4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 * Made mandatory locking a mount option. Default is not to allow mandatory
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fdtable.h>
120 #include <linux/fs.h>
121 #include <linux/init.h>
122 #include <linux/module.h>
123 #include <linux/security.h>
124 #include <linux/slab.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
128 #include <linux/pid_namespace.h>
129 #include <linux/hashtable.h>
130 #include <linux/percpu.h>
131 #include <linux/lglock.h>
133 #include <asm/uaccess.h>
135 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
136 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
137 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
138 #define IS_FILE_PVT(fl) (fl->fl_flags & FL_FILE_PVT)
140 static bool lease_breaking(struct file_lock *fl)
142 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
145 static int target_leasetype(struct file_lock *fl)
147 if (fl->fl_flags & FL_UNLOCK_PENDING)
149 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
154 int leases_enable = 1;
155 int lease_break_time = 45;
157 #define for_each_lock(inode, lockp) \
158 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
161 * The global file_lock_list is only used for displaying /proc/locks, so we
162 * keep a list on each CPU, with each list protected by its own spinlock via
163 * the file_lock_lglock. Note that alterations to the list also require that
164 * the relevant i_lock is held.
166 DEFINE_STATIC_LGLOCK(file_lock_lglock);
167 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
170 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
171 * It is protected by blocked_lock_lock.
173 * We hash locks by lockowner in order to optimize searching for the lock a
174 * particular lockowner is waiting on.
176 * FIXME: make this value scale via some heuristic? We generally will want more
177 * buckets when we have more lockowners holding locks, but that's a little
178 * difficult to determine without knowing what the workload will look like.
180 #define BLOCKED_HASH_BITS 7
181 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
184 * This lock protects the blocked_hash. Generally, if you're accessing it, you
185 * want to be holding this lock.
187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
188 * pointer for file_lock structures that are acting as lock requests (in
189 * contrast to those that are acting as records of acquired locks).
191 * Note that when we acquire this lock in order to change the above fields,
192 * we often hold the i_lock as well. In certain cases, when reading the fields
193 * protected by this lock, we can skip acquiring it iff we already hold the
196 * In particular, adding an entry to the fl_block list requires that you hold
197 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
198 * an entry from the list however only requires the file_lock_lock.
200 static DEFINE_SPINLOCK(blocked_lock_lock);
202 static struct kmem_cache *filelock_cache __read_mostly;
204 static void locks_init_lock_heads(struct file_lock *fl)
206 INIT_HLIST_NODE(&fl->fl_link);
207 INIT_LIST_HEAD(&fl->fl_block);
208 init_waitqueue_head(&fl->fl_wait);
211 /* Allocate an empty lock structure. */
212 struct file_lock *locks_alloc_lock(void)
214 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
217 locks_init_lock_heads(fl);
221 EXPORT_SYMBOL_GPL(locks_alloc_lock);
223 void locks_release_private(struct file_lock *fl)
226 if (fl->fl_ops->fl_release_private)
227 fl->fl_ops->fl_release_private(fl);
233 EXPORT_SYMBOL_GPL(locks_release_private);
235 /* Free a lock which is not in use. */
236 void locks_free_lock(struct file_lock *fl)
238 BUG_ON(waitqueue_active(&fl->fl_wait));
239 BUG_ON(!list_empty(&fl->fl_block));
240 BUG_ON(!hlist_unhashed(&fl->fl_link));
242 locks_release_private(fl);
243 kmem_cache_free(filelock_cache, fl);
245 EXPORT_SYMBOL(locks_free_lock);
247 void locks_init_lock(struct file_lock *fl)
249 memset(fl, 0, sizeof(struct file_lock));
250 locks_init_lock_heads(fl);
253 EXPORT_SYMBOL(locks_init_lock);
255 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
258 if (fl->fl_ops->fl_copy_lock)
259 fl->fl_ops->fl_copy_lock(new, fl);
260 new->fl_ops = fl->fl_ops;
263 new->fl_lmops = fl->fl_lmops;
267 * Initialize a new lock from an existing file_lock structure.
269 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
271 new->fl_owner = fl->fl_owner;
272 new->fl_pid = fl->fl_pid;
274 new->fl_flags = fl->fl_flags;
275 new->fl_type = fl->fl_type;
276 new->fl_start = fl->fl_start;
277 new->fl_end = fl->fl_end;
279 new->fl_lmops = NULL;
281 EXPORT_SYMBOL(__locks_copy_lock);
283 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
285 locks_release_private(new);
287 __locks_copy_lock(new, fl);
288 new->fl_file = fl->fl_file;
289 new->fl_ops = fl->fl_ops;
290 new->fl_lmops = fl->fl_lmops;
292 locks_copy_private(new, fl);
295 EXPORT_SYMBOL(locks_copy_lock);
297 static inline int flock_translate_cmd(int cmd) {
299 return cmd & (LOCK_MAND | LOCK_RW);
311 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
312 static int flock_make_lock(struct file *filp, struct file_lock **lock,
315 struct file_lock *fl;
316 int type = flock_translate_cmd(cmd);
320 fl = locks_alloc_lock();
325 fl->fl_pid = current->tgid;
326 fl->fl_flags = FL_FLOCK;
328 fl->fl_end = OFFSET_MAX;
334 static int assign_type(struct file_lock *fl, long type)
348 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
351 switch (l->l_whence) {
356 fl->fl_start = filp->f_pos;
359 fl->fl_start = i_size_read(file_inode(filp));
364 if (l->l_start > OFFSET_MAX - fl->fl_start)
366 fl->fl_start += l->l_start;
367 if (fl->fl_start < 0)
370 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
371 POSIX-2001 defines it. */
373 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
375 fl->fl_end = fl->fl_start + l->l_len - 1;
377 } else if (l->l_len < 0) {
378 if (fl->fl_start + l->l_len < 0)
380 fl->fl_end = fl->fl_start - 1;
381 fl->fl_start += l->l_len;
383 fl->fl_end = OFFSET_MAX;
385 fl->fl_owner = current->files;
386 fl->fl_pid = current->tgid;
388 fl->fl_flags = FL_POSIX;
392 /* Ensure that fl->fl_filp has compatible f_mode */
395 if (!(filp->f_mode & FMODE_READ))
399 if (!(filp->f_mode & FMODE_WRITE))
404 return assign_type(fl, l->l_type);
407 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
410 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
413 struct flock64 ll = {
415 .l_whence = l->l_whence,
416 .l_start = l->l_start,
420 return flock64_to_posix_lock(filp, fl, &ll);
423 /* default lease lock manager operations */
424 static void lease_break_callback(struct file_lock *fl)
426 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
429 static const struct lock_manager_operations lease_manager_ops = {
430 .lm_break = lease_break_callback,
431 .lm_change = lease_modify,
435 * Initialize a lease, use the default lock manager operations
437 static int lease_init(struct file *filp, long type, struct file_lock *fl)
439 if (assign_type(fl, type) != 0)
442 fl->fl_owner = current->files;
443 fl->fl_pid = current->tgid;
446 fl->fl_flags = FL_LEASE;
448 fl->fl_end = OFFSET_MAX;
450 fl->fl_lmops = &lease_manager_ops;
454 /* Allocate a file_lock initialised to this type of lease */
455 static struct file_lock *lease_alloc(struct file *filp, long type)
457 struct file_lock *fl = locks_alloc_lock();
461 return ERR_PTR(error);
463 error = lease_init(filp, type, fl);
466 return ERR_PTR(error);
471 /* Check if two locks overlap each other.
