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 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
350 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
355 switch (l->l_whence) {
363 start = i_size_read(file_inode(filp));
369 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
370 POSIX-2001 defines it. */
374 fl->fl_end = OFFSET_MAX;
376 end = start + l->l_len - 1;
378 } else if (l->l_len < 0) {
385 fl->fl_start = start; /* we record the absolute position */
386 if (fl->fl_end < fl->fl_start)
389 fl->fl_owner = current->files;
390 fl->fl_pid = current->tgid;
392 fl->fl_flags = FL_POSIX;
396 return assign_type(fl, l->l_type);
399 #if BITS_PER_LONG == 32
400 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
405 switch (l->l_whence) {
413 start = i_size_read(file_inode(filp));
422 fl->fl_end = OFFSET_MAX;
424 fl->fl_end = start + l->l_len - 1;
425 } else if (l->l_len < 0) {
426 fl->fl_end = start - 1;
431 fl->fl_start = start; /* we record the absolute position */
432 if (fl->fl_end < fl->fl_start)
435 fl->fl_owner = current->files;
436 fl->fl_pid = current->tgid;
438 fl->fl_flags = FL_POSIX;
442 return assign_type(fl, l->l_type);
446 /* default lease lock manager operations */
447 static void lease_break_callback(struct file_lock *fl)
449 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
452 static const struct lock_manager_operations lease_manager_ops = {
453 .lm_break = lease_break_callback,
454 .lm_change = lease_modify,
458 * Initialize a lease, use the default lock manager operations
460 static int lease_init(struct file *filp, long type, struct file_lock *fl)
462 if (assign_type(fl, type) != 0)
465 fl->fl_owner = current->files;
466 fl->fl_pid = current->tgid;
469 fl->fl_flags = FL_LEASE;
471 fl->fl_end = OFFSET_MAX;
473 fl->fl_lmops = &lease_manager_ops;
477 /* Allocate a file_lock initialised to this type of lease */
478 static struct file_lock *lease_alloc(struct file *filp, long type)
480 struct file_lock *fl = locks_alloc_lock();
484 return ERR_PTR(error);
486 error = lease_init(filp, type, fl);
489 return ERR_PTR(error);
494 /* Check if two locks overlap each other.
496 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
498 return ((fl1->fl_end >= fl2->fl_start) &&
499 (fl2->fl_end >= fl1->fl_start));
503 * Check whether two locks have the same owner.
505 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
507 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
508 return fl2->fl_lmops == fl1->fl_lmops &&
509 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
510 return fl1->fl_owner == fl2->fl_owner;
513 /* Must be called with the i_lock held! */
514 static void locks_insert_global_locks(struct file_lock *fl)
516 lg_local_lock(&file_lock_lglock);
517 fl->fl_link_cpu = smp_processor_id();
518 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
519 lg_local_unlock(&file_lock_lglock);
522 /* Must be called with the i_lock held! */
523 static void locks_delete_global_locks(struct file_lock *fl)
526 * Avoid taking lock if already unhashed. This is safe since this check
527 * is done while holding the i_lock, and new insertions into the list
528 * also require that it be held.
530 if (hlist_unhashed(&fl->fl_link))
532 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
533 hlist_del_init(&fl->fl_link);
534 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
538 posix_owner_key(struct file_lock *fl)
540 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
541 return fl->fl_lmops->lm_owner_key(fl);
542 return (unsigned long)fl->fl_owner;
545 static void locks_insert_global_blocked(struct file_lock *waiter)
547 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
550 static void locks_delete_global_blocked(struct file_lock *waiter)
552 hash_del(&waiter->fl_link);
555 /* Remove waiter from blocker's block list.
556 * When blocker ends up pointing to itself then the list is empty.
558 * Must be called with blocked_lock_lock held.
560 static void __locks_delete_block(struct file_lock *waiter)
562 locks_delete_global_blocked(waiter);
563 list_del_init(&waiter->fl_block);
564 waiter->fl_next = NULL;
567 static void locks_delete_block(struct file_lock *waiter)
569 spin_lock(&blocked_lock_lock);
570 __locks_delete_block(waiter);
571 spin_unlock(&blocked_lock_lock);
574 /* Insert waiter into blocker's block list.
575 * We use a circular list so that processes can be easily woken up in
576 * the order they blocked. The documentation doesn't require this but
577 * it seems like the reasonable thing to do.
579 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
580 * list itself is protected by the blocked_lock_lock, but by ensuring that the
581 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
582 * in some cases when we see that the fl_block list is empty.
584 static void __locks_insert_block(struct file_lock *blocker,
585 struct file_lock *waiter)
587 BUG_ON(!list_empty(&waiter->fl_block));
588 waiter->fl_next = blocker;
589 list_add_tail(&waiter->fl_block, &blocker->fl_block);
590 if (IS_POSIX(blocker))
591 locks_insert_global_blocked(waiter);
594 /* Must be called with i_lock held. */
595 static void locks_insert_block(struct file_lock *blocker,
596 struct file_lock *waiter)
598 spin_lock(&blocked_lock_lock);
599 __locks_insert_block(blocker, waiter);
600 spin_unlock(&blocked_lock_lock);
604 * Wake up processes blocked waiting for blocker.
606 * Must be called with the inode->i_lock held!
608 static void locks_wake_up_blocks(struct file_lock *blocker)
611 * Avoid taking global lock if list is empty. This is safe since new
612 * blocked requests are only added to the list under the i_lock, and
613 * the i_lock is always held here. Note that removal from the fl_block
614 * list does not require the i_lock, so we must recheck list_empty()
615 * after acquiring the blocked_lock_lock.
617 if (list_empty(&blocker->fl_block))
620 spin_lock(&blocked_lock_lock);
621 while (!list_empty(&blocker->fl_block)) {
622 struct file_lock *waiter;
624 waiter = list_first_entry(&blocker->fl_block,
625 struct file_lock, fl_block);
626 __locks_delete_block(waiter);
627 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
628 waiter->fl_lmops->lm_notify(waiter);
630 wake_up(&waiter->fl_wait);
632 spin_unlock(&blocked_lock_lock);
635 /* Insert file lock fl into an inode's lock list at the position indicated
636 * by pos. At the same time add the lock to the global file lock list.
