1 // SPDX-License-Identifier: GPL-2.0-only
5 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
6 * Doug Evans (dje@spiff.uucp), August 07, 1992
8 * Deadlock detection added.
9 * FIXME: one thing isn't handled yet:
10 * - mandatory locks (requires lots of changes elsewhere)
11 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
13 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
14 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
16 * Converted file_lock_table to a linked list from an array, which eliminates
17 * the limits on how many active file locks are open.
18 * Chad Page (pageone@netcom.com), November 27, 1994
20 * Removed dependency on file descriptors. dup()'ed file descriptors now
21 * get the same locks as the original file descriptors, and a close() on
22 * any file descriptor removes ALL the locks on the file for the current
23 * process. Since locks still depend on the process id, locks are inherited
24 * after an exec() but not after a fork(). This agrees with POSIX, and both
25 * BSD and SVR4 practice.
26 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
28 * Scrapped free list which is redundant now that we allocate locks
29 * dynamically with kmalloc()/kfree().
30 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
32 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
34 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
35 * fcntl() system call. They have the semantics described above.
37 * FL_FLOCK locks are created with calls to flock(), through the flock()
38 * system call, which is new. Old C libraries implement flock() via fcntl()
39 * and will continue to use the old, broken implementation.
41 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
42 * with a file pointer (filp). As a result they can be shared by a parent
43 * process and its children after a fork(). They are removed when the last
44 * file descriptor referring to the file pointer is closed (unless explicitly
47 * FL_FLOCK locks never deadlock, an existing lock is always removed before
48 * upgrading from shared to exclusive (or vice versa). When this happens
49 * any processes blocked by the current lock are woken up and allowed to
50 * run before the new lock is applied.
51 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
53 * Removed some race conditions in flock_lock_file(), marked other possible
54 * races. Just grep for FIXME to see them.
55 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
57 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
58 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
59 * once we've checked for blocking and deadlocking.
60 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
62 * Initial implementation of mandatory locks. SunOS turned out to be
63 * a rotten model, so I implemented the "obvious" semantics.
64 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
65 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
67 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
68 * check if a file has mandatory locks, used by mmap(), open() and creat() to
69 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
71 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
73 * Tidied up block list handling. Added '/proc/locks' interface.
74 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
76 * Fixed deadlock condition for pathological code that mixes calls to
77 * flock() and fcntl().
78 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
80 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
81 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
82 * guarantee sensible behaviour in the case where file system modules might
83 * be compiled with different options than the kernel itself.
84 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
86 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
87 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
88 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
90 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
91 * locks. Changed process synchronisation to avoid dereferencing locks that
92 * have already been freed.
93 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
95 * Made the block list a circular list to minimise searching in the list.
96 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
98 * Made mandatory locking a mount option. Default is not to allow mandatory
100 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
102 * Some adaptations for NFS support.
103 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
105 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
106 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
108 * Use slab allocator instead of kmalloc/kfree.
109 * Use generic list implementation from <linux/list.h>.
110 * Sped up posix_locks_deadlock by only considering blocked locks.
111 * Matthew Wilcox <willy@debian.org>, March, 2000.
113 * Leases and LOCK_MAND
114 * Matthew Wilcox <willy@debian.org>, June, 2000.
115 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 * Locking conflicts and dependencies:
118 * If multiple threads attempt to lock the same byte (or flock the same file)
119 * only one can be granted the lock, and other must wait their turn.
120 * The first lock has been "applied" or "granted", the others are "waiting"
121 * and are "blocked" by the "applied" lock..
123 * Waiting and applied locks are all kept in trees whose properties are:
125 * - the root of a tree may be an applied or waiting lock.
126 * - every other node in the tree is a waiting lock that
127 * conflicts with every ancestor of that node.
129 * Every such tree begins life as a waiting singleton which obviously
130 * satisfies the above properties.
132 * The only ways we modify trees preserve these properties:
134 * 1. We may add a new leaf node, but only after first verifying that it
135 * conflicts with all of its ancestors.
136 * 2. We may remove the root of a tree, creating a new singleton
137 * tree from the root and N new trees rooted in the immediate
139 * 3. If the root of a tree is not currently an applied lock, we may
140 * apply it (if possible).