473 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
475 return ((fl1->fl_end >= fl2->fl_start) &&
476 (fl2->fl_end >= fl1->fl_start));
480 * Check whether two locks have the same owner.
482 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
484 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
485 return fl2->fl_lmops == fl1->fl_lmops &&
486 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
487 return fl1->fl_owner == fl2->fl_owner;
490 /* Must be called with the i_lock held! */
491 static void locks_insert_global_locks(struct file_lock *fl)
493 lg_local_lock(&file_lock_lglock);
494 fl->fl_link_cpu = smp_processor_id();
495 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
496 lg_local_unlock(&file_lock_lglock);
499 /* Must be called with the i_lock held! */
500 static void locks_delete_global_locks(struct file_lock *fl)
503 * Avoid taking lock if already unhashed. This is safe since this check
504 * is done while holding the i_lock, and new insertions into the list
505 * also require that it be held.
507 if (hlist_unhashed(&fl->fl_link))
509 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
510 hlist_del_init(&fl->fl_link);
511 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
515 posix_owner_key(struct file_lock *fl)
517 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
518 return fl->fl_lmops->lm_owner_key(fl);
519 return (unsigned long)fl->fl_owner;
522 static void locks_insert_global_blocked(struct file_lock *waiter)
524 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
527 static void locks_delete_global_blocked(struct file_lock *waiter)
529 hash_del(&waiter->fl_link);
532 /* Remove waiter from blocker's block list.
533 * When blocker ends up pointing to itself then the list is empty.
535 * Must be called with blocked_lock_lock held.
537 static void __locks_delete_block(struct file_lock *waiter)
539 locks_delete_global_blocked(waiter);
540 list_del_init(&waiter->fl_block);
541 waiter->fl_next = NULL;
544 static void locks_delete_block(struct file_lock *waiter)
546 spin_lock(&blocked_lock_lock);
547 __locks_delete_block(waiter);
548 spin_unlock(&blocked_lock_lock);
551 /* Insert waiter into blocker's block list.
552 * We use a circular list so that processes can be easily woken up in
553 * the order they blocked. The documentation doesn't require this but
554 * it seems like the reasonable thing to do.
556 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
557 * list itself is protected by the blocked_lock_lock, but by ensuring that the
558 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
559 * in some cases when we see that the fl_block list is empty.
561 static void __locks_insert_block(struct file_lock *blocker,
562 struct file_lock *waiter)
564 BUG_ON(!list_empty(&waiter->fl_block));
565 waiter->fl_next = blocker;
566 list_add_tail(&waiter->fl_block, &blocker->fl_block);
567 if (IS_POSIX(blocker) && !IS_FILE_PVT(blocker))
568 locks_insert_global_blocked(waiter);
571 /* Must be called with i_lock held. */
572 static void locks_insert_block(struct file_lock *blocker,
573 struct file_lock *waiter)
575 spin_lock(&blocked_lock_lock);
576 __locks_insert_block(blocker, waiter);
577 spin_unlock(&blocked_lock_lock);
581 * Wake up processes blocked waiting for blocker.
583 * Must be called with the inode->i_lock held!
585 static void locks_wake_up_blocks(struct file_lock *blocker)
588 * Avoid taking global lock if list is empty. This is safe since new
589 * blocked requests are only added to the list under the i_lock, and
590 * the i_lock is always held here. Note that removal from the fl_block
591 * list does not require the i_lock, so we must recheck list_empty()
592 * after acquiring the blocked_lock_lock.
594 if (list_empty(&blocker->fl_block))
597 spin_lock(&blocked_lock_lock);
598 while (!list_empty(&blocker->fl_block)) {
599 struct file_lock *waiter;
601 waiter = list_first_entry(&blocker->fl_block,
602 struct file_lock, fl_block);
603 __locks_delete_block(waiter);
604 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
605 waiter->fl_lmops->lm_notify(waiter);
607 wake_up(&waiter->fl_wait);
609 spin_unlock(&blocked_lock_lock);
612 /* Insert file lock fl into an inode's lock list at the position indicated
613 * by pos. At the same time add the lock to the global file lock list.
615 * Must be called with the i_lock held!
617 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
619 fl->fl_nspid = get_pid(task_tgid(current));
621 /* insert into file's list */
625 locks_insert_global_locks(fl);
629 * locks_delete_lock - Delete a lock and then free it.
630 * @thisfl_p: pointer that points to the fl_next field of the previous
631 * inode->i_flock list entry
633 * Unlink a lock from all lists and free the namespace reference, but don't
634 * free it yet. Wake up processes that are blocked waiting for this lock and
635 * notify the FS that the lock has been cleared.
637 * Must be called with the i_lock held!
639 static void locks_unlink_lock(struct file_lock **thisfl_p)
641 struct file_lock *fl = *thisfl_p;
643 locks_delete_global_locks(fl);
645 *thisfl_p = fl->fl_next;
649 put_pid(fl->fl_nspid);
653 locks_wake_up_blocks(fl);
657 * Unlink a lock from all lists and free it.
659 * Must be called with i_lock held!
661 static void locks_delete_lock(struct file_lock **thisfl_p)
663 struct file_lock *fl = *thisfl_p;
665 locks_unlink_lock(thisfl_p);
669 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
670 * checks for shared/exclusive status of overlapping locks.
672 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
674 if (sys_fl->fl_type == F_WRLCK)
676 if (caller_fl->fl_type == F_WRLCK)
681 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
682 * checking before calling the locks_conflict().
684 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
686 /* POSIX locks owned by the same process do not conflict with
689 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
692 /* Check whether they overlap */
693 if (!locks_overlap(caller_fl, sys_fl))
696 return (locks_conflict(caller_fl, sys_fl));
699 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
700 * checking before calling the locks_conflict().
702 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
704 /* FLOCK locks referring to the same filp do not conflict with
707 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
709 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
712 return (locks_conflict(caller_fl, sys_fl));
716 posix_test_lock(struct file *filp, struct file_lock *fl)
718 struct file_lock *cfl;
719 struct inode *inode = file_inode(filp);
721 spin_lock(&inode->i_lock);
722 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
725 if (posix_locks_conflict(fl, cfl))
729 __locks_copy_lock(fl, cfl);
731 fl->fl_pid = pid_vnr(cfl->fl_nspid);
733 fl->fl_type = F_UNLCK;
734 spin_unlock(&inode->i_lock);
737 EXPORT_SYMBOL(posix_test_lock);
740 * Deadlock detection:
742 * We attempt to detect deadlocks that are due purely to posix file
745 * We assume that a task can be waiting for at most one lock at a time.