638 * Must be called with the i_lock held!
640 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
642 fl->fl_nspid = get_pid(task_tgid(current));
644 /* insert into file's list */
648 locks_insert_global_locks(fl);
652 * locks_delete_lock - Delete a lock and then free it.
653 * @thisfl_p: pointer that points to the fl_next field of the previous
654 * inode->i_flock list entry
656 * Unlink a lock from all lists and free the namespace reference, but don't
657 * free it yet. Wake up processes that are blocked waiting for this lock and
658 * notify the FS that the lock has been cleared.
660 * Must be called with the i_lock held!
662 static void locks_unlink_lock(struct file_lock **thisfl_p)
664 struct file_lock *fl = *thisfl_p;
666 locks_delete_global_locks(fl);
668 *thisfl_p = fl->fl_next;
672 put_pid(fl->fl_nspid);
676 locks_wake_up_blocks(fl);
680 * Unlink a lock from all lists and free it.
682 * Must be called with i_lock held!
684 static void locks_delete_lock(struct file_lock **thisfl_p)
686 struct file_lock *fl = *thisfl_p;
688 locks_unlink_lock(thisfl_p);
692 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
693 * checks for shared/exclusive status of overlapping locks.
695 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
697 if (sys_fl->fl_type == F_WRLCK)
699 if (caller_fl->fl_type == F_WRLCK)
704 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
705 * checking before calling the locks_conflict().
707 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
709 /* POSIX locks owned by the same process do not conflict with
712 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
715 /* Check whether they overlap */
716 if (!locks_overlap(caller_fl, sys_fl))
719 return (locks_conflict(caller_fl, sys_fl));
722 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
723 * checking before calling the locks_conflict().
725 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
727 /* FLOCK locks referring to the same filp do not conflict with
730 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
732 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
735 return (locks_conflict(caller_fl, sys_fl));
739 posix_test_lock(struct file *filp, struct file_lock *fl)
741 struct file_lock *cfl;
742 struct inode *inode = file_inode(filp);
744 spin_lock(&inode->i_lock);
745 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
748 if (posix_locks_conflict(fl, cfl))
752 __locks_copy_lock(fl, cfl);
754 fl->fl_pid = pid_vnr(cfl->fl_nspid);
756 fl->fl_type = F_UNLCK;
757 spin_unlock(&inode->i_lock);
760 EXPORT_SYMBOL(posix_test_lock);
763 * Deadlock detection:
765 * We attempt to detect deadlocks that are due purely to posix file
768 * We assume that a task can be waiting for at most one lock at a time.
769 * So for any acquired lock, the process holding that lock may be
770 * waiting on at most one other lock. That lock in turns may be held by
771 * someone waiting for at most one other lock. Given a requested lock
772 * caller_fl which is about to wait for a conflicting lock block_fl, we
773 * follow this chain of waiters to ensure we are not about to create a
776 * Since we do this before we ever put a process to sleep on a lock, we
777 * are ensured that there is never a cycle; that is what guarantees that
778 * the while() loop in posix_locks_deadlock() eventually completes.
780 * Note: the above assumption may not be true when handling lock
781 * requests from a broken NFS client. It may also fail in the presence
782 * of tasks (such as posix threads) sharing the same open file table.
784 * To handle those cases, we just bail out after a few iterations.
787 #define MAX_DEADLK_ITERATIONS 10
789 /* Find a lock that the owner of the given block_fl is blocking on. */
790 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
792 struct file_lock *fl;
794 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
795 if (posix_same_owner(fl, block_fl))
801 /* Must be called with the blocked_lock_lock held! */
802 static int posix_locks_deadlock(struct file_lock *caller_fl,
803 struct file_lock *block_fl)
807 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
808 if (i++ > MAX_DEADLK_ITERATIONS)
810 if (posix_same_owner(caller_fl, block_fl))
816 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
817 * after any leases, but before any posix locks.
819 * Note that if called with an FL_EXISTS argument, the caller may determine
820 * whether or not a lock was successfully freed by testing the return
823 static int flock_lock_file(struct file *filp, struct file_lock *request)
825 struct file_lock *new_fl = NULL;
826 struct file_lock **before;
827 struct inode * inode = file_inode(filp);
831 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
832 new_fl = locks_alloc_lock();
837 spin_lock(&inode->i_lock);
838 if (request->fl_flags & FL_ACCESS)
841 for_each_lock(inode, before) {
842 struct file_lock *fl = *before;
847 if (filp != fl->fl_file)
849 if (request->fl_type == fl->fl_type)
852 locks_delete_lock(before);
856 if (request->fl_type == F_UNLCK) {
857 if ((request->fl_flags & FL_EXISTS) && !found)
863 * If a higher-priority process was blocked on the old file lock,
864 * give it the opportunity to lock the file.
867 spin_unlock(&inode->i_lock);
869 spin_lock(&inode->i_lock);
873 for_each_lock(inode, before) {
874 struct file_lock *fl = *before;
879 if (!flock_locks_conflict(request, fl))
882 if (!(request->fl_flags & FL_SLEEP))
884 error = FILE_LOCK_DEFERRED;
885 locks_insert_block(fl, request);
888 if (request->fl_flags & FL_ACCESS)
890 locks_copy_lock(new_fl, request);
891 locks_insert_lock(before, new_fl);
896 spin_unlock(&inode->i_lock);
898 locks_free_lock(new_fl);
902 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
904 struct file_lock *fl;
905 struct file_lock *new_fl = NULL;
906 struct file_lock *new_fl2 = NULL;
907 struct file_lock *left = NULL;
908 struct file_lock *right = NULL;
909 struct file_lock **before;
914 * We may need two file_lock structures for this operation,
915 * so we get them in advance to avoid races.
917 * In some cases we can be sure, that no new locks will be needed
919 if (!(request->fl_flags & FL_ACCESS) &&
920 (request->fl_type != F_UNLCK ||
921 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
922 new_fl = locks_alloc_lock();
923 new_fl2 = locks_alloc_lock();
926 spin_lock(&inode->i_lock);
928 * New lock request. Walk all POSIX locks and look for conflicts. If
929 * there are any, either return error or put the request on the
930 * blocker's list of waiters and the global blocked_hash.
