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/filesystems/mandatory-locking.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/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
128 #include <linux/pid_namespace.h>
129 #include <linux/hashtable.h>
130 #include <linux/percpu.h>
131 #include <linux/lglock.h>
133 #include <asm/uaccess.h>
135 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
136 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
137 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
138 #define IS_FILE_PVT(fl) (fl->fl_flags & FL_FILE_PVT)
140 static bool lease_breaking(struct file_lock *fl)
142 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
145 static int target_leasetype(struct file_lock *fl)
147 if (fl->fl_flags & FL_UNLOCK_PENDING)
149 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
154 int leases_enable = 1;
155 int lease_break_time = 45;
157 #define for_each_lock(inode, lockp) \
158 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
161 * The global file_lock_list is only used for displaying /proc/locks, so we
162 * keep a list on each CPU, with each list protected by its own spinlock via
163 * the file_lock_lglock. Note that alterations to the list also require that
164 * the relevant i_lock is held.
166 DEFINE_STATIC_LGLOCK(file_lock_lglock);
167 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
170 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
171 * It is protected by blocked_lock_lock.
173 * We hash locks by lockowner in order to optimize searching for the lock a
174 * particular lockowner is waiting on.
176 * FIXME: make this value scale via some heuristic? We generally will want more
177 * buckets when we have more lockowners holding locks, but that's a little
178 * difficult to determine without knowing what the workload will look like.
180 #define BLOCKED_HASH_BITS 7
181 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
184 * This lock protects the blocked_hash. Generally, if you're accessing it, you
185 * want to be holding this lock.
187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
188 * pointer for file_lock structures that are acting as lock requests (in
189 * contrast to those that are acting as records of acquired locks).
191 * Note that when we acquire this lock in order to change the above fields,
192 * we often hold the i_lock as well. In certain cases, when reading the fields
193 * protected by this lock, we can skip acquiring it iff we already hold the
196 * In particular, adding an entry to the fl_block list requires that you hold
197 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
198 * an entry from the list however only requires the file_lock_lock.
200 static DEFINE_SPINLOCK(blocked_lock_lock);
202 static struct kmem_cache *filelock_cache __read_mostly;
204 static void locks_init_lock_heads(struct file_lock *fl)
206 INIT_HLIST_NODE(&fl->fl_link);
207 INIT_LIST_HEAD(&fl->fl_block);
208 init_waitqueue_head(&fl->fl_wait);
211 /* Allocate an empty lock structure. */
212 struct file_lock *locks_alloc_lock(void)
214 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
217 locks_init_lock_heads(fl);
221 EXPORT_SYMBOL_GPL(locks_alloc_lock);
223 void locks_release_private(struct file_lock *fl)
226 if (fl->fl_ops->fl_release_private)
227 fl->fl_ops->fl_release_private(fl);
233 EXPORT_SYMBOL_GPL(locks_release_private);
235 /* Free a lock which is not in use. */
236 void locks_free_lock(struct file_lock *fl)
238 BUG_ON(waitqueue_active(&fl->fl_wait));
239 BUG_ON(!list_empty(&fl->fl_block));
240 BUG_ON(!hlist_unhashed(&fl->fl_link));
242 locks_release_private(fl);
243 kmem_cache_free(filelock_cache, fl);
245 EXPORT_SYMBOL(locks_free_lock);
247 void locks_init_lock(struct file_lock *fl)
249 memset(fl, 0, sizeof(struct file_lock));
250 locks_init_lock_heads(fl);
253 EXPORT_SYMBOL(locks_init_lock);
255 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
258 if (fl->fl_ops->fl_copy_lock)
259 fl->fl_ops->fl_copy_lock(new, fl);
260 new->fl_ops = fl->fl_ops;
263 new->fl_lmops = fl->fl_lmops;
267 * Initialize a new lock from an existing file_lock structure.
269 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
271 new->fl_owner = fl->fl_owner;
272 new->fl_pid = fl->fl_pid;
274 new->fl_flags = fl->fl_flags;
275 new->fl_type = fl->fl_type;
276 new->fl_start = fl->fl_start;
277 new->fl_end = fl->fl_end;
279 new->fl_lmops = NULL;
281 EXPORT_SYMBOL(__locks_copy_lock);
283 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
285 locks_release_private(new);
287 __locks_copy_lock(new, fl);
288 new->fl_file = fl->fl_file;
289 new->fl_ops = fl->fl_ops;
290 new->fl_lmops = fl->fl_lmops;
292 locks_copy_private(new, fl);
295 EXPORT_SYMBOL(locks_copy_lock);
297 static inline int flock_translate_cmd(int cmd) {
299 return cmd & (LOCK_MAND | LOCK_RW);
311 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
312 static int flock_make_lock(struct file *filp, struct file_lock **lock,
315 struct file_lock *fl;
316 int type = flock_translate_cmd(cmd);
320 fl = locks_alloc_lock();
325 fl->fl_pid = current->tgid;
326 fl->fl_flags = FL_FLOCK;
328 fl->fl_end = OFFSET_MAX;
334 static int assign_type(struct file_lock *fl, long type)
348 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
351 switch (l->l_whence) {
356 fl->fl_start = filp->f_pos;
359 fl->fl_start = i_size_read(file_inode(filp));
364 if (l->l_start > OFFSET_MAX - fl->fl_start)
366 fl->fl_start += l->l_start;
367 if (fl->fl_start < 0)
370 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
371 POSIX-2001 defines it. */
373 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
375 fl->fl_end = fl->fl_start + l->l_len - 1;
377 } else if (l->l_len < 0) {
378 if (fl->fl_start + l->l_len < 0)
380 fl->fl_end = fl->fl_start - 1;
381 fl->fl_start += l->l_len;
383 fl->fl_end = OFFSET_MAX;
385 fl->fl_owner = current->files;
386 fl->fl_pid = current->tgid;
388 fl->fl_flags = FL_POSIX;
392 /* Ensure that fl->fl_filp has compatible f_mode */
395 if (!(filp->f_mode & FMODE_READ))
399 if (!(filp->f_mode & FMODE_WRITE))
404 return assign_type(fl, l->l_type);
407 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
410 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
413 struct flock64 ll = {
415 .l_whence = l->l_whence,
416 .l_start = l->l_start,
420 return flock64_to_posix_lock(filp, fl, &ll);
423 /* default lease lock manager operations */
424 static void lease_break_callback(struct file_lock *fl)
426 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
429 static const struct lock_manager_operations lease_manager_ops = {
430 .lm_break = lease_break_callback,
431 .lm_change = lease_modify,
435 * Initialize a lease, use the default lock manager operations
437 static int lease_init(struct file *filp, long type, struct file_lock *fl)
439 if (assign_type(fl, type) != 0)
442 fl->fl_owner = current->files;
443 fl->fl_pid = current->tgid;
446 fl->fl_flags = FL_LEASE;
448 fl->fl_end = OFFSET_MAX;
450 fl->fl_lmops = &lease_manager_ops;
454 /* Allocate a file_lock initialised to this type of lease */
455 static struct file_lock *lease_alloc(struct file *filp, long type)
457 struct file_lock *fl = locks_alloc_lock();
461 return ERR_PTR(error);
463 error = lease_init(filp, type, fl);
466 return ERR_PTR(error);
471 /* Check if two locks overlap each other.