746 * So for any acquired lock, the process holding that lock may be
747 * waiting on at most one other lock. That lock in turns may be held by
748 * someone waiting for at most one other lock. Given a requested lock
749 * caller_fl which is about to wait for a conflicting lock block_fl, we
750 * follow this chain of waiters to ensure we are not about to create a
753 * Since we do this before we ever put a process to sleep on a lock, we
754 * are ensured that there is never a cycle; that is what guarantees that
755 * the while() loop in posix_locks_deadlock() eventually completes.
757 * Note: the above assumption may not be true when handling lock
758 * requests from a broken NFS client. It may also fail in the presence
759 * of tasks (such as posix threads) sharing the same open file table.
760 * To handle those cases, we just bail out after a few iterations.
762 * For FL_FILE_PVT locks, the owner is the filp, not the files_struct.
763 * Because the owner is not even nominally tied to a thread of
764 * execution, the deadlock detection below can't reasonably work well. Just
767 * In principle, we could do a more limited deadlock detection on FL_FILE_PVT
768 * locks that just checks for the case where two tasks are attempting to
769 * upgrade from read to write locks on the same inode.
772 #define MAX_DEADLK_ITERATIONS 10
774 /* Find a lock that the owner of the given block_fl is blocking on. */
775 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
777 struct file_lock *fl;
779 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
780 if (posix_same_owner(fl, block_fl))
786 /* Must be called with the blocked_lock_lock held! */
787 static int posix_locks_deadlock(struct file_lock *caller_fl,
788 struct file_lock *block_fl)
793 * This deadlock detector can't reasonably detect deadlocks with
794 * FL_FILE_PVT locks, since they aren't owned by a process, per-se.
796 if (IS_FILE_PVT(caller_fl))
799 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
800 if (i++ > MAX_DEADLK_ITERATIONS)
802 if (posix_same_owner(caller_fl, block_fl))
808 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
809 * after any leases, but before any posix locks.
811 * Note that if called with an FL_EXISTS argument, the caller may determine
812 * whether or not a lock was successfully freed by testing the return
815 static int flock_lock_file(struct file *filp, struct file_lock *request)
817 struct file_lock *new_fl = NULL;
818 struct file_lock **before;
819 struct inode * inode = file_inode(filp);
823 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
824 new_fl = locks_alloc_lock();
829 spin_lock(&inode->i_lock);
830 if (request->fl_flags & FL_ACCESS)
833 for_each_lock(inode, before) {
834 struct file_lock *fl = *before;
839 if (filp != fl->fl_file)
841 if (request->fl_type == fl->fl_type)
844 locks_delete_lock(before);
848 if (request->fl_type == F_UNLCK) {
849 if ((request->fl_flags & FL_EXISTS) && !found)
855 * If a higher-priority process was blocked on the old file lock,
856 * give it the opportunity to lock the file.
859 spin_unlock(&inode->i_lock);
861 spin_lock(&inode->i_lock);
865 for_each_lock(inode, before) {
866 struct file_lock *fl = *before;
871 if (!flock_locks_conflict(request, fl))
874 if (!(request->fl_flags & FL_SLEEP))
876 error = FILE_LOCK_DEFERRED;
877 locks_insert_block(fl, request);
880 if (request->fl_flags & FL_ACCESS)
882 locks_copy_lock(new_fl, request);
883 locks_insert_lock(before, new_fl);
888 spin_unlock(&inode->i_lock);
890 locks_free_lock(new_fl);
894 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
896 struct file_lock *fl;
897 struct file_lock *new_fl = NULL;
898 struct file_lock *new_fl2 = NULL;
899 struct file_lock *left = NULL;
900 struct file_lock *right = NULL;
901 struct file_lock **before;
906 * We may need two file_lock structures for this operation,
907 * so we get them in advance to avoid races.
909 * In some cases we can be sure, that no new locks will be needed
911 if (!(request->fl_flags & FL_ACCESS) &&
912 (request->fl_type != F_UNLCK ||
913 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
914 new_fl = locks_alloc_lock();
915 new_fl2 = locks_alloc_lock();
918 spin_lock(&inode->i_lock);
920 * New lock request. Walk all POSIX locks and look for conflicts. If
921 * there are any, either return error or put the request on the
922 * blocker's list of waiters and the global blocked_hash.
924 if (request->fl_type != F_UNLCK) {
925 for_each_lock(inode, before) {
929 if (!posix_locks_conflict(request, fl))
932 __locks_copy_lock(conflock, fl);
934 if (!(request->fl_flags & FL_SLEEP))
937 * Deadlock detection and insertion into the blocked
938 * locks list must be done while holding the same lock!
941 spin_lock(&blocked_lock_lock);
942 if (likely(!posix_locks_deadlock(request, fl))) {
943 error = FILE_LOCK_DEFERRED;
944 __locks_insert_block(fl, request);
946 spin_unlock(&blocked_lock_lock);
951 /* If we're just looking for a conflict, we're done. */
953 if (request->fl_flags & FL_ACCESS)
957 * Find the first old lock with the same owner as the new lock.
960 before = &inode->i_flock;
962 /* First skip locks owned by other processes. */
963 while ((fl = *before) && (!IS_POSIX(fl) ||
964 !posix_same_owner(request, fl))) {
965 before = &fl->fl_next;
968 /* Process locks with this owner. */
969 while ((fl = *before) && posix_same_owner(request, fl)) {
970 /* Detect adjacent or overlapping regions (if same lock type)
972 if (request->fl_type == fl->fl_type) {
973 /* In all comparisons of start vs end, use
974 * "start - 1" rather than "end + 1". If end
975 * is OFFSET_MAX, end + 1 will become negative.
977 if (fl->fl_end < request->fl_start - 1)
979 /* If the next lock in the list has entirely bigger
980 * addresses than the new one, insert the lock here.
982 if (fl->fl_start - 1 > request->fl_end)
985 /* If we come here, the new and old lock are of the
986 * same type and adjacent or overlapping. Make one
987 * lock yielding from the lower start address of both
988 * locks to the higher end address.
990 if (fl->fl_start > request->fl_start)
991 fl->fl_start = request->fl_start;
993 request->fl_start = fl->fl_start;
994 if (fl->fl_end < request->fl_end)
995 fl->fl_end = request->fl_end;
997 request->fl_end = fl->fl_end;
999 locks_delete_lock(before);
1006 /* Processing for different lock types is a bit
1009 if (fl->fl_end < request->fl_start)
1011 if (fl->fl_start > request->fl_end)
1013 if (request->fl_type == F_UNLCK)
1015 if (fl->fl_start < request->fl_start)
1017 /* If the next lock in the list has a higher end
1018 * address than the new one, insert the new one here.
1020 if (fl->fl_end > request->fl_end) {
1024 if (fl->fl_start >= request->fl_start) {
1025 /* The new lock completely replaces an old
1026 * one (This may happen several times).
1029 locks_delete_lock(before);
1032 /* Replace the old lock with the new one.
1033 * Wake up anybody waiting for the old one,
1034 * as the change in lock type might satisfy
1037 locks_wake_up_blocks(fl);
1038 fl->fl_start = request->fl_start;
1039 fl->fl_end = request->fl_end;
1040 fl->fl_type = request->fl_type;
1041 locks_release_private(fl);
1042 locks_copy_private(fl, request);
1047 /* Go on to next lock.
