file_lock fl;
1120 locks_init_lock(&fl);
1121 fl.fl_owner = current->files;
1122 fl.fl_pid = current->tgid;
1124 fl.fl_flags = FL_POSIX | FL_ACCESS;
1125 if (filp && !(filp->f_flags & O_NONBLOCK))
1126 fl.fl_flags |= FL_SLEEP;
1127 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1128 fl.fl_start = offset;
1129 fl.fl_end = offset + count - 1;
1132 error = __posix_lock_file(inode, &fl, NULL);
1133 if (error != FILE_LOCK_DEFERRED)
1135 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1138 * If we've been sleeping someone might have
1139 * changed the permissions behind our back.
1141 if (__mandatory_lock(inode))
1145 locks_delete_block(&fl);
1152 EXPORT_SYMBOL(locks_mandatory_area);
1154 /* We already had a lease on this file; just change its type */
1155 int lease_modify(struct file_lock **before, int arg)
1157 struct file_lock *fl = *before;
1158 int error = assign_type(fl, arg);
1162 locks_wake_up_blocks(fl);
1164 locks_delete_lock(before);
1168 EXPORT_SYMBOL(lease_modify);
1170 static void time_out_leases(struct inode *inode)
1172 struct file_lock **before;
1173 struct file_lock *fl;
1175 before = &inode->i_flock;
1176 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1177 if ((fl->fl_break_time == 0)
1178 || time_before(jiffies, fl->fl_break_time)) {
1179 before = &fl->fl_next;
1182 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1183 if (fl == *before) /* lease_modify may have freed fl */
1184 before = &fl->fl_next;
1189 * __break_lease - revoke all outstanding leases on file
1190 * @inode: the inode of the file to return
1191 * @mode: the open mode (read or write)
1193 * break_lease (inlined for speed) has checked there already is at least
1194 * some kind of lock (maybe a lease) on this file. Leases are broken on
1195 * a call to open() or truncate(). This function can sleep unless you
1196 * specified %O_NONBLOCK to your open().
1198 int __break_lease(struct inode *inode, unsigned int mode)
1200 int error = 0, future;
1201 struct file_lock *new_fl, *flock;
1202 struct file_lock *fl;
1203 unsigned long break_time;
1204 int i_have_this_lease = 0;
1205 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1207 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1211 time_out_leases(inode);
1213 flock = inode->i_flock;
1214 if ((flock == NULL) || !IS_LEASE(flock))
1217 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1218 if (fl->fl_owner == current->files)
1219 i_have_this_lease = 1;
1222 /* If we want write access, we have to revoke any lease. */
1223 future = F_UNLCK | F_INPROGRESS;
1224 } else if (flock->fl_type & F_INPROGRESS) {
1225 /* If the lease is already being broken, we just leave it */
1226 future = flock->fl_type;
1227 } else if (flock->fl_type & F_WRLCK) {
1228 /* Downgrade the exclusive lease to a read-only lease. */
1229 future = F_RDLCK | F_INPROGRESS;
1231 /* the existing lease was read-only, so we can read too. */
1235 if (IS_ERR(new_fl) && !i_have_this_lease
1236 && ((mode & O_NONBLOCK) == 0)) {
1237 error = PTR_ERR(new_fl);
1242 if (lease_break_time > 0) {
1243 break_time = jiffies + lease_break_time * HZ;
1244 if (break_time == 0)
1245 break_time++; /* so that 0 means no break time */
1248 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1249 if (fl->fl_type != future) {
1250 fl->fl_type = future;
1251 fl->fl_break_time = break_time;
1252 /* lease must have lmops break callback */
1253 fl->fl_lmops->fl_break(fl);
1257 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1258 error = -EWOULDBLOCK;
1263 break_time = flock->fl_break_time;
1264 if (break_time != 0) {
1265 break_time -= jiffies;
1266 if (break_time == 0)
1269 locks_insert_block(flock, new_fl);
1271 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1272 !new_fl->fl_next, break_time);
1274 __locks_delete_block(new_fl);
1277 time_out_leases(inode);
1278 /* Wait for the next lease that has not been broken yet */
1279 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1280 flock = flock->fl_next) {
1281 if (flock->fl_type & F_INPROGRESS)
1289 if (!IS_ERR(new_fl))
1290 locks_free_lock(new_fl);
1294 EXPORT_SYMBOL(__break_lease);
1297 * lease_get_mtime - get the last modified time of an inode
1299 * @time: pointer to a timespec which will contain the last modified time
1301 * This is to force NFS clients to flush their caches for files with
1302 * exclusive leases. The justification is that if someone has an
1303 * exclusive lease, then they could be modifying it.