141 * 4. We may upgrade the root of the tree (either extend its range,
142 * or upgrade its entire range from read to write).
144 * When an applied lock is modified in a way that reduces or downgrades any
145 * part of its range, we remove all its children (2 above). This particularly
146 * happens when a lock is unlocked.
148 * For each of those child trees we "wake up" the thread which is
149 * waiting for the lock so it can continue handling as follows: if the
150 * root of the tree applies, we do so (3). If it doesn't, it must
151 * conflict with some applied lock. We remove (wake up) all of its children
152 * (2), and add it is a new leaf to the tree rooted in the applied
153 * lock (1). We then repeat the process recursively with those
158 #include <linux/capability.h>
159 #include <linux/file.h>
160 #include <linux/fdtable.h>
161 #include <linux/fs.h>
162 #include <linux/init.h>
163 #include <linux/security.h>
164 #include <linux/slab.h>
165 #include <linux/syscalls.h>
166 #include <linux/time.h>
167 #include <linux/rcupdate.h>
168 #include <linux/pid_namespace.h>
169 #include <linux/hashtable.h>
170 #include <linux/percpu.h>
172 #define CREATE_TRACE_POINTS
173 #include <trace/events/filelock.h>
175 #include <linux/uaccess.h>
177 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
178 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
179 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
180 #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
181 #define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
183 static bool lease_breaking(struct file_lock *fl)
185 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
188 static int target_leasetype(struct file_lock *fl)
190 if (fl->fl_flags & FL_UNLOCK_PENDING)
192 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
197 int leases_enable = 1;
198 int lease_break_time = 45;
201 * The global file_lock_list is only used for displaying /proc/locks, so we
202 * keep a list on each CPU, with each list protected by its own spinlock.
203 * Global serialization is done using file_rwsem.
205 * Note that alterations to the list also require that the relevant flc_lock is
208 struct file_lock_list_struct {
210 struct hlist_head hlist;
212 static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
213 DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
216 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
217 * It is protected by blocked_lock_lock.
219 * We hash locks by lockowner in order to optimize searching for the lock a
220 * particular lockowner is waiting on.
222 * FIXME: make this value scale via some heuristic? We generally will want more
223 * buckets when we have more lockowners holding locks, but that's a little
224 * difficult to determine without knowing what the workload will look like.
226 #define BLOCKED_HASH_BITS 7
227 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
230 * This lock protects the blocked_hash. Generally, if you're accessing it, you
231 * want to be holding this lock.
233 * In addition, it also protects the fl->fl_blocked_requests list, and the
234 * fl->fl_blocker pointer for file_lock structures that are acting as lock
235 * requests (in contrast to those that are acting as records of acquired locks).
237 * Note that when we acquire this lock in order to change the above fields,
238 * we often hold the flc_lock as well. In certain cases, when reading the fields
239 * protected by this lock, we can skip acquiring it iff we already hold the
242 static DEFINE_SPINLOCK(blocked_lock_lock);
244 static struct kmem_cache *flctx_cache __read_mostly;
245 static struct kmem_cache *filelock_cache __read_mostly;
247 static struct file_lock_context *
248 locks_get_lock_context(struct inode *inode, int type)
250 struct file_lock_context *ctx;
252 /* paired with cmpxchg() below */
253 ctx = smp_load_acquire(&inode->i_flctx);
254 if (likely(ctx) || type == F_UNLCK)
257 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
261 spin_lock_init(&ctx->flc_lock);
262 INIT_LIST_HEAD(&ctx->flc_flock);
263 INIT_LIST_HEAD(&ctx->flc_posix);
264 INIT_LIST_HEAD(&ctx->flc_lease);
267 * Assign the pointer if it's not already assigned. If it is, then
268 * free the context we just allocated.