1050 before = &fl->fl_next;
1054 * The above code only modifies existing locks in case of merging or
1055 * replacing. If new lock(s) need to be inserted all modifications are
1056 * done below this, so it's safe yet to bail out.
1058 error = -ENOLCK; /* "no luck" */
1059 if (right && left == right && !new_fl2)
1064 if (request->fl_type == F_UNLCK) {
1065 if (request->fl_flags & FL_EXISTS)
1074 locks_copy_lock(new_fl, request);
1075 locks_insert_lock(before, new_fl);
1079 if (left == right) {
1080 /* The new lock breaks the old one in two pieces,
1081 * so we have to use the second new lock.
1085 locks_copy_lock(left, right);
1086 locks_insert_lock(before, left);
1088 right->fl_start = request->fl_end + 1;
1089 locks_wake_up_blocks(right);
1092 left->fl_end = request->fl_start - 1;
1093 locks_wake_up_blocks(left);
1096 spin_unlock(&inode->i_lock);
1098 * Free any unused locks.
1101 locks_free_lock(new_fl);
1103 locks_free_lock(new_fl2);
1108 * posix_lock_file - Apply a POSIX-style lock to a file
1109 * @filp: The file to apply the lock to
1110 * @fl: The lock to be applied
1111 * @conflock: Place to return a copy of the conflicting lock, if found.
1113 * Add a POSIX style lock to a file.
1114 * We merge adjacent & overlapping locks whenever possible.
1115 * POSIX locks are sorted by owner task, then by starting address
1117 * Note that if called with an FL_EXISTS argument, the caller may determine
1118 * whether or not a lock was successfully freed by testing the return
1119 * value for -ENOENT.
1121 int posix_lock_file(struct file *filp, struct file_lock *fl,
1122 struct file_lock *conflock)
1124 return __posix_lock_file(file_inode(filp), fl, conflock);
1126 EXPORT_SYMBOL(posix_lock_file);
1129 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1130 * @filp: The file to apply the lock to
1131 * @fl: The lock to be applied
1133 * Add a POSIX style lock to a file.
1134 * We merge adjacent & overlapping locks whenever possible.
1135 * POSIX locks are sorted by owner task, then by starting address
1137 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1142 error = posix_lock_file(filp, fl, NULL);
1143 if (error != FILE_LOCK_DEFERRED)
1145 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1149 locks_delete_block(fl);
1154 EXPORT_SYMBOL(posix_lock_file_wait);
1157 * locks_mandatory_locked - Check for an active lock
1158 * @file: the file to check
1160 * Searches the inode's list of locks to find any POSIX locks which conflict.
1161 * This function is called from locks_verify_locked() only.
1163 int locks_mandatory_locked(struct file *file)
1165 struct inode *inode = file_inode(file);
1166 fl_owner_t owner = current->files;
1167 struct file_lock *fl;
1170 * Search the lock list for this inode for any POSIX locks.
1172 spin_lock(&inode->i_lock);
1173 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1176 if (fl->fl_owner != owner && fl->fl_owner != (fl_owner_t)file)
1179 spin_unlock(&inode->i_lock);
1180 return fl ? -EAGAIN : 0;
1184 * locks_mandatory_area - Check for a conflicting lock
1185 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1187 * @inode: the file to check
1188 * @filp: how the file was opened (if it was)
1189 * @offset: start of area to check
1190 * @count: length of area to check
1192 * Searches the inode's list of locks to find any POSIX locks which conflict.
1193 * This function is called from rw_verify_area() and
1194 * locks_verify_truncate().
1196 int locks_mandatory_area(int read_write, struct inode *inode,
1197 struct file *filp, loff_t offset,
1200 struct file_lock fl;
1203 locks_init_lock(&fl);
1204 fl.fl_owner = current->files;
1205 fl.fl_pid = current->tgid;
1207 fl.fl_flags = FL_POSIX | FL_ACCESS;
1208 if (filp && !(filp->f_flags & O_NONBLOCK))
1209 fl.fl_flags |= FL_SLEEP;
1210 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1211 fl.fl_start = offset;
1212 fl.fl_end = offset + count - 1;
1215 error = __posix_lock_file(inode, &fl, NULL);
1216 if (error != FILE_LOCK_DEFERRED)
1218 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1221 * If we've been sleeping someone might have
1222 * changed the permissions behind our back.
1224 if (__mandatory_lock(inode))
1228 locks_delete_block(&fl);
1235 EXPORT_SYMBOL(locks_mandatory_area);
1237 static void lease_clear_pending(struct file_lock *fl, int arg)
1241 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1244 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1248 /* We already had a lease on this file; just change its type */
1249 int lease_modify(struct file_lock **before, int arg)
1251 struct file_lock *fl = *before;
1252 int error = assign_type(fl, arg);
1256 lease_clear_pending(fl, arg);
1257 locks_wake_up_blocks(fl);
1258 if (arg == F_UNLCK) {
1259 struct file *filp = fl->fl_file;
1262 filp->f_owner.signum = 0;
1263 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1264 if (fl->fl_fasync != NULL) {
1265 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1266 fl->fl_fasync = NULL;
1268 locks_delete_lock(before);
1273 EXPORT_SYMBOL(lease_modify);
1275 static bool past_time(unsigned long then)
1278 /* 0 is a special value meaning "this never expires": */
1280 return time_after(jiffies, then);
1283 static void time_out_leases(struct inode *inode)
1285 struct file_lock **before;
1286 struct file_lock *fl;
1288 before = &inode->i_flock;
1289 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1290 if (past_time(fl->fl_downgrade_time))
1291 lease_modify(before, F_RDLCK);
1292 if (past_time(fl->fl_break_time))
1293 lease_modify(before, F_UNLCK);
1294 if (fl == *before) /* lease_modify may have freed fl */
1295 before = &fl->fl_next;
1299 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1301 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1303 return locks_conflict(breaker, lease);
1307 * __break_lease - revoke all outstanding leases on file
1308 * @inode: the inode of the file to return
1309 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1311 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1314 * break_lease (inlined for speed) has checked there already is at least
1315 * some kind of lock (maybe a lease) on this file. Leases are broken on
1316 * a call to open() or truncate(). This function can sleep unless you
1317 * specified %O_NONBLOCK to your open().