932 if (request->fl_type != F_UNLCK) {
933 for_each_lock(inode, before) {
937 if (!posix_locks_conflict(request, fl))
940 __locks_copy_lock(conflock, fl);
942 if (!(request->fl_flags & FL_SLEEP))
945 * Deadlock detection and insertion into the blocked
946 * locks list must be done while holding the same lock!
949 spin_lock(&blocked_lock_lock);
950 if (likely(!posix_locks_deadlock(request, fl))) {
951 error = FILE_LOCK_DEFERRED;
952 __locks_insert_block(fl, request);
954 spin_unlock(&blocked_lock_lock);
959 /* If we're just looking for a conflict, we're done. */
961 if (request->fl_flags & FL_ACCESS)
965 * Find the first old lock with the same owner as the new lock.
968 before = &inode->i_flock;
970 /* First skip locks owned by other processes. */
971 while ((fl = *before) && (!IS_POSIX(fl) ||
972 !posix_same_owner(request, fl))) {
973 before = &fl->fl_next;
976 /* Process locks with this owner. */
977 while ((fl = *before) && posix_same_owner(request, fl)) {
978 /* Detect adjacent or overlapping regions (if same lock type)
980 if (request->fl_type == fl->fl_type) {
981 /* In all comparisons of start vs end, use
982 * "start - 1" rather than "end + 1". If end
983 * is OFFSET_MAX, end + 1 will become negative.
985 if (fl->fl_end < request->fl_start - 1)
987 /* If the next lock in the list has entirely bigger
988 * addresses than the new one, insert the lock here.
990 if (fl->fl_start - 1 > request->fl_end)
993 /* If we come here, the new and old lock are of the
994 * same type and adjacent or overlapping. Make one
995 * lock yielding from the lower start address of both
996 * locks to the higher end address.
998 if (fl->fl_start > request->fl_start)
999 fl->fl_start = request->fl_start;
1001 request->fl_start = fl->fl_start;
1002 if (fl->fl_end < request->fl_end)
1003 fl->fl_end = request->fl_end;
1005 request->fl_end = fl->fl_end;
1007 locks_delete_lock(before);
1014 /* Processing for different lock types is a bit
1017 if (fl->fl_end < request->fl_start)
1019 if (fl->fl_start > request->fl_end)
1021 if (request->fl_type == F_UNLCK)
1023 if (fl->fl_start < request->fl_start)
1025 /* If the next lock in the list has a higher end
1026 * address than the new one, insert the new one here.
1028 if (fl->fl_end > request->fl_end) {
1032 if (fl->fl_start >= request->fl_start) {
1033 /* The new lock completely replaces an old
1034 * one (This may happen several times).
1037 locks_delete_lock(before);
1040 /* Replace the old lock with the new one.
1041 * Wake up anybody waiting for the old one,
1042 * as the change in lock type might satisfy
1045 locks_wake_up_blocks(fl);
1046 fl->fl_start = request->fl_start;
1047 fl->fl_end = request->fl_end;
1048 fl->fl_type = request->fl_type;
1049 locks_release_private(fl);
1050 locks_copy_private(fl, request);
1055 /* Go on to next lock.
1058 before = &fl->fl_next;
1062 * The above code only modifies existing locks in case of merging or
1063 * replacing. If new lock(s) need to be inserted all modifications are
1064 * done below this, so it's safe yet to bail out.
1066 error = -ENOLCK; /* "no luck" */
1067 if (right && left == right && !new_fl2)
1072 if (request->fl_type == F_UNLCK) {
1073 if (request->fl_flags & FL_EXISTS)
1082 locks_copy_lock(new_fl, request);
1083 locks_insert_lock(before, new_fl);
1087 if (left == right) {
1088 /* The new lock breaks the old one in two pieces,
1089 * so we have to use the second new lock.
1093 locks_copy_lock(left, right);
1094 locks_insert_lock(before, left);
1096 right->fl_start = request->fl_end + 1;
1097 locks_wake_up_blocks(right);
1100 left->fl_end = request->fl_start - 1;
1101 locks_wake_up_blocks(left);
1104 spin_unlock(&inode->i_lock);
1106 * Free any unused locks.
1109 locks_free_lock(new_fl);
1111 locks_free_lock(new_fl2);
1116 * posix_lock_file - Apply a POSIX-style lock to a file
1117 * @filp: The file to apply the lock to
1118 * @fl: The lock to be applied
1119 * @conflock: Place to return a copy of the conflicting lock, if found.
1121 * Add a POSIX style lock to a file.
1122 * We merge adjacent & overlapping locks whenever possible.
1123 * POSIX locks are sorted by owner task, then by starting address
1125 * Note that if called with an FL_EXISTS argument, the caller may determine
1126 * whether or not a lock was successfully freed by testing the return
1127 * value for -ENOENT.
1129 int posix_lock_file(struct file *filp, struct file_lock *fl,
1130 struct file_lock *conflock)
1132 return __posix_lock_file(file_inode(filp), fl, conflock);
1134 EXPORT_SYMBOL(posix_lock_file);
1137 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1138 * @filp: The file to apply the lock to
1139 * @fl: The lock to be applied
1141 * Add a POSIX style lock to a file.
1142 * We merge adjacent & overlapping locks whenever possible.
1143 * POSIX locks are sorted by owner task, then by starting address
1145 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1150 error = posix_lock_file(filp, fl, NULL);
1151 if (error != FILE_LOCK_DEFERRED)
1153 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1157 locks_delete_block(fl);
1162 EXPORT_SYMBOL(posix_lock_file_wait);
1165 * locks_mandatory_locked - Check for an active lock
1166 * @inode: the file to check
1168 * Searches the inode's list of locks to find any POSIX locks which conflict.
1169 * This function is called from locks_verify_locked() only.
1171 int locks_mandatory_locked(struct inode *inode)
1173 fl_owner_t owner = current->files;
1174 struct file_lock *fl;
1177 * Search the lock list for this inode for any POSIX locks.
