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))
139 static bool lease_breaking(struct file_lock *fl)
141 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
144 static int target_leasetype(struct file_lock *fl)
146 if (fl->fl_flags & FL_UNLOCK_PENDING)
148 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
153 int leases_enable = 1;
154 int lease_break_time = 45;
156 #define for_each_lock(inode, lockp) \
157 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
160 * The global file_lock_list is only used for displaying /proc/locks, so we
161 * keep a list on each CPU, with each list protected by its own spinlock via
162 * the file_lock_lglock. Note that alterations to the list also require that
163 * the relevant i_lock is held.
165 DEFINE_STATIC_LGLOCK(file_lock_lglock);
166 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
169 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
170 * It is protected by blocked_lock_lock.
172 * We hash locks by lockowner in order to optimize searching for the lock a
173 * particular lockowner is waiting on.
175 * FIXME: make this value scale via some heuristic? We generally will want more
176 * buckets when we have more lockowners holding locks, but that's a little
177 * difficult to determine without knowing what the workload will look like.
179 #define BLOCKED_HASH_BITS 7
180 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
183 * This lock protects the blocked_hash. Generally, if you're accessing it, you
184 * want to be holding this lock.
186 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
187 * pointer for file_lock structures that are acting as lock requests (in
188 * contrast to those that are acting as records of acquired locks).
190 * Note that when we acquire this lock in order to change the above fields,
191 * we often hold the i_lock as well. In certain cases, when reading the fields
192 * protected by this lock, we can skip acquiring it iff we already hold the
195 * In particular, adding an entry to the fl_block list requires that you hold
196 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
197 * an entry from the list however only requires the file_lock_lock.
199 static DEFINE_SPINLOCK(blocked_lock_lock);
201 static struct kmem_cache *filelock_cache __read_mostly;
203 static void locks_init_lock_heads(struct file_lock *fl)
205 INIT_HLIST_NODE(&fl->fl_link);
206 INIT_LIST_HEAD(&fl->fl_block);
207 init_waitqueue_head(&fl->fl_wait);
210 /* Allocate an empty lock structure. */
211 struct file_lock *locks_alloc_lock(void)
213 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
216 locks_init_lock_heads(fl);
220 EXPORT_SYMBOL_GPL(locks_alloc_lock);
222 void locks_release_private(struct file_lock *fl)
225 if (fl->fl_ops->fl_release_private)
226 fl->fl_ops->fl_release_private(fl);
232 EXPORT_SYMBOL_GPL(locks_release_private);
234 /* Free a lock which is not in use. */
235 void locks_free_lock(struct file_lock *fl)
237 BUG_ON(waitqueue_active(&fl->fl_wait));
238 BUG_ON(!list_empty(&fl->fl_block));
239 BUG_ON(!hlist_unhashed(&fl->fl_link));
241 locks_release_private(fl);
242 kmem_cache_free(filelock_cache, fl);
244 EXPORT_SYMBOL(locks_free_lock);
246 void locks_init_lock(struct file_lock *fl)
248 memset(fl, 0, sizeof(struct file_lock));
249 locks_init_lock_heads(fl);
252 EXPORT_SYMBOL(locks_init_lock);
254 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
257 if (fl->fl_ops->fl_copy_lock)
258 fl->fl_ops->fl_copy_lock(new, fl);
259 new->fl_ops = fl->fl_ops;
262 new->fl_lmops = fl->fl_lmops;
266 * Initialize a new lock from an existing file_lock structure.
268 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
270 new->fl_owner = fl->fl_owner;
271 new->fl_pid = fl->fl_pid;
273 new->fl_flags = fl->fl_flags;
274 new->fl_type = fl->fl_type;
275 new->fl_start = fl->fl_start;
276 new->fl_end = fl->fl_end;
278 new->fl_lmops = NULL;
280 EXPORT_SYMBOL(__locks_copy_lock);
282 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
284 locks_release_private(new);
286 __locks_copy_lock(new, fl);
287 new->fl_file = fl->fl_file;
288 new->fl_ops = fl->fl_ops;
289 new->fl_lmops = fl->fl_lmops;
291 locks_copy_private(new, fl);
294 EXPORT_SYMBOL(locks_copy_lock);
296 static inline int flock_translate_cmd(int cmd) {
298 return cmd & (LOCK_MAND | LOCK_RW);
310 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
311 static int flock_make_lock(struct file *filp, struct file_lock **lock,
314 struct file_lock *fl;
315 int type = flock_translate_cmd(cmd);
319 fl = locks_alloc_lock();
324 fl->fl_pid = current->tgid;
325 fl->fl_flags = FL_FLOCK;
327 fl->fl_end = OFFSET_MAX;
333 static int assign_type(struct file_lock *fl, long type)
347 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
350 switch (l->l_whence) {
355 fl->fl_start = filp->f_pos;
358 fl->fl_start = i_size_read(file_inode(filp));
363 if (l->l_start > OFFSET_MAX - fl->fl_start)
365 fl->fl_start += l->l_start;
366 if (fl->fl_start < 0)
369 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
370 POSIX-2001 defines it. */
372 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
374 fl->fl_end = fl->fl_start + l->l_len - 1;
376 } else if (l->l_len < 0) {
377 if (fl->fl_start + l->l_len < 0)
379 fl->fl_end = fl->fl_start - 1;
380 fl->fl_start += l->l_len;
382 fl->fl_end = OFFSET_MAX;
384 fl->fl_owner = current->files;
385 fl->fl_pid = current->tgid;
387 fl->fl_flags = FL_POSIX;
391 /* Ensure that fl->fl_filp has compatible f_mode */
394 if (!(filp->f_mode & FMODE_READ))
398 if (!(filp->f_mode & FMODE_WRITE))
403 return assign_type(fl, l->l_type);
406 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
409 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
412 struct flock64 ll = {
414 .l_whence = l->l_whence,
415 .l_start = l->l_start,
419 return flock64_to_posix_lock(filp, fl, &ll);
422 /* default lease lock manager operations */
423 static void lease_break_callback(struct file_lock *fl)
425 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
428 static const struct lock_manager_operations lease_manager_ops = {
429 .lm_break = lease_break_callback,
430 .lm_change = lease_modify,
434 * Initialize a lease, use the default lock manager operations
436 static int lease_init(struct file *filp, long type, struct file_lock *fl)
438 if (assign_type(fl, type) != 0)
441 fl->fl_owner = current->files;
442 fl->fl_pid = current->tgid;
445 fl->fl_flags = FL_LEASE;
447 fl->fl_end = OFFSET_MAX;
449 fl->fl_lmops = &lease_manager_ops;
453 /* Allocate a file_lock initialised to this type of lease */
454 static struct file_lock *lease_alloc(struct file *filp, long type)
456 struct file_lock *fl = locks_alloc_lock();
460 return ERR_PTR(error);
462 error = lease_init(filp, type, fl);
465 return ERR_PTR(error);
470 /* Check if two locks overlap each other.