473 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
475 return ((fl1->fl_end >= fl2->fl_start) &&
476 (fl2->fl_end >= fl1->fl_start));
480 * Check whether two locks have the same owner.
482 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
484 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
485 return fl2->fl_lmops == fl1->fl_lmops &&
486 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
487 return fl1->fl_owner == fl2->fl_owner;
490 /* Must be called with the i_lock held! */
491 static void locks_insert_global_locks(struct file_lock *fl)
493 lg_local_lock(&file_lock_lglock);
494 fl->fl_link_cpu = smp_processor_id();
495 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
496 lg_local_unlock(&file_lock_lglock);
499 /* Must be called with the i_lock held! */
500 static void locks_delete_global_locks(struct file_lock *fl)
503 * Avoid taking lock if already unhashed. This is safe since this check
504 * is done while holding the i_lock, and new insertions into the list
505 * also require that it be held.
507 if (hlist_unhashed(&fl->fl_link))
509 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
510 hlist_del_init(&fl->fl_link);
511 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
515 posix_owner_key(struct file_lock *fl)
517 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
518 return fl->fl_lmops->lm_owner_key(fl);
519 return (unsigned long)fl->fl_owner;
522 static void locks_insert_global_blocked(struct file_lock *waiter)
524 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
527 static void locks_delete_global_blocked(struct file_lock *waiter)
529 hash_del(&waiter->fl_link);
532 /* Remove waiter from blocker's block list.
533 * When blocker ends up pointing to itself then the list is empty.
535 * Must be called with blocked_lock_lock held.
537 static void __locks_delete_block(struct file_lock *waiter)
539 locks_delete_global_blocked(waiter);
540 list_del_init(&waiter->fl_block);
541 waiter->fl_next = NULL;
544 static void locks_delete_block(struct file_lock *waiter)
546 spin_lock(&blocked_lock_lock);
547 __locks_delete_block(waiter);
548 spin_unlock(&blocked_lock_lock);
551 /* Insert waiter into blocker's block list.
552 * We use a circular list so that processes can be easily woken up in
553 * the order they blocked. The documentation doesn't require this but
554 * it seems like the reasonable thing to do.
556 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
557 * list itself is protected by the blocked_lock_lock, but by ensuring that the
558 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
559 * in some cases when we see that the fl_block list is empty.
561 static void __locks_insert_block(struct file_lock *blocker,
562 struct file_lock *waiter)
564 BUG_ON(!list_empty(&waiter->fl_block));
565 waiter->fl_next = blocker;
566 list_add_tail(&waiter->fl_block, &blocker->fl_block);
567 if (IS_POSIX(blocker))
568 locks_insert_global_blocked(waiter);
571 /* Must be called with i_lock held. */
572 static void locks_insert_block(struct file_lock *blocker,
573 struct file_lock *waiter)
575 spin_lock(&blocked_lock_lock);
576 __locks_insert_block(blocker, waiter);
577 spin_unlock(&blocked_lock_lock);
581 * Wake up processes blocked waiting for blocker.
583 * Must be called with the inode->i_lock held!
585 static void locks_wake_up_blocks(struct file_lock *blocker)
588 * Avoid taking global lock if list is empty. This is safe since new
589 * blocked requests are only added to the list under the i_lock, and
590 * the i_lock is always held here. Note that removal from the fl_block
591 * list does not require the i_lock, so we must recheck list_empty()
592 * after acquiring the blocked_lock_lock.
594 if (list_empty(&blocker->fl_block))
597 spin_lock(&blocked_lock_lock);
598 while (!list_empty(&blocker->fl_block)) {
599 struct file_lock *waiter;
601 waiter = list_first_entry(&blocker->fl_block,
602 struct file_lock, fl_block);
603 __locks_delete_block(waiter);
604 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
605 waiter->fl_lmops->lm_notify(waiter);
607 wake_up(&waiter->fl_wait);
609 spin_unlock(&blocked_lock_lock);
612 /* Insert file lock fl into an inode's lock list at the position indicated
613 * by pos. At the same time add the lock to the global file lock list.
615 * Must be called with the i_lock held!
617 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
619 fl->fl_nspid = get_pid(task_tgid(current));
621 /* insert into file's list */
625 locks_insert_global_locks(fl);
629 * locks_delete_lock - Delete a lock and then free it.
630 * @thisfl_p: pointer that points to the fl_next field of the previous
631 * inode->i_flock list entry
633 * Unlink a lock from all lists and free the namespace reference, but don't
634 * free it yet. Wake up processes that are blocked waiting for this lock and
635 * notify the FS that the lock has been cleared.
637 * Must be called with the i_lock held!
639 static void locks_unlink_lock(struct file_lock **thisfl_p)
641 struct file_lock *fl = *thisfl_p;
643 locks_delete_global_locks(fl);
645 *thisfl_p = fl->fl_next;
649 put_pid(fl->fl_nspid);
653 locks_wake_up_blocks(fl);
657 * Unlink a lock from all lists and free it.
659 * Must be called with i_lock held!
661 static void locks_delete_lock(struct file_lock **thisfl_p)
663 struct file_lock *fl = *thisfl_p;
665 locks_unlink_lock(thisfl_p);
669 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
670 * checks for shared/exclusive status of overlapping locks.
672 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
674 if (sys_fl->fl_type == F_WRLCK)
676 if (caller_fl->fl_type == F_WRLCK)
681 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
682 * checking before calling the locks_conflict().
684 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
686 /* POSIX locks owned by the same process do not conflict with
689 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
692 /* Check whether they overlap */
693 if (!locks_overlap(caller_fl, sys_fl))
696 return (locks_conflict(caller_fl, sys_fl));
699 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
700 * checking before calling the locks_conflict().
702 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
704 /* FLOCK locks referring to the same filp do not conflict with
707 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
709 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
712 return (locks_conflict(caller_fl, sys_fl));
716 posix_test_lock(struct file *filp, struct file_lock *fl)
718 struct file_lock *cfl;
719 struct inode *inode = file_inode(filp);
721 spin_lock(&inode->i_lock);
722 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
725 if (posix_locks_conflict(fl, cfl))
729 __locks_copy_lock(fl, cfl);
731 fl->fl_pid = pid_vnr(cfl->fl_nspid);
733 fl->fl_type = F_UNLCK;
734 spin_unlock(&inode->i_lock);
737 EXPORT_SYMBOL(posix_test_lock);
740 * Deadlock detection:
742 * We attempt to detect deadlocks that are due purely to posix file
745 * We assume that a task can be waiting for at most one lock at a time.