1179 spin_lock(&inode->i_lock);
1180 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1183 if (fl->fl_owner != owner)
1186 spin_unlock(&inode->i_lock);
1187 return fl ? -EAGAIN : 0;
1191 * locks_mandatory_area - Check for a conflicting lock
1192 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1194 * @inode: the file to check
1195 * @filp: how the file was opened (if it was)
1196 * @offset: start of area to check
1197 * @count: length of area to check
1199 * Searches the inode's list of locks to find any POSIX locks which conflict.
1200 * This function is called from rw_verify_area() and
1201 * locks_verify_truncate().
1203 int locks_mandatory_area(int read_write, struct inode *inode,
1204 struct file *filp, loff_t offset,
1207 struct file_lock fl;
1210 locks_init_lock(&fl);
1211 fl.fl_owner = current->files;
1212 fl.fl_pid = current->tgid;
1214 fl.fl_flags = FL_POSIX | FL_ACCESS;
1215 if (filp && !(filp->f_flags & O_NONBLOCK))
1216 fl.fl_flags |= FL_SLEEP;
1217 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1218 fl.fl_start = offset;
1219 fl.fl_end = offset + count - 1;
1222 error = __posix_lock_file(inode, &fl, NULL);
1223 if (error != FILE_LOCK_DEFERRED)
1225 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1228 * If we've been sleeping someone might have
1229 * changed the permissions behind our back.
1231 if (__mandatory_lock(inode))
1235 locks_delete_block(&fl);
1242 EXPORT_SYMBOL(locks_mandatory_area);
1244 static void lease_clear_pending(struct file_lock *fl, int arg)
1248 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1251 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1255 /* We already had a lease on this file; just change its type */
1256 int lease_modify(struct file_lock **before, int arg)
1258 struct file_lock *fl = *before;
1259 int error = assign_type(fl, arg);
1263 lease_clear_pending(fl, arg);
1264 locks_wake_up_blocks(fl);
1265 if (arg == F_UNLCK) {
1266 struct file *filp = fl->fl_file;
1269 filp->f_owner.signum = 0;
1270 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1271 if (fl->fl_fasync != NULL) {
1272 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1273 fl->fl_fasync = NULL;
1275 locks_delete_lock(before);
1280 EXPORT_SYMBOL(lease_modify);
1282 static bool past_time(unsigned long then)
1285 /* 0 is a special value meaning "this never expires": */
1287 return time_after(jiffies, then);
1290 static void time_out_leases(struct inode *inode)
1292 struct file_lock **before;
1293 struct file_lock *fl;
1295 before = &inode->i_flock;
1296 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1297 if (past_time(fl->fl_downgrade_time))
1298 lease_modify(before, F_RDLCK);
1299 if (past_time(fl->fl_break_time))
1300 lease_modify(before, F_UNLCK);
1301 if (fl == *before) /* lease_modify may have freed fl */
1302 before = &fl->fl_next;
1306 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1308 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1310 return locks_conflict(breaker, lease);
1314 * __break_lease - revoke all outstanding leases on file
1315 * @inode: the inode of the file to return
1316 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1318 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1321 * break_lease (inlined for speed) has checked there already is at least
1322 * some kind of lock (maybe a lease) on this file. Leases are broken on
1323 * a call to open() or truncate(). This function can sleep unless you
1324 * specified %O_NONBLOCK to your open().
1326 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1329 struct file_lock *new_fl, *flock;
1330 struct file_lock *fl;
1331 unsigned long break_time;
1332 int i_have_this_lease = 0;
1333 bool lease_conflict = false;
1334 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1336 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1338 return PTR_ERR(new_fl);
1339 new_fl->fl_flags = type;
1341 spin_lock(&inode->i_lock);
1343 time_out_leases(inode);
1345 flock = inode->i_flock;
1346 if ((flock == NULL) || !IS_LEASE(flock))
1349 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1350 if (leases_conflict(fl, new_fl)) {
1351 lease_conflict = true;
1352 if (fl->fl_owner == current->files)
1353 i_have_this_lease = 1;
1356 if (!lease_conflict)
1360 if (lease_break_time > 0) {
1361 break_time = jiffies + lease_break_time * HZ;
1362 if (break_time == 0)
1363 break_time++; /* so that 0 means no break time */
1366 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1367 if (!leases_conflict(fl, new_fl))
1370 if (fl->fl_flags & FL_UNLOCK_PENDING)
1372 fl->fl_flags |= FL_UNLOCK_PENDING;
1373 fl->fl_break_time = break_time;
1375 if (lease_breaking(flock))
1377 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1378 fl->fl_downgrade_time = break_time;
1380 fl->fl_lmops->lm_break(fl);
1383 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1384 error = -EWOULDBLOCK;
1389 break_time = flock->fl_break_time;
1390 if (break_time != 0) {
1391 break_time -= jiffies;
1392 if (break_time == 0)
1395 locks_insert_block(flock, new_fl);
1396 spin_unlock(&inode->i_lock);
1397 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1398 !new_fl->fl_next, break_time);
1399 spin_lock(&inode->i_lock);
1400 locks_delete_block(new_fl);
1403 time_out_leases(inode);
1405 * Wait for the next conflicting lease that has not been
1408 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1409 flock = flock->fl_next) {
1410 if (leases_conflict(new_fl, flock))
1417 spin_unlock(&inode->i_lock);
1418 locks_free_lock(new_fl);
1422 EXPORT_SYMBOL(__break_lease);
1425 * lease_get_mtime - get the last modified time of an inode
1427 * @time: pointer to a timespec which will contain the last modified time
1429 * This is to force NFS clients to flush their caches for files with
1430 * exclusive leases. The justification is that if someone has an
1431 * exclusive lease, then they could be modifying it.
1433 void lease_get_mtime(struct inode *inode, struct timespec *time)
1435 struct file_lock *flock = inode->i_flock;
1436 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1437 *time = current_fs_time(inode->i_sb);
1439 *time = inode->i_mtime;
1442 EXPORT_SYMBOL(lease_get_mtime);
1445 * fcntl_getlease - Enquire what lease is currently active
1448 * The value returned by this function will be one of
1449 * (if no lease break is pending):
1451 * %F_RDLCK to indicate a shared lease is held.
1453 * %F_WRLCK to indicate an exclusive lease is held.