472 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
474 return ((fl1->fl_end >= fl2->fl_start) &&
475 (fl2->fl_end >= fl1->fl_start));
479 * Check whether two locks have the same owner.
481 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
483 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
484 return fl2->fl_lmops == fl1->fl_lmops &&
485 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
486 return fl1->fl_owner == fl2->fl_owner;
489 /* Must be called with the i_lock held! */
490 static void locks_insert_global_locks(struct file_lock *fl)
492 lg_local_lock(&file_lock_lglock);
493 fl->fl_link_cpu = smp_processor_id();
494 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
495 lg_local_unlock(&file_lock_lglock);
498 /* Must be called with the i_lock held! */
499 static void locks_delete_global_locks(struct file_lock *fl)
502 * Avoid taking lock if already unhashed. This is safe since this check
503 * is done while holding the i_lock, and new insertions into the list
504 * also require that it be held.
506 if (hlist_unhashed(&fl->fl_link))
508 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
509 hlist_del_init(&fl->fl_link);
510 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
514 posix_owner_key(struct file_lock *fl)
516 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
517 return fl->fl_lmops->lm_owner_key(fl);
518 return (unsigned long)fl->fl_owner;
521 static void locks_insert_global_blocked(struct file_lock *waiter)
523 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
526 static void locks_delete_global_blocked(struct file_lock *waiter)
528 hash_del(&waiter->fl_link);
531 /* Remove waiter from blocker's block list.
532 * When blocker ends up pointing to itself then the list is empty.
534 * Must be called with blocked_lock_lock held.
536 static void __locks_delete_block(struct file_lock *waiter)
538 locks_delete_global_blocked(waiter);
539 list_del_init(&waiter->fl_block);
540 waiter->fl_next = NULL;
543 static void locks_delete_block(struct file_lock *waiter)
545 spin_lock(&blocked_lock_lock);
546 __locks_delete_block(waiter);
547 spin_unlock(&blocked_lock_lock);
550 /* Insert waiter into blocker's block list.
551 * We use a circular list so that processes can be easily woken up in
552 * the order they blocked. The documentation doesn't require this but
553 * it seems like the reasonable thing to do.
555 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
556 * list itself is protected by the blocked_lock_lock, but by ensuring that the
557 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
558 * in some cases when we see that the fl_block list is empty.
560 static void __locks_insert_block(struct file_lock *blocker,
561 struct file_lock *waiter)
563 BUG_ON(!list_empty(&waiter->fl_block));
564 waiter->fl_next = blocker;
565 list_add_tail(&waiter->fl_block, &blocker->fl_block);
566 if (IS_POSIX(blocker))
567 locks_insert_global_blocked(waiter);
570 /* Must be called with i_lock held. */
571 static void locks_insert_block(struct file_lock *blocker,
572 struct file_lock *waiter)
574 spin_lock(&blocked_lock_lock);
575 __locks_insert_block(blocker, waiter);
576 spin_unlock(&blocked_lock_lock);
580 * Wake up processes blocked waiting for blocker.
582 * Must be called with the inode->i_lock held!
584 static void locks_wake_up_blocks(struct file_lock *blocker)
587 * Avoid taking global lock if list is empty. This is safe since new
588 * blocked requests are only added to the list under the i_lock, and
589 * the i_lock is always held here. Note that removal from the fl_block
590 * list does not require the i_lock, so we must recheck list_empty()
591 * after acquiring the blocked_lock_lock.
593 if (list_empty(&blocker->fl_block))
596 spin_lock(&blocked_lock_lock);
597 while (!list_empty(&blocker->fl_block)) {
598 struct file_lock *waiter;
600 waiter = list_first_entry(&blocker->fl_block,
601 struct file_lock, fl_block);
602 __locks_delete_block(waiter);
603 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
604 waiter->fl_lmops->lm_notify(waiter);
606 wake_up(&waiter->fl_wait);
608 spin_unlock(&blocked_lock_lock);
611 /* Insert file lock fl into an inode's lock list at the position indicated
612 * by pos. At the same time add the lock to the global file lock list.
614 * Must be called with the i_lock held!
616 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
618 fl->fl_nspid = get_pid(task_tgid(current));
620 /* insert into file's list */
624 locks_insert_global_locks(fl);
628 * locks_delete_lock - Delete a lock and then free it.
629 * @thisfl_p: pointer that points to the fl_next field of the previous
630 * inode->i_flock list entry
632 * Unlink a lock from all lists and free the namespace reference, but don't
633 * free it yet. Wake up processes that are blocked waiting for this lock and
634 * notify the FS that the lock has been cleared.
636 * Must be called with the i_lock held!
638 static void locks_unlink_lock(struct file_lock **thisfl_p)
640 struct file_lock *fl = *thisfl_p;
642 locks_delete_global_locks(fl);
644 *thisfl_p = fl->fl_next;
648 put_pid(fl->fl_nspid);
652 locks_wake_up_blocks(fl);
656 * Unlink a lock from all lists and free it.
658 * Must be called with i_lock held!
660 static void locks_delete_lock(struct file_lock **thisfl_p)
662 struct file_lock *fl = *thisfl_p;
664 locks_unlink_lock(thisfl_p);
668 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
669 * checks for shared/exclusive status of overlapping locks.
671 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
673 if (sys_fl->fl_type == F_WRLCK)
675 if (caller_fl->fl_type == F_WRLCK)
680 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
681 * checking before calling the locks_conflict().
683 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
685 /* POSIX locks owned by the same process do not conflict with
688 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
691 /* Check whether they overlap */
692 if (!locks_overlap(caller_fl, sys_fl))
695 return (locks_conflict(caller_fl, sys_fl));
698 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
699 * checking before calling the locks_conflict().
701 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
703 /* FLOCK locks referring to the same filp do not conflict with
706 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
708 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
711 return (locks_conflict(caller_fl, sys_fl));
715 posix_test_lock(struct file *filp, struct file_lock *fl)
717 struct file_lock *cfl;
718 struct inode *inode = file_inode(filp);
720 spin_lock(&inode->i_lock);
721 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
724 if (posix_locks_conflict(fl, cfl))
728 __locks_copy_lock(fl, cfl);
730 fl->fl_pid = pid_vnr(cfl->fl_nspid);
732 fl->fl_type = F_UNLCK;
733 spin_unlock(&inode->i_lock);
736 EXPORT_SYMBOL(posix_test_lock);
739 * Deadlock detection:
741 * We attempt to detect deadlocks that are due purely to posix file
744 * We assume that a task can be waiting for at most one lock at a time.
