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 #define CREATE_TRACE_POINTS
134 #include <trace/events/filelock.h>
136 #include <asm/uaccess.h>
138 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
139 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
140 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
141 #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
143 static bool lease_breaking(struct file_lock *fl)
145 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
148 static int target_leasetype(struct file_lock *fl)
150 if (fl->fl_flags & FL_UNLOCK_PENDING)
152 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
157 int leases_enable = 1;
158 int lease_break_time = 45;
160 #define for_each_lock(inode, lockp) \
161 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
164 * The global file_lock_list is only used for displaying /proc/locks, so we
165 * keep a list on each CPU, with each list protected by its own spinlock via
166 * the file_lock_lglock. Note that alterations to the list also require that
167 * the relevant i_lock is held.
169 DEFINE_STATIC_LGLOCK(file_lock_lglock);
170 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
173 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
174 * It is protected by blocked_lock_lock.
176 * We hash locks by lockowner in order to optimize searching for the lock a
177 * particular lockowner is waiting on.
179 * FIXME: make this value scale via some heuristic? We generally will want more
180 * buckets when we have more lockowners holding locks, but that's a little
181 * difficult to determine without knowing what the workload will look like.
183 #define BLOCKED_HASH_BITS 7
184 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
187 * This lock protects the blocked_hash. Generally, if you're accessing it, you
188 * want to be holding this lock.
190 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
191 * pointer for file_lock structures that are acting as lock requests (in
192 * contrast to those that are acting as records of acquired locks).
194 * Note that when we acquire this lock in order to change the above fields,
195 * we often hold the i_lock as well. In certain cases, when reading the fields
196 * protected by this lock, we can skip acquiring it iff we already hold the
199 * In particular, adding an entry to the fl_block list requires that you hold
200 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
201 * an entry from the list however only requires the file_lock_lock.
203 static DEFINE_SPINLOCK(blocked_lock_lock);
205 static struct kmem_cache *filelock_cache __read_mostly;
207 static void locks_init_lock_heads(struct file_lock *fl)
209 INIT_HLIST_NODE(&fl->fl_link);
210 INIT_LIST_HEAD(&fl->fl_block);
211 init_waitqueue_head(&fl->fl_wait);
214 /* Allocate an empty lock structure. */
215 struct file_lock *locks_alloc_lock(void)
217 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
220 locks_init_lock_heads(fl);
224 EXPORT_SYMBOL_GPL(locks_alloc_lock);
226 void locks_release_private(struct file_lock *fl)
229 if (fl->fl_ops->fl_release_private)
230 fl->fl_ops->fl_release_private(fl);
235 if (fl->fl_lmops->lm_put_owner)
236 fl->fl_lmops->lm_put_owner(fl);
240 EXPORT_SYMBOL_GPL(locks_release_private);
242 /* Free a lock which is not in use. */
243 void locks_free_lock(struct file_lock *fl)
245 BUG_ON(waitqueue_active(&fl->fl_wait));
246 BUG_ON(!list_empty(&fl->fl_block));
247 BUG_ON(!hlist_unhashed(&fl->fl_link));
249 locks_release_private(fl);
250 kmem_cache_free(filelock_cache, fl);
252 EXPORT_SYMBOL(locks_free_lock);
255 locks_dispose_list(struct list_head *dispose)
257 struct file_lock *fl;
259 while (!list_empty(dispose)) {
260 fl = list_first_entry(dispose, struct file_lock, fl_block);
261 list_del_init(&fl->fl_block);
266 void locks_init_lock(struct file_lock *fl)
268 memset(fl, 0, sizeof(struct file_lock));
269 locks_init_lock_heads(fl);
272 EXPORT_SYMBOL(locks_init_lock);
275 * Initialize a new lock from an existing file_lock structure.
277 void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
279 new->fl_owner = fl->fl_owner;
280 new->fl_pid = fl->fl_pid;
282 new->fl_flags = fl->fl_flags;
283 new->fl_type = fl->fl_type;
284 new->fl_start = fl->fl_start;
285 new->fl_end = fl->fl_end;
286 new->fl_lmops = fl->fl_lmops;
290 if (fl->fl_lmops->lm_get_owner)
291 fl->fl_lmops->lm_get_owner(new, fl);
294 EXPORT_SYMBOL(locks_copy_conflock);
296 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
298 /* "new" must be a freshly-initialized lock */
299 WARN_ON_ONCE(new->fl_ops);
301 locks_copy_conflock(new, fl);
303 new->fl_file = fl->fl_file;
304 new->fl_ops = fl->fl_ops;
307 if (fl->fl_ops->fl_copy_lock)
308 fl->fl_ops->fl_copy_lock(new, fl);
312 EXPORT_SYMBOL(locks_copy_lock);
314 static inline int flock_translate_cmd(int cmd) {
316 return cmd & (LOCK_MAND | LOCK_RW);
328 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
329 static int flock_make_lock(struct file *filp, struct file_lock **lock,
332 struct file_lock *fl;
333 int type = flock_translate_cmd(cmd);
337 fl = locks_alloc_lock();
343 fl->fl_pid = current->tgid;
344 fl->fl_flags = FL_FLOCK;
346 fl->fl_end = OFFSET_MAX;
352 static int assign_type(struct file_lock *fl, long type)
366 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
369 switch (l->l_whence) {
374 fl->fl_start = filp->f_pos;
377 fl->fl_start = i_size_read(file_inode(filp));
382 if (l->l_start > OFFSET_MAX - fl->fl_start)
384 fl->fl_start += l->l_start;
385 if (fl->fl_start < 0)
388 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
389 POSIX-2001 defines it. */
391 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
393 fl->fl_end = fl->fl_start + l->l_len - 1;
395 } else if (l->l_len < 0) {
396 if (fl->fl_start + l->l_len < 0)
398 fl->fl_end = fl->fl_start - 1;
399 fl->fl_start += l->l_len;
401 fl->fl_end = OFFSET_MAX;
403 fl->fl_owner = current->files;
404 fl->fl_pid = current->tgid;
406 fl->fl_flags = FL_POSIX;
410 return assign_type(fl, l->l_type);
413 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
416 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
419 struct flock64 ll = {
421 .l_whence = l->l_whence,
422 .l_start = l->l_start,
426 return flock64_to_posix_lock(filp, fl, &ll);
429 /* default lease lock manager operations */
430 static void lease_break_callback(struct file_lock *fl)
432 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
436 lease_setup(struct file_lock *fl, void **priv)
438 struct file *filp = fl->fl_file;
439 struct fasync_struct *fa = *priv;
442 * fasync_insert_entry() returns the old entry if any. If there was no
443 * old entry, then it used "priv" and inserted it into the fasync list.
444 * Clear the pointer to indicate that it shouldn't be freed.