1305 void lease_get_mtime(struct inode *inode, struct timespec *time)
1307 struct file_lock *flock = inode->i_flock;
1308 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1309 *time = current_fs_time(inode->i_sb);
1311 *time = inode->i_mtime;
1314 EXPORT_SYMBOL(lease_get_mtime);
1317 * fcntl_getlease - Enquire what lease is currently active
1320 * The value returned by this function will be one of
1321 * (if no lease break is pending):
1323 * %F_RDLCK to indicate a shared lease is held.
1325 * %F_WRLCK to indicate an exclusive lease is held.
1327 * %F_UNLCK to indicate no lease is held.
1329 * (if a lease break is pending):
1331 * %F_RDLCK to indicate an exclusive lease needs to be
1332 * changed to a shared lease (or removed).
1334 * %F_UNLCK to indicate the lease needs to be removed.
1336 * XXX: sfr & willy disagree over whether F_INPROGRESS
1337 * should be returned to userspace.
1339 int fcntl_getlease(struct file *filp)
1341 struct file_lock *fl;
1345 time_out_leases(filp->f_path.dentry->d_inode);
1346 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1348 if (fl->fl_file == filp) {
1349 type = fl->fl_type & ~F_INPROGRESS;
1358 * generic_setlease - sets a lease on an open file
1359 * @filp: file pointer
1360 * @arg: type of lease to obtain
1361 * @flp: input - file_lock to use, output - file_lock inserted
1363 * The (input) flp->fl_lmops->fl_break function is required
1366 * Called with file_lock_lock held.
1368 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1370 struct file_lock *fl, **before, **my_before = NULL, *lease;
1371 struct dentry *dentry = filp->f_path.dentry;
1372 struct inode *inode = dentry->d_inode;
1373 int error, rdlease_count = 0, wrlease_count = 0;
1378 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
1381 if (!S_ISREG(inode->i_mode))
1383 error = security_file_lock(filp, arg);
1387 time_out_leases(inode);
1389 BUG_ON(!(*flp)->fl_lmops->fl_break);
1391 if (arg != F_UNLCK) {
1393 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1395 if ((arg == F_WRLCK)
1396 && ((atomic_read(&dentry->d_count) > 1)
1397 || (atomic_read(&inode->i_count) > 1)))
1402 * At this point, we know that if there is an exclusive
1403 * lease on this file, then we hold it on this filp
1404 * (otherwise our open of this file would have blocked).
1405 * And if we are trying to acquire an exclusive lease,
1406 * then the file is not open by anyone (including us)
1407 * except for this filp.
1409 for (before = &inode->i_flock;
1410 ((fl = *before) != NULL) && IS_LEASE(fl);
1411 before = &fl->fl_next) {
1412 if (lease->fl_lmops->fl_mylease(fl, lease))
1414 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1416 * Someone is in the process of opening this
1417 * file for writing so we may not take an
1418 * exclusive lease on it.