746 * So for any acquired lock, the process holding that lock may be
747 * waiting on at most one other lock. That lock in turns may be held by
748 * someone waiting for at most one other lock. Given a requested lock
749 * caller_fl which is about to wait for a conflicting lock block_fl, we
750 * follow this chain of waiters to ensure we are not about to create a
753 * Since we do this before we ever put a process to sleep on a lock, we
754 * are ensured that there is never a cycle; that is what guarantees that
755 * the while() loop in posix_locks_deadlock() eventually completes.
757 * Note: the above assumption may not be true when handling lock
758 * requests from a broken NFS client. It may also fail in the presence
759 * of tasks (such as posix threads) sharing the same open file table.
761 * To handle those cases, we just bail out after a few iterations.
764 #define MAX_DEADLK_ITERATIONS 10
766 /* Find a lock that the owner of the given block_fl is blocking on. */
767 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
769 struct file_lock *fl;
771 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
772 if (posix_same_owner(fl, block_fl))
778 /* Must be called with the blocked_lock_lock held! */
779 static int posix_locks_deadlock(struct file_lock *caller_fl,
780 struct file_lock *block_fl)
784 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
785 if (i++ > MAX_DEADLK_ITERATIONS)
787 if (posix_same_owner(caller_fl, block_fl))
793 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
794 * after any leases, but before any posix locks.
796 * Note that if called with an FL_EXISTS argument, the caller may determine
797 * whether or not a lock was successfully freed by testing the return
800 static int flock_lock_file(struct file *filp, struct file_lock *request)
802 struct file_lock *new_fl = NULL;
803 struct file_lock **before;
804 struct inode * inode = file_inode(filp);
808 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
809 new_fl = locks_alloc_lock();
814 spin_lock(&inode->i_lock);
815 if (request->fl_flags & FL_ACCESS)
818 for_each_lock(inode, before) {
819 struct file_lock *fl = *before;
824 if (filp != fl->fl_file)
826 if (request->fl_type == fl->fl_type)
829 locks_delete_lock(before);
833 if (request->fl_type == F_UNLCK) {
834 if ((request->fl_flags & FL_EXISTS) && !found)
840 * If a higher-priority process was blocked on the old file lock,
841 * give it the opportunity to lock the file.
844 spin_unlock(&inode->i_lock);
846 spin_lock(&inode->i_lock);
850 for_each_lock(inode, before) {
851 struct file_lock *fl = *before;
856 if (!flock_locks_conflict(request, fl))
859 if (!(request->fl_flags & FL_SLEEP))
861 error = FILE_LOCK_DEFERRED;
862 locks_insert_block(fl, request);
865 if (request->fl_flags & FL_ACCESS)
867 locks_copy_lock(new_fl, request);
868 locks_insert_lock(before, new_fl);
873 spin_unlock(&inode->i_lock);
875 locks_free_lock(new_fl);
879 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
881 struct file_lock *fl;
882 struct file_lock *new_fl = NULL;
883 struct file_lock *new_fl2 = NULL;
884 struct file_lock *left = NULL;
885 struct file_lock *right = NULL;
886 struct file_lock **before;
891 * We may need two file_lock structures for this operation,
892 * so we get them in advance to avoid races.
894 * In some cases we can be sure, that no new locks will be needed
896 if (!(request->fl_flags & FL_ACCESS) &&
897 (request->fl_type != F_UNLCK ||
898 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
899 new_fl = locks_alloc_lock();
900 new_fl2 = locks_alloc_lock();
903 spin_lock(&inode->i_lock);
905 * New lock request. Walk all POSIX locks and look for conflicts. If
906 * there are any, either return error or put the request on the
907 * blocker's list of waiters and the global blocked_hash.
909 if (request->fl_type != F_UNLCK) {
910 for_each_lock(inode, before) {
914 if (!posix_locks_conflict(request, fl))
917 __locks_copy_lock(conflock, fl);
919 if (!(request->fl_flags & FL_SLEEP))
922 * Deadlock detection and insertion into the blocked
923 * locks list must be done while holding the same lock!
926 spin_lock(&blocked_lock_lock);
927 if (likely(!posix_locks_deadlock(request, fl))) {
928 error = FILE_LOCK_DEFERRED;
929 __locks_insert_block(fl, request);
931 spin_unlock(&blocked_lock_lock);
936 /* If we're just looking for a conflict, we're done. */
938 if (request->fl_flags & FL_ACCESS)
942 * Find the first old lock with the same owner as the new lock.
945 before = &inode->i_flock;
947 /* First skip locks owned by other processes. */
948 while ((fl = *before) && (!IS_POSIX(fl) ||
949 !posix_same_owner(request, fl))) {
950 before = &fl->fl_next;
953 /* Process locks with this owner. */
954 while ((fl = *before) && posix_same_owner(request, fl)) {
955 /* Detect adjacent or overlapping regions (if same lock type)
957 if (request->fl_type == fl->fl_type) {
958 /* In all comparisons of start vs end, use
959 * "start - 1" rather than "end + 1". If end
960 * is OFFSET_MAX, end + 1 will become negative.
962 if (fl->fl_end < request->fl_start - 1)
964 /* If the next lock in the list has entirely bigger
965 * addresses than the new one, insert the lock here.
967 if (fl->fl_start - 1 > request->fl_end)
970 /* If we come here, the new and old lock are of the
971 * same type and adjacent or overlapping. Make one
972 * lock yielding from the lower start address of both
973 * locks to the higher end address.
975 if (fl->fl_start > request->fl_start)
976 fl->fl_start = request->fl_start;
978 request->fl_start = fl->fl_start;
979 if (fl->fl_end < request->fl_end)
980 fl->fl_end = request->fl_end;
982 request->fl_end = fl->fl_end;
984 locks_delete_lock(before);
991 /* Processing for different lock types is a bit
994 if (fl->fl_end < request->fl_start)
996 if (fl->fl_start > request->fl_end)
998 if (request->fl_type == F_UNLCK)
1000 if (fl->fl_start < request->fl_start)
1002 /* If the next lock in the list has a higher end
1003 * address than the new one, insert the new one here.
1005 if (fl->fl_end > request->fl_end) {
1009 if (fl->fl_start >= request->fl_start) {
1010 /* The new lock completely replaces an old
1011 * one (This may happen several times).
1014 locks_delete_lock(before);
1017 /* Replace the old lock with the new one.
1018 * Wake up anybody waiting for the old one,
1019 * as the change in lock type might satisfy
1022 locks_wake_up_blocks(fl);
1023 fl->fl_start = request->fl_start;
1024 fl->fl_end = request->fl_end;
1025 fl->fl_type = request->fl_type;
1026 locks_release_private(fl);
1027 locks_copy_private(fl, request);
1032 /* Go on to next lock.