446 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
449 __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
452 static const struct lock_manager_operations lease_manager_ops = {
453 .lm_break = lease_break_callback,
454 .lm_change = lease_modify,
455 .lm_setup = lease_setup,
459 * Initialize a lease, use the default lock manager operations
461 static int lease_init(struct file *filp, long type, struct file_lock *fl)
463 if (assign_type(fl, type) != 0)
466 fl->fl_owner = current->files;
467 fl->fl_pid = current->tgid;
470 fl->fl_flags = FL_LEASE;
472 fl->fl_end = OFFSET_MAX;
474 fl->fl_lmops = &lease_manager_ops;
478 /* Allocate a file_lock initialised to this type of lease */
479 static struct file_lock *lease_alloc(struct file *filp, long type)
481 struct file_lock *fl = locks_alloc_lock();
485 return ERR_PTR(error);
487 error = lease_init(filp, type, fl);
490 return ERR_PTR(error);
495 /* Check if two locks overlap each other.
497 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
499 return ((fl1->fl_end >= fl2->fl_start) &&
500 (fl2->fl_end >= fl1->fl_start));
504 * Check whether two locks have the same owner.
506 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
508 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
509 return fl2->fl_lmops == fl1->fl_lmops &&
510 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
511 return fl1->fl_owner == fl2->fl_owner;
514 /* Must be called with the i_lock held! */
515 static void locks_insert_global_locks(struct file_lock *fl)
517 lg_local_lock(&file_lock_lglock);
518 fl->fl_link_cpu = smp_processor_id();
519 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
520 lg_local_unlock(&file_lock_lglock);
523 /* Must be called with the i_lock held! */
524 static void locks_delete_global_locks(struct file_lock *fl)
527 * Avoid taking lock if already unhashed. This is safe since this check
528 * is done while holding the i_lock, and new insertions into the list
529 * also require that it be held.
531 if (hlist_unhashed(&fl->fl_link))
533 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
534 hlist_del_init(&fl->fl_link);
535 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
539 posix_owner_key(struct file_lock *fl)
541 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
542 return fl->fl_lmops->lm_owner_key(fl);
543 return (unsigned long)fl->fl_owner;
546 static void locks_insert_global_blocked(struct file_lock *waiter)
548 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
551 static void locks_delete_global_blocked(struct file_lock *waiter)
553 hash_del(&waiter->fl_link);
556 /* Remove waiter from blocker's block list.
557 * When blocker ends up pointing to itself then the list is empty.
559 * Must be called with blocked_lock_lock held.
561 static void __locks_delete_block(struct file_lock *waiter)
563 locks_delete_global_blocked(waiter);
564 list_del_init(&waiter->fl_block);
565 waiter->fl_next = NULL;
568 static void locks_delete_block(struct file_lock *waiter)
570 spin_lock(&blocked_lock_lock);
571 __locks_delete_block(waiter);
572 spin_unlock(&blocked_lock_lock);
575 /* Insert waiter into blocker's block list.
576 * We use a circular list so that processes can be easily woken up in
577 * the order they blocked. The documentation doesn't require this but
578 * it seems like the reasonable thing to do.
580 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
581 * list itself is protected by the blocked_lock_lock, but by ensuring that the
582 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
583 * in some cases when we see that the fl_block list is empty.
585 static void __locks_insert_block(struct file_lock *blocker,
586 struct file_lock *waiter)
588 BUG_ON(!list_empty(&waiter->fl_block));
589 waiter->fl_next = blocker;
590 list_add_tail(&waiter->fl_block, &blocker->fl_block);
591 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
592 locks_insert_global_blocked(waiter);
595 /* Must be called with i_lock held. */
596 static void locks_insert_block(struct file_lock *blocker,
597 struct file_lock *waiter)
599 spin_lock(&blocked_lock_lock);
600 __locks_insert_block(blocker, waiter);
601 spin_unlock(&blocked_lock_lock);
605 * Wake up processes blocked waiting for blocker.
607 * Must be called with the inode->i_lock held!
609 static void locks_wake_up_blocks(struct file_lock *blocker)
612 * Avoid taking global lock if list is empty. This is safe since new
613 * blocked requests are only added to the list under the i_lock, and
614 * the i_lock is always held here. Note that removal from the fl_block
615 * list does not require the i_lock, so we must recheck list_empty()
616 * after acquiring the blocked_lock_lock.
618 if (list_empty(&blocker->fl_block))
621 spin_lock(&blocked_lock_lock);
622 while (!list_empty(&blocker->fl_block)) {
623 struct file_lock *waiter;
625 waiter = list_first_entry(&blocker->fl_block,
626 struct file_lock, fl_block);
627 __locks_delete_block(waiter);
628 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
629 waiter->fl_lmops->lm_notify(waiter);
631 wake_up(&waiter->fl_wait);
633 spin_unlock(&blocked_lock_lock);
636 /* Insert file lock fl into an inode's lock list at the position indicated
637 * by pos. At the same time add the lock to the global file lock list.
639 * Must be called with the i_lock held!
641 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
643 fl->fl_nspid = get_pid(task_tgid(current));
645 /* insert into file's list */
649 locks_insert_global_locks(fl);
653 * locks_delete_lock - Delete a lock and then free it.
654 * @thisfl_p: pointer that points to the fl_next field of the previous
655 * inode->i_flock list entry
657 * Unlink a lock from all lists and free the namespace reference, but don't
658 * free it yet. Wake up processes that are blocked waiting for this lock and
659 * notify the FS that the lock has been cleared.
661 * Must be called with the i_lock held!
663 static void locks_unlink_lock(struct file_lock **thisfl_p)
665 struct file_lock *fl = *thisfl_p;
667 locks_delete_global_locks(fl);
669 *thisfl_p = fl->fl_next;
673 put_pid(fl->fl_nspid);
677 locks_wake_up_blocks(fl);
681 * Unlink a lock from all lists and free it.
683 * Must be called with i_lock held!
685 static void locks_delete_lock(struct file_lock **thisfl_p,
686 struct list_head *dispose)
688 struct file_lock *fl = *thisfl_p;
690 locks_unlink_lock(thisfl_p);
692 list_add(&fl->fl_block, dispose);
697 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
698 * checks for shared/exclusive status of overlapping locks.
700 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
702 if (sys_fl->fl_type == F_WRLCK)
704 if (caller_fl->fl_type == F_WRLCK)
709 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
710 * checking before calling the locks_conflict().
712 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
714 /* POSIX locks owned by the same process do not conflict with
717 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
720 /* Check whether they overlap */
721 if (!locks_overlap(caller_fl, sys_fl))
724 return (locks_conflict(caller_fl, sys_fl));
727 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
728 * checking before calling the locks_conflict().
730 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
732 /* FLOCK locks referring to the same filp do not conflict with
735 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
737 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
740 return (locks_conflict(caller_fl, sys_fl));
744 posix_test_lock(struct file *filp, struct file_lock *fl)
746 struct file_lock *cfl;
747 struct inode *inode = file_inode(filp);
749 spin_lock(&inode->i_lock);
750 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
753 if (posix_locks_conflict(fl, cfl))
757 locks_copy_conflock(fl, cfl);
759 fl->fl_pid = pid_vnr(cfl->fl_nspid);
761 fl->fl_type = F_UNLCK;
762 spin_unlock(&inode->i_lock);
765 EXPORT_SYMBOL(posix_test_lock);
768 * Deadlock detection:
770 * We attempt to detect deadlocks that are due purely to posix file
773 * We assume that a task can be waiting for at most one lock at a time.