1426 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1427 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1430 if (my_before != NULL) {
1431 error = lease->fl_lmops->fl_change(my_before, arg);
1444 locks_insert_lock(before, lease);
1450 EXPORT_SYMBOL(generic_setlease);
1452 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1454 if (filp->f_op && filp->f_op->setlease)
1455 return filp->f_op->setlease(filp, arg, lease);
1457 return generic_setlease(filp, arg, lease);
1461 * vfs_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 (*lease)->fl_lmops->fl_break operation must be set; if not,
1468 * break_lease will oops!
1470 * This will call the filesystem's setlease file method, if
1471 * defined. Note that there is no getlease method; instead, the
1472 * filesystem setlease method should call back to setlease() to
1473 * add a lease to the inode's lease list, where fcntl_getlease() can
1474 * find it. Since fcntl_getlease() only reports whether the current
1475 * task holds a lease, a cluster filesystem need only do this for
1476 * leases held by processes on this node.
1478 * There is also no break_lease method; filesystems that
1479 * handle their own leases should break leases themselves from the
1480 * filesystem's open, create, and (on truncate) setattr methods.
1482 * Warning: the only current setlease methods exist only to disable
1483 * leases in certain cases. More vfs changes may be required to
1484 * allow a full filesystem lease implementation.
1487 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1492 error = __vfs_setlease(filp, arg, lease);
1497 EXPORT_SYMBOL_GPL(vfs_setlease);
1499 static int do_fcntl_delete_lease(struct file *filp)
1501 struct file_lock fl, *flp = &fl;
1503 lease_init(filp, F_UNLCK, flp);
1505 return vfs_setlease(filp, F_UNLCK, &flp);
1508 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1510 struct file_lock *fl;
1511 struct fasync_struct *new;
1512 struct inode *inode = filp->f_path.dentry->d_inode;
1515 fl = lease_alloc(filp, arg);
1519 new = fasync_alloc();
1521 locks_free_lock(fl);
1525 error = __vfs_setlease(filp, arg, &fl);
1528 locks_free_lock(fl);
1529 goto out_free_fasync;
1533 * fasync_insert_entry() returns the old entry if any.
1534 * If there was no old entry, then it used 'new' and
1535 * inserted it into the fasync list. Clear new so that
1536 * we don't release it here.
1538 if (!fasync_insert_entry(fd, filp, &fl->fl_fasync, new))
1542 /* remove lease just inserted by setlease */
1543 fl->fl_type = F_UNLCK | F_INPROGRESS;
1544 fl->fl_break_time = jiffies - 10;
1545 time_out_leases(inode);
1547 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1558 * fcntl_setlease - sets a lease on an open file
1559 * @fd: open file descriptor
1560 * @filp: file pointer
1561 * @arg: type of lease to obtain
1563 * Call this fcntl to establish a lease on the file.
1564 * Note that you also need to call %F_SETSIG to
1565 * receive a signal when the lease is broken.
1567 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1570 return do_fcntl_delete_lease(filp);
1571 return do_fcntl_add_lease(fd, filp, arg);
1575 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1576 * @filp: The file to apply the lock to
1577 * @fl: The lock to be applied
1579 * Add a FLOCK style lock to a file.
1581 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1586 error = flock_lock_file(filp, fl);
1587 if (error != FILE_LOCK_DEFERRED)
1589 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1593 locks_delete_block(fl);
1599 EXPORT_SYMBOL(flock_lock_file_wait);
1602 * sys_flock: - flock() system call.
1603 * @fd: the file descriptor to lock.
1604 * @cmd: the type of lock to apply.
1606 * Apply a %FL_FLOCK style lock to an open file descriptor.
1607 * The @cmd can be one of
1609 * %LOCK_SH -- a shared lock.
1611 * %LOCK_EX -- an exclusive lock.
1613 * %LOCK_UN -- remove an existing lock.
1615 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1617 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1618 * processes read and write access respectively.