270 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
271 kmem_cache_free(flctx_cache, ctx);
272 ctx = smp_load_acquire(&inode->i_flctx);
275 trace_locks_get_lock_context(inode, type, ctx);
280 locks_dump_ctx_list(struct list_head *list, char *list_type)
282 struct file_lock *fl;
284 list_for_each_entry(fl, list, fl_list) {
285 pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
290 locks_check_ctx_lists(struct inode *inode)
292 struct file_lock_context *ctx = inode->i_flctx;
294 if (unlikely(!list_empty(&ctx->flc_flock) ||
295 !list_empty(&ctx->flc_posix) ||
296 !list_empty(&ctx->flc_lease))) {
297 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
298 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
300 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
301 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
302 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
307 locks_check_ctx_file_list(struct file *filp, struct list_head *list,
310 struct file_lock *fl;
311 struct inode *inode = locks_inode(filp);
313 list_for_each_entry(fl, list, fl_list)
314 if (fl->fl_file == filp)
315 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
316 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
317 list_type, MAJOR(inode->i_sb->s_dev),
318 MINOR(inode->i_sb->s_dev), inode->i_ino,
319 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
323 locks_free_lock_context(struct inode *inode)
325 struct file_lock_context *ctx = inode->i_flctx;
328 locks_check_ctx_lists(inode);
329 kmem_cache_free(flctx_cache, ctx);
333 static void locks_init_lock_heads(struct file_lock *fl)
335 INIT_HLIST_NODE(&fl->fl_link);
336 INIT_LIST_HEAD(&fl->fl_list);
337 INIT_LIST_HEAD(&fl->fl_blocked_requests);
338 INIT_LIST_HEAD(&fl->fl_blocked_member);
339 init_waitqueue_head(&fl->fl_wait);
342 /* Allocate an empty lock structure. */
343 struct file_lock *locks_alloc_lock(void)
345 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
348 locks_init_lock_heads(fl);
352 EXPORT_SYMBOL_GPL(locks_alloc_lock);
354 void locks_release_private(struct file_lock *fl)
356 BUG_ON(waitqueue_active(&fl->fl_wait));
357 BUG_ON(!list_empty(&fl->fl_list));
358 BUG_ON(!list_empty(&fl->fl_blocked_requests));
359 BUG_ON(!list_empty(&fl->fl_blocked_member));
360 BUG_ON(!hlist_unhashed(&fl->fl_link));
363 if (fl->fl_ops->fl_release_private)
364 fl->fl_ops->fl_release_private(fl);
369 if (fl->fl_lmops->lm_put_owner) {
370 fl->fl_lmops->lm_put_owner(fl->fl_owner);
376 EXPORT_SYMBOL_GPL(locks_release_private);
378 /* Free a lock which is not in use. */
379 void locks_free_lock(struct file_lock *fl)
381 locks_release_private(fl);
382 kmem_cache_free(filelock_cache, fl);
384 EXPORT_SYMBOL(locks_free_lock);
387 locks_dispose_list(struct list_head *dispose)
389 struct file_lock *fl;
391 while (!list_empty(dispose)) {
392 fl = list_first_entry(dispose, struct file_lock, fl_list);
393 list_del_init(&fl->fl_list);
398 void locks_init_lock(struct file_lock *fl)
400 memset(fl, 0, sizeof(struct file_lock));
401 locks_init_lock_heads(fl);
403 EXPORT_SYMBOL(locks_init_lock);
406 * Initialize a new lock from an existing file_lock structure.