774 * So for any acquired lock, the process holding that lock may be
775 * waiting on at most one other lock. That lock in turns may be held by
776 * someone waiting for at most one other lock. Given a requested lock
777 * caller_fl which is about to wait for a conflicting lock block_fl, we
778 * follow this chain of waiters to ensure we are not about to create a
781 * Since we do this before we ever put a process to sleep on a lock, we
782 * are ensured that there is never a cycle; that is what guarantees that
783 * the while() loop in posix_locks_deadlock() eventually completes.
785 * Note: the above assumption may not be true when handling lock
786 * requests from a broken NFS client. It may also fail in the presence
787 * of tasks (such as posix threads) sharing the same open file table.
788 * To handle those cases, we just bail out after a few iterations.
790 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
791 * Because the owner is not even nominally tied to a thread of
792 * execution, the deadlock detection below can't reasonably work well. Just
795 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
796 * locks that just checks for the case where two tasks are attempting to
797 * upgrade from read to write locks on the same inode.
800 #define MAX_DEADLK_ITERATIONS 10
802 /* Find a lock that the owner of the given block_fl is blocking on. */
803 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
805 struct file_lock *fl;
807 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
808 if (posix_same_owner(fl, block_fl))
814 /* Must be called with the blocked_lock_lock held! */
815 static int posix_locks_deadlock(struct file_lock *caller_fl,
816 struct file_lock *block_fl)
821 * This deadlock detector can't reasonably detect deadlocks with
822 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
824 if (IS_OFDLCK(caller_fl))
827 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
828 if (i++ > MAX_DEADLK_ITERATIONS)
830 if (posix_same_owner(caller_fl, block_fl))
836 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
837 * after any leases, but before any posix locks.
839 * Note that if called with an FL_EXISTS argument, the caller may determine
840 * whether or not a lock was successfully freed by testing the return
843 static int flock_lock_file(struct file *filp, struct file_lock *request)
845 struct file_lock *new_fl = NULL;
846 struct file_lock **before;
847 struct inode * inode = file_inode(filp);
852 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
853 new_fl = locks_alloc_lock();
858 spin_lock(&inode->i_lock);
859 if (request->fl_flags & FL_ACCESS)
862 for_each_lock(inode, before) {
863 struct file_lock *fl = *before;
868 if (filp != fl->fl_file)
870 if (request->fl_type == fl->fl_type)
873 locks_delete_lock(before, &dispose);
877 if (request->fl_type == F_UNLCK) {
878 if ((request->fl_flags & FL_EXISTS) && !found)
884 * If a higher-priority process was blocked on the old file lock,
885 * give it the opportunity to lock the file.
888 spin_unlock(&inode->i_lock);
890 spin_lock(&inode->i_lock);
894 for_each_lock(inode, before) {
895 struct file_lock *fl = *before;
900 if (!flock_locks_conflict(request, fl))
903 if (!(request->fl_flags & FL_SLEEP))
905 error = FILE_LOCK_DEFERRED;
906 locks_insert_block(fl, request);
909 if (request->fl_flags & FL_ACCESS)
911 locks_copy_lock(new_fl, request);
912 locks_insert_lock(before, new_fl);
917 spin_unlock(&inode->i_lock);
919 locks_free_lock(new_fl);
920 locks_dispose_list(&dispose);
924 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
926 struct file_lock *fl;
927 struct file_lock *new_fl = NULL;
928 struct file_lock *new_fl2 = NULL;
929 struct file_lock *left = NULL;
930 struct file_lock *right = NULL;
931 struct file_lock **before;
937 * We may need two file_lock structures for this operation,
938 * so we get them in advance to avoid races.
940 * In some cases we can be sure, that no new locks will be needed
942 if (!(request->fl_flags & FL_ACCESS) &&
943 (request->fl_type != F_UNLCK ||
944 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
945 new_fl = locks_alloc_lock();
946 new_fl2 = locks_alloc_lock();
949 spin_lock(&inode->i_lock);
951 * New lock request. Walk all POSIX locks and look for conflicts. If
952 * there are any, either return error or put the request on the
953 * blocker's list of waiters and the global blocked_hash.
955 if (request->fl_type != F_UNLCK) {
956 for_each_lock(inode, before) {
960 if (!posix_locks_conflict(request, fl))
963 locks_copy_conflock(conflock, fl);
965 if (!(request->fl_flags & FL_SLEEP))
968 * Deadlock detection and insertion into the blocked
969 * locks list must be done while holding the same lock!
972 spin_lock(&blocked_lock_lock);
973 if (likely(!posix_locks_deadlock(request, fl))) {
974 error = FILE_LOCK_DEFERRED;
975 __locks_insert_block(fl, request);
977 spin_unlock(&blocked_lock_lock);
982 /* If we're just looking for a conflict, we're done. */
984 if (request->fl_flags & FL_ACCESS)
988 * Find the first old lock with the same owner as the new lock.
991 before = &inode->i_flock;
993 /* First skip locks owned by other processes. */
994 while ((fl = *before) && (!IS_POSIX(fl) ||
995 !posix_same_owner(request, fl))) {
996 before = &fl->fl_next;
999 /* Process locks with this owner. */
1000 while ((fl = *before) && posix_same_owner(request, fl)) {
1001 /* Detect adjacent or overlapping regions (if same lock type)
1003 if (request->fl_type == fl->fl_type) {
1004 /* In all comparisons of start vs end, use
1005 * "start - 1" rather than "end + 1". If end
1006 * is OFFSET_MAX, end + 1 will become negative.
1008 if (fl->fl_end < request->fl_start - 1)
1010 /* If the next lock in the list has entirely bigger
1011 * addresses than the new one, insert the lock here.
1013 if (fl->fl_start - 1 > request->fl_end)
1016 /* If we come here, the new and old lock are of the
1017 * same type and adjacent or overlapping. Make one
1018 * lock yielding from the lower start address of both
1019 * locks to the higher end address.
1021 if (fl->fl_start > request->fl_start)
1022 fl->fl_start = request->fl_start;
1024 request->fl_start = fl->fl_start;
1025 if (fl->fl_end < request->fl_end)
1026 fl->fl_end = request->fl_end;
1028 request->fl_end = fl->fl_end;
1030 locks_delete_lock(before, &dispose);
1037 /* Processing for different lock types is a bit
1040 if (fl->fl_end < request->fl_start)
1042 if (fl->fl_start > request->fl_end)
1044 if (request->fl_type == F_UNLCK)
1046 if (fl->fl_start < request->fl_start)
1048 /* If the next lock in the list has a higher end
1049 * address than the new one, insert the new one here.