1319 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1322 struct file_lock *new_fl, *flock;
1323 struct file_lock *fl;
1324 unsigned long break_time;
1325 int i_have_this_lease = 0;
1326 bool lease_conflict = false;
1327 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1329 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1331 return PTR_ERR(new_fl);
1332 new_fl->fl_flags = type;
1334 spin_lock(&inode->i_lock);
1336 time_out_leases(inode);
1338 flock = inode->i_flock;
1339 if ((flock == NULL) || !IS_LEASE(flock))
1342 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1343 if (leases_conflict(fl, new_fl)) {
1344 lease_conflict = true;
1345 if (fl->fl_owner == current->files)
1346 i_have_this_lease = 1;
1349 if (!lease_conflict)
1353 if (lease_break_time > 0) {
1354 break_time = jiffies + lease_break_time * HZ;
1355 if (break_time == 0)
1356 break_time++; /* so that 0 means no break time */
1359 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1360 if (!leases_conflict(fl, new_fl))
1363 if (fl->fl_flags & FL_UNLOCK_PENDING)
1365 fl->fl_flags |= FL_UNLOCK_PENDING;
1366 fl->fl_break_time = break_time;
1368 if (lease_breaking(flock))
1370 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1371 fl->fl_downgrade_time = break_time;
1373 fl->fl_lmops->lm_break(fl);
1376 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1377 error = -EWOULDBLOCK;
1382 break_time = flock->fl_break_time;
1383 if (break_time != 0) {
1384 break_time -= jiffies;
1385 if (break_time == 0)
1388 locks_insert_block(flock, new_fl);
1389 spin_unlock(&inode->i_lock);
1390 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1391 !new_fl->fl_next, break_time);
1392 spin_lock(&inode->i_lock);
1393 locks_delete_block(new_fl);
1396 time_out_leases(inode);
1398 * Wait for the next conflicting lease that has not been
1401 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1402 flock = flock->fl_next) {
1403 if (leases_conflict(new_fl, flock))
1410 spin_unlock(&inode->i_lock);
1411 locks_free_lock(new_fl);
1415 EXPORT_SYMBOL(__break_lease);
1418 * lease_get_mtime - get the last modified time of an inode
1420 * @time: pointer to a timespec which will contain the last modified time
1422 * This is to force NFS clients to flush their caches for files with
1423 * exclusive leases. The justification is that if someone has an
1424 * exclusive lease, then they could be modifying it.
1426 void lease_get_mtime(struct inode *inode, struct timespec *time)
1428 struct file_lock *flock = inode->i_flock;
1429 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1430 *time = current_fs_time(inode->i_sb);
1432 *time = inode->i_mtime;
1435 EXPORT_SYMBOL(lease_get_mtime);
1438 * fcntl_getlease - Enquire what lease is currently active
1441 * The value returned by this function will be one of
1442 * (if no lease break is pending):
1444 * %F_RDLCK to indicate a shared lease is held.
1446 * %F_WRLCK to indicate an exclusive lease is held.
1448 * %F_UNLCK to indicate no lease is held.
1450 * (if a lease break is pending):
1452 * %F_RDLCK to indicate an exclusive lease needs to be
1453 * changed to a shared lease (or removed).
1455 * %F_UNLCK to indicate the lease needs to be removed.
1457 * XXX: sfr & willy disagree over whether F_INPROGRESS
1458 * should be returned to userspace.
1460 int fcntl_getlease(struct file *filp)
1462 struct file_lock *fl;
1463 struct inode *inode = file_inode(filp);
1466 spin_lock(&inode->i_lock);
1467 time_out_leases(file_inode(filp));
1468 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1470 if (fl->fl_file == filp) {
1471 type = target_leasetype(fl);
1475 spin_unlock(&inode->i_lock);
1480 * check_conflicting_open - see if the given dentry points to a file that has
1481 * an existing open that would conflict with the
1483 * @dentry: dentry to check
1484 * @arg: type of lease that we're trying to acquire
1486 * Check to see if there's an existing open fd on this file that would
1487 * conflict with the lease we're trying to set.
1490 check_conflicting_open(const struct dentry *dentry, const long arg)
1493 struct inode *inode = dentry->d_inode;
1495 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1498 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1499 (atomic_read(&inode->i_count) > 1)))
1505 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1507 struct file_lock *fl, **before, **my_before = NULL, *lease;
1508 struct dentry *dentry = filp->f_path.dentry;
1509 struct inode *inode = dentry->d_inode;
1510 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1515 * In the delegation case we need mutual exclusion with
1516 * a number of operations that take the i_mutex. We trylock
1517 * because delegations are an optional optimization, and if
1518 * there's some chance of a conflict--we'd rather not
1519 * bother, maybe that's a sign this just isn't a good file to
1520 * hand out a delegation on.
1522 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1525 if (is_deleg && arg == F_WRLCK) {
1526 /* Write delegations are not currently supported: */
1527 mutex_unlock(&inode->i_mutex);
1532 error = check_conflicting_open(dentry, arg);
1537 * At this point, we know that if there is an exclusive
1538 * lease on this file, then we hold it on this filp
1539 * (otherwise our open of this file would have blocked).
1540 * And if we are trying to acquire an exclusive lease,
1541 * then the file is not open by anyone (including us)
1542 * except for this filp.
1545 for (before = &inode->i_flock;
1546 ((fl = *before) != NULL) && IS_LEASE(fl);
1547 before = &fl->fl_next) {
1548 if (fl->fl_file == filp) {
1553 * No exclusive leases if someone else has a lease on
1559 * Modifying our existing lease is OK, but no getting a
1560 * new lease if someone else is opening for write:
1562 if (fl->fl_flags & FL_UNLOCK_PENDING)
1566 if (my_before != NULL) {
1567 error = lease->fl_lmops->lm_change(my_before, arg);
1577 locks_insert_lock(before, lease);
1579 * The check in break_lease() is lockless. It's possible for another
1580 * open to race in after we did the earlier check for a conflicting
1581 * open but before the lease was inserted. Check again for a
1582 * conflicting open and cancel the lease if there is one.
1584 * We also add a barrier here to ensure that the insertion of the lock
1585 * precedes these checks.
1588 error = check_conflicting_open(dentry, arg);
1590 locks_unlink_lock(flp);
1593 mutex_unlock(&inode->i_mutex);
1597 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1599 struct file_lock *fl, **before;
1600 struct dentry *dentry = filp->f_path.dentry;
1601 struct inode *inode = dentry->d_inode;
1603 for (before = &inode->i_flock;
1604 ((fl = *before) != NULL) && IS_LEASE(fl);
1605 before = &fl->fl_next) {
1606 if (fl->fl_file != filp)
1608 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1614 * generic_setlease - sets a lease on an open file
1615 * @filp: file pointer
1616 * @arg: type of lease to obtain
1617 * @flp: input - file_lock to use, output - file_lock inserted
1619 * The (input) flp->fl_lmops->lm_break function is required
1622 * Called with inode->i_lock held.
1624 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1626 struct dentry *dentry = filp->f_path.dentry;
1627 struct inode *inode = dentry->d_inode;
1630 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1632 if (!S_ISREG(inode->i_mode))
1634 error = security_file_lock(filp, arg);
1638 time_out_leases(inode);
1640 BUG_ON(!(*flp)->fl_lmops->lm_break);
1644 return generic_delete_lease(filp, flp);
1647 return generic_add_lease(filp, arg, flp);
1652 EXPORT_SYMBOL(generic_setlease);
1654 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1656 if (filp->f_op->setlease)
1657 return filp->f_op->setlease(filp, arg, lease);
1659 return generic_setlease(filp, arg, lease);
1663 * vfs_setlease - sets a lease on an open file
1664 * @filp: file pointer
1665 * @arg: type of lease to obtain
1666 * @lease: file_lock to use
1668 * Call this to establish a lease on the file.
1669 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1670 * break_lease will oops!