1035 before = &fl->fl_next;
1039 * The above code only modifies existing locks in case of merging or
1040 * replacing. If new lock(s) need to be inserted all modifications are
1041 * done below this, so it's safe yet to bail out.
1043 error = -ENOLCK; /* "no luck" */
1044 if (right && left == right && !new_fl2)
1049 if (request->fl_type == F_UNLCK) {
1050 if (request->fl_flags & FL_EXISTS)
1059 locks_copy_lock(new_fl, request);
1060 locks_insert_lock(before, new_fl);
1064 if (left == right) {
1065 /* The new lock breaks the old one in two pieces,
1066 * so we have to use the second new lock.
1070 locks_copy_lock(left, right);
1071 locks_insert_lock(before, left);
1073 right->fl_start = request->fl_end + 1;
1074 locks_wake_up_blocks(right);
1077 left->fl_end = request->fl_start - 1;
1078 locks_wake_up_blocks(left);
1081 spin_unlock(&inode->i_lock);
1083 * Free any unused locks.
1086 locks_free_lock(new_fl);
1088 locks_free_lock(new_fl2);
1093 * posix_lock_file - Apply a POSIX-style lock to a file
1094 * @filp: The file to apply the lock to
1095 * @fl: The lock to be applied
1096 * @conflock: Place to return a copy of the conflicting lock, if found.
1098 * Add a POSIX style lock to a file.
1099 * We merge adjacent & overlapping locks whenever possible.
1100 * POSIX locks are sorted by owner task, then by starting address
1102 * Note that if called with an FL_EXISTS argument, the caller may determine
1103 * whether or not a lock was successfully freed by testing the return
1104 * value for -ENOENT.
1106 int posix_lock_file(struct file *filp, struct file_lock *fl,
1107 struct file_lock *conflock)
1109 return __posix_lock_file(file_inode(filp), fl, conflock);
1111 EXPORT_SYMBOL(posix_lock_file);
1114 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1115 * @filp: The file to apply the lock to
1116 * @fl: The lock to be applied
1118 * Add a POSIX style lock to a file.
1119 * We merge adjacent & overlapping locks whenever possible.
1120 * POSIX locks are sorted by owner task, then by starting address
1122 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1127 error = posix_lock_file(filp, fl, NULL);
1128 if (error != FILE_LOCK_DEFERRED)
1130 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1134 locks_delete_block(fl);
1139 EXPORT_SYMBOL(posix_lock_file_wait);
1142 * locks_mandatory_locked - Check for an active lock
1143 * @inode: the file to check
1145 * Searches the inode's list of locks to find any POSIX locks which conflict.
1146 * This function is called from locks_verify_locked() only.
1148 int locks_mandatory_locked(struct inode *inode)
1150 fl_owner_t owner = current->files;
1151 struct file_lock *fl;
1154 * Search the lock list for this inode for any POSIX locks.
1156 spin_lock(&inode->i_lock);
1157 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1160 if (fl->fl_owner != owner)
1163 spin_unlock(&inode->i_lock);
1164 return fl ? -EAGAIN : 0;
1168 * locks_mandatory_area - Check for a conflicting lock
1169 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1171 * @inode: the file to check
1172 * @filp: how the file was opened (if it was)
1173 * @offset: start of area to check
1174 * @count: length of area to check
1176 * Searches the inode's list of locks to find any POSIX locks which conflict.
1177 * This function is called from rw_verify_area() and
1178 * locks_verify_truncate().
1180 int locks_mandatory_area(int read_write, struct inode *inode,
1181 struct file *filp, loff_t offset,
1184 struct file_lock fl;
1187 locks_init_lock(&fl);
1188 fl.fl_owner = current->files;
1189 fl.fl_pid = current->tgid;
1191 fl.fl_flags = FL_POSIX | FL_ACCESS;
1192 if (filp && !(filp->f_flags & O_NONBLOCK))
1193 fl.fl_flags |= FL_SLEEP;
1194 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1195 fl.fl_start = offset;
1196 fl.fl_end = offset + count - 1;
1199 error = __posix_lock_file(inode, &fl, NULL);
1200 if (error != FILE_LOCK_DEFERRED)
1202 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1205 * If we've been sleeping someone might have
1206 * changed the permissions behind our back.
1208 if (__mandatory_lock(inode))
1212 locks_delete_block(&fl);
1219 EXPORT_SYMBOL(locks_mandatory_area);
1221 static void lease_clear_pending(struct file_lock *fl, int arg)
1225 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1228 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1232 /* We already had a lease on this file; just change its type */
1233 int lease_modify(struct file_lock **before, int arg)
1235 struct file_lock *fl = *before;
1236 int error = assign_type(fl, arg);
1240 lease_clear_pending(fl, arg);
1241 locks_wake_up_blocks(fl);
1242 if (arg == F_UNLCK) {
1243 struct file *filp = fl->fl_file;
1246 filp->f_owner.signum = 0;
1247 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1248 if (fl->fl_fasync != NULL) {
1249 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1250 fl->fl_fasync = NULL;
1252 locks_delete_lock(before);
1257 EXPORT_SYMBOL(lease_modify);
1259 static bool past_time(unsigned long then)
1262 /* 0 is a special value meaning "this never expires": */
1264 return time_after(jiffies, then);
1267 static void time_out_leases(struct inode *inode)
1269 struct file_lock **before;
1270 struct file_lock *fl;
1272 before = &inode->i_flock;
1273 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1274 if (past_time(fl->fl_downgrade_time))
1275 lease_modify(before, F_RDLCK);
1276 if (past_time(fl->fl_break_time))
1277 lease_modify(before, F_UNLCK);
1278 if (fl == *before) /* lease_modify may have freed fl */
1279 before = &fl->fl_next;
1283 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1285 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1287 return locks_conflict(breaker, lease);
1291 * __break_lease - revoke all outstanding leases on file
1292 * @inode: the inode of the file to return
1293 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1295 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1298 * break_lease (inlined for speed) has checked there already is at least
1299 * some kind of lock (maybe a lease) on this file. Leases are broken on
1300 * a call to open() or truncate(). This function can sleep unless you
1301 * specified %O_NONBLOCK to your open().