408 void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
410 new->fl_owner = fl->fl_owner;
411 new->fl_pid = fl->fl_pid;
413 new->fl_flags = fl->fl_flags;
414 new->fl_type = fl->fl_type;
415 new->fl_start = fl->fl_start;
416 new->fl_end = fl->fl_end;
417 new->fl_lmops = fl->fl_lmops;
421 if (fl->fl_lmops->lm_get_owner)
422 fl->fl_lmops->lm_get_owner(fl->fl_owner);
425 EXPORT_SYMBOL(locks_copy_conflock);
427 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
429 /* "new" must be a freshly-initialized lock */
430 WARN_ON_ONCE(new->fl_ops);
432 locks_copy_conflock(new, fl);
434 new->fl_file = fl->fl_file;
435 new->fl_ops = fl->fl_ops;
438 if (fl->fl_ops->fl_copy_lock)
439 fl->fl_ops->fl_copy_lock(new, fl);
442 EXPORT_SYMBOL(locks_copy_lock);
444 static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
449 * As ctx->flc_lock is held, new requests cannot be added to
450 * ->fl_blocked_requests, so we don't need a lock to check if it
453 if (list_empty(&fl->fl_blocked_requests))
455 spin_lock(&blocked_lock_lock);
456 list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
457 list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
459 spin_unlock(&blocked_lock_lock);
462 static inline int flock_translate_cmd(int cmd) {
464 return cmd & (LOCK_MAND | LOCK_RW);
476 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
477 static struct file_lock *
478 flock_make_lock(struct file *filp, unsigned int cmd, struct file_lock *fl)
480 int type = flock_translate_cmd(cmd);
483 return ERR_PTR(type);
486 fl = locks_alloc_lock();
488 return ERR_PTR(-ENOMEM);
495 fl->fl_pid = current->tgid;
496 fl->fl_flags = FL_FLOCK;
498 fl->fl_end = OFFSET_MAX;
503 static int assign_type(struct file_lock *fl, long type)
517 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
520 switch (l->l_whence) {
525 fl->fl_start = filp->f_pos;
528 fl->fl_start = i_size_read(file_inode(filp));
533 if (l->l_start > OFFSET_MAX - fl->fl_start)
535 fl->fl_start += l->l_start;
536 if (fl->fl_start < 0)
539 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
540 POSIX-2001 defines it. */
542 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
544 fl->fl_end = fl->fl_start + l->l_len - 1;
546 } else if (l->l_len < 0) {
547 if (fl->fl_start + l->l_len < 0)
549 fl->fl_end = fl->fl_start - 1;
550 fl->fl_start += l->l_len;
552 fl->fl_end = OFFSET_MAX;
554 fl->fl_owner = current->files;
555 fl->fl_pid = current->tgid;
557 fl->fl_flags = FL_POSIX;
561 return assign_type(fl, l->l_type);
564 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
567 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
570 struct flock64 ll = {
572 .l_whence = l->l_whence,
573 .l_start = l->l_start,
577 return flock64_to_posix_lock(filp, fl, &ll);
580 /* default lease lock manager operations */
582 lease_break_callback(struct file_lock *fl)
584 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
589 lease_setup(struct file_lock *fl, void **priv)
591 struct file *filp = fl->fl_file;
592 struct fasync_struct *fa = *priv;
595 * fasync_insert_entry() returns the old entry if any. If there was no
596 * old entry, then it used "priv" and inserted it into the fasync list.
597 * Clear the pointer to indicate that it shouldn't be freed.
599 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
602 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
605 static const struct lock_manager_operations lease_manager_ops = {
606 .lm_break = lease_break_callback,
607 .lm_change = lease_modify,
608 .lm_setup = lease_setup,
612 * Initialize a lease, use the default lock manager operations
614 static int lease_init(struct file *filp, long type, struct file_lock *fl)
616 if (assign_type(fl, type) != 0)
620 fl->fl_pid = current->tgid;
623 fl->fl_flags = FL_LEASE;
625 fl->fl_end = OFFSET_MAX;
627 fl->fl_lmops = &lease_manager_ops;
631 /* Allocate a file_lock initialised to this type of lease */
632 static struct file_lock *lease_alloc(struct file *filp, long type)
634 struct file_lock *fl = locks_alloc_lock();
638 return ERR_PTR(error);
640 error = lease_init(filp, type, fl);
643 return ERR_PTR(error);
648 /* Check if two locks overlap each other.
650 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
652 return ((fl1->fl_end >= fl2->fl_start) &&
653 (fl2->fl_end >= fl1->fl_start));
657 * Check whether two locks have the same owner.