1051 if (fl->fl_end > request->fl_end) {
1055 if (fl->fl_start >= request->fl_start) {
1056 /* The new lock completely replaces an old
1057 * one (This may happen several times).
1060 locks_delete_lock(before, &dispose);
1064 * Replace the old lock with new_fl, and
1065 * remove the old one. It's safe to do the
1066 * insert here since we know that we won't be
1067 * using new_fl later, and that the lock is
1068 * just replacing an existing lock.
1073 locks_copy_lock(new_fl, request);
1076 locks_delete_lock(before, &dispose);
1077 locks_insert_lock(before, request);
1081 /* Go on to next lock.
1084 before = &fl->fl_next;
1088 * The above code only modifies existing locks in case of merging or
1089 * replacing. If new lock(s) need to be inserted all modifications are
1090 * done below this, so it's safe yet to bail out.
1092 error = -ENOLCK; /* "no luck" */
1093 if (right && left == right && !new_fl2)
1098 if (request->fl_type == F_UNLCK) {
1099 if (request->fl_flags & FL_EXISTS)
1108 locks_copy_lock(new_fl, request);
1109 locks_insert_lock(before, new_fl);
1113 if (left == right) {
1114 /* The new lock breaks the old one in two pieces,
1115 * so we have to use the second new lock.
1119 locks_copy_lock(left, right);
1120 locks_insert_lock(before, left);
1122 right->fl_start = request->fl_end + 1;
1123 locks_wake_up_blocks(right);
1126 left->fl_end = request->fl_start - 1;
1127 locks_wake_up_blocks(left);
1130 spin_unlock(&inode->i_lock);
1132 * Free any unused locks.
1135 locks_free_lock(new_fl);
1137 locks_free_lock(new_fl2);
1138 locks_dispose_list(&dispose);
1143 * posix_lock_file - Apply a POSIX-style lock to a file
1144 * @filp: The file to apply the lock to
1145 * @fl: The lock to be applied
1146 * @conflock: Place to return a copy of the conflicting lock, if found.
1148 * Add a POSIX style lock to a file.
1149 * We merge adjacent & overlapping locks whenever possible.
1150 * POSIX locks are sorted by owner task, then by starting address
1152 * Note that if called with an FL_EXISTS argument, the caller may determine
1153 * whether or not a lock was successfully freed by testing the return
1154 * value for -ENOENT.
1156 int posix_lock_file(struct file *filp, struct file_lock *fl,
1157 struct file_lock *conflock)
1159 return __posix_lock_file(file_inode(filp), fl, conflock);
1161 EXPORT_SYMBOL(posix_lock_file);
1164 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1165 * @filp: The file to apply the lock to
1166 * @fl: The lock to be applied
1168 * Add a POSIX style lock to a file.
1169 * We merge adjacent & overlapping locks whenever possible.
1170 * POSIX locks are sorted by owner task, then by starting address
1172 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1177 error = posix_lock_file(filp, fl, NULL);
1178 if (error != FILE_LOCK_DEFERRED)
1180 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1184 locks_delete_block(fl);
1189 EXPORT_SYMBOL(posix_lock_file_wait);
1192 * locks_mandatory_locked - Check for an active lock
1193 * @file: the file to check
1195 * Searches the inode's list of locks to find any POSIX locks which conflict.
1196 * This function is called from locks_verify_locked() only.
1198 int locks_mandatory_locked(struct file *file)
1200 struct inode *inode = file_inode(file);
1201 struct file_lock *fl;
1204 * Search the lock list for this inode for any POSIX locks.
1206 spin_lock(&inode->i_lock);
1207 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1210 if (fl->fl_owner != current->files &&
1211 fl->fl_owner != file)
1214 spin_unlock(&inode->i_lock);
1215 return fl ? -EAGAIN : 0;
1219 * locks_mandatory_area - Check for a conflicting lock
1220 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1222 * @inode: the file to check
1223 * @filp: how the file was opened (if it was)
1224 * @offset: start of area to check
1225 * @count: length of area to check
1227 * Searches the inode's list of locks to find any POSIX locks which conflict.
1228 * This function is called from rw_verify_area() and
1229 * locks_verify_truncate().
1231 int locks_mandatory_area(int read_write, struct inode *inode,
1232 struct file *filp, loff_t offset,
1235 struct file_lock fl;
1239 locks_init_lock(&fl);
1240 fl.fl_pid = current->tgid;
1242 fl.fl_flags = FL_POSIX | FL_ACCESS;
1243 if (filp && !(filp->f_flags & O_NONBLOCK))
1245 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1246 fl.fl_start = offset;
1247 fl.fl_end = offset + count - 1;
1252 fl.fl_flags &= ~FL_SLEEP;
1253 error = __posix_lock_file(inode, &fl, NULL);
1259 fl.fl_flags |= FL_SLEEP;
1260 fl.fl_owner = current->files;
1261 error = __posix_lock_file(inode, &fl, NULL);
1262 if (error != FILE_LOCK_DEFERRED)
1264 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1267 * If we've been sleeping someone might have
1268 * changed the permissions behind our back.
1270 if (__mandatory_lock(inode))
1274 locks_delete_block(&fl);
1281 EXPORT_SYMBOL(locks_mandatory_area);
1283 static void lease_clear_pending(struct file_lock *fl, int arg)
1287 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1290 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1294 /* We already had a lease on this file; just change its type */
1295 int lease_modify(struct file_lock **before, int arg, struct list_head *dispose)
1297 struct file_lock *fl = *before;
1298 int error = assign_type(fl, arg);
1302 lease_clear_pending(fl, arg);
1303 locks_wake_up_blocks(fl);
1304 if (arg == F_UNLCK) {
1305 struct file *filp = fl->fl_file;
1308 filp->f_owner.signum = 0;
1309 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1310 if (fl->fl_fasync != NULL) {
1311 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1312 fl->fl_fasync = NULL;
1314 locks_delete_lock(before, dispose);
1318 EXPORT_SYMBOL(lease_modify);
1320 static bool past_time(unsigned long then)
1323 /* 0 is a special value meaning "this never expires": */
1325 return time_after(jiffies, then);
1328 static void time_out_leases(struct inode *inode, struct list_head *dispose)
1330 struct file_lock **before;
1331 struct file_lock *fl;
1333 lockdep_assert_held(&inode->i_lock);
1335 before = &inode->i_flock;
1336 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1337 trace_time_out_leases(inode, fl);
1338 if (past_time(fl->fl_downgrade_time))
1339 lease_modify(before, F_RDLCK, dispose);
1340 if (past_time(fl->fl_break_time))
1341 lease_modify(before, F_UNLCK, dispose);
1342 if (fl == *before) /* lease_modify may have freed fl */
1343 before = &fl->fl_next;
1347 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1349 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1351 return locks_conflict(breaker, lease);
1355 any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1357 struct file_lock *fl;
1359 lockdep_assert_held(&inode->i_lock);
1361 for (fl = inode->i_flock ; fl && IS_LEASE(fl); fl = fl->fl_next) {
1362 if (leases_conflict(fl, breaker))
1369 * __break_lease - revoke all outstanding leases on file
1370 * @inode: the inode of the file to return
1371 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1373 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1376 * break_lease (inlined for speed) has checked there already is at least
1377 * some kind of lock (maybe a lease) on this file. Leases are broken on
1378 * a call to open() or truncate(). This function can sleep unless you
1379 * specified %O_NONBLOCK to your open().