1620 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1623 struct file_lock *lock;
1624 int can_sleep, unlock;
1632 can_sleep = !(cmd & LOCK_NB);
1634 unlock = (cmd == LOCK_UN);
1636 if (!unlock && !(cmd & LOCK_MAND) &&
1637 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
1640 error = flock_make_lock(filp, &lock, cmd);
1644 lock->fl_flags |= FL_SLEEP;
1646 error = security_file_lock(filp, lock->fl_type);
1650 if (filp->f_op && filp->f_op->flock)
1651 error = filp->f_op->flock(filp,
1652 (can_sleep) ? F_SETLKW : F_SETLK,
1655 error = flock_lock_file_wait(filp, lock);
1658 locks_free_lock(lock);
1667 * vfs_test_lock - test file byte range lock
1668 * @filp: The file to test lock for
1669 * @fl: The lock to test; also used to hold result
1671 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1672 * setting conf->fl_type to something other than F_UNLCK.
1674 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1676 if (filp->f_op && filp->f_op->lock)
1677 return filp->f_op->lock(filp, F_GETLK, fl);
1678 posix_test_lock(filp, fl);
1681 EXPORT_SYMBOL_GPL(vfs_test_lock);
1683 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1685 flock->l_pid = fl->fl_pid;
1686 #if BITS_PER_LONG == 32
1688 * Make sure we can represent the posix lock via
1689 * legacy 32bit flock.
1691 if (fl->fl_start > OFFT_OFFSET_MAX)
1693 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1696 flock->l_start = fl->fl_start;
1697 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1698 fl->fl_end - fl->fl_start + 1;
1699 flock->l_whence = 0;
1700 flock->l_type = fl->fl_type;
1704 #if BITS_PER_LONG == 32
1705 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1707 flock->l_pid = fl->fl_pid;
1708 flock->l_start = fl->fl_start;
1709 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1710 fl->fl_end - fl->fl_start + 1;
1711 flock->l_whence = 0;
1712 flock->l_type = fl->fl_type;
1716 /* Report the first existing lock that would conflict with l.
1717 * This implements the F_GETLK command of fcntl().
1719 int fcntl_getlk(struct file *filp, struct flock __user *l)
1721 struct file_lock file_lock;
1726 if (copy_from_user(&flock, l, sizeof(flock)))
1729 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1732 error = flock_to_posix_lock(filp, &file_lock, &flock);
1736 error = vfs_test_lock(filp, &file_lock);
1740 flock.l_type = file_lock.fl_type;
1741 if (file_lock.fl_type != F_UNLCK) {
1742 error = posix_lock_to_flock(&flock, &file_lock);
1747 if (!copy_to_user(l, &flock, sizeof(flock)))
1754 * vfs_lock_file - file byte range lock
1755 * @filp: The file to apply the lock to
1756 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1757 * @fl: The lock to be applied
1758 * @conf: Place to return a copy of the conflicting lock, if found.
1760 * A caller that doesn't care about the conflicting lock may pass NULL
1761 * as the final argument.
1763 * If the filesystem defines a private ->lock() method, then @conf will
1764 * be left unchanged; so a caller that cares should initialize it to
1765 * some acceptable default.
1767 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1768 * locks, the ->lock() interface may return asynchronously, before the lock has
1769 * been granted or denied by the underlying filesystem, if (and only if)
1770 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1771 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1772 * the request is for a blocking lock. When ->lock() does return asynchronously,
1773 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
1774 * request completes.
1775 * If the request is for non-blocking lock the file system should return
1776 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1777 * with the result. If the request timed out the callback routine will return a
1778 * nonzero return code and the file system should release the lock. The file
1779 * system is also responsible to keep a corresponding posix lock when it
1780 * grants a lock so the VFS can find out which locks are locally held and do
1781 * the correct lock cleanup when required.