1672 * This will call the filesystem's setlease file method, if
1673 * defined. Note that there is no getlease method; instead, the
1674 * filesystem setlease method should call back to setlease() to
1675 * add a lease to the inode's lease list, where fcntl_getlease() can
1676 * find it. Since fcntl_getlease() only reports whether the current
1677 * task holds a lease, a cluster filesystem need only do this for
1678 * leases held by processes on this node.
1680 * There is also no break_lease method; filesystems that
1681 * handle their own leases should break leases themselves from the
1682 * filesystem's open, create, and (on truncate) setattr methods.
1684 * Warning: the only current setlease methods exist only to disable
1685 * leases in certain cases. More vfs changes may be required to
1686 * allow a full filesystem lease implementation.
1689 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1691 struct inode *inode = file_inode(filp);
1694 spin_lock(&inode->i_lock);
1695 error = __vfs_setlease(filp, arg, lease);
1696 spin_unlock(&inode->i_lock);
1700 EXPORT_SYMBOL_GPL(vfs_setlease);
1702 static int do_fcntl_delete_lease(struct file *filp)
1704 struct file_lock fl, *flp = &fl;
1706 lease_init(filp, F_UNLCK, flp);
1708 return vfs_setlease(filp, F_UNLCK, &flp);
1711 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1713 struct file_lock *fl, *ret;
1714 struct inode *inode = file_inode(filp);
1715 struct fasync_struct *new;
1718 fl = lease_alloc(filp, arg);
1722 new = fasync_alloc();
1724 locks_free_lock(fl);
1728 spin_lock(&inode->i_lock);
1729 error = __vfs_setlease(filp, arg, &ret);
1731 spin_unlock(&inode->i_lock);
1732 locks_free_lock(fl);
1733 goto out_free_fasync;
1736 locks_free_lock(fl);
1739 * fasync_insert_entry() returns the old entry if any.
1740 * If there was no old entry, then it used 'new' and
1741 * inserted it into the fasync list. Clear new so that
1742 * we don't release it here.
1744 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1747 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1748 spin_unlock(&inode->i_lock);
1757 * fcntl_setlease - sets a lease on an open file
1758 * @fd: open file descriptor
1759 * @filp: file pointer
1760 * @arg: type of lease to obtain
1762 * Call this fcntl to establish a lease on the file.
1763 * Note that you also need to call %F_SETSIG to
1764 * receive a signal when the lease is broken.
1766 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1769 return do_fcntl_delete_lease(filp);
1770 return do_fcntl_add_lease(fd, filp, arg);
1774 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1775 * @filp: The file to apply the lock to
1776 * @fl: The lock to be applied
1778 * Add a FLOCK style lock to a file.
1780 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1785 error = flock_lock_file(filp, fl);
1786 if (error != FILE_LOCK_DEFERRED)
1788 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1792 locks_delete_block(fl);
1798 EXPORT_SYMBOL(flock_lock_file_wait);
1801 * sys_flock: - flock() system call.
1802 * @fd: the file descriptor to lock.
1803 * @cmd: the type of lock to apply.
1805 * Apply a %FL_FLOCK style lock to an open file descriptor.
1806 * The @cmd can be one of
1808 * %LOCK_SH -- a shared lock.
1810 * %LOCK_EX -- an exclusive lock.
1812 * %LOCK_UN -- remove an existing lock.
1814 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1816 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1817 * processes read and write access respectively.
1819 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1821 struct fd f = fdget(fd);
1822 struct file_lock *lock;
1823 int can_sleep, unlock;
1830 can_sleep = !(cmd & LOCK_NB);
1832 unlock = (cmd == LOCK_UN);
1834 if (!unlock && !(cmd & LOCK_MAND) &&
1835 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1838 error = flock_make_lock(f.file, &lock, cmd);
1842 lock->fl_flags |= FL_SLEEP;
1844 error = security_file_lock(f.file, lock->fl_type);
1848 if (f.file->f_op->flock)
1849 error = f.file->f_op->flock(f.file,
1850 (can_sleep) ? F_SETLKW : F_SETLK,
1853 error = flock_lock_file_wait(f.file, lock);
1856 locks_free_lock(lock);
1865 * vfs_test_lock - test file byte range lock
1866 * @filp: The file to test lock for
1867 * @fl: The lock to test; also used to hold result
1869 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1870 * setting conf->fl_type to something other than F_UNLCK.
1872 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1874 if (filp->f_op->lock)
1875 return filp->f_op->lock(filp, F_GETLK, fl);
1876 posix_test_lock(filp, fl);
1879 EXPORT_SYMBOL_GPL(vfs_test_lock);
1881 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1883 flock->l_pid = IS_FILE_PVT(fl) ? -1 : fl->fl_pid;
1884 #if BITS_PER_LONG == 32
1886 * Make sure we can represent the posix lock via
1887 * legacy 32bit flock.
1889 if (fl->fl_start > OFFT_OFFSET_MAX)
1891 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1894 flock->l_start = fl->fl_start;
1895 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1896 fl->fl_end - fl->fl_start + 1;
1897 flock->l_whence = 0;
1898 flock->l_type = fl->fl_type;
1902 #if BITS_PER_LONG == 32
1903 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1905 flock->l_pid = IS_FILE_PVT(fl) ? -1 : fl->fl_pid;
1906 flock->l_start = fl->fl_start;
1907 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1908 fl->fl_end - fl->fl_start + 1;
1909 flock->l_whence = 0;
1910 flock->l_type = fl->fl_type;
1914 /* Report the first existing lock that would conflict with l.
1915 * This implements the F_GETLK command of fcntl().
1917 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
1919 struct file_lock file_lock;
1924 if (copy_from_user(&flock, l, sizeof(flock)))
1927 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1930 error = flock_to_posix_lock(filp, &file_lock, &flock);
1934 if (cmd == F_GETLKP) {
1936 if (flock.l_pid != 0)
1940 file_lock.fl_flags |= FL_FILE_PVT;
1941 file_lock.fl_owner = (fl_owner_t)filp;
1944 error = vfs_test_lock(filp, &file_lock);
1948 flock.l_type = file_lock.fl_type;
1949 if (file_lock.fl_type != F_UNLCK) {
1950 error = posix_lock_to_flock(&flock, &file_lock);
1955 if (!copy_to_user(l, &flock, sizeof(flock)))
1962 * vfs_lock_file - file byte range lock
1963 * @filp: The file to apply the lock to
1964 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1965 * @fl: The lock to be applied
1966 * @conf: Place to return a copy of the conflicting lock, if found.
1968 * A caller that doesn't care about the conflicting lock may pass NULL
1969 * as the final argument.
1971 * If the filesystem defines a private ->lock() method, then @conf will
1972 * be left unchanged; so a caller that cares should initialize it to
1973 * some acceptable default.
1975 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1976 * locks, the ->lock() interface may return asynchronously, before the lock has
1977 * been granted or denied by the underlying filesystem, if (and only if)
1978 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1979 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1980 * the request is for a blocking lock. When ->lock() does return asynchronously,
1981 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1982 * request completes.