745 * So for any acquired lock, the process holding that lock may be
746 * waiting on at most one other lock. That lock in turns may be held by
747 * someone waiting for at most one other lock. Given a requested lock
748 * caller_fl which is about to wait for a conflicting lock block_fl, we
749 * follow this chain of waiters to ensure we are not about to create a
752 * Since we do this before we ever put a process to sleep on a lock, we
753 * are ensured that there is never a cycle; that is what guarantees that
754 * the while() loop in posix_locks_deadlock() eventually completes.
756 * Note: the above assumption may not be true when handling lock
757 * requests from a broken NFS client. It may also fail in the presence
758 * of tasks (such as posix threads) sharing the same open file table.
760 * To handle those cases, we just bail out after a few iterations.
763 #define MAX_DEADLK_ITERATIONS 10
765 /* Find a lock that the owner of the given block_fl is blocking on. */
766 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
768 struct file_lock *fl;
770 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
771 if (posix_same_owner(fl, block_fl))
777 /* Must be called with the blocked_lock_lock held! */
778 static int posix_locks_deadlock(struct file_lock *caller_fl,
779 struct file_lock *block_fl)
783 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
784 if (i++ > MAX_DEADLK_ITERATIONS)
786 if (posix_same_owner(caller_fl, block_fl))
792 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
793 * after any leases, but before any posix locks.
795 * Note that if called with an FL_EXISTS argument, the caller may determine
796 * whether or not a lock was successfully freed by testing the return
799 static int flock_lock_file(struct file *filp, struct file_lock *request)
801 struct file_lock *new_fl = NULL;
802 struct file_lock **before;
803 struct inode * inode = file_inode(filp);
807 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
808 new_fl = locks_alloc_lock();
813 spin_lock(&inode->i_lock);
814 if (request->fl_flags & FL_ACCESS)
817 for_each_lock(inode, before) {
818 struct file_lock *fl = *before;
823 if (filp != fl->fl_file)
825 if (request->fl_type == fl->fl_type)
828 locks_delete_lock(before);
832 if (request->fl_type == F_UNLCK) {
833 if ((request->fl_flags & FL_EXISTS) && !found)
839 * If a higher-priority process was blocked on the old file lock,
840 * give it the opportunity to lock the file.
843 spin_unlock(&inode->i_lock);
845 spin_lock(&inode->i_lock);
849 for_each_lock(inode, before) {
850 struct file_lock *fl = *before;
855 if (!flock_locks_conflict(request, fl))
858 if (!(request->fl_flags & FL_SLEEP))
860 error = FILE_LOCK_DEFERRED;
861 locks_insert_block(fl, request);
864 if (request->fl_flags & FL_ACCESS)
866 locks_copy_lock(new_fl, request);
867 locks_insert_lock(before, new_fl);
872 spin_unlock(&inode->i_lock);
874 locks_free_lock(new_fl);
878 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
880 struct file_lock *fl;
881 struct file_lock *new_fl = NULL;
882 struct file_lock *new_fl2 = NULL;
883 struct file_lock *left = NULL;
884 struct file_lock *right = NULL;
885 struct file_lock **before;
890 * We may need two file_lock structures for this operation,
891 * so we get them in advance to avoid races.
893 * In some cases we can be sure, that no new locks will be needed
895 if (!(request->fl_flags & FL_ACCESS) &&
896 (request->fl_type != F_UNLCK ||
897 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
898 new_fl = locks_alloc_lock();
899 new_fl2 = locks_alloc_lock();
902 spin_lock(&inode->i_lock);
904 * New lock request. Walk all POSIX locks and look for conflicts. If
905 * there are any, either return error or put the request on the
906 * blocker's list of waiters and the global blocked_hash.
908 if (request->fl_type != F_UNLCK) {
909 for_each_lock(inode, before) {
913 if (!posix_locks_conflict(request, fl))
916 __locks_copy_lock(conflock, fl);
918 if (!(request->fl_flags & FL_SLEEP))
921 * Deadlock detection and insertion into the blocked
922 * locks list must be done while holding the same lock!
925 spin_lock(&blocked_lock_lock);
926 if (likely(!posix_locks_deadlock(request, fl))) {
927 error = FILE_LOCK_DEFERRED;
928 __locks_insert_block(fl, request);
930 spin_unlock(&blocked_lock_lock);
935 /* If we're just looking for a conflict, we're done. */
937 if (request->fl_flags & FL_ACCESS)
941 * Find the first old lock with the same owner as the new lock.
944 before = &inode->i_flock;
946 /* First skip locks owned by other processes. */
947 while ((fl = *before) && (!IS_POSIX(fl) ||
948 !posix_same_owner(request, fl))) {
949 before = &fl->fl_next;
952 /* Process locks with this owner. */
953 while ((fl = *before) && posix_same_owner(request, fl)) {
954 /* Detect adjacent or overlapping regions (if same lock type)
956 if (request->fl_type == fl->fl_type) {
957 /* In all comparisons of start vs end, use
958 * "start - 1" rather than "end + 1". If end
959 * is OFFSET_MAX, end + 1 will become negative.
961 if (fl->fl_end < request->fl_start - 1)
963 /* If the next lock in the list has entirely bigger
964 * addresses than the new one, insert the lock here.
966 if (fl->fl_start - 1 > request->fl_end)
969 /* If we come here, the new and old lock are of the
970 * same type and adjacent or overlapping. Make one
971 * lock yielding from the lower start address of both
972 * locks to the higher end address.
974 if (fl->fl_start > request->fl_start)
975 fl->fl_start = request->fl_start;
977 request->fl_start = fl->fl_start;
978 if (fl->fl_end < request->fl_end)
979 fl->fl_end = request->fl_end;
981 request->fl_end = fl->fl_end;
983 locks_delete_lock(before);
990 /* Processing for different lock types is a bit
993 if (fl->fl_end < request->fl_start)
995 if (fl->fl_start > request->fl_end)
997 if (request->fl_type == F_UNLCK)
999 if (fl->fl_start < request->fl_start)
1001 /* If the next lock in the list has a higher end
1002 * address than the new one, insert the new one here.
1004 if (fl->fl_end > request->fl_end) {
1008 if (fl->fl_start >= request->fl_start) {
1009 /* The new lock completely replaces an old
1010 * one (This may happen several times).
1013 locks_delete_lock(before);
1016 /* Replace the old lock with the new one.
1017 * Wake up anybody waiting for the old one,
1018 * as the change in lock type might satisfy
1021 locks_wake_up_blocks(fl);
1022 fl->fl_start = request->fl_start;
1023 fl->fl_end = request->fl_end;
1024 fl->fl_type = request->fl_type;
1025 locks_release_private(fl);
1026 locks_copy_private(fl, request);
1031 /* Go on to next lock.