1381 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1384 struct file_lock *new_fl;
1385 struct file_lock *fl;
1386 unsigned long break_time;
1387 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1390 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1392 return PTR_ERR(new_fl);
1393 new_fl->fl_flags = type;
1395 spin_lock(&inode->i_lock);
1397 time_out_leases(inode, &dispose);
1399 if (!any_leases_conflict(inode, new_fl))
1403 if (lease_break_time > 0) {
1404 break_time = jiffies + lease_break_time * HZ;
1405 if (break_time == 0)
1406 break_time++; /* so that 0 means no break time */
1409 for (fl = inode->i_flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1410 if (!leases_conflict(fl, new_fl))
1413 if (fl->fl_flags & FL_UNLOCK_PENDING)
1415 fl->fl_flags |= FL_UNLOCK_PENDING;
1416 fl->fl_break_time = break_time;
1418 if (lease_breaking(inode->i_flock))
1420 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1421 fl->fl_downgrade_time = break_time;
1423 fl->fl_lmops->lm_break(fl);
1426 if (mode & O_NONBLOCK) {
1427 trace_break_lease_noblock(inode, new_fl);
1428 error = -EWOULDBLOCK;
1433 break_time = inode->i_flock->fl_break_time;
1434 if (break_time != 0)
1435 break_time -= jiffies;
1436 if (break_time == 0)
1438 locks_insert_block(inode->i_flock, new_fl);
1439 trace_break_lease_block(inode, new_fl);
1440 spin_unlock(&inode->i_lock);
1441 locks_dispose_list(&dispose);
1442 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1443 !new_fl->fl_next, break_time);
1444 spin_lock(&inode->i_lock);
1445 trace_break_lease_unblock(inode, new_fl);
1446 locks_delete_block(new_fl);
1449 * Wait for the next conflicting lease that has not been
1453 time_out_leases(inode, &dispose);
1454 if (any_leases_conflict(inode, new_fl))
1461 spin_unlock(&inode->i_lock);
1462 locks_dispose_list(&dispose);
1463 locks_free_lock(new_fl);
1467 EXPORT_SYMBOL(__break_lease);
1470 * lease_get_mtime - get the last modified time of an inode
1472 * @time: pointer to a timespec which will contain the last modified time
1474 * This is to force NFS clients to flush their caches for files with
1475 * exclusive leases. The justification is that if someone has an
1476 * exclusive lease, then they could be modifying it.
1478 void lease_get_mtime(struct inode *inode, struct timespec *time)
1480 bool has_lease = false;
1481 struct file_lock *flock;
1483 if (inode->i_flock) {
1484 spin_lock(&inode->i_lock);
1485 flock = inode->i_flock;
1486 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1488 spin_unlock(&inode->i_lock);
1492 *time = current_fs_time(inode->i_sb);
1494 *time = inode->i_mtime;
1497 EXPORT_SYMBOL(lease_get_mtime);
1500 * fcntl_getlease - Enquire what lease is currently active
1503 * The value returned by this function will be one of
1504 * (if no lease break is pending):
1506 * %F_RDLCK to indicate a shared lease is held.
1508 * %F_WRLCK to indicate an exclusive lease is held.
1510 * %F_UNLCK to indicate no lease is held.
1512 * (if a lease break is pending):
1514 * %F_RDLCK to indicate an exclusive lease needs to be
1515 * changed to a shared lease (or removed).
1517 * %F_UNLCK to indicate the lease needs to be removed.
1519 * XXX: sfr & willy disagree over whether F_INPROGRESS
1520 * should be returned to userspace.
1522 int fcntl_getlease(struct file *filp)
1524 struct file_lock *fl;
1525 struct inode *inode = file_inode(filp);
1529 spin_lock(&inode->i_lock);
1530 time_out_leases(file_inode(filp), &dispose);
1531 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1533 if (fl->fl_file == filp) {
1534 type = target_leasetype(fl);
1538 spin_unlock(&inode->i_lock);
1539 locks_dispose_list(&dispose);
1544 * check_conflicting_open - see if the given dentry points to a file that has
1545 * an existing open that would conflict with the
1547 * @dentry: dentry to check
1548 * @arg: type of lease that we're trying to acquire
1550 * Check to see if there's an existing open fd on this file that would
1551 * conflict with the lease we're trying to set.
1554 check_conflicting_open(const struct dentry *dentry, const long arg)
1557 struct inode *inode = dentry->d_inode;
1559 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1562 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1563 (atomic_read(&inode->i_count) > 1)))
1570 generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1572 struct file_lock *fl, **before, **my_before = NULL, *lease;
1573 struct dentry *dentry = filp->f_path.dentry;
1574 struct inode *inode = dentry->d_inode;
1575 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1580 trace_generic_add_lease(inode, lease);
1583 * In the delegation case we need mutual exclusion with
1584 * a number of operations that take the i_mutex. We trylock
1585 * because delegations are an optional optimization, and if
1586 * there's some chance of a conflict--we'd rather not
1587 * bother, maybe that's a sign this just isn't a good file to
1588 * hand out a delegation on.
1590 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1593 if (is_deleg && arg == F_WRLCK) {
1594 /* Write delegations are not currently supported: */
1595 mutex_unlock(&inode->i_mutex);
1600 spin_lock(&inode->i_lock);
1601 time_out_leases(inode, &dispose);
1602 error = check_conflicting_open(dentry, arg);
1607 * At this point, we know that if there is an exclusive
1608 * lease on this file, then we hold it on this filp
1609 * (otherwise our open of this file would have blocked).
1610 * And if we are trying to acquire an exclusive lease,
1611 * then the file is not open by anyone (including us)
1612 * except for this filp.
1615 for (before = &inode->i_flock;
1616 ((fl = *before) != NULL) && IS_LEASE(fl);
1617 before = &fl->fl_next) {
1618 if (fl->fl_file == filp) {
1623 * No exclusive leases if someone else has a lease on
1629 * Modifying our existing lease is OK, but no getting a
1630 * new lease if someone else is opening for write:
1632 if (fl->fl_flags & FL_UNLOCK_PENDING)
1636 if (my_before != NULL) {
1638 error = lease->fl_lmops->lm_change(my_before, arg, &dispose);
1648 locks_insert_lock(before, lease);
1650 * The check in break_lease() is lockless. It's possible for another
1651 * open to race in after we did the earlier check for a conflicting
1652 * open but before the lease was inserted. Check again for a
1653 * conflicting open and cancel the lease if there is one.
1655 * We also add a barrier here to ensure that the insertion of the lock
1656 * precedes these checks.