1455 * %F_UNLCK to indicate no lease is held.
1457 * (if a lease break is pending):
1459 * %F_RDLCK to indicate an exclusive lease needs to be
1460 * changed to a shared lease (or removed).
1462 * %F_UNLCK to indicate the lease needs to be removed.
1464 * XXX: sfr & willy disagree over whether F_INPROGRESS
1465 * should be returned to userspace.
1467 int fcntl_getlease(struct file *filp)
1469 struct file_lock *fl;
1470 struct inode *inode = file_inode(filp);
1473 spin_lock(&inode->i_lock);
1474 time_out_leases(file_inode(filp));
1475 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1477 if (fl->fl_file == filp) {
1478 type = target_leasetype(fl);
1482 spin_unlock(&inode->i_lock);
1487 * check_conflicting_open - see if the given dentry points to a file that has
1488 * an existing open that would conflict with the
1490 * @dentry: dentry to check
1491 * @arg: type of lease that we're trying to acquire
1493 * Check to see if there's an existing open fd on this file that would
1494 * conflict with the lease we're trying to set.
1497 check_conflicting_open(const struct dentry *dentry, const long arg)
1500 struct inode *inode = dentry->d_inode;
1502 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1505 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1506 (atomic_read(&inode->i_count) > 1)))
1512 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1514 struct file_lock *fl, **before, **my_before = NULL, *lease;
1515 struct dentry *dentry = filp->f_path.dentry;
1516 struct inode *inode = dentry->d_inode;
1517 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1522 * In the delegation case we need mutual exclusion with
1523 * a number of operations that take the i_mutex. We trylock
1524 * because delegations are an optional optimization, and if
1525 * there's some chance of a conflict--we'd rather not
1526 * bother, maybe that's a sign this just isn't a good file to
1527 * hand out a delegation on.
1529 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1532 if (is_deleg && arg == F_WRLCK) {
1533 /* Write delegations are not currently supported: */
1534 mutex_unlock(&inode->i_mutex);
1539 error = check_conflicting_open(dentry, arg);
1544 * At this point, we know that if there is an exclusive
1545 * lease on this file, then we hold it on this filp
1546 * (otherwise our open of this file would have blocked).
1547 * And if we are trying to acquire an exclusive lease,
1548 * then the file is not open by anyone (including us)
1549 * except for this filp.
1552 for (before = &inode->i_flock;
1553 ((fl = *before) != NULL) && IS_LEASE(fl);
1554 before = &fl->fl_next) {
1555 if (fl->fl_file == filp) {
1560 * No exclusive leases if someone else has a lease on
1566 * Modifying our existing lease is OK, but no getting a
1567 * new lease if someone else is opening for write:
1569 if (fl->fl_flags & FL_UNLOCK_PENDING)
1573 if (my_before != NULL) {
1574 error = lease->fl_lmops->lm_change(my_before, arg);
1584 locks_insert_lock(before, lease);
1586 * The check in break_lease() is lockless. It's possible for another
1587 * open to race in after we did the earlier check for a conflicting
1588 * open but before the lease was inserted. Check again for a
1589 * conflicting open and cancel the lease if there is one.
1591 * We also add a barrier here to ensure that the insertion of the lock
1592 * precedes these checks.
1595 error = check_conflicting_open(dentry, arg);
1597 locks_unlink_lock(flp);
1600 mutex_unlock(&inode->i_mutex);
1604 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1606 struct file_lock *fl, **before;
1607 struct dentry *dentry = filp->f_path.dentry;
1608 struct inode *inode = dentry->d_inode;
1610 for (before = &inode->i_flock;
1611 ((fl = *before) != NULL) && IS_LEASE(fl);
1612 before = &fl->fl_next) {
1613 if (fl->fl_file != filp)
1615 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1621 * generic_setlease - sets a lease on an open file
1622 * @filp: file pointer
1623 * @arg: type of lease to obtain
1624 * @flp: input - file_lock to use, output - file_lock inserted
1626 * The (input) flp->fl_lmops->lm_break function is required
1629 * Called with inode->i_lock held.
1631 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1633 struct dentry *dentry = filp->f_path.dentry;
1634 struct inode *inode = dentry->d_inode;
1637 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1639 if (!S_ISREG(inode->i_mode))
1641 error = security_file_lock(filp, arg);
1645 time_out_leases(inode);
1647 BUG_ON(!(*flp)->fl_lmops->lm_break);
1651 return generic_delete_lease(filp, flp);
1654 return generic_add_lease(filp, arg, flp);
1659 EXPORT_SYMBOL(generic_setlease);
1661 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1663 if (filp->f_op->setlease)
1664 return filp->f_op->setlease(filp, arg, lease);
1666 return generic_setlease(filp, arg, lease);
1670 * vfs_setlease - sets a lease on an open file
1671 * @filp: file pointer
1672 * @arg: type of lease to obtain
1673 * @lease: file_lock to use
1675 * Call this to establish a lease on the file.
1676 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1677 * break_lease will oops!
1679 * This will call the filesystem's setlease file method, if
1680 * defined. Note that there is no getlease method; instead, the
1681 * filesystem setlease method should call back to setlease() to
1682 * add a lease to the inode's lease list, where fcntl_getlease() can
1683 * find it. Since fcntl_getlease() only reports whether the current
1684 * task holds a lease, a cluster filesystem need only do this for
1685 * leases held by processes on this node.
1687 * There is also no break_lease method; filesystems that
1688 * handle their own leases should break leases themselves from the
1689 * filesystem's open, create, and (on truncate) setattr methods.
1691 * Warning: the only current setlease methods exist only to disable
1692 * leases in certain cases. More vfs changes may be required to
1693 * allow a full filesystem lease implementation.