1303 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1306 struct file_lock *new_fl, *flock;
1307 struct file_lock *fl;
1308 unsigned long break_time;
1309 int i_have_this_lease = 0;
1310 bool lease_conflict = false;
1311 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1313 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1315 return PTR_ERR(new_fl);
1316 new_fl->fl_flags = type;
1318 spin_lock(&inode->i_lock);
1320 time_out_leases(inode);
1322 flock = inode->i_flock;
1323 if ((flock == NULL) || !IS_LEASE(flock))
1326 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1327 if (leases_conflict(fl, new_fl)) {
1328 lease_conflict = true;
1329 if (fl->fl_owner == current->files)
1330 i_have_this_lease = 1;
1333 if (!lease_conflict)
1337 if (lease_break_time > 0) {
1338 break_time = jiffies + lease_break_time * HZ;
1339 if (break_time == 0)
1340 break_time++; /* so that 0 means no break time */
1343 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1344 if (!leases_conflict(fl, new_fl))
1347 if (fl->fl_flags & FL_UNLOCK_PENDING)
1349 fl->fl_flags |= FL_UNLOCK_PENDING;
1350 fl->fl_break_time = break_time;
1352 if (lease_breaking(flock))
1354 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1355 fl->fl_downgrade_time = break_time;
1357 fl->fl_lmops->lm_break(fl);
1360 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1361 error = -EWOULDBLOCK;
1366 break_time = flock->fl_break_time;
1367 if (break_time != 0) {
1368 break_time -= jiffies;
1369 if (break_time == 0)
1372 locks_insert_block(flock, new_fl);
1373 spin_unlock(&inode->i_lock);
1374 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1375 !new_fl->fl_next, break_time);
1376 spin_lock(&inode->i_lock);
1377 locks_delete_block(new_fl);
1380 time_out_leases(inode);
1382 * Wait for the next conflicting lease that has not been
1385 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1386 flock = flock->fl_next) {
1387 if (leases_conflict(new_fl, flock))
1394 spin_unlock(&inode->i_lock);
1395 locks_free_lock(new_fl);
1399 EXPORT_SYMBOL(__break_lease);
1402 * lease_get_mtime - get the last modified time of an inode
1404 * @time: pointer to a timespec which will contain the last modified time
1406 * This is to force NFS clients to flush their caches for files with
1407 * exclusive leases. The justification is that if someone has an
1408 * exclusive lease, then they could be modifying it.
1410 void lease_get_mtime(struct inode *inode, struct timespec *time)
1412 struct file_lock *flock = inode->i_flock;
1413 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1414 *time = current_fs_time(inode->i_sb);
1416 *time = inode->i_mtime;
1419 EXPORT_SYMBOL(lease_get_mtime);
1422 * fcntl_getlease - Enquire what lease is currently active
1425 * The value returned by this function will be one of
1426 * (if no lease break is pending):
1428 * %F_RDLCK to indicate a shared lease is held.
1430 * %F_WRLCK to indicate an exclusive lease is held.
1432 * %F_UNLCK to indicate no lease is held.
1434 * (if a lease break is pending):
1436 * %F_RDLCK to indicate an exclusive lease needs to be
1437 * changed to a shared lease (or removed).
1439 * %F_UNLCK to indicate the lease needs to be removed.
1441 * XXX: sfr & willy disagree over whether F_INPROGRESS
1442 * should be returned to userspace.
1444 int fcntl_getlease(struct file *filp)
1446 struct file_lock *fl;
1447 struct inode *inode = file_inode(filp);
1450 spin_lock(&inode->i_lock);
1451 time_out_leases(file_inode(filp));
1452 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1454 if (fl->fl_file == filp) {
1455 type = target_leasetype(fl);
1459 spin_unlock(&inode->i_lock);
1464 * check_conflicting_open - see if the given dentry points to a file that has
1465 * an existing open that would conflict with the
1467 * @dentry: dentry to check
1468 * @arg: type of lease that we're trying to acquire
1470 * Check to see if there's an existing open fd on this file that would
1471 * conflict with the lease we're trying to set.
1474 check_conflicting_open(const struct dentry *dentry, const long arg)
1477 struct inode *inode = dentry->d_inode;
1479 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1482 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1483 (atomic_read(&inode->i_count) > 1)))
1489 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1491 struct file_lock *fl, **before, **my_before = NULL, *lease;
1492 struct dentry *dentry = filp->f_path.dentry;
1493 struct inode *inode = dentry->d_inode;
1494 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1499 * In the delegation case we need mutual exclusion with
1500 * a number of operations that take the i_mutex. We trylock
1501 * because delegations are an optional optimization, and if
1502 * there's some chance of a conflict--we'd rather not
1503 * bother, maybe that's a sign this just isn't a good file to
1504 * hand out a delegation on.
1506 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1509 if (is_deleg && arg == F_WRLCK) {
1510 /* Write delegations are not currently supported: */
1511 mutex_unlock(&inode->i_mutex);
1516 error = check_conflicting_open(dentry, arg);
1521 * At this point, we know that if there is an exclusive
1522 * lease on this file, then we hold it on this filp
1523 * (otherwise our open of this file would have blocked).
1524 * And if we are trying to acquire an exclusive lease,
1525 * then the file is not open by anyone (including us)
1526 * except for this filp.
1529 for (before = &inode->i_flock;
1530 ((fl = *before) != NULL) && IS_LEASE(fl);
1531 before = &fl->fl_next) {
1532 if (fl->fl_file == filp) {
1537 * No exclusive leases if someone else has a lease on
1543 * Modifying our existing lease is OK, but no getting a
1544 * new lease if someone else is opening for write:
1546 if (fl->fl_flags & FL_UNLOCK_PENDING)
1550 if (my_before != NULL) {
1551 error = lease->fl_lmops->lm_change(my_before, arg);
1561 locks_insert_lock(before, lease);
1563 * The check in break_lease() is lockless. It's possible for another
1564 * open to race in after we did the earlier check for a conflicting
1565 * open but before the lease was inserted. Check again for a
1566 * conflicting open and cancel the lease if there is one.
1568 * We also add a barrier here to ensure that the insertion of the lock
1569 * precedes these checks.
1572 error = check_conflicting_open(dentry, arg);
1574 locks_unlink_lock(flp);
1577 mutex_unlock(&inode->i_mutex);
1581 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1583 struct file_lock *fl, **before;
1584 struct dentry *dentry = filp->f_path.dentry;
1585 struct inode *inode = dentry->d_inode;
1587 for (before = &inode->i_flock;
1588 ((fl = *before) != NULL) && IS_LEASE(fl);
1589 before = &fl->fl_next) {
1590 if (fl->fl_file != filp)
1592 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1598 * generic_setlease - sets a lease on an open file
1599 * @filp: file pointer
1600 * @arg: type of lease to obtain
1601 * @flp: input - file_lock to use, output - file_lock inserted
1603 * The (input) flp->fl_lmops->lm_break function is required
1606 * Called with inode->i_lock held.
1608 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1610 struct dentry *dentry = filp->f_path.dentry;
1611 struct inode *inode = dentry->d_inode;
1614 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1616 if (!S_ISREG(inode->i_mode))
1618 error = security_file_lock(filp, arg);
1622 time_out_leases(inode);
1624 BUG_ON(!(*flp)->fl_lmops->lm_break);
1628 return generic_delete_lease(filp, flp);
1631 return generic_add_lease(filp, arg, flp);
1636 EXPORT_SYMBOL(generic_setlease);
1638 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1640 if (filp->f_op->setlease)
1641 return filp->f_op->setlease(filp, arg, lease);
1643 return generic_setlease(filp, arg, lease);
1647 * vfs_setlease - sets a lease on an open file
1648 * @filp: file pointer
1649 * @arg: type of lease to obtain
1650 * @lease: file_lock to use
1652 * Call this to establish a lease on the file.