1034 before = &fl->fl_next;
1038 * The above code only modifies existing locks in case of merging or
1039 * replacing. If new lock(s) need to be inserted all modifications are
1040 * done below this, so it's safe yet to bail out.
1042 error = -ENOLCK; /* "no luck" */
1043 if (right && left == right && !new_fl2)
1048 if (request->fl_type == F_UNLCK) {
1049 if (request->fl_flags & FL_EXISTS)
1058 locks_copy_lock(new_fl, request);
1059 locks_insert_lock(before, new_fl);
1063 if (left == right) {
1064 /* The new lock breaks the old one in two pieces,
1065 * so we have to use the second new lock.
1069 locks_copy_lock(left, right);
1070 locks_insert_lock(before, left);
1072 right->fl_start = request->fl_end + 1;
1073 locks_wake_up_blocks(right);
1076 left->fl_end = request->fl_start - 1;
1077 locks_wake_up_blocks(left);
1080 spin_unlock(&inode->i_lock);
1082 * Free any unused locks.
1085 locks_free_lock(new_fl);
1087 locks_free_lock(new_fl2);
1092 * posix_lock_file - Apply a POSIX-style lock to a file
1093 * @filp: The file to apply the lock to
1094 * @fl: The lock to be applied
1095 * @conflock: Place to return a copy of the conflicting lock, if found.
1097 * Add a POSIX style lock to a file.
1098 * We merge adjacent & overlapping locks whenever possible.
1099 * POSIX locks are sorted by owner task, then by starting address
1101 * Note that if called with an FL_EXISTS argument, the caller may determine
1102 * whether or not a lock was successfully freed by testing the return
1103 * value for -ENOENT.
1105 int posix_lock_file(struct file *filp, struct file_lock *fl,
1106 struct file_lock *conflock)
1108 return __posix_lock_file(file_inode(filp), fl, conflock);
1110 EXPORT_SYMBOL(posix_lock_file);
1113 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1114 * @filp: The file to apply the lock to
1115 * @fl: The lock to be applied
1117 * Add a POSIX style lock to a file.
1118 * We merge adjacent & overlapping locks whenever possible.
1119 * POSIX locks are sorted by owner task, then by starting address
1121 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1126 error = posix_lock_file(filp, fl, NULL);
1127 if (error != FILE_LOCK_DEFERRED)
1129 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1133 locks_delete_block(fl);
1138 EXPORT_SYMBOL(posix_lock_file_wait);
1141 * locks_mandatory_locked - Check for an active lock
1142 * @inode: the file to check
1144 * Searches the inode's list of locks to find any POSIX locks which conflict.
1145 * This function is called from locks_verify_locked() only.
1147 int locks_mandatory_locked(struct inode *inode)
1149 fl_owner_t owner = current->files;
1150 struct file_lock *fl;
1153 * Search the lock list for this inode for any POSIX locks.
1155 spin_lock(&inode->i_lock);
1156 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1159 if (fl->fl_owner != owner)
1162 spin_unlock(&inode->i_lock);
1163 return fl ? -EAGAIN : 0;
1167 * locks_mandatory_area - Check for a conflicting lock
1168 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1170 * @inode: the file to check
1171 * @filp: how the file was opened (if it was)
1172 * @offset: start of area to check
1173 * @count: length of area to check
1175 * Searches the inode's list of locks to find any POSIX locks which conflict.
1176 * This function is called from rw_verify_area() and
1177 * locks_verify_truncate().
1179 int locks_mandatory_area(int read_write, struct inode *inode,
1180 struct file *filp, loff_t offset,
1183 struct file_lock fl;
1186 locks_init_lock(&fl);
1187 fl.fl_owner = current->files;
1188 fl.fl_pid = current->tgid;
1190 fl.fl_flags = FL_POSIX | FL_ACCESS;
1191 if (filp && !(filp->f_flags & O_NONBLOCK))
1192 fl.fl_flags |= FL_SLEEP;
1193 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1194 fl.fl_start = offset;
1195 fl.fl_end = offset + count - 1;
1198 error = __posix_lock_file(inode, &fl, NULL);
1199 if (error != FILE_LOCK_DEFERRED)
1201 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1204 * If we've been sleeping someone might have
1205 * changed the permissions behind our back.
1207 if (__mandatory_lock(inode))
1211 locks_delete_block(&fl);
1218 EXPORT_SYMBOL(locks_mandatory_area);
1220 static void lease_clear_pending(struct file_lock *fl, int arg)
1224 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1227 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1231 /* We already had a lease on this file; just change its type */
1232 int lease_modify(struct file_lock **before, int arg)
1234 struct file_lock *fl = *before;
1235 int error = assign_type(fl, arg);
1239 lease_clear_pending(fl, arg);
1240 locks_wake_up_blocks(fl);
1241 if (arg == F_UNLCK) {
1242 struct file *filp = fl->fl_file;
1245 filp->f_owner.signum = 0;
1246 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1247 if (fl->fl_fasync != NULL) {
1248 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1249 fl->fl_fasync = NULL;
1251 locks_delete_lock(before);
1256 EXPORT_SYMBOL(lease_modify);
1258 static bool past_time(unsigned long then)
1261 /* 0 is a special value meaning "this never expires": */
1263 return time_after(jiffies, then);
1266 static void time_out_leases(struct inode *inode)
1268 struct file_lock **before;
1269 struct file_lock *fl;
1271 before = &inode->i_flock;
1272 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1273 if (past_time(fl->fl_downgrade_time))
1274 lease_modify(before, F_RDLCK);
1275 if (past_time(fl->fl_break_time))
1276 lease_modify(before, F_UNLCK);
1277 if (fl == *before) /* lease_modify may have freed fl */
1278 before = &fl->fl_next;
1282 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1284 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1286 return locks_conflict(breaker, lease);
1290 * __break_lease - revoke all outstanding leases on file
1291 * @inode: the inode of the file to return
1292 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1294 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1297 * break_lease (inlined for speed) has checked there already is at least
1298 * some kind of lock (maybe a lease) on this file. Leases are broken on
1299 * a call to open() or truncate(). This function can sleep unless you
1300 * specified %O_NONBLOCK to your open().