1659 error = check_conflicting_open(dentry, arg);
1664 if (lease->fl_lmops->lm_setup)
1665 lease->fl_lmops->lm_setup(lease, priv);
1667 spin_unlock(&inode->i_lock);
1668 locks_dispose_list(&dispose);
1670 mutex_unlock(&inode->i_mutex);
1671 if (!error && !my_before)
1675 locks_unlink_lock(before);
1679 static int generic_delete_lease(struct file *filp)
1681 int error = -EAGAIN;
1682 struct file_lock *fl, **before;
1683 struct dentry *dentry = filp->f_path.dentry;
1684 struct inode *inode = dentry->d_inode;
1687 spin_lock(&inode->i_lock);
1688 time_out_leases(inode, &dispose);
1689 for (before = &inode->i_flock;
1690 ((fl = *before) != NULL) && IS_LEASE(fl);
1691 before = &fl->fl_next) {
1692 if (fl->fl_file == filp)
1695 trace_generic_delete_lease(inode, fl);
1697 error = fl->fl_lmops->lm_change(before, F_UNLCK, &dispose);
1698 spin_unlock(&inode->i_lock);
1699 locks_dispose_list(&dispose);
1704 * generic_setlease - sets a lease on an open file
1705 * @filp: file pointer
1706 * @arg: type of lease to obtain
1707 * @flp: input - file_lock to use, output - file_lock inserted
1708 * @priv: private data for lm_setup (may be NULL if lm_setup
1709 * doesn't require it)
1711 * The (input) flp->fl_lmops->lm_break function is required
1714 int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1717 struct dentry *dentry = filp->f_path.dentry;
1718 struct inode *inode = dentry->d_inode;
1721 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1723 if (!S_ISREG(inode->i_mode))
1725 error = security_file_lock(filp, arg);
1731 return generic_delete_lease(filp);
1734 if (!(*flp)->fl_lmops->lm_break) {
1738 return generic_add_lease(filp, arg, flp, priv);
1743 EXPORT_SYMBOL(generic_setlease);
1746 * vfs_setlease - sets a lease on an open file
1747 * @filp: file pointer
1748 * @arg: type of lease to obtain
1749 * @lease: file_lock to use when adding a lease
1750 * @priv: private info for lm_setup when adding a lease (may be
1751 * NULL if lm_setup doesn't require it)
1753 * Call this to establish a lease on the file. The "lease" argument is not
1754 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1755 * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1756 * if not, this function will return -ENOLCK (and generate a scary-looking
1759 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1760 * may be NULL if the lm_setup operation doesn't require it.
1763 vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1765 if (filp->f_op->setlease)
1766 return filp->f_op->setlease(filp, arg, lease, priv);
1768 return generic_setlease(filp, arg, lease, priv);
1770 EXPORT_SYMBOL_GPL(vfs_setlease);
1772 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1774 struct file_lock *fl;
1775 struct fasync_struct *new;
1778 fl = lease_alloc(filp, arg);
1782 new = fasync_alloc();
1784 locks_free_lock(fl);
1789 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1791 locks_free_lock(fl);
1798 * fcntl_setlease - sets a lease on an open file
1799 * @fd: open file descriptor
1800 * @filp: file pointer
1801 * @arg: type of lease to obtain
1803 * Call this fcntl to establish a lease on the file.
1804 * Note that you also need to call %F_SETSIG to
1805 * receive a signal when the lease is broken.
1807 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1810 return vfs_setlease(filp, F_UNLCK, NULL, NULL);
1811 return do_fcntl_add_lease(fd, filp, arg);
1815 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1816 * @filp: The file to apply the lock to
1817 * @fl: The lock to be applied
1819 * Add a FLOCK style lock to a file.
1821 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1826 error = flock_lock_file(filp, fl);
1827 if (error != FILE_LOCK_DEFERRED)
1829 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1833 locks_delete_block(fl);
1839 EXPORT_SYMBOL(flock_lock_file_wait);
1842 * sys_flock: - flock() system call.
1843 * @fd: the file descriptor to lock.
1844 * @cmd: the type of lock to apply.
1846 * Apply a %FL_FLOCK style lock to an open file descriptor.
1847 * The @cmd can be one of
1849 * %LOCK_SH -- a shared lock.
1851 * %LOCK_EX -- an exclusive lock.
1853 * %LOCK_UN -- remove an existing lock.
1855 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1857 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1858 * processes read and write access respectively.
1860 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1862 struct fd f = fdget(fd);
1863 struct file_lock *lock;
1864 int can_sleep, unlock;
1871 can_sleep = !(cmd & LOCK_NB);
1873 unlock = (cmd == LOCK_UN);
1875 if (!unlock && !(cmd & LOCK_MAND) &&
1876 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1879 error = flock_make_lock(f.file, &lock, cmd);
1883 lock->fl_flags |= FL_SLEEP;
1885 error = security_file_lock(f.file, lock->fl_type);
1889 if (f.file->f_op->flock)
1890 error = f.file->f_op->flock(f.file,
1891 (can_sleep) ? F_SETLKW : F_SETLK,
1894 error = flock_lock_file_wait(f.file, lock);
1897 locks_free_lock(lock);
1906 * vfs_test_lock - test file byte range lock
1907 * @filp: The file to test lock for
1908 * @fl: The lock to test; also used to hold result
1910 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1911 * setting conf->fl_type to something other than F_UNLCK.
1913 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1915 if (filp->f_op->lock)
1916 return filp->f_op->lock(filp, F_GETLK, fl);
1917 posix_test_lock(filp, fl);
1920 EXPORT_SYMBOL_GPL(vfs_test_lock);
1922 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1924 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
1925 #if BITS_PER_LONG == 32
1927 * Make sure we can represent the posix lock via
1928 * legacy 32bit flock.
1930 if (fl->fl_start > OFFT_OFFSET_MAX)
1932 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1935 flock->l_start = fl->fl_start;
1936 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1937 fl->fl_end - fl->fl_start + 1;
1938 flock->l_whence = 0;
1939 flock->l_type = fl->fl_type;
1943 #if BITS_PER_LONG == 32
1944 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1946 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
1947 flock->l_start = fl->fl_start;
1948 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1949 fl->fl_end - fl->fl_start + 1;
1950 flock->l_whence = 0;
1951 flock->l_type = fl->fl_type;
1955 /* Report the first existing lock that would conflict with l.
1956 * This implements the F_GETLK command of fcntl().
1958 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
1960 struct file_lock file_lock;
1965 if (copy_from_user(&flock, l, sizeof(flock)))
1968 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1971 error = flock_to_posix_lock(filp, &file_lock, &flock);
1975 if (cmd == F_OFD_GETLK) {
1977 if (flock.l_pid != 0)
1981 file_lock.fl_flags |= FL_OFDLCK;
1982 file_lock.fl_owner = filp;
1985 error = vfs_test_lock(filp, &file_lock);
1989 flock.l_type = file_lock.fl_type;
1990 if (file_lock.fl_type != F_UNLCK) {
1991 error = posix_lock_to_flock(&flock, &file_lock);
1996 if (!copy_to_user(l, &flock, sizeof(flock)))
1999 locks_release_private(&file_lock);
2005 * vfs_lock_file - file byte range lock
2006 * @filp: The file to apply the lock to
2007 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2008 * @fl: The lock to be applied
2009 * @conf: Place to return a copy of the conflicting lock, if found.