1782 * The underlying filesystem must not drop the kernel lock or call
1783 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1786 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1788 if (filp->f_op && filp->f_op->lock)
1789 return filp->f_op->lock(filp, cmd, fl);
1791 return posix_lock_file(filp, fl, conf);
1793 EXPORT_SYMBOL_GPL(vfs_lock_file);
1795 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1796 struct file_lock *fl)
1800 error = security_file_lock(filp, fl->fl_type);
1805 error = vfs_lock_file(filp, cmd, fl, NULL);
1806 if (error != FILE_LOCK_DEFERRED)
1808 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1812 locks_delete_block(fl);
1819 /* Apply the lock described by l to an open file descriptor.
1820 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1822 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1823 struct flock __user *l)
1825 struct file_lock *file_lock = locks_alloc_lock();
1827 struct inode *inode;
1831 if (file_lock == NULL)
1835 * This might block, so we do it before checking the inode.
1838 if (copy_from_user(&flock, l, sizeof(flock)))
1841 inode = filp->f_path.dentry->d_inode;
1843 /* Don't allow mandatory locks on files that may be memory mapped
1846 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1852 error = flock_to_posix_lock(filp, file_lock, &flock);
1855 if (cmd == F_SETLKW) {
1856 file_lock->fl_flags |= FL_SLEEP;
1860 switch (flock.l_type) {
1862 if (!(filp->f_mode & FMODE_READ))
1866 if (!(filp->f_mode & FMODE_WRITE))
1876 error = do_lock_file_wait(filp, cmd, file_lock);
1879 * Attempt to detect a close/fcntl race and recover by
1880 * releasing the lock that was just acquired.
1883 * we need that spin_lock here - it prevents reordering between
1884 * update of inode->i_flock and check for it done in close().
1885 * rcu_read_lock() wouldn't do.
1887 spin_lock(¤t->files->file_lock);
1889 spin_unlock(¤t->files->file_lock);
1890 if (!error && f != filp && flock.l_type != F_UNLCK) {
1891 flock.l_type = F_UNLCK;
1896 locks_free_lock(file_lock);
1900 #if BITS_PER_LONG == 32
1901 /* Report the first existing lock that would conflict with l.
1902 * This implements the F_GETLK command of fcntl().
1904 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1906 struct file_lock file_lock;
1907 struct flock64 flock;
1911 if (copy_from_user(&flock, l, sizeof(flock)))
1914 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1917 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1921 error = vfs_test_lock(filp, &file_lock);
1925 flock.l_type = file_lock.fl_type;
1926 if (file_lock.fl_type != F_UNLCK)
1927 posix_lock_to_flock64(&flock, &file_lock);
1930 if (!copy_to_user(l, &flock, sizeof(flock)))
1937 /* Apply the lock described by l to an open file descriptor.
1938 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1940 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1941 struct flock64 __user *l)
1943 struct file_lock *file_lock = locks_alloc_lock();
1944 struct flock64 flock;
1945 struct inode *inode;
1949 if (file_lock == NULL)
1953 * This might block, so we do it before checking the inode.
1956 if (copy_from_user(&flock, l, sizeof(flock)))
1959 inode = filp->f_path.dentry->d_inode;
1961 /* Don't allow mandatory locks on files that may be memory mapped
1964 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1970 error = flock64_to_posix_lock(filp, file_lock, &flock);
1973 if (cmd == F_SETLKW64) {
1974 file_lock->fl_flags |= FL_SLEEP;
1978 switch (flock.l_type) {
1980 if (!(filp->f_mode & FMODE_READ))
1984 if (!(filp->f_mode & FMODE_WRITE))
1994 error = do_lock_file_wait(filp, cmd, file_lock);
1997 * Attempt to detect a close/fcntl race and recover by
1998 * releasing the lock that was just acquired.
2000 spin_lock(¤t->files->file_lock);
2002 spin_unlock(¤t->files->file_lock);
2003 if (!error && f != filp && flock.l_type != F_UNLCK) {
2004 flock.l_type = F_UNLCK;
2009 locks_free_lock(file_lock);
2012 #endif /* BITS_PER_LONG == 32 */
2015 * This function is called when the file is being removed
2016 * from the task's fd array. POSIX locks belonging to this task
2017 * are deleted at this time.