1696 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1698 struct inode *inode = file_inode(filp);
1701 spin_lock(&inode->i_lock);
1702 error = __vfs_setlease(filp, arg, lease);
1703 spin_unlock(&inode->i_lock);
1707 EXPORT_SYMBOL_GPL(vfs_setlease);
1709 static int do_fcntl_delete_lease(struct file *filp)
1711 struct file_lock fl, *flp = &fl;
1713 lease_init(filp, F_UNLCK, flp);
1715 return vfs_setlease(filp, F_UNLCK, &flp);
1718 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1720 struct file_lock *fl, *ret;
1721 struct inode *inode = file_inode(filp);
1722 struct fasync_struct *new;
1725 fl = lease_alloc(filp, arg);
1729 new = fasync_alloc();
1731 locks_free_lock(fl);
1735 spin_lock(&inode->i_lock);
1736 error = __vfs_setlease(filp, arg, &ret);
1738 spin_unlock(&inode->i_lock);
1739 locks_free_lock(fl);
1740 goto out_free_fasync;
1743 locks_free_lock(fl);
1746 * fasync_insert_entry() returns the old entry if any.
1747 * If there was no old entry, then it used 'new' and
1748 * inserted it into the fasync list. Clear new so that
1749 * we don't release it here.
1751 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1754 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1755 spin_unlock(&inode->i_lock);
1764 * fcntl_setlease - sets a lease on an open file
1765 * @fd: open file descriptor
1766 * @filp: file pointer
1767 * @arg: type of lease to obtain
1769 * Call this fcntl to establish a lease on the file.
1770 * Note that you also need to call %F_SETSIG to
1771 * receive a signal when the lease is broken.
1773 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1776 return do_fcntl_delete_lease(filp);
1777 return do_fcntl_add_lease(fd, filp, arg);
1781 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1782 * @filp: The file to apply the lock to
1783 * @fl: The lock to be applied
1785 * Add a FLOCK style lock to a file.
1787 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1792 error = flock_lock_file(filp, fl);
1793 if (error != FILE_LOCK_DEFERRED)
1795 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1799 locks_delete_block(fl);
1805 EXPORT_SYMBOL(flock_lock_file_wait);
1808 * sys_flock: - flock() system call.
1809 * @fd: the file descriptor to lock.
1810 * @cmd: the type of lock to apply.
1812 * Apply a %FL_FLOCK style lock to an open file descriptor.
1813 * The @cmd can be one of
1815 * %LOCK_SH -- a shared lock.
1817 * %LOCK_EX -- an exclusive lock.
1819 * %LOCK_UN -- remove an existing lock.
1821 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1823 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1824 * processes read and write access respectively.
1826 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1828 struct fd f = fdget(fd);
1829 struct file_lock *lock;
1830 int can_sleep, unlock;
1837 can_sleep = !(cmd & LOCK_NB);
1839 unlock = (cmd == LOCK_UN);
1841 if (!unlock && !(cmd & LOCK_MAND) &&
1842 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1845 error = flock_make_lock(f.file, &lock, cmd);
1849 lock->fl_flags |= FL_SLEEP;
1851 error = security_file_lock(f.file, lock->fl_type);
1855 if (f.file->f_op->flock)
1856 error = f.file->f_op->flock(f.file,
1857 (can_sleep) ? F_SETLKW : F_SETLK,
1860 error = flock_lock_file_wait(f.file, lock);
1863 locks_free_lock(lock);
1872 * vfs_test_lock - test file byte range lock
1873 * @filp: The file to test lock for
1874 * @fl: The lock to test; also used to hold result
1876 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1877 * setting conf->fl_type to something other than F_UNLCK.
1879 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1881 if (filp->f_op->lock)
1882 return filp->f_op->lock(filp, F_GETLK, fl);
1883 posix_test_lock(filp, fl);
1886 EXPORT_SYMBOL_GPL(vfs_test_lock);
1888 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1890 flock->l_pid = fl->fl_pid;
1891 #if BITS_PER_LONG == 32
1893 * Make sure we can represent the posix lock via
1894 * legacy 32bit flock.
1896 if (fl->fl_start > OFFT_OFFSET_MAX)
1898 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1901 flock->l_start = fl->fl_start;
1902 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1903 fl->fl_end - fl->fl_start + 1;
1904 flock->l_whence = 0;
1905 flock->l_type = fl->fl_type;
1909 #if BITS_PER_LONG == 32
1910 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1912 flock->l_pid = fl->fl_pid;
1913 flock->l_start = fl->fl_start;
1914 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1915 fl->fl_end - fl->fl_start + 1;
1916 flock->l_whence = 0;
1917 flock->l_type = fl->fl_type;
1921 /* Report the first existing lock that would conflict with l.
1922 * This implements the F_GETLK command of fcntl().
1924 int fcntl_getlk(struct file *filp, struct flock __user *l)
1926 struct file_lock file_lock;
1931 if (copy_from_user(&flock, l, sizeof(flock)))
1934 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1937 error = flock_to_posix_lock(filp, &file_lock, &flock);
1941 error = vfs_test_lock(filp, &file_lock);
1945 flock.l_type = file_lock.fl_type;
1946 if (file_lock.fl_type != F_UNLCK) {
1947 error = posix_lock_to_flock(&flock, &file_lock);
1952 if (!copy_to_user(l, &flock, sizeof(flock)))
1959 * vfs_lock_file - file byte range lock
1960 * @filp: The file to apply the lock to
1961 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1962 * @fl: The lock to be applied
1963 * @conf: Place to return a copy of the conflicting lock, if found.
1965 * A caller that doesn't care about the conflicting lock may pass NULL
1966 * as the final argument.
1968 * If the filesystem defines a private ->lock() method, then @conf will
1969 * be left unchanged; so a caller that cares should initialize it to
1970 * some acceptable default.
1972 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1973 * locks, the ->lock() interface may return asynchronously, before the lock has
1974 * been granted or denied by the underlying filesystem, if (and only if)
1975 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1976 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1977 * the request is for a blocking lock. When ->lock() does return asynchronously,
1978 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1979 * request completes.
1980 * If the request is for non-blocking lock the file system should return
1981 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1982 * with the result. If the request timed out the callback routine will return a
1983 * nonzero return code and the file system should release the lock. The file
1984 * system is also responsible to keep a corresponding posix lock when it
1985 * grants a lock so the VFS can find out which locks are locally held and do
1986 * the correct lock cleanup when required.