2011 * A caller that doesn't care about the conflicting lock may pass NULL
2012 * as the final argument.
2014 * If the filesystem defines a private ->lock() method, then @conf will
2015 * be left unchanged; so a caller that cares should initialize it to
2016 * some acceptable default.
2018 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2019 * locks, the ->lock() interface may return asynchronously, before the lock has
2020 * been granted or denied by the underlying filesystem, if (and only if)
2021 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2022 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2023 * the request is for a blocking lock. When ->lock() does return asynchronously,
2024 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2025 * request completes.
2026 * If the request is for non-blocking lock the file system should return
2027 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2028 * with the result. If the request timed out the callback routine will return a
2029 * nonzero return code and the file system should release the lock. The file
2030 * system is also responsible to keep a corresponding posix lock when it
2031 * grants a lock so the VFS can find out which locks are locally held and do
2032 * the correct lock cleanup when required.
2033 * The underlying filesystem must not drop the kernel lock or call
2034 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2037 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2039 if (filp->f_op->lock)
2040 return filp->f_op->lock(filp, cmd, fl);
2042 return posix_lock_file(filp, fl, conf);
2044 EXPORT_SYMBOL_GPL(vfs_lock_file);
2046 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2047 struct file_lock *fl)
2051 error = security_file_lock(filp, fl->fl_type);
2056 error = vfs_lock_file(filp, cmd, fl, NULL);
2057 if (error != FILE_LOCK_DEFERRED)
2059 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2063 locks_delete_block(fl);
2070 /* Ensure that fl->fl_filp has compatible f_mode for F_SETLK calls */
2072 check_fmode_for_setlk(struct file_lock *fl)
2074 switch (fl->fl_type) {
2076 if (!(fl->fl_file->f_mode & FMODE_READ))
2080 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2086 /* Apply the lock described by l to an open file descriptor.
2087 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2089 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2090 struct flock __user *l)
2092 struct file_lock *file_lock = locks_alloc_lock();
2094 struct inode *inode;
2098 if (file_lock == NULL)
2102 * This might block, so we do it before checking the inode.
2105 if (copy_from_user(&flock, l, sizeof(flock)))
2108 inode = file_inode(filp);
2110 /* Don't allow mandatory locks on files that may be memory mapped
2113 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2119 error = flock_to_posix_lock(filp, file_lock, &flock);
2123 error = check_fmode_for_setlk(file_lock);
2128 * If the cmd is requesting file-private locks, then set the
2129 * FL_OFDLCK flag and override the owner.
2134 if (flock.l_pid != 0)
2138 file_lock->fl_flags |= FL_OFDLCK;
2139 file_lock->fl_owner = filp;
2143 if (flock.l_pid != 0)
2147 file_lock->fl_flags |= FL_OFDLCK;
2148 file_lock->fl_owner = filp;
2151 file_lock->fl_flags |= FL_SLEEP;
2154 error = do_lock_file_wait(filp, cmd, file_lock);
2157 * Attempt to detect a close/fcntl race and recover by
2158 * releasing the lock that was just acquired.
2161 * we need that spin_lock here - it prevents reordering between
2162 * update of inode->i_flock and check for it done in close().
2163 * rcu_read_lock() wouldn't do.
2165 spin_lock(¤t->files->file_lock);
2167 spin_unlock(¤t->files->file_lock);
2168 if (!error && f != filp && flock.l_type != F_UNLCK) {
2169 flock.l_type = F_UNLCK;
2174 locks_free_lock(file_lock);
2178 #if BITS_PER_LONG == 32
2179 /* Report the first existing lock that would conflict with l.
2180 * This implements the F_GETLK command of fcntl().
2182 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
2184 struct file_lock file_lock;
2185 struct flock64 flock;
2189 if (copy_from_user(&flock, l, sizeof(flock)))
2192 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2195 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2199 if (cmd == F_OFD_GETLK) {
2201 if (flock.l_pid != 0)
2205 file_lock.fl_flags |= FL_OFDLCK;
2206 file_lock.fl_owner = filp;
2209 error = vfs_test_lock(filp, &file_lock);
2213 flock.l_type = file_lock.fl_type;
2214 if (file_lock.fl_type != F_UNLCK)
2215 posix_lock_to_flock64(&flock, &file_lock);
2218 if (!copy_to_user(l, &flock, sizeof(flock)))
2221 locks_release_private(&file_lock);
2226 /* Apply the lock described by l to an open file descriptor.
2227 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2229 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2230 struct flock64 __user *l)
2232 struct file_lock *file_lock = locks_alloc_lock();
2233 struct flock64 flock;
2234 struct inode *inode;
2238 if (file_lock == NULL)
2242 * This might block, so we do it before checking the inode.
2245 if (copy_from_user(&flock, l, sizeof(flock)))
2248 inode = file_inode(filp);
2250 /* Don't allow mandatory locks on files that may be memory mapped
2253 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2259 error = flock64_to_posix_lock(filp, file_lock, &flock);
2263 error = check_fmode_for_setlk(file_lock);
2268 * If the cmd is requesting file-private locks, then set the
2269 * FL_OFDLCK flag and override the owner.
2274 if (flock.l_pid != 0)
2278 file_lock->fl_flags |= FL_OFDLCK;
2279 file_lock->fl_owner = filp;
2283 if (flock.l_pid != 0)
2287 file_lock->fl_flags |= FL_OFDLCK;
2288 file_lock->fl_owner = filp;
2291 file_lock->fl_flags |= FL_SLEEP;
2294 error = do_lock_file_wait(filp, cmd, file_lock);
2297 * Attempt to detect a close/fcntl race and recover by
2298 * releasing the lock that was just acquired.
2300 spin_lock(¤t->files->file_lock);
2302 spin_unlock(¤t->files->file_lock);
2303 if (!error && f != filp && flock.l_type != F_UNLCK) {
2304 flock.l_type = F_UNLCK;
2309 locks_free_lock(file_lock);
2312 #endif /* BITS_PER_LONG == 32 */
2315 * This function is called when the file is being removed
2316 * from the task's fd array. POSIX locks belonging to this task
2317 * are deleted at this time.
2319 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2321 struct file_lock lock;
2324 * If there are no locks held on this file, we don't need to call
2325 * posix_lock_file(). Another process could be setting a lock on this
2326 * file at the same time, but we wouldn't remove that lock anyway.