1302 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1305 struct file_lock *new_fl, *flock;
1306 struct file_lock *fl;
1307 unsigned long break_time;
1308 int i_have_this_lease = 0;
1309 bool lease_conflict = false;
1310 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1312 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1314 return PTR_ERR(new_fl);
1315 new_fl->fl_flags = type;
1317 spin_lock(&inode->i_lock);
1319 time_out_leases(inode);
1321 flock = inode->i_flock;
1322 if ((flock == NULL) || !IS_LEASE(flock))
1325 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1326 if (leases_conflict(fl, new_fl)) {
1327 lease_conflict = true;
1328 if (fl->fl_owner == current->files)
1329 i_have_this_lease = 1;
1332 if (!lease_conflict)
1336 if (lease_break_time > 0) {
1337 break_time = jiffies + lease_break_time * HZ;
1338 if (break_time == 0)
1339 break_time++; /* so that 0 means no break time */
1342 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1343 if (!leases_conflict(fl, new_fl))
1346 if (fl->fl_flags & FL_UNLOCK_PENDING)
1348 fl->fl_flags |= FL_UNLOCK_PENDING;
1349 fl->fl_break_time = break_time;
1351 if (lease_breaking(flock))
1353 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1354 fl->fl_downgrade_time = break_time;
1356 fl->fl_lmops->lm_break(fl);
1359 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1360 error = -EWOULDBLOCK;
1365 break_time = flock->fl_break_time;
1366 if (break_time != 0) {
1367 break_time -= jiffies;
1368 if (break_time == 0)
1371 locks_insert_block(flock, new_fl);
1372 spin_unlock(&inode->i_lock);
1373 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1374 !new_fl->fl_next, break_time);
1375 spin_lock(&inode->i_lock);
1376 locks_delete_block(new_fl);
1379 time_out_leases(inode);
1381 * Wait for the next conflicting lease that has not been
1384 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1385 flock = flock->fl_next) {
1386 if (leases_conflict(new_fl, flock))
1393 spin_unlock(&inode->i_lock);
1394 locks_free_lock(new_fl);
1398 EXPORT_SYMBOL(__break_lease);
1401 * lease_get_mtime - get the last modified time of an inode
1403 * @time: pointer to a timespec which will contain the last modified time
1405 * This is to force NFS clients to flush their caches for files with
1406 * exclusive leases. The justification is that if someone has an
1407 * exclusive lease, then they could be modifying it.
1409 void lease_get_mtime(struct inode *inode, struct timespec *time)
1411 struct file_lock *flock = inode->i_flock;
1412 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1413 *time = current_fs_time(inode->i_sb);
1415 *time = inode->i_mtime;
1418 EXPORT_SYMBOL(lease_get_mtime);
1421 * fcntl_getlease - Enquire what lease is currently active
1424 * The value returned by this function will be one of
1425 * (if no lease break is pending):
1427 * %F_RDLCK to indicate a shared lease is held.
1429 * %F_WRLCK to indicate an exclusive lease is held.
1431 * %F_UNLCK to indicate no lease is held.
1433 * (if a lease break is pending):
1435 * %F_RDLCK to indicate an exclusive lease needs to be
1436 * changed to a shared lease (or removed).
1438 * %F_UNLCK to indicate the lease needs to be removed.
1440 * XXX: sfr & willy disagree over whether F_INPROGRESS
1441 * should be returned to userspace.
1443 int fcntl_getlease(struct file *filp)
1445 struct file_lock *fl;
1446 struct inode *inode = file_inode(filp);
1449 spin_lock(&inode->i_lock);
1450 time_out_leases(file_inode(filp));
1451 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1453 if (fl->fl_file == filp) {
1454 type = target_leasetype(fl);
1458 spin_unlock(&inode->i_lock);
1463 * check_conflicting_open - see if the given dentry points to a file that has
1464 * an existing open that would conflict with the
1466 * @dentry: dentry to check
1467 * @arg: type of lease that we're trying to acquire
1469 * Check to see if there's an existing open fd on this file that would
1470 * conflict with the lease we're trying to set.
1473 check_conflicting_open(const struct dentry *dentry, const long arg)
1476 struct inode *inode = dentry->d_inode;
1478 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1481 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1482 (atomic_read(&inode->i_count) > 1)))
1488 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1490 struct file_lock *fl, **before, **my_before = NULL, *lease;
1491 struct dentry *dentry = filp->f_path.dentry;
1492 struct inode *inode = dentry->d_inode;
1493 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1498 * In the delegation case we need mutual exclusion with
1499 * a number of operations that take the i_mutex. We trylock
1500 * because delegations are an optional optimization, and if
1501 * there's some chance of a conflict--we'd rather not
1502 * bother, maybe that's a sign this just isn't a good file to
1503 * hand out a delegation on.
1505 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1508 if (is_deleg && arg == F_WRLCK) {
1509 /* Write delegations are not currently supported: */
1510 mutex_unlock(&inode->i_mutex);
1515 error = check_conflicting_open(dentry, arg);
1520 * At this point, we know that if there is an exclusive
1521 * lease on this file, then we hold it on this filp
1522 * (otherwise our open of this file would have blocked).
1523 * And if we are trying to acquire an exclusive lease,
1524 * then the file is not open by anyone (including us)
1525 * except for this filp.
1528 for (before = &inode->i_flock;
1529 ((fl = *before) != NULL) && IS_LEASE(fl);
1530 before = &fl->fl_next) {
1531 if (fl->fl_file == filp) {
1536 * No exclusive leases if someone else has a lease on
1542 * Modifying our existing lease is OK, but no getting a
1543 * new lease if someone else is opening for write:
1545 if (fl->fl_flags & FL_UNLOCK_PENDING)
1549 if (my_before != NULL) {
1550 error = lease->fl_lmops->lm_change(my_before, arg);
1560 locks_insert_lock(before, lease);
1562 * The check in break_lease() is lockless. It's possible for another
1563 * open to race in after we did the earlier check for a conflicting
1564 * open but before the lease was inserted. Check again for a
1565 * conflicting open and cancel the lease if there is one.
1567 * We also add a barrier here to ensure that the insertion of the lock
1568 * precedes these checks.
1571 error = check_conflicting_open(dentry, arg);
1573 locks_unlink_lock(flp);
1576 mutex_unlock(&inode->i_mutex);
1580 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1582 struct file_lock *fl, **before;
1583 struct dentry *dentry = filp->f_path.dentry;
1584 struct inode *inode = dentry->d_inode;
1586 for (before = &inode->i_flock;
1587 ((fl = *before) != NULL) && IS_LEASE(fl);
1588 before = &fl->fl_next) {
1589 if (fl->fl_file != filp)
1591 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1597 * generic_setlease - sets a lease on an open file
1598 * @filp: file pointer
1599 * @arg: type of lease to obtain
1600 * @flp: input - file_lock to use, output - file_lock inserted
1602 * The (input) flp->fl_lmops->lm_break function is required
1605 * Called with inode->i_lock held.