1987 * The underlying filesystem must not drop the kernel lock or call
1988 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1991 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1993 if (filp->f_op->lock)
1994 return filp->f_op->lock(filp, cmd, fl);
1996 return posix_lock_file(filp, fl, conf);
1998 EXPORT_SYMBOL_GPL(vfs_lock_file);
2000 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2001 struct file_lock *fl)
2005 error = security_file_lock(filp, fl->fl_type);
2010 error = vfs_lock_file(filp, cmd, fl, NULL);
2011 if (error != FILE_LOCK_DEFERRED)
2013 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2017 locks_delete_block(fl);
2024 /* Apply the lock described by l to an open file descriptor.
2025 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2027 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2028 struct flock __user *l)
2030 struct file_lock *file_lock = locks_alloc_lock();
2032 struct inode *inode;
2036 if (file_lock == NULL)
2040 * This might block, so we do it before checking the inode.
2043 if (copy_from_user(&flock, l, sizeof(flock)))
2046 inode = file_inode(filp);
2048 /* Don't allow mandatory locks on files that may be memory mapped
2051 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2057 error = flock_to_posix_lock(filp, file_lock, &flock);
2060 if (cmd == F_SETLKW) {
2061 file_lock->fl_flags |= FL_SLEEP;
2065 switch (flock.l_type) {
2067 if (!(filp->f_mode & FMODE_READ))
2071 if (!(filp->f_mode & FMODE_WRITE))
2081 error = do_lock_file_wait(filp, cmd, file_lock);
2084 * Attempt to detect a close/fcntl race and recover by
2085 * releasing the lock that was just acquired.
2088 * we need that spin_lock here - it prevents reordering between
2089 * update of inode->i_flock and check for it done in close().
2090 * rcu_read_lock() wouldn't do.
2092 spin_lock(¤t->files->file_lock);
2094 spin_unlock(¤t->files->file_lock);
2095 if (!error && f != filp && flock.l_type != F_UNLCK) {
2096 flock.l_type = F_UNLCK;
2101 locks_free_lock(file_lock);
2105 #if BITS_PER_LONG == 32
2106 /* Report the first existing lock that would conflict with l.
2107 * This implements the F_GETLK command of fcntl().
2109 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
2111 struct file_lock file_lock;
2112 struct flock64 flock;
2116 if (copy_from_user(&flock, l, sizeof(flock)))
2119 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2122 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2126 error = vfs_test_lock(filp, &file_lock);
2130 flock.l_type = file_lock.fl_type;
2131 if (file_lock.fl_type != F_UNLCK)
2132 posix_lock_to_flock64(&flock, &file_lock);
2135 if (!copy_to_user(l, &flock, sizeof(flock)))
2142 /* Apply the lock described by l to an open file descriptor.
2143 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2145 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2146 struct flock64 __user *l)
2148 struct file_lock *file_lock = locks_alloc_lock();
2149 struct flock64 flock;
2150 struct inode *inode;
2154 if (file_lock == NULL)
2158 * This might block, so we do it before checking the inode.
2161 if (copy_from_user(&flock, l, sizeof(flock)))
2164 inode = file_inode(filp);
2166 /* Don't allow mandatory locks on files that may be memory mapped
2169 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2175 error = flock64_to_posix_lock(filp, file_lock, &flock);
2178 if (cmd == F_SETLKW64) {
2179 file_lock->fl_flags |= FL_SLEEP;
2183 switch (flock.l_type) {
2185 if (!(filp->f_mode & FMODE_READ))
2189 if (!(filp->f_mode & FMODE_WRITE))
2199 error = do_lock_file_wait(filp, cmd, file_lock);
2202 * Attempt to detect a close/fcntl race and recover by
2203 * releasing the lock that was just acquired.
2205 spin_lock(¤t->files->file_lock);
2207 spin_unlock(¤t->files->file_lock);
2208 if (!error && f != filp && flock.l_type != F_UNLCK) {
2209 flock.l_type = F_UNLCK;
2214 locks_free_lock(file_lock);
2217 #endif /* BITS_PER_LONG == 32 */
2220 * This function is called when the file is being removed
2221 * from the task's fd array. POSIX locks belonging to this task
2222 * are deleted at this time.
2224 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2226 struct file_lock lock;
2229 * If there are no locks held on this file, we don't need to call
2230 * posix_lock_file(). Another process could be setting a lock on this
2231 * file at the same time, but we wouldn't remove that lock anyway.
2233 if (!file_inode(filp)->i_flock)
2236 lock.fl_type = F_UNLCK;
2237 lock.fl_flags = FL_POSIX | FL_CLOSE;
2239 lock.fl_end = OFFSET_MAX;
2240 lock.fl_owner = owner;
2241 lock.fl_pid = current->tgid;
2242 lock.fl_file = filp;
2244 lock.fl_lmops = NULL;
2246 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2248 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2249 lock.fl_ops->fl_release_private(&lock);
2252 EXPORT_SYMBOL(locks_remove_posix);
2255 * This function is called on the last close of an open file.
2257 void locks_remove_flock(struct file *filp)
2259 struct inode * inode = file_inode(filp);
2260 struct file_lock *fl;
2261 struct file_lock **before;
2263 if (!inode->i_flock)
2266 if (filp->f_op->flock) {
2267 struct file_lock fl = {
2268 .fl_pid = current->tgid,
2270 .fl_flags = FL_FLOCK,
2272 .fl_end = OFFSET_MAX,
2274 filp->f_op->flock(filp, F_SETLKW, &fl);
2275 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2276 fl.fl_ops->fl_release_private(&fl);
2279 spin_lock(&inode->i_lock);
2280 before = &inode->i_flock;
2282 while ((fl = *before) != NULL) {
2283 if (fl->fl_file == filp) {
2285 locks_delete_lock(before);
2289 lease_modify(before, F_UNLCK);
2295 before = &fl->fl_next;
2297 spin_unlock(&inode->i_lock);
2301 * posix_unblock_lock - stop waiting for a file lock
2302 * @waiter: the lock which was waiting
2304 * lockd needs to block waiting for locks.