1983 * If the request is for non-blocking lock the file system should return
1984 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1985 * with the result. If the request timed out the callback routine will return a
1986 * nonzero return code and the file system should release the lock. The file
1987 * system is also responsible to keep a corresponding posix lock when it
1988 * grants a lock so the VFS can find out which locks are locally held and do
1989 * the correct lock cleanup when required.
1990 * The underlying filesystem must not drop the kernel lock or call
1991 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1994 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1996 if (filp->f_op->lock)
1997 return filp->f_op->lock(filp, cmd, fl);
1999 return posix_lock_file(filp, fl, conf);
2001 EXPORT_SYMBOL_GPL(vfs_lock_file);
2003 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2004 struct file_lock *fl)
2008 error = security_file_lock(filp, fl->fl_type);
2013 error = vfs_lock_file(filp, cmd, fl, NULL);
2014 if (error != FILE_LOCK_DEFERRED)
2016 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2020 locks_delete_block(fl);
2027 /* Apply the lock described by l to an open file descriptor.
2028 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2030 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2031 struct flock __user *l)
2033 struct file_lock *file_lock = locks_alloc_lock();
2035 struct inode *inode;
2039 if (file_lock == NULL)
2043 * This might block, so we do it before checking the inode.
2046 if (copy_from_user(&flock, l, sizeof(flock)))
2049 inode = file_inode(filp);
2051 /* Don't allow mandatory locks on files that may be memory mapped
2054 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2060 error = flock_to_posix_lock(filp, file_lock, &flock);
2065 * If the cmd is requesting file-private locks, then set the
2066 * FL_FILE_PVT flag and override the owner.
2071 if (flock.l_pid != 0)
2075 file_lock->fl_flags |= FL_FILE_PVT;
2076 file_lock->fl_owner = (fl_owner_t)filp;
2080 if (flock.l_pid != 0)
2084 file_lock->fl_flags |= FL_FILE_PVT;
2085 file_lock->fl_owner = (fl_owner_t)filp;
2088 file_lock->fl_flags |= FL_SLEEP;
2091 error = do_lock_file_wait(filp, cmd, file_lock);
2094 * Attempt to detect a close/fcntl race and recover by
2095 * releasing the lock that was just acquired.
2098 * we need that spin_lock here - it prevents reordering between
2099 * update of inode->i_flock and check for it done in close().
2100 * rcu_read_lock() wouldn't do.
2102 spin_lock(¤t->files->file_lock);
2104 spin_unlock(¤t->files->file_lock);
2105 if (!error && f != filp && flock.l_type != F_UNLCK) {
2106 flock.l_type = F_UNLCK;
2111 locks_free_lock(file_lock);
2115 #if BITS_PER_LONG == 32
2116 /* Report the first existing lock that would conflict with l.
2117 * This implements the F_GETLK command of fcntl().
2119 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
2121 struct file_lock file_lock;
2122 struct flock64 flock;
2126 if (copy_from_user(&flock, l, sizeof(flock)))
2129 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2132 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2136 if (cmd == F_GETLKP) {
2138 if (flock.l_pid != 0)
2142 file_lock.fl_flags |= FL_FILE_PVT;
2143 file_lock.fl_owner = (fl_owner_t)filp;
2146 error = vfs_test_lock(filp, &file_lock);
2150 flock.l_type = file_lock.fl_type;
2151 if (file_lock.fl_type != F_UNLCK)
2152 posix_lock_to_flock64(&flock, &file_lock);
2155 if (!copy_to_user(l, &flock, sizeof(flock)))
2162 /* Apply the lock described by l to an open file descriptor.
2163 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2165 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2166 struct flock64 __user *l)
2168 struct file_lock *file_lock = locks_alloc_lock();
2169 struct flock64 flock;
2170 struct inode *inode;
2174 if (file_lock == NULL)
2178 * This might block, so we do it before checking the inode.
2181 if (copy_from_user(&flock, l, sizeof(flock)))
2184 inode = file_inode(filp);
2186 /* Don't allow mandatory locks on files that may be memory mapped
2189 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2195 error = flock64_to_posix_lock(filp, file_lock, &flock);
2200 * If the cmd is requesting file-private locks, then set the
2201 * FL_FILE_PVT flag and override the owner.
2206 if (flock.l_pid != 0)
2210 file_lock->fl_flags |= FL_FILE_PVT;
2211 file_lock->fl_owner = (fl_owner_t)filp;
2215 if (flock.l_pid != 0)
2219 file_lock->fl_flags |= FL_FILE_PVT;
2220 file_lock->fl_owner = (fl_owner_t)filp;
2223 file_lock->fl_flags |= FL_SLEEP;
2226 error = do_lock_file_wait(filp, cmd, file_lock);
2229 * Attempt to detect a close/fcntl race and recover by
2230 * releasing the lock that was just acquired.
2232 spin_lock(¤t->files->file_lock);
2234 spin_unlock(¤t->files->file_lock);
2235 if (!error && f != filp && flock.l_type != F_UNLCK) {
2236 flock.l_type = F_UNLCK;
2241 locks_free_lock(file_lock);
2244 #endif /* BITS_PER_LONG == 32 */
2247 * This function is called when the file is being removed
2248 * from the task's fd array. POSIX locks belonging to this task
2249 * are deleted at this time.
2251 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2253 struct file_lock lock;
2256 * If there are no locks held on this file, we don't need to call
2257 * posix_lock_file(). Another process could be setting a lock on this
2258 * file at the same time, but we wouldn't remove that lock anyway.
2260 if (!file_inode(filp)->i_flock)
2263 lock.fl_type = F_UNLCK;
2264 lock.fl_flags = FL_POSIX | FL_CLOSE;
2266 lock.fl_end = OFFSET_MAX;
2267 lock.fl_owner = owner;
2268 lock.fl_pid = current->tgid;
2269 lock.fl_file = filp;
2271 lock.fl_lmops = NULL;
2273 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2275 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2276 lock.fl_ops->fl_release_private(&lock);
2279 EXPORT_SYMBOL(locks_remove_posix);
2282 * This function is called on the last close of an open file.
2284 void locks_remove_file(struct file *filp)
2286 struct inode * inode = file_inode(filp);
2287 struct file_lock *fl;
2288 struct file_lock **before;
2290 if (!inode->i_flock)
2293 locks_remove_posix(filp, (fl_owner_t)filp);
2295 if (filp->f_op->flock) {
2296 struct file_lock fl = {
2297 .fl_pid = current->tgid,
2299 .fl_flags = FL_FLOCK,
2301 .fl_end = OFFSET_MAX,
2303 filp->f_op->flock(filp, F_SETLKW, &fl);
2304 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2305 fl.fl_ops->fl_release_private(&fl);
2308 spin_lock(&inode->i_lock);
2309 before = &inode->i_flock;
2311 while ((fl = *before) != NULL) {
2312 if (fl->fl_file == filp) {
2314 lease_modify(before, F_UNLCK);
2319 * There's a leftover lock on the list of a type that
2320 * we didn't expect to see. Most likely a classic
2321 * POSIX lock that ended up not getting released
2322 * properly, or that raced onto the list somehow. Log
2323 * some info about it and then just remove it from
2327 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2328 MAJOR(inode->i_sb->s_dev),
2329 MINOR(inode->i_sb->s_dev), inode->i_ino,
2330 fl->fl_type, fl->fl_flags,
2331 fl->fl_start, fl->fl_end);
2333 locks_delete_lock(before);
2336 before = &fl->fl_next;
2338 spin_unlock(&inode->i_lock);
2342 * posix_unblock_lock - stop waiting for a file lock
2343 * @waiter: the lock which was waiting
2345 * lockd needs to block waiting for locks.