2019 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2021 struct file_lock lock;
2024 * If there are no locks held on this file, we don't need to call
2025 * posix_lock_file(). Another process could be setting a lock on this
2026 * file at the same time, but we wouldn't remove that lock anyway.
2028 if (!filp->f_path.dentry->d_inode->i_flock)
2031 lock.fl_type = F_UNLCK;
2032 lock.fl_flags = FL_POSIX | FL_CLOSE;
2034 lock.fl_end = OFFSET_MAX;
2035 lock.fl_owner = owner;
2036 lock.fl_pid = current->tgid;
2037 lock.fl_file = filp;
2039 lock.fl_lmops = NULL;
2041 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2043 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2044 lock.fl_ops->fl_release_private(&lock);
2047 EXPORT_SYMBOL(locks_remove_posix);
2050 * This function is called on the last close of an open file.
2052 void locks_remove_flock(struct file *filp)
2054 struct inode * inode = filp->f_path.dentry->d_inode;
2055 struct file_lock *fl;
2056 struct file_lock **before;
2058 if (!inode->i_flock)
2061 if (filp->f_op && filp->f_op->flock) {
2062 struct file_lock fl = {
2063 .fl_pid = current->tgid,
2065 .fl_flags = FL_FLOCK,
2067 .fl_end = OFFSET_MAX,
2069 filp->f_op->flock(filp, F_SETLKW, &fl);
2070 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2071 fl.fl_ops->fl_release_private(&fl);
2075 before = &inode->i_flock;
2077 while ((fl = *before) != NULL) {
2078 if (fl->fl_file == filp) {
2080 locks_delete_lock(before);
2084 lease_modify(before, F_UNLCK);
2090 before = &fl->fl_next;
2096 * posix_unblock_lock - stop waiting for a file lock
2097 * @filp: how the file was opened
2098 * @waiter: the lock which was waiting
2100 * lockd needs to block waiting for locks.
2103 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2108 if (waiter->fl_next)
2109 __locks_delete_block(waiter);
2116 EXPORT_SYMBOL(posix_unblock_lock);
2119 * vfs_cancel_lock - file byte range unblock lock
2120 * @filp: The file to apply the unblock to
2121 * @fl: The lock to be unblocked
2123 * Used by lock managers to cancel blocked requests
2125 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2127 if (filp->f_op && filp->f_op->lock)
2128 return filp->f_op->lock(filp, F_CANCELLK, fl);
2132 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2134 #ifdef CONFIG_PROC_FS
2135 #include <linux/proc_fs.h>
2136 #include <linux/seq_file.h>
2138 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2139 loff_t id, char *pfx)
2141 struct inode *inode = NULL;
2142 unsigned int fl_pid;
2145 fl_pid = pid_vnr(fl->fl_nspid);
2147 fl_pid = fl->fl_pid;
2149 if (fl->fl_file != NULL)
2150 inode = fl->fl_file->f_path.dentry->d_inode;
2152 seq_printf(f, "%lld:%s ", id, pfx);
2154 seq_printf(f, "%6s %s ",
2155 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2156 (inode == NULL) ? "*NOINODE*" :
2157 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2158 } else if (IS_FLOCK(fl)) {
2159 if (fl->fl_type & LOCK_MAND) {
2160 seq_printf(f, "FLOCK MSNFS ");
2162 seq_printf(f, "FLOCK ADVISORY ");
2164 } else if (IS_LEASE(fl)) {
2165 seq_printf(f, "LEASE ");
2166 if (fl->fl_type & F_INPROGRESS)
2167 seq_printf(f, "BREAKING ");
2168 else if (fl->fl_file)
2169 seq_printf(f, "ACTIVE ");
2171 seq_printf(f, "BREAKER ");
2173 seq_printf(f, "UNKNOWN UNKNOWN ");
2175 if (fl->fl_type & LOCK_MAND) {
2176 seq_printf(f, "%s ",
2177 (fl->fl_type & LOCK_READ)
2178 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2179 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2181 seq_printf(f, "%s ",
2182 (fl->fl_type & F_INPROGRESS)