1607 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1609 struct dentry *dentry = filp->f_path.dentry;
1610 struct inode *inode = dentry->d_inode;
1613 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1615 if (!S_ISREG(inode->i_mode))
1617 error = security_file_lock(filp, arg);
1621 time_out_leases(inode);
1623 BUG_ON(!(*flp)->fl_lmops->lm_break);
1627 return generic_delete_lease(filp, flp);
1630 return generic_add_lease(filp, arg, flp);
1635 EXPORT_SYMBOL(generic_setlease);
1637 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1639 if (filp->f_op->setlease)
1640 return filp->f_op->setlease(filp, arg, lease);
1642 return generic_setlease(filp, arg, lease);
1646 * vfs_setlease - sets a lease on an open file
1647 * @filp: file pointer
1648 * @arg: type of lease to obtain
1649 * @lease: file_lock to use
1651 * Call this to establish a lease on the file.
1652 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1653 * break_lease will oops!
1655 * This will call the filesystem's setlease file method, if
1656 * defined. Note that there is no getlease method; instead, the
1657 * filesystem setlease method should call back to setlease() to
1658 * add a lease to the inode's lease list, where fcntl_getlease() can
1659 * find it. Since fcntl_getlease() only reports whether the current
1660 * task holds a lease, a cluster filesystem need only do this for
1661 * leases held by processes on this node.
1663 * There is also no break_lease method; filesystems that
1664 * handle their own leases should break leases themselves from the
1665 * filesystem's open, create, and (on truncate) setattr methods.
1667 * Warning: the only current setlease methods exist only to disable
1668 * leases in certain cases. More vfs changes may be required to
1669 * allow a full filesystem lease implementation.
1672 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1674 struct inode *inode = file_inode(filp);
1677 spin_lock(&inode->i_lock);
1678 error = __vfs_setlease(filp, arg, lease);
1679 spin_unlock(&inode->i_lock);
1683 EXPORT_SYMBOL_GPL(vfs_setlease);
1685 static int do_fcntl_delete_lease(struct file *filp)
1687 struct file_lock fl, *flp = &fl;
1689 lease_init(filp, F_UNLCK, flp);
1691 return vfs_setlease(filp, F_UNLCK, &flp);
1694 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1696 struct file_lock *fl, *ret;
1697 struct inode *inode = file_inode(filp);
1698 struct fasync_struct *new;
1701 fl = lease_alloc(filp, arg);
1705 new = fasync_alloc();
1707 locks_free_lock(fl);
1711 spin_lock(&inode->i_lock);
1712 error = __vfs_setlease(filp, arg, &ret);
1714 spin_unlock(&inode->i_lock);
1715 locks_free_lock(fl);
1716 goto out_free_fasync;
1719 locks_free_lock(fl);
1722 * fasync_insert_entry() returns the old entry if any.
1723 * If there was no old entry, then it used 'new' and
1724 * inserted it into the fasync list. Clear new so that
1725 * we don't release it here.
1727 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1730 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1731 spin_unlock(&inode->i_lock);
1740 * fcntl_setlease - sets a lease on an open file
1741 * @fd: open file descriptor
1742 * @filp: file pointer
1743 * @arg: type of lease to obtain
1745 * Call this fcntl to establish a lease on the file.
1746 * Note that you also need to call %F_SETSIG to
1747 * receive a signal when the lease is broken.
1749 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1752 return do_fcntl_delete_lease(filp);
1753 return do_fcntl_add_lease(fd, filp, arg);
1757 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1758 * @filp: The file to apply the lock to
1759 * @fl: The lock to be applied
1761 * Add a FLOCK style lock to a file.
1763 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1768 error = flock_lock_file(filp, fl);
1769 if (error != FILE_LOCK_DEFERRED)
1771 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1775 locks_delete_block(fl);
1781 EXPORT_SYMBOL(flock_lock_file_wait);
1784 * sys_flock: - flock() system call.
1785 * @fd: the file descriptor to lock.
1786 * @cmd: the type of lock to apply.
1788 * Apply a %FL_FLOCK style lock to an open file descriptor.
1789 * The @cmd can be one of
1791 * %LOCK_SH -- a shared lock.
1793 * %LOCK_EX -- an exclusive lock.
1795 * %LOCK_UN -- remove an existing lock.
1797 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1799 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1800 * processes read and write access respectively.
1802 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1804 struct fd f = fdget(fd);
1805 struct file_lock *lock;
1806 int can_sleep, unlock;
1813 can_sleep = !(cmd & LOCK_NB);
1815 unlock = (cmd == LOCK_UN);
1817 if (!unlock && !(cmd & LOCK_MAND) &&
1818 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1821 error = flock_make_lock(f.file, &lock, cmd);
1825 lock->fl_flags |= FL_SLEEP;
1827 error = security_file_lock(f.file, lock->fl_type);
1831 if (f.file->f_op->flock)
1832 error = f.file->f_op->flock(f.file,
1833 (can_sleep) ? F_SETLKW : F_SETLK,
1836 error = flock_lock_file_wait(f.file, lock);
1839 locks_free_lock(lock);
1848 * vfs_test_lock - test file byte range lock
1849 * @filp: The file to test lock for
1850 * @fl: The lock to test; also used to hold result
1852 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1853 * setting conf->fl_type to something other than F_UNLCK.
1855 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1857 if (filp->f_op->lock)
1858 return filp->f_op->lock(filp, F_GETLK, fl);
1859 posix_test_lock(filp, fl);
1862 EXPORT_SYMBOL_GPL(vfs_test_lock);
1864 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1866 flock->l_pid = fl->fl_pid;
1867 #if BITS_PER_LONG == 32
1869 * Make sure we can represent the posix lock via
1870 * legacy 32bit flock.
1872 if (fl->fl_start > OFFT_OFFSET_MAX)
1874 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1877 flock->l_start = fl->fl_start;
1878 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1879 fl->fl_end - fl->fl_start + 1;
1880 flock->l_whence = 0;
1881 flock->l_type = fl->fl_type;
1885 #if BITS_PER_LONG == 32
1886 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1888 flock->l_pid = fl->fl_pid;
1889 flock->l_start = fl->fl_start;
1890 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1891 fl->fl_end - fl->fl_start + 1;
1892 flock->l_whence = 0;
1893 flock->l_type = fl->fl_type;
1897 /* Report the first existing lock that would conflict with l.
1898 * This implements the F_GETLK command of fcntl().
1900 int fcntl_getlk(struct file *filp, struct flock __user *l)
1902 struct file_lock file_lock;
1907 if (copy_from_user(&flock, l, sizeof(flock)))
1910 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1913 error = flock_to_posix_lock(filp, &file_lock, &flock);
1917 error = vfs_test_lock(filp, &file_lock);
1921 flock.l_type = file_lock.fl_type;
1922 if (file_lock.fl_type != F_UNLCK) {
1923 error = posix_lock_to_flock(&flock, &file_lock);
1928 if (!copy_to_user(l, &flock, sizeof(flock)))
1935 * vfs_lock_file - file byte range lock
1936 * @filp: The file to apply the lock to
1937 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1938 * @fl: The lock to be applied
1939 * @conf: Place to return a copy of the conflicting lock, if found.