1653 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1654 * break_lease will oops!
1656 * This will call the filesystem's setlease file method, if
1657 * defined. Note that there is no getlease method; instead, the
1658 * filesystem setlease method should call back to setlease() to
1659 * add a lease to the inode's lease list, where fcntl_getlease() can
1660 * find it. Since fcntl_getlease() only reports whether the current
1661 * task holds a lease, a cluster filesystem need only do this for
1662 * leases held by processes on this node.
1664 * There is also no break_lease method; filesystems that
1665 * handle their own leases should break leases themselves from the
1666 * filesystem's open, create, and (on truncate) setattr methods.
1668 * Warning: the only current setlease methods exist only to disable
1669 * leases in certain cases. More vfs changes may be required to
1670 * allow a full filesystem lease implementation.
1673 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1675 struct inode *inode = file_inode(filp);
1678 spin_lock(&inode->i_lock);
1679 error = __vfs_setlease(filp, arg, lease);
1680 spin_unlock(&inode->i_lock);
1684 EXPORT_SYMBOL_GPL(vfs_setlease);
1686 static int do_fcntl_delete_lease(struct file *filp)
1688 struct file_lock fl, *flp = &fl;
1690 lease_init(filp, F_UNLCK, flp);
1692 return vfs_setlease(filp, F_UNLCK, &flp);
1695 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1697 struct file_lock *fl, *ret;
1698 struct inode *inode = file_inode(filp);
1699 struct fasync_struct *new;
1702 fl = lease_alloc(filp, arg);
1706 new = fasync_alloc();
1708 locks_free_lock(fl);
1712 spin_lock(&inode->i_lock);
1713 error = __vfs_setlease(filp, arg, &ret);
1715 spin_unlock(&inode->i_lock);
1716 locks_free_lock(fl);
1717 goto out_free_fasync;
1720 locks_free_lock(fl);
1723 * fasync_insert_entry() returns the old entry if any.
1724 * If there was no old entry, then it used 'new' and
1725 * inserted it into the fasync list. Clear new so that
1726 * we don't release it here.
1728 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1731 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1732 spin_unlock(&inode->i_lock);
1741 * fcntl_setlease - sets a lease on an open file
1742 * @fd: open file descriptor
1743 * @filp: file pointer
1744 * @arg: type of lease to obtain
1746 * Call this fcntl to establish a lease on the file.
1747 * Note that you also need to call %F_SETSIG to
1748 * receive a signal when the lease is broken.
1750 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1753 return do_fcntl_delete_lease(filp);
1754 return do_fcntl_add_lease(fd, filp, arg);
1758 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1759 * @filp: The file to apply the lock to
1760 * @fl: The lock to be applied
1762 * Add a FLOCK style lock to a file.
1764 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1769 error = flock_lock_file(filp, fl);
1770 if (error != FILE_LOCK_DEFERRED)
1772 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1776 locks_delete_block(fl);
1782 EXPORT_SYMBOL(flock_lock_file_wait);
1785 * sys_flock: - flock() system call.
1786 * @fd: the file descriptor to lock.
1787 * @cmd: the type of lock to apply.
1789 * Apply a %FL_FLOCK style lock to an open file descriptor.
1790 * The @cmd can be one of
1792 * %LOCK_SH -- a shared lock.
1794 * %LOCK_EX -- an exclusive lock.
1796 * %LOCK_UN -- remove an existing lock.
1798 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1800 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1801 * processes read and write access respectively.
1803 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1805 struct fd f = fdget(fd);
1806 struct file_lock *lock;
1807 int can_sleep, unlock;
1814 can_sleep = !(cmd & LOCK_NB);
1816 unlock = (cmd == LOCK_UN);
1818 if (!unlock && !(cmd & LOCK_MAND) &&
1819 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1822 error = flock_make_lock(f.file, &lock, cmd);
1826 lock->fl_flags |= FL_SLEEP;
1828 error = security_file_lock(f.file, lock->fl_type);
1832 if (f.file->f_op->flock)
1833 error = f.file->f_op->flock(f.file,
1834 (can_sleep) ? F_SETLKW : F_SETLK,
1837 error = flock_lock_file_wait(f.file, lock);
1840 locks_free_lock(lock);
1849 * vfs_test_lock - test file byte range lock
1850 * @filp: The file to test lock for
1851 * @fl: The lock to test; also used to hold result
1853 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1854 * setting conf->fl_type to something other than F_UNLCK.
1856 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1858 if (filp->f_op->lock)
1859 return filp->f_op->lock(filp, F_GETLK, fl);
1860 posix_test_lock(filp, fl);
1863 EXPORT_SYMBOL_GPL(vfs_test_lock);
1865 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1867 flock->l_pid = IS_FILE_PVT(fl) ? -1 : fl->fl_pid;
1868 #if BITS_PER_LONG == 32
1870 * Make sure we can represent the posix lock via
1871 * legacy 32bit flock.
1873 if (fl->fl_start > OFFT_OFFSET_MAX)
1875 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1878 flock->l_start = fl->fl_start;
1879 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1880 fl->fl_end - fl->fl_start + 1;
1881 flock->l_whence = 0;
1882 flock->l_type = fl->fl_type;
1886 #if BITS_PER_LONG == 32
1887 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1889 flock->l_pid = IS_FILE_PVT(fl) ? -1 : fl->fl_pid;
1890 flock->l_start = fl->fl_start;
1891 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1892 fl->fl_end - fl->fl_start + 1;
1893 flock->l_whence = 0;
1894 flock->l_type = fl->fl_type;
1898 /* Report the first existing lock that would conflict with l.
1899 * This implements the F_GETLK command of fcntl().
1901 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
1903 struct file_lock file_lock;
1908 if (copy_from_user(&flock, l, sizeof(flock)))
1911 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1914 error = flock_to_posix_lock(filp, &file_lock, &flock);
1918 error = vfs_test_lock(filp, &file_lock);
1922 flock.l_type = file_lock.fl_type;
1923 if (file_lock.fl_type != F_UNLCK) {
1924 error = posix_lock_to_flock(&flock, &file_lock);
1929 if (!copy_to_user(l, &flock, sizeof(flock)))
1936 * vfs_lock_file - file byte range lock
1937 * @filp: The file to apply the lock to
1938 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1939 * @fl: The lock to be applied
1940 * @conf: Place to return a copy of the conflicting lock, if found.
1942 * A caller that doesn't care about the conflicting lock may pass NULL
1943 * as the final argument.