2328 if (!file_inode(filp)->i_flock)
2331 lock.fl_type = F_UNLCK;
2332 lock.fl_flags = FL_POSIX | FL_CLOSE;
2334 lock.fl_end = OFFSET_MAX;
2335 lock.fl_owner = owner;
2336 lock.fl_pid = current->tgid;
2337 lock.fl_file = filp;
2339 lock.fl_lmops = NULL;
2341 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2343 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2344 lock.fl_ops->fl_release_private(&lock);
2347 EXPORT_SYMBOL(locks_remove_posix);
2350 * This function is called on the last close of an open file.
2352 void locks_remove_file(struct file *filp)
2354 struct inode * inode = file_inode(filp);
2355 struct file_lock *fl;
2356 struct file_lock **before;
2359 if (!inode->i_flock)
2362 locks_remove_posix(filp, filp);
2364 if (filp->f_op->flock) {
2365 struct file_lock fl = {
2367 .fl_pid = current->tgid,
2369 .fl_flags = FL_FLOCK,
2371 .fl_end = OFFSET_MAX,
2373 filp->f_op->flock(filp, F_SETLKW, &fl);
2374 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2375 fl.fl_ops->fl_release_private(&fl);
2378 spin_lock(&inode->i_lock);
2379 before = &inode->i_flock;
2381 while ((fl = *before) != NULL) {
2382 if (fl->fl_file == filp) {
2384 lease_modify(before, F_UNLCK, &dispose);
2389 * There's a leftover lock on the list of a type that
2390 * we didn't expect to see. Most likely a classic
2391 * POSIX lock that ended up not getting released
2392 * properly, or that raced onto the list somehow. Log
2393 * some info about it and then just remove it from
2397 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2398 MAJOR(inode->i_sb->s_dev),
2399 MINOR(inode->i_sb->s_dev), inode->i_ino,
2400 fl->fl_type, fl->fl_flags,
2401 fl->fl_start, fl->fl_end);
2403 locks_delete_lock(before, &dispose);
2406 before = &fl->fl_next;
2408 spin_unlock(&inode->i_lock);
2409 locks_dispose_list(&dispose);
2413 * posix_unblock_lock - stop waiting for a file lock
2414 * @waiter: the lock which was waiting
2416 * lockd needs to block waiting for locks.
2419 posix_unblock_lock(struct file_lock *waiter)
2423 spin_lock(&blocked_lock_lock);
2424 if (waiter->fl_next)
2425 __locks_delete_block(waiter);
2428 spin_unlock(&blocked_lock_lock);
2431 EXPORT_SYMBOL(posix_unblock_lock);
2434 * vfs_cancel_lock - file byte range unblock lock
2435 * @filp: The file to apply the unblock to
2436 * @fl: The lock to be unblocked
2438 * Used by lock managers to cancel blocked requests
2440 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2442 if (filp->f_op->lock)
2443 return filp->f_op->lock(filp, F_CANCELLK, fl);
2447 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2449 #ifdef CONFIG_PROC_FS
2450 #include <linux/proc_fs.h>
2451 #include <linux/seq_file.h>
2453 struct locks_iterator {
2458 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2459 loff_t id, char *pfx)
2461 struct inode *inode = NULL;
2462 unsigned int fl_pid;
2465 fl_pid = pid_vnr(fl->fl_nspid);
2467 fl_pid = fl->fl_pid;
2469 if (fl->fl_file != NULL)
2470 inode = file_inode(fl->fl_file);
2472 seq_printf(f, "%lld:%s ", id, pfx);
2474 if (fl->fl_flags & FL_ACCESS)
2475 seq_puts(f, "ACCESS");
2476 else if (IS_OFDLCK(fl))
2477 seq_puts(f, "OFDLCK");
2479 seq_puts(f, "POSIX ");
2481 seq_printf(f, " %s ",
2482 (inode == NULL) ? "*NOINODE*" :
2483 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2484 } else if (IS_FLOCK(fl)) {
2485 if (fl->fl_type & LOCK_MAND) {
2486 seq_puts(f, "FLOCK MSNFS ");
2488 seq_puts(f, "FLOCK ADVISORY ");
2490 } else if (IS_LEASE(fl)) {
2491 if (fl->fl_flags & FL_DELEG)
2492 seq_puts(f, "DELEG ");
2494 seq_puts(f, "LEASE ");
2496 if (lease_breaking(fl))
2497 seq_puts(f, "BREAKING ");
2498 else if (fl->fl_file)
2499 seq_puts(f, "ACTIVE ");
2501 seq_puts(f, "BREAKER ");
2503 seq_puts(f, "UNKNOWN UNKNOWN ");
2505 if (fl->fl_type & LOCK_MAND) {
2506 seq_printf(f, "%s ",
2507 (fl->fl_type & LOCK_READ)
2508 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2509 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2511 seq_printf(f, "%s ",
2512 (lease_breaking(fl))
2513 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2514 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2517 #ifdef WE_CAN_BREAK_LSLK_NOW
2518 seq_printf(f, "%d %s:%ld ", fl_pid,
2519 inode->i_sb->s_id, inode->i_ino);
2521 /* userspace relies on this representation of dev_t ;-( */
2522 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2523 MAJOR(inode->i_sb->s_dev),
2524 MINOR(inode->i_sb->s_dev), inode->i_ino);
2527 seq_printf(f, "%d <none>:0 ", fl_pid);
2530 if (fl->fl_end == OFFSET_MAX)
2531 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2533 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2535 seq_puts(f, "0 EOF\n");
2539 static int locks_show(struct seq_file *f, void *v)
2541 struct locks_iterator *iter = f->private;
2542 struct file_lock *fl, *bfl;
2544 fl = hlist_entry(v, struct file_lock, fl_link);
2546 lock_get_status(f, fl, iter->li_pos, "");
2548 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2549 lock_get_status(f, bfl, iter->li_pos, " ->");
2554 static void *locks_start(struct seq_file *f, loff_t *pos)
2555 __acquires(&blocked_lock_lock)
2557 struct locks_iterator *iter = f->private;
2559 iter->li_pos = *pos + 1;
2560 lg_global_lock(&file_lock_lglock);
2561 spin_lock(&blocked_lock_lock);
2562 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2565 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2567 struct locks_iterator *iter = f->private;
2570 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2573 static void locks_stop(struct seq_file *f, void *v)
2574 __releases(&blocked_lock_lock)
2576 spin_unlock(&blocked_lock_lock);
2577 lg_global_unlock(&file_lock_lglock);
2580 static const struct seq_operations locks_seq_operations = {
2581 .start = locks_start,
2587 static int locks_open(struct inode *inode, struct file *filp)
2589 return seq_open_private(filp, &locks_seq_operations,
2590 sizeof(struct locks_iterator));
2593 static const struct file_operations proc_locks_operations = {
2596 .llseek = seq_lseek,
2597 .release = seq_release_private,
2600 static int __init proc_locks_init(void)
2602 proc_create("locks", 0, NULL, &proc_locks_operations);
2605 module_init(proc_locks_init);
2608 static int __init filelock_init(void)
2612 filelock_cache = kmem_cache_create("file_lock_cache",
2613 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2615 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2617 for_each_possible_cpu(i)
2618 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2623 core_initcall(filelock_init);