659 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
661 return fl1->fl_owner == fl2->fl_owner;
664 /* Must be called with the flc_lock held! */
665 static void locks_insert_global_locks(struct file_lock *fl)
667 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
669 percpu_rwsem_assert_held(&file_rwsem);
671 spin_lock(&fll->lock);
672 fl->fl_link_cpu = smp_processor_id();
673 hlist_add_head(&fl->fl_link, &fll->hlist);
674 spin_unlock(&fll->lock);
677 /* Must be called with the flc_lock held! */
678 static void locks_delete_global_locks(struct file_lock *fl)
680 struct file_lock_list_struct *fll;
682 percpu_rwsem_assert_held(&file_rwsem);
685 * Avoid taking lock if already unhashed. This is safe since this check
686 * is done while holding the flc_lock, and new insertions into the list
687 * also require that it be held.
689 if (hlist_unhashed(&fl->fl_link))
692 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
693 spin_lock(&fll->lock);
694 hlist_del_init(&fl->fl_link);
695 spin_unlock(&fll->lock);
699 posix_owner_key(struct file_lock *fl)
701 return (unsigned long)fl->fl_owner;
704 static void locks_insert_global_blocked(struct file_lock *waiter)
706 lockdep_assert_held(&blocked_lock_lock);
708 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
711 static void locks_delete_global_blocked(struct file_lock *waiter)
713 lockdep_assert_held(&blocked_lock_lock);
715 hash_del(&waiter->fl_link);
718 /* Remove waiter from blocker's block list.
719 * When blocker ends up pointing to itself then the list is empty.
721 * Must be called with blocked_lock_lock held.
723 static void __locks_delete_block(struct file_lock *waiter)
725 locks_delete_global_blocked(waiter);
726 list_del_init(&waiter->fl_blocked_member);
727 waiter->fl_blocker = NULL;
730 static void __locks_wake_up_blocks(struct file_lock *blocker)
732 while (!list_empty(&blocker->fl_blocked_requests)) {
733 struct file_lock *waiter;
735 waiter = list_first_entry(&blocker->fl_blocked_requests,
736 struct file_lock, fl_blocked_member);
737 __locks_delete_block(waiter);
738 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
739 waiter->fl_lmops->lm_notify(waiter);
741 wake_up(&waiter->fl_wait);
746 * locks_delete_lock - stop waiting for a file lock
747 * @waiter: the lock which was waiting
749 * lockd/nfsd need to disconnect the lock while working on it.
751 int locks_delete_block(struct file_lock *waiter)
753 int status = -ENOENT;
756 * If fl_blocker is NULL, it won't be set again as this thread
757 * "owns" the lock and is the only one that might try to claim
758 * the lock. So it is safe to test fl_blocker locklessly.
759 * Also if fl_blocker is NULL, this waiter is not listed on
760 * fl_blocked_requests for some lock, so no other request can
761 * be added to the list of fl_blocked_requests for this
762 * request. So if fl_blocker is NULL, it is safe to
763 * locklessly check if fl_blocked_requests is empty. If both
764 * of these checks succeed, there is no need to take the lock.
766 if (waiter->fl_blocker == NULL &&
767 list_empty(&waiter->fl_blocked_requests))
769 spin_lock(&blocked_lock_lock);
770 if (waiter->fl_blocker)
772 __locks_wake_up_blocks(waiter);
773 __locks_delete_block(waiter);
774 spin_unlock(&blocked_lock_lock);
777 EXPORT_SYMBOL(locks_delete_block);
779 /* Insert waiter into blocker's block list.
780 * We use a circular list so that processes can be easily woken up in
781 * the order they blocked. The documentation doesn't require this but
782 * it seems like the reasonable thing to do.
784 * Must be called with both the flc_lock and blocked_lock_lock held. The
785 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
786 * but by ensuring that the flc_lock is also held on insertions we can avoid
787 * taking the blocked_lock_lock in some cases when we see that the
788 * fl_blocked_requests list is empty.
790 * Rather than just adding to the list, we check for conflicts with any existing
791 * waiters, and add beneath any waiter that blocks the new waiter.
792 * Thus wakeups don't happen until needed.
794 static void __locks_insert_block(struct file_lock *blocker,
795 struct file_lock *waiter,
796 bool conflict(struct file_lock *,
799 struct file_lock *fl;
800 BUG_ON(!list_empty(&waiter->fl_blocked_member));
803 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
804 if (conflict(fl, waiter)) {
808 waiter->fl_blocker = blocker;
809 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
810 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
811 locks_insert_global_blocked(waiter);
813 /* The requests in waiter->fl_blocked are known to conflict with
814 * waiter, but might not conflict with blocker, or the requests
815 * and lock which block it. So they all need to be woken.