2307 posix_unblock_lock(struct file_lock *waiter)
2311 spin_lock(&blocked_lock_lock);
2312 if (waiter->fl_next)
2313 __locks_delete_block(waiter);
2316 spin_unlock(&blocked_lock_lock);
2319 EXPORT_SYMBOL(posix_unblock_lock);
2322 * vfs_cancel_lock - file byte range unblock lock
2323 * @filp: The file to apply the unblock to
2324 * @fl: The lock to be unblocked
2326 * Used by lock managers to cancel blocked requests
2328 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2330 if (filp->f_op->lock)
2331 return filp->f_op->lock(filp, F_CANCELLK, fl);
2335 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2337 #ifdef CONFIG_PROC_FS
2338 #include <linux/proc_fs.h>
2339 #include <linux/seq_file.h>
2341 struct locks_iterator {
2346 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2347 loff_t id, char *pfx)
2349 struct inode *inode = NULL;
2350 unsigned int fl_pid;
2353 fl_pid = pid_vnr(fl->fl_nspid);
2355 fl_pid = fl->fl_pid;
2357 if (fl->fl_file != NULL)
2358 inode = file_inode(fl->fl_file);
2360 seq_printf(f, "%lld:%s ", id, pfx);
2362 seq_printf(f, "%6s %s ",
2363 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2364 (inode == NULL) ? "*NOINODE*" :
2365 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2366 } else if (IS_FLOCK(fl)) {
2367 if (fl->fl_type & LOCK_MAND) {
2368 seq_printf(f, "FLOCK MSNFS ");
2370 seq_printf(f, "FLOCK ADVISORY ");
2372 } else if (IS_LEASE(fl)) {
2373 seq_printf(f, "LEASE ");
2374 if (lease_breaking(fl))
2375 seq_printf(f, "BREAKING ");
2376 else if (fl->fl_file)
2377 seq_printf(f, "ACTIVE ");
2379 seq_printf(f, "BREAKER ");
2381 seq_printf(f, "UNKNOWN UNKNOWN ");
2383 if (fl->fl_type & LOCK_MAND) {
2384 seq_printf(f, "%s ",
2385 (fl->fl_type & LOCK_READ)
2386 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2387 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2389 seq_printf(f, "%s ",
2390 (lease_breaking(fl))
2391 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2392 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2395 #ifdef WE_CAN_BREAK_LSLK_NOW
2396 seq_printf(f, "%d %s:%ld ", fl_pid,
2397 inode->i_sb->s_id, inode->i_ino);
2399 /* userspace relies on this representation of dev_t ;-( */
2400 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2401 MAJOR(inode->i_sb->s_dev),
2402 MINOR(inode->i_sb->s_dev), inode->i_ino);
2405 seq_printf(f, "%d <none>:0 ", fl_pid);
2408 if (fl->fl_end == OFFSET_MAX)
2409 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2411 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2413 seq_printf(f, "0 EOF\n");
2417 static int locks_show(struct seq_file *f, void *v)
2419 struct locks_iterator *iter = f->private;
2420 struct file_lock *fl, *bfl;
2422 fl = hlist_entry(v, struct file_lock, fl_link);
2424 lock_get_status(f, fl, iter->li_pos, "");
2426 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2427 lock_get_status(f, bfl, iter->li_pos, " ->");
2432 static void *locks_start(struct seq_file *f, loff_t *pos)
2433 __acquires(&blocked_lock_lock)
2435 struct locks_iterator *iter = f->private;
2437 iter->li_pos = *pos + 1;
2438 lg_global_lock(&file_lock_lglock);
2439 spin_lock(&blocked_lock_lock);
2440 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2443 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2445 struct locks_iterator *iter = f->private;
2448 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2451 static void locks_stop(struct seq_file *f, void *v)
2452 __releases(&blocked_lock_lock)
2454 spin_unlock(&blocked_lock_lock);
2455 lg_global_unlock(&file_lock_lglock);
2458 static const struct seq_operations locks_seq_operations = {
2459 .start = locks_start,
2465 static int locks_open(struct inode *inode, struct file *filp)
2467 return seq_open_private(filp, &locks_seq_operations,
2468 sizeof(struct locks_iterator));
2471 static const struct file_operations proc_locks_operations = {
2474 .llseek = seq_lseek,
2475 .release = seq_release_private,
2478 static int __init proc_locks_init(void)
2480 proc_create("locks", 0, NULL, &proc_locks_operations);
2483 module_init(proc_locks_init);
2487 * lock_may_read - checks that the region is free of locks
2488 * @inode: the inode that is being read
2489 * @start: the first byte to read
2490 * @len: the number of bytes to read
2492 * Emulates Windows locking requirements. Whole-file
2493 * mandatory locks (share modes) can prohibit a read and
2494 * byte-range POSIX locks can prohibit a read if they overlap.
2496 * N.B. this function is only ever called
2497 * from knfsd and ownership of locks is never checked.
2499 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2501 struct file_lock *fl;
2504 spin_lock(&inode->i_lock);
2505 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2507 if (fl->fl_type == F_RDLCK)
2509 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2511 } else if (IS_FLOCK(fl)) {
2512 if (!(fl->fl_type & LOCK_MAND))
2514 if (fl->fl_type & LOCK_READ)
2521 spin_unlock(&inode->i_lock);
2525 EXPORT_SYMBOL(lock_may_read);
2528 * lock_may_write - checks that the region is free of locks
2529 * @inode: the inode that is being written
2530 * @start: the first byte to write
2531 * @len: the number of bytes to write
2533 * Emulates Windows locking requirements. Whole-file
2534 * mandatory locks (share modes) can prohibit a write and
2535 * byte-range POSIX locks can prohibit a write if they overlap.
2537 * N.B. this function is only ever called
2538 * from knfsd and ownership of locks is never checked.
2540 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2542 struct file_lock *fl;
2545 spin_lock(&inode->i_lock);
2546 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2548 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2550 } else if (IS_FLOCK(fl)) {
2551 if (!(fl->fl_type & LOCK_MAND))
2553 if (fl->fl_type & LOCK_WRITE)
2560 spin_unlock(&inode->i_lock);
2564 EXPORT_SYMBOL(lock_may_write);
2566 static int __init filelock_init(void)
2570 filelock_cache = kmem_cache_create("file_lock_cache",
2571 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2573 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2575 for_each_possible_cpu(i)
2576 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2581 core_initcall(filelock_init);