1941 * A caller that doesn't care about the conflicting lock may pass NULL
1942 * as the final argument.
1944 * If the filesystem defines a private ->lock() method, then @conf will
1945 * be left unchanged; so a caller that cares should initialize it to
1946 * some acceptable default.
1948 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1949 * locks, the ->lock() interface may return asynchronously, before the lock has
1950 * been granted or denied by the underlying filesystem, if (and only if)
1951 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1952 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1953 * the request is for a blocking lock. When ->lock() does return asynchronously,
1954 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1955 * request completes.
1956 * If the request is for non-blocking lock the file system should return
1957 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1958 * with the result. If the request timed out the callback routine will return a
1959 * nonzero return code and the file system should release the lock. The file
1960 * system is also responsible to keep a corresponding posix lock when it
1961 * grants a lock so the VFS can find out which locks are locally held and do
1962 * the correct lock cleanup when required.
1963 * The underlying filesystem must not drop the kernel lock or call
1964 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1967 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1969 if (filp->f_op->lock)
1970 return filp->f_op->lock(filp, cmd, fl);
1972 return posix_lock_file(filp, fl, conf);
1974 EXPORT_SYMBOL_GPL(vfs_lock_file);
1976 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1977 struct file_lock *fl)
1981 error = security_file_lock(filp, fl->fl_type);
1986 error = vfs_lock_file(filp, cmd, fl, NULL);
1987 if (error != FILE_LOCK_DEFERRED)
1989 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1993 locks_delete_block(fl);
2000 /* Apply the lock described by l to an open file descriptor.
2001 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2003 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2004 struct flock __user *l)
2006 struct file_lock *file_lock = locks_alloc_lock();
2008 struct inode *inode;
2012 if (file_lock == NULL)
2016 * This might block, so we do it before checking the inode.
2019 if (copy_from_user(&flock, l, sizeof(flock)))
2022 inode = file_inode(filp);
2024 /* Don't allow mandatory locks on files that may be memory mapped
2027 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2033 error = flock_to_posix_lock(filp, file_lock, &flock);
2036 if (cmd == F_SETLKW) {
2037 file_lock->fl_flags |= FL_SLEEP;
2040 error = do_lock_file_wait(filp, cmd, file_lock);
2043 * Attempt to detect a close/fcntl race and recover by
2044 * releasing the lock that was just acquired.
2047 * we need that spin_lock here - it prevents reordering between
2048 * update of inode->i_flock and check for it done in close().
2049 * rcu_read_lock() wouldn't do.
2051 spin_lock(¤t->files->file_lock);
2053 spin_unlock(¤t->files->file_lock);
2054 if (!error && f != filp && flock.l_type != F_UNLCK) {
2055 flock.l_type = F_UNLCK;
2060 locks_free_lock(file_lock);
2064 #if BITS_PER_LONG == 32
2065 /* Report the first existing lock that would conflict with l.
2066 * This implements the F_GETLK command of fcntl().
2068 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
2070 struct file_lock file_lock;
2071 struct flock64 flock;
2075 if (copy_from_user(&flock, l, sizeof(flock)))
2078 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2081 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2085 error = vfs_test_lock(filp, &file_lock);
2089 flock.l_type = file_lock.fl_type;
2090 if (file_lock.fl_type != F_UNLCK)
2091 posix_lock_to_flock64(&flock, &file_lock);
2094 if (!copy_to_user(l, &flock, sizeof(flock)))
2101 /* Apply the lock described by l to an open file descriptor.
2102 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2104 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2105 struct flock64 __user *l)
2107 struct file_lock *file_lock = locks_alloc_lock();
2108 struct flock64 flock;
2109 struct inode *inode;
2113 if (file_lock == NULL)
2117 * This might block, so we do it before checking the inode.
2120 if (copy_from_user(&flock, l, sizeof(flock)))
2123 inode = file_inode(filp);
2125 /* Don't allow mandatory locks on files that may be memory mapped
2128 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2134 error = flock64_to_posix_lock(filp, file_lock, &flock);
2137 if (cmd == F_SETLKW64) {
2138 file_lock->fl_flags |= FL_SLEEP;
2141 error = do_lock_file_wait(filp, cmd, file_lock);
2144 * Attempt to detect a close/fcntl race and recover by
2145 * releasing the lock that was just acquired.
2147 spin_lock(¤t->files->file_lock);
2149 spin_unlock(¤t->files->file_lock);
2150 if (!error && f != filp && flock.l_type != F_UNLCK) {
2151 flock.l_type = F_UNLCK;
2156 locks_free_lock(file_lock);
2159 #endif /* BITS_PER_LONG == 32 */
2162 * This function is called when the file is being removed
2163 * from the task's fd array. POSIX locks belonging to this task
2164 * are deleted at this time.
2166 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2168 struct file_lock lock;
2171 * If there are no locks held on this file, we don't need to call
2172 * posix_lock_file(). Another process could be setting a lock on this
2173 * file at the same time, but we wouldn't remove that lock anyway.
2175 if (!file_inode(filp)->i_flock)
2178 lock.fl_type = F_UNLCK;
2179 lock.fl_flags = FL_POSIX | FL_CLOSE;
2181 lock.fl_end = OFFSET_MAX;
2182 lock.fl_owner = owner;
2183 lock.fl_pid = current->tgid;
2184 lock.fl_file = filp;
2186 lock.fl_lmops = NULL;
2188 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2190 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2191 lock.fl_ops->fl_release_private(&lock);
2194 EXPORT_SYMBOL(locks_remove_posix);
2197 * This function is called on the last close of an open file.
2199 void locks_remove_flock(struct file *filp)
2201 struct inode * inode = file_inode(filp);
2202 struct file_lock *fl;
2203 struct file_lock **before;
2205 if (!inode->i_flock)
2208 if (filp->f_op->flock) {
2209 struct file_lock fl = {
2210 .fl_pid = current->tgid,
2212 .fl_flags = FL_FLOCK,
2214 .fl_end = OFFSET_MAX,
2216 filp->f_op->flock(filp, F_SETLKW, &fl);
2217 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2218 fl.fl_ops->fl_release_private(&fl);
2221 spin_lock(&inode->i_lock);
2222 before = &inode->i_flock;
2224 while ((fl = *before) != NULL) {
2225 if (fl->fl_file == filp) {
2227 lease_modify(before, F_UNLCK);
2232 * There's a leftover lock on the list of a type that
2233 * we didn't expect to see. Most likely a classic
2234 * POSIX lock that ended up not getting released
2235 * properly, or that raced onto the list somehow. Log
2236 * some info about it and then just remove it from
2240 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2241 MAJOR(inode->i_sb->s_dev),
2242 MINOR(inode->i_sb->s_dev), inode->i_ino,
2243 fl->fl_type, fl->fl_flags,
2244 fl->fl_start, fl->fl_end);
2246 locks_delete_lock(before);
2249 before = &fl->fl_next;
2251 spin_unlock(&inode->i_lock);
2255 * posix_unblock_lock - stop waiting for a file lock
2256 * @waiter: the lock which was waiting
2258 * lockd needs to block waiting for locks.