817 __locks_wake_up_blocks(waiter);
820 /* Must be called with flc_lock held. */
821 static void locks_insert_block(struct file_lock *blocker,
822 struct file_lock *waiter,
823 bool conflict(struct file_lock *,
826 spin_lock(&blocked_lock_lock);
827 __locks_insert_block(blocker, waiter, conflict);
828 spin_unlock(&blocked_lock_lock);
832 * Wake up processes blocked waiting for blocker.
834 * Must be called with the inode->flc_lock held!
836 static void locks_wake_up_blocks(struct file_lock *blocker)
839 * Avoid taking global lock if list is empty. This is safe since new
840 * blocked requests are only added to the list under the flc_lock, and
841 * the flc_lock is always held here. Note that removal from the
842 * fl_blocked_requests list does not require the flc_lock, so we must
843 * recheck list_empty() after acquiring the blocked_lock_lock.
845 if (list_empty(&blocker->fl_blocked_requests))
848 spin_lock(&blocked_lock_lock);
849 __locks_wake_up_blocks(blocker);
850 spin_unlock(&blocked_lock_lock);
854 locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
856 list_add_tail(&fl->fl_list, before);
857 locks_insert_global_locks(fl);
861 locks_unlink_lock_ctx(struct file_lock *fl)
863 locks_delete_global_locks(fl);
864 list_del_init(&fl->fl_list);
865 locks_wake_up_blocks(fl);
869 locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
871 locks_unlink_lock_ctx(fl);
873 list_add(&fl->fl_list, dispose);
878 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
879 * checks for shared/exclusive status of overlapping locks.
881 static bool locks_conflict(struct file_lock *caller_fl,
882 struct file_lock *sys_fl)
884 if (sys_fl->fl_type == F_WRLCK)
886 if (caller_fl->fl_type == F_WRLCK)
891 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
892 * checking before calling the locks_conflict().
894 static bool posix_locks_conflict(struct file_lock *caller_fl,
895 struct file_lock *sys_fl)
897 /* POSIX locks owned by the same process do not conflict with
900 if (posix_same_owner(caller_fl, sys_fl))
903 /* Check whether they overlap */
904 if (!locks_overlap(caller_fl, sys_fl))
907 return locks_conflict(caller_fl, sys_fl);
910 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
911 * checking before calling the locks_conflict().
913 static bool flock_locks_conflict(struct file_lock *caller_fl,
914 struct file_lock *sys_fl)
916 /* FLOCK locks referring to the same filp do not conflict with
919 if (caller_fl->fl_file == sys_fl->fl_file)
921 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
924 return locks_conflict(caller_fl, sys_fl);
928 posix_test_lock(struct file *filp, struct file_lock *fl)
930 struct file_lock *cfl;
931 struct file_lock_context *ctx;
932 struct inode *inode = locks_inode(filp);
934 ctx = smp_load_acquire(&inode->i_flctx);
935 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
936 fl->fl_type = F_UNLCK;
940 spin_lock(&ctx->flc_lock);
941 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
942 if (posix_locks_conflict(fl, cfl)) {
943 locks_copy_conflock(fl, cfl);
947 fl->fl_type = F_UNLCK;
949 spin_unlock(&ctx->flc_lock);
952 EXPORT_SYMBOL(posix_test_lock);
955 * Deadlock detection:
957 * We attempt to detect deadlocks that are due purely to posix file
960 * We assume that a task can be waiting for at most one lock at a time.
961 * So for any acquired lock, the process holding that lock may be
962 * waiting on at most one other lock. That lock in turns may be held by
963 * someone waiting for at most one other lock. Given a requested lock
964 * caller_fl which is about to wait for a conflicting lock block_fl, we
965 * follow this chain of waiters to ensure we are not about to create a
968 * Since we do this before we ever put a process to sleep on a lock, we
969 * are ensured that there is never a cycle; that is what guarantees that
970 * the while() loop in posix_locks_deadlock() eventually completes.