2348 posix_unblock_lock(struct file_lock *waiter)
2352 spin_lock(&blocked_lock_lock);
2353 if (waiter->fl_next)
2354 __locks_delete_block(waiter);
2357 spin_unlock(&blocked_lock_lock);
2360 EXPORT_SYMBOL(posix_unblock_lock);
2363 * vfs_cancel_lock - file byte range unblock lock
2364 * @filp: The file to apply the unblock to
2365 * @fl: The lock to be unblocked
2367 * Used by lock managers to cancel blocked requests
2369 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2371 if (filp->f_op->lock)
2372 return filp->f_op->lock(filp, F_CANCELLK, fl);
2376 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2378 #ifdef CONFIG_PROC_FS
2379 #include <linux/proc_fs.h>
2380 #include <linux/seq_file.h>
2382 struct locks_iterator {
2387 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2388 loff_t id, char *pfx)
2390 struct inode *inode = NULL;
2391 unsigned int fl_pid;
2394 fl_pid = pid_vnr(fl->fl_nspid);
2396 fl_pid = fl->fl_pid;
2398 if (fl->fl_file != NULL)
2399 inode = file_inode(fl->fl_file);
2401 seq_printf(f, "%lld:%s ", id, pfx);
2403 if (fl->fl_flags & FL_ACCESS)
2404 seq_printf(f, "ACCESS");
2405 else if (IS_FILE_PVT(fl))
2406 seq_printf(f, "FLPVT ");
2408 seq_printf(f, "POSIX ");
2410 seq_printf(f, " %s ",
2411 (inode == NULL) ? "*NOINODE*" :
2412 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2413 } else if (IS_FLOCK(fl)) {
2414 if (fl->fl_type & LOCK_MAND) {
2415 seq_printf(f, "FLOCK MSNFS ");
2417 seq_printf(f, "FLOCK ADVISORY ");
2419 } else if (IS_LEASE(fl)) {
2420 seq_printf(f, "LEASE ");
2421 if (lease_breaking(fl))
2422 seq_printf(f, "BREAKING ");
2423 else if (fl->fl_file)
2424 seq_printf(f, "ACTIVE ");
2426 seq_printf(f, "BREAKER ");
2428 seq_printf(f, "UNKNOWN UNKNOWN ");
2430 if (fl->fl_type & LOCK_MAND) {
2431 seq_printf(f, "%s ",
2432 (fl->fl_type & LOCK_READ)
2433 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2434 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2436 seq_printf(f, "%s ",
2437 (lease_breaking(fl))
2438 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2439 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2442 #ifdef WE_CAN_BREAK_LSLK_NOW
2443 seq_printf(f, "%d %s:%ld ", fl_pid,
2444 inode->i_sb->s_id, inode->i_ino);
2446 /* userspace relies on this representation of dev_t ;-( */
2447 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2448 MAJOR(inode->i_sb->s_dev),
2449 MINOR(inode->i_sb->s_dev), inode->i_ino);
2452 seq_printf(f, "%d <none>:0 ", fl_pid);
2455 if (fl->fl_end == OFFSET_MAX)
2456 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2458 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2460 seq_printf(f, "0 EOF\n");
2464 static int locks_show(struct seq_file *f, void *v)
2466 struct locks_iterator *iter = f->private;
2467 struct file_lock *fl, *bfl;
2469 fl = hlist_entry(v, struct file_lock, fl_link);
2471 lock_get_status(f, fl, iter->li_pos, "");
2473 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2474 lock_get_status(f, bfl, iter->li_pos, " ->");
2479 static void *locks_start(struct seq_file *f, loff_t *pos)
2480 __acquires(&blocked_lock_lock)
2482 struct locks_iterator *iter = f->private;
2484 iter->li_pos = *pos + 1;
2485 lg_global_lock(&file_lock_lglock);
2486 spin_lock(&blocked_lock_lock);
2487 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2490 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2492 struct locks_iterator *iter = f->private;
2495 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2498 static void locks_stop(struct seq_file *f, void *v)
2499 __releases(&blocked_lock_lock)
2501 spin_unlock(&blocked_lock_lock);
2502 lg_global_unlock(&file_lock_lglock);
2505 static const struct seq_operations locks_seq_operations = {
2506 .start = locks_start,
2512 static int locks_open(struct inode *inode, struct file *filp)
2514 return seq_open_private(filp, &locks_seq_operations,
2515 sizeof(struct locks_iterator));
2518 static const struct file_operations proc_locks_operations = {
2521 .llseek = seq_lseek,
2522 .release = seq_release_private,
2525 static int __init proc_locks_init(void)
2527 proc_create("locks", 0, NULL, &proc_locks_operations);
2530 module_init(proc_locks_init);
2534 * lock_may_read - checks that the region is free of locks
2535 * @inode: the inode that is being read
2536 * @start: the first byte to read
2537 * @len: the number of bytes to read
2539 * Emulates Windows locking requirements. Whole-file
2540 * mandatory locks (share modes) can prohibit a read and
2541 * byte-range POSIX locks can prohibit a read if they overlap.
2543 * N.B. this function is only ever called
2544 * from knfsd and ownership of locks is never checked.
2546 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2548 struct file_lock *fl;
2551 spin_lock(&inode->i_lock);
2552 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2554 if (fl->fl_type == F_RDLCK)
2556 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2558 } else if (IS_FLOCK(fl)) {
2559 if (!(fl->fl_type & LOCK_MAND))
2561 if (fl->fl_type & LOCK_READ)
2568 spin_unlock(&inode->i_lock);
2572 EXPORT_SYMBOL(lock_may_read);
2575 * lock_may_write - checks that the region is free of locks
2576 * @inode: the inode that is being written
2577 * @start: the first byte to write
2578 * @len: the number of bytes to write
2580 * Emulates Windows locking requirements. Whole-file
2581 * mandatory locks (share modes) can prohibit a write and
2582 * byte-range POSIX locks can prohibit a write if they overlap.
2584 * N.B. this function is only ever called
2585 * from knfsd and ownership of locks is never checked.
2587 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2589 struct file_lock *fl;
2592 spin_lock(&inode->i_lock);
2593 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2595 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2597 } else if (IS_FLOCK(fl)) {
2598 if (!(fl->fl_type & LOCK_MAND))
2600 if (fl->fl_type & LOCK_WRITE)
2607 spin_unlock(&inode->i_lock);
2611 EXPORT_SYMBOL(lock_may_write);
2613 static int __init filelock_init(void)
2617 filelock_cache = kmem_cache_create("file_lock_cache",
2618 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2620 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2622 for_each_possible_cpu(i)
2623 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2628 core_initcall(filelock_init);