1945 * If the filesystem defines a private ->lock() method, then @conf will
1946 * be left unchanged; so a caller that cares should initialize it to
1947 * some acceptable default.
1949 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1950 * locks, the ->lock() interface may return asynchronously, before the lock has
1951 * been granted or denied by the underlying filesystem, if (and only if)
1952 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1953 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1954 * the request is for a blocking lock. When ->lock() does return asynchronously,
1955 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1956 * request completes.
1957 * If the request is for non-blocking lock the file system should return
1958 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1959 * with the result. If the request timed out the callback routine will return a
1960 * nonzero return code and the file system should release the lock. The file
1961 * system is also responsible to keep a corresponding posix lock when it
1962 * grants a lock so the VFS can find out which locks are locally held and do
1963 * the correct lock cleanup when required.
1964 * The underlying filesystem must not drop the kernel lock or call
1965 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1968 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1970 if (filp->f_op->lock)
1971 return filp->f_op->lock(filp, cmd, fl);
1973 return posix_lock_file(filp, fl, conf);
1975 EXPORT_SYMBOL_GPL(vfs_lock_file);
1977 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1978 struct file_lock *fl)
1982 error = security_file_lock(filp, fl->fl_type);
1987 error = vfs_lock_file(filp, cmd, fl, NULL);
1988 if (error != FILE_LOCK_DEFERRED)
1990 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1994 locks_delete_block(fl);
2001 /* Apply the lock described by l to an open file descriptor.
2002 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2004 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2005 struct flock __user *l)
2007 struct file_lock *file_lock = locks_alloc_lock();
2009 struct inode *inode;
2013 if (file_lock == NULL)
2017 * This might block, so we do it before checking the inode.
2020 if (copy_from_user(&flock, l, sizeof(flock)))
2023 inode = file_inode(filp);
2025 /* Don't allow mandatory locks on files that may be memory mapped
2028 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2034 error = flock_to_posix_lock(filp, file_lock, &flock);
2037 if (cmd == F_SETLKW) {
2038 file_lock->fl_flags |= FL_SLEEP;
2041 error = do_lock_file_wait(filp, cmd, file_lock);
2044 * Attempt to detect a close/fcntl race and recover by
2045 * releasing the lock that was just acquired.
2048 * we need that spin_lock here - it prevents reordering between
2049 * update of inode->i_flock and check for it done in close().
2050 * rcu_read_lock() wouldn't do.
2052 spin_lock(¤t->files->file_lock);
2054 spin_unlock(¤t->files->file_lock);
2055 if (!error && f != filp && flock.l_type != F_UNLCK) {
2056 flock.l_type = F_UNLCK;
2061 locks_free_lock(file_lock);
2065 #if BITS_PER_LONG == 32
2066 /* Report the first existing lock that would conflict with l.
2067 * This implements the F_GETLK command of fcntl().
2069 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
2071 struct file_lock file_lock;
2072 struct flock64 flock;
2076 if (copy_from_user(&flock, l, sizeof(flock)))
2079 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2082 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2086 error = vfs_test_lock(filp, &file_lock);
2090 flock.l_type = file_lock.fl_type;
2091 if (file_lock.fl_type != F_UNLCK)
2092 posix_lock_to_flock64(&flock, &file_lock);
2095 if (!copy_to_user(l, &flock, sizeof(flock)))
2102 /* Apply the lock described by l to an open file descriptor.
2103 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2105 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2106 struct flock64 __user *l)
2108 struct file_lock *file_lock = locks_alloc_lock();
2109 struct flock64 flock;
2110 struct inode *inode;
2114 if (file_lock == NULL)
2118 * This might block, so we do it before checking the inode.
2121 if (copy_from_user(&flock, l, sizeof(flock)))
2124 inode = file_inode(filp);
2126 /* Don't allow mandatory locks on files that may be memory mapped
2129 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2135 error = flock64_to_posix_lock(filp, file_lock, &flock);
2138 if (cmd == F_SETLKW64) {
2139 file_lock->fl_flags |= FL_SLEEP;
2142 error = do_lock_file_wait(filp, cmd, file_lock);
2145 * Attempt to detect a close/fcntl race and recover by
2146 * releasing the lock that was just acquired.
2148 spin_lock(¤t->files->file_lock);
2150 spin_unlock(¤t->files->file_lock);
2151 if (!error && f != filp && flock.l_type != F_UNLCK) {
2152 flock.l_type = F_UNLCK;
2157 locks_free_lock(file_lock);
2160 #endif /* BITS_PER_LONG == 32 */
2163 * This function is called when the file is being removed
2164 * from the task's fd array. POSIX locks belonging to this task
2165 * are deleted at this time.
2167 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2169 struct file_lock lock;
2172 * If there are no locks held on this file, we don't need to call
2173 * posix_lock_file(). Another process could be setting a lock on this
2174 * file at the same time, but we wouldn't remove that lock anyway.
2176 if (!file_inode(filp)->i_flock)
2179 lock.fl_type = F_UNLCK;
2180 lock.fl_flags = FL_POSIX | FL_CLOSE;
2182 lock.fl_end = OFFSET_MAX;
2183 lock.fl_owner = owner;
2184 lock.fl_pid = current->tgid;
2185 lock.fl_file = filp;
2187 lock.fl_lmops = NULL;
2189 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2191 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2192 lock.fl_ops->fl_release_private(&lock);
2195 EXPORT_SYMBOL(locks_remove_posix);
2198 * This function is called on the last close of an open file.
2200 void locks_remove_file(struct file *filp)
2202 struct inode * inode = file_inode(filp);
2203 struct file_lock *fl;
2204 struct file_lock **before;
2206 if (!inode->i_flock)
2209 if (filp->f_op->flock) {
2210 struct file_lock fl = {
2211 .fl_pid = current->tgid,
2213 .fl_flags = FL_FLOCK,
2215 .fl_end = OFFSET_MAX,
2217 filp->f_op->flock(filp, F_SETLKW, &fl);
2218 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2219 fl.fl_ops->fl_release_private(&fl);
2222 spin_lock(&inode->i_lock);
2223 before = &inode->i_flock;
2225 while ((fl = *before) != NULL) {
2226 if (fl->fl_file == filp) {
2228 lease_modify(before, F_UNLCK);
2233 * There's a leftover lock on the list of a type that
2234 * we didn't expect to see. Most likely a classic
2235 * POSIX lock that ended up not getting released
2236 * properly, or that raced onto the list somehow. Log
2237 * some info about it and then just remove it from
2241 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2242 MAJOR(inode->i_sb->s_dev),
2243 MINOR(inode->i_sb->s_dev), inode->i_ino,
2244 fl->fl_type, fl->fl_flags,
2245 fl->fl_start, fl->fl_end);
2247 locks_delete_lock(before);
2250 before = &fl->fl_next;
2252 spin_unlock(&inode->i_lock);
2256 * posix_unblock_lock - stop waiting for a file lock
2257 * @waiter: the lock which was waiting
2259 * lockd needs to block waiting for locks.