2261 posix_unblock_lock(struct file_lock *waiter)
2265 spin_lock(&blocked_lock_lock);
2266 if (waiter->fl_next)
2267 __locks_delete_block(waiter);
2270 spin_unlock(&blocked_lock_lock);
2273 EXPORT_SYMBOL(posix_unblock_lock);
2276 * vfs_cancel_lock - file byte range unblock lock
2277 * @filp: The file to apply the unblock to
2278 * @fl: The lock to be unblocked
2280 * Used by lock managers to cancel blocked requests
2282 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2284 if (filp->f_op->lock)
2285 return filp->f_op->lock(filp, F_CANCELLK, fl);
2289 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2291 #ifdef CONFIG_PROC_FS
2292 #include <linux/proc_fs.h>
2293 #include <linux/seq_file.h>
2295 struct locks_iterator {
2300 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2301 loff_t id, char *pfx)
2303 struct inode *inode = NULL;
2304 unsigned int fl_pid;
2307 fl_pid = pid_vnr(fl->fl_nspid);
2309 fl_pid = fl->fl_pid;
2311 if (fl->fl_file != NULL)
2312 inode = file_inode(fl->fl_file);
2314 seq_printf(f, "%lld:%s ", id, pfx);
2316 seq_printf(f, "%6s %s ",
2317 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2318 (inode == NULL) ? "*NOINODE*" :
2319 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2320 } else if (IS_FLOCK(fl)) {
2321 if (fl->fl_type & LOCK_MAND) {
2322 seq_printf(f, "FLOCK MSNFS ");
2324 seq_printf(f, "FLOCK ADVISORY ");
2326 } else if (IS_LEASE(fl)) {
2327 seq_printf(f, "LEASE ");
2328 if (lease_breaking(fl))
2329 seq_printf(f, "BREAKING ");
2330 else if (fl->fl_file)
2331 seq_printf(f, "ACTIVE ");
2333 seq_printf(f, "BREAKER ");
2335 seq_printf(f, "UNKNOWN UNKNOWN ");
2337 if (fl->fl_type & LOCK_MAND) {
2338 seq_printf(f, "%s ",
2339 (fl->fl_type & LOCK_READ)
2340 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2341 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2343 seq_printf(f, "%s ",
2344 (lease_breaking(fl))
2345 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2346 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2349 #ifdef WE_CAN_BREAK_LSLK_NOW
2350 seq_printf(f, "%d %s:%ld ", fl_pid,
2351 inode->i_sb->s_id, inode->i_ino);
2353 /* userspace relies on this representation of dev_t ;-( */
2354 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2355 MAJOR(inode->i_sb->s_dev),
2356 MINOR(inode->i_sb->s_dev), inode->i_ino);
2359 seq_printf(f, "%d <none>:0 ", fl_pid);
2362 if (fl->fl_end == OFFSET_MAX)
2363 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2365 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2367 seq_printf(f, "0 EOF\n");
2371 static int locks_show(struct seq_file *f, void *v)
2373 struct locks_iterator *iter = f->private;
2374 struct file_lock *fl, *bfl;
2376 fl = hlist_entry(v, struct file_lock, fl_link);
2378 lock_get_status(f, fl, iter->li_pos, "");
2380 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2381 lock_get_status(f, bfl, iter->li_pos, " ->");
2386 static void *locks_start(struct seq_file *f, loff_t *pos)
2387 __acquires(&blocked_lock_lock)
2389 struct locks_iterator *iter = f->private;
2391 iter->li_pos = *pos + 1;
2392 lg_global_lock(&file_lock_lglock);
2393 spin_lock(&blocked_lock_lock);
2394 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2397 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2399 struct locks_iterator *iter = f->private;
2402 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2405 static void locks_stop(struct seq_file *f, void *v)
2406 __releases(&blocked_lock_lock)
2408 spin_unlock(&blocked_lock_lock);
2409 lg_global_unlock(&file_lock_lglock);
2412 static const struct seq_operations locks_seq_operations = {
2413 .start = locks_start,
2419 static int locks_open(struct inode *inode, struct file *filp)
2421 return seq_open_private(filp, &locks_seq_operations,
2422 sizeof(struct locks_iterator));
2425 static const struct file_operations proc_locks_operations = {
2428 .llseek = seq_lseek,
2429 .release = seq_release_private,
2432 static int __init proc_locks_init(void)
2434 proc_create("locks", 0, NULL, &proc_locks_operations);
2437 module_init(proc_locks_init);
2441 * lock_may_read - checks that the region is free of locks
2442 * @inode: the inode that is being read
2443 * @start: the first byte to read
2444 * @len: the number of bytes to read
2446 * Emulates Windows locking requirements. Whole-file
2447 * mandatory locks (share modes) can prohibit a read and
2448 * byte-range POSIX locks can prohibit a read if they overlap.
2450 * N.B. this function is only ever called
2451 * from knfsd and ownership of locks is never checked.
2453 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2455 struct file_lock *fl;
2458 spin_lock(&inode->i_lock);
2459 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2461 if (fl->fl_type == F_RDLCK)
2463 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2465 } else if (IS_FLOCK(fl)) {
2466 if (!(fl->fl_type & LOCK_MAND))
2468 if (fl->fl_type & LOCK_READ)
2475 spin_unlock(&inode->i_lock);
2479 EXPORT_SYMBOL(lock_may_read);
2482 * lock_may_write - checks that the region is free of locks
2483 * @inode: the inode that is being written
2484 * @start: the first byte to write
2485 * @len: the number of bytes to write
2487 * Emulates Windows locking requirements. Whole-file
2488 * mandatory locks (share modes) can prohibit a write and
2489 * byte-range POSIX locks can prohibit a write if they overlap.
2491 * N.B. this function is only ever called
2492 * from knfsd and ownership of locks is never checked.
2494 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2496 struct file_lock *fl;
2499 spin_lock(&inode->i_lock);
2500 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2502 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2504 } else if (IS_FLOCK(fl)) {
2505 if (!(fl->fl_type & LOCK_MAND))
2507 if (fl->fl_type & LOCK_WRITE)
2514 spin_unlock(&inode->i_lock);
2518 EXPORT_SYMBOL(lock_may_write);
2520 static int __init filelock_init(void)
2524 filelock_cache = kmem_cache_create("file_lock_cache",
2525 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2527 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2529 for_each_possible_cpu(i)
2530 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2535 core_initcall(filelock_init);