2183 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2184 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2187 #ifdef WE_CAN_BREAK_LSLK_NOW
2188 seq_printf(f, "%d %s:%ld ", fl_pid,
2189 inode->i_sb->s_id, inode->i_ino);
2191 /* userspace relies on this representation of dev_t ;-( */
2192 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2193 MAJOR(inode->i_sb->s_dev),
2194 MINOR(inode->i_sb->s_dev), inode->i_ino);
2197 seq_printf(f, "%d <none>:0 ", fl_pid);
2200 if (fl->fl_end == OFFSET_MAX)
2201 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2203 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2205 seq_printf(f, "0 EOF\n");
2209 static int locks_show(struct seq_file *f, void *v)
2211 struct file_lock *fl, *bfl;
2213 fl = list_entry(v, struct file_lock, fl_link);
2215 lock_get_status(f, fl, *((loff_t *)f->private), "");
2217 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2218 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2223 static void *locks_start(struct seq_file *f, loff_t *pos)
2225 loff_t *p = f->private;
2229 return seq_list_start(&file_lock_list, *pos);
2232 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2234 loff_t *p = f->private;
2236 return seq_list_next(v, &file_lock_list, pos);
2239 static void locks_stop(struct seq_file *f, void *v)
2244 static const struct seq_operations locks_seq_operations = {
2245 .start = locks_start,
2251 static int locks_open(struct inode *inode, struct file *filp)
2253 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
2256 static const struct file_operations proc_locks_operations = {
2259 .llseek = seq_lseek,
2260 .release = seq_release_private,
2263 static int __init proc_locks_init(void)
2265 proc_create("locks", 0, NULL, &proc_locks_operations);
2268 module_init(proc_locks_init);
2272 * lock_may_read - checks that the region is free of locks
2273 * @inode: the inode that is being read
2274 * @start: the first byte to read
2275 * @len: the number of bytes to read
2277 * Emulates Windows locking requirements. Whole-file
2278 * mandatory locks (share modes) can prohibit a read and
2279 * byte-range POSIX locks can prohibit a read if they overlap.
2281 * N.B. this function is only ever called
2282 * from knfsd and ownership of locks is never checked.
2284 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2286 struct file_lock *fl;
2289 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2291 if (fl->fl_type == F_RDLCK)
2293 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2295 } else if (IS_FLOCK(fl)) {
2296 if (!(fl->fl_type & LOCK_MAND))
2298 if (fl->fl_type & LOCK_READ)
2309 EXPORT_SYMBOL(lock_may_read);
2312 * lock_may_write - checks that the region is free of locks
2313 * @inode: the inode that is being written
2314 * @start: the first byte to write
2315 * @len: the number of bytes to write
2317 * Emulates Windows locking requirements. Whole-file
2318 * mandatory locks (share modes) can prohibit a write and
2319 * byte-range POSIX locks can prohibit a write if they overlap.
2321 * N.B. this function is only ever called
2322 * from knfsd and ownership of locks is never checked.
2324 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2326 struct file_lock *fl;
2329 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2331 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2333 } else if (IS_FLOCK(fl)) {
2334 if (!(fl->fl_type & LOCK_MAND))
2336 if (fl->fl_type & LOCK_WRITE)
2347 EXPORT_SYMBOL(lock_may_write);
2349 static int __init filelock_init(void)
2351 filelock_cache = kmem_cache_create("file_lock_cache",
2352 sizeof(struct file_lock), 0, SLAB_PANIC,
2357 core_initcall(filelock_init);