2262 posix_unblock_lock(struct file_lock *waiter)
2266 spin_lock(&blocked_lock_lock);
2267 if (waiter->fl_next)
2268 __locks_delete_block(waiter);
2271 spin_unlock(&blocked_lock_lock);
2274 EXPORT_SYMBOL(posix_unblock_lock);
2277 * vfs_cancel_lock - file byte range unblock lock
2278 * @filp: The file to apply the unblock to
2279 * @fl: The lock to be unblocked
2281 * Used by lock managers to cancel blocked requests
2283 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2285 if (filp->f_op->lock)
2286 return filp->f_op->lock(filp, F_CANCELLK, fl);
2290 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2292 #ifdef CONFIG_PROC_FS
2293 #include <linux/proc_fs.h>
2294 #include <linux/seq_file.h>
2296 struct locks_iterator {
2301 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2302 loff_t id, char *pfx)
2304 struct inode *inode = NULL;
2305 unsigned int fl_pid;
2308 fl_pid = pid_vnr(fl->fl_nspid);
2310 fl_pid = fl->fl_pid;
2312 if (fl->fl_file != NULL)
2313 inode = file_inode(fl->fl_file);
2315 seq_printf(f, "%lld:%s ", id, pfx);
2317 if (fl->fl_flags & FL_ACCESS)
2318 seq_printf(f, "ACCESS");
2319 else if (IS_FILE_PVT(fl))
2320 seq_printf(f, "FLPVT ");
2322 seq_printf(f, "POSIX ");
2324 seq_printf(f, " %s ",
2325 (inode == NULL) ? "*NOINODE*" :
2326 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2327 } else if (IS_FLOCK(fl)) {
2328 if (fl->fl_type & LOCK_MAND) {
2329 seq_printf(f, "FLOCK MSNFS ");
2331 seq_printf(f, "FLOCK ADVISORY ");
2333 } else if (IS_LEASE(fl)) {
2334 seq_printf(f, "LEASE ");
2335 if (lease_breaking(fl))
2336 seq_printf(f, "BREAKING ");
2337 else if (fl->fl_file)
2338 seq_printf(f, "ACTIVE ");
2340 seq_printf(f, "BREAKER ");
2342 seq_printf(f, "UNKNOWN UNKNOWN ");
2344 if (fl->fl_type & LOCK_MAND) {
2345 seq_printf(f, "%s ",
2346 (fl->fl_type & LOCK_READ)
2347 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2348 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2350 seq_printf(f, "%s ",
2351 (lease_breaking(fl))
2352 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2353 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2356 #ifdef WE_CAN_BREAK_LSLK_NOW
2357 seq_printf(f, "%d %s:%ld ", fl_pid,
2358 inode->i_sb->s_id, inode->i_ino);
2360 /* userspace relies on this representation of dev_t ;-( */
2361 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2362 MAJOR(inode->i_sb->s_dev),
2363 MINOR(inode->i_sb->s_dev), inode->i_ino);
2366 seq_printf(f, "%d <none>:0 ", fl_pid);
2369 if (fl->fl_end == OFFSET_MAX)
2370 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2372 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2374 seq_printf(f, "0 EOF\n");
2378 static int locks_show(struct seq_file *f, void *v)
2380 struct locks_iterator *iter = f->private;
2381 struct file_lock *fl, *bfl;
2383 fl = hlist_entry(v, struct file_lock, fl_link);
2385 lock_get_status(f, fl, iter->li_pos, "");
2387 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2388 lock_get_status(f, bfl, iter->li_pos, " ->");
2393 static void *locks_start(struct seq_file *f, loff_t *pos)
2394 __acquires(&blocked_lock_lock)
2396 struct locks_iterator *iter = f->private;
2398 iter->li_pos = *pos + 1;
2399 lg_global_lock(&file_lock_lglock);
2400 spin_lock(&blocked_lock_lock);
2401 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2404 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2406 struct locks_iterator *iter = f->private;
2409 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2412 static void locks_stop(struct seq_file *f, void *v)
2413 __releases(&blocked_lock_lock)
2415 spin_unlock(&blocked_lock_lock);
2416 lg_global_unlock(&file_lock_lglock);
2419 static const struct seq_operations locks_seq_operations = {
2420 .start = locks_start,
2426 static int locks_open(struct inode *inode, struct file *filp)
2428 return seq_open_private(filp, &locks_seq_operations,
2429 sizeof(struct locks_iterator));
2432 static const struct file_operations proc_locks_operations = {
2435 .llseek = seq_lseek,
2436 .release = seq_release_private,
2439 static int __init proc_locks_init(void)
2441 proc_create("locks", 0, NULL, &proc_locks_operations);
2444 module_init(proc_locks_init);
2448 * lock_may_read - checks that the region is free of locks
2449 * @inode: the inode that is being read
2450 * @start: the first byte to read
2451 * @len: the number of bytes to read
2453 * Emulates Windows locking requirements. Whole-file
2454 * mandatory locks (share modes) can prohibit a read and
2455 * byte-range POSIX locks can prohibit a read if they overlap.
2457 * N.B. this function is only ever called
2458 * from knfsd and ownership of locks is never checked.
2460 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2462 struct file_lock *fl;
2465 spin_lock(&inode->i_lock);
2466 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2468 if (fl->fl_type == F_RDLCK)
2470 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2472 } else if (IS_FLOCK(fl)) {
2473 if (!(fl->fl_type & LOCK_MAND))
2475 if (fl->fl_type & LOCK_READ)
2482 spin_unlock(&inode->i_lock);
2486 EXPORT_SYMBOL(lock_may_read);
2489 * lock_may_write - checks that the region is free of locks
2490 * @inode: the inode that is being written
2491 * @start: the first byte to write
2492 * @len: the number of bytes to write
2494 * Emulates Windows locking requirements. Whole-file
2495 * mandatory locks (share modes) can prohibit a write and
2496 * byte-range POSIX locks can prohibit a write if they overlap.
2498 * N.B. this function is only ever called
2499 * from knfsd and ownership of locks is never checked.
2501 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2503 struct file_lock *fl;
2506 spin_lock(&inode->i_lock);
2507 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2509 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2511 } else if (IS_FLOCK(fl)) {
2512 if (!(fl->fl_type & LOCK_MAND))
2514 if (fl->fl_type & LOCK_WRITE)
2521 spin_unlock(&inode->i_lock);
2525 EXPORT_SYMBOL(lock_may_write);
2527 static int __init filelock_init(void)
2531 filelock_cache = kmem_cache_create("file_lock_cache",
2532 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2534 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2536 for_each_possible_cpu(i)
2537 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2542 core_initcall(filelock_init);