972 * Note: the above assumption may not be true when handling lock
973 * requests from a broken NFS client. It may also fail in the presence
974 * of tasks (such as posix threads) sharing the same open file table.
975 * To handle those cases, we just bail out after a few iterations.
977 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
978 * Because the owner is not even nominally tied to a thread of
979 * execution, the deadlock detection below can't reasonably work well. Just
982 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
983 * locks that just checks for the case where two tasks are attempting to
984 * upgrade from read to write locks on the same inode.
987 #define MAX_DEADLK_ITERATIONS 10
989 /* Find a lock that the owner of the given block_fl is blocking on. */
990 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
992 struct file_lock *fl;
994 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
995 if (posix_same_owner(fl, block_fl)) {
996 while (fl->fl_blocker)
1004 /* Must be called with the blocked_lock_lock held! */
1005 static int posix_locks_deadlock(struct file_lock *caller_fl,
1006 struct file_lock *block_fl)
1010 lockdep_assert_held(&blocked_lock_lock);
1013 * This deadlock detector can't reasonably detect deadlocks with
1014 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1016 if (IS_OFDLCK(caller_fl))
1019 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
1020 if (i++ > MAX_DEADLK_ITERATIONS)
1022 if (posix_same_owner(caller_fl, block_fl))
1028 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1029 * after any leases, but before any posix locks.
1031 * Note that if called with an FL_EXISTS argument, the caller may determine
1032 * whether or not a lock was successfully freed by testing the return
1033 * value for -ENOENT.
1035 static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1037 struct file_lock *new_fl = NULL;
1038 struct file_lock *fl;
1039 struct file_lock_context *ctx;
1044 ctx = locks_get_lock_context(inode, request->fl_type);
1046 if (request->fl_type != F_UNLCK)
1048 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1051 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1052 new_fl = locks_alloc_lock();
1057 percpu_down_read(&file_rwsem);
1058 spin_lock(&ctx->flc_lock);
1059 if (request->fl_flags & FL_ACCESS)
1062 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1063 if (request->fl_file != fl->fl_file)
1065 if (request->fl_type == fl->fl_type)
1068 locks_delete_lock_ctx(fl, &dispose);
1072 if (request->fl_type == F_UNLCK) {
1073 if ((request->fl_flags & FL_EXISTS) && !found)
1079 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1080 if (!flock_locks_conflict(request, fl))
1083 if (!(request->fl_flags & FL_SLEEP))
1085 error = FILE_LOCK_DEFERRED;
1086 locks_insert_block(fl, request, flock_locks_conflict);
1089 if (request->fl_flags & FL_ACCESS)
1091 locks_copy_lock(new_fl, request);
1092 locks_move_blocks(new_fl, request);
1093 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1098 spin_unlock(&ctx->flc_lock);
1099 percpu_up_read(&file_rwsem);
1101 locks_free_lock(new_fl);
1102 locks_dispose_list(&dispose);
1103 trace_flock_lock_inode(inode, request, error);
1107 static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1108 struct file_lock *conflock)
1110 struct file_lock *fl, *tmp;
1111 struct file_lock *new_fl = NULL;
1112 struct file_lock *new_fl2 = NULL;
1113 struct file_lock *left = NULL;
1114 struct file_lock *right = NULL;
1115 struct file_lock_context *ctx;
1120 ctx = locks_get_lock_context(inode, request->fl_type);
1122 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1125 * We may need two file_lock structures for this operation,
1126 * so we get them in advance to avoid races.
1128 * In some cases we can be sure, that no new locks will be needed
1130 if (!(request->fl_flags & FL_ACCESS) &&
1131 (request->fl_type != F_UNLCK ||
1132 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1133 new_fl = locks_alloc_lock();
1134 new_fl2 = locks_alloc_lock();
1137 percpu_down_read(&file_rwsem);
1138 spin_lock(&ctx->flc_lock);
1140 * New lock request. Walk all POSIX locks and look for conflicts. If
1141 * there are any, either return error or put the request on the
1142 * blocker's list of waiters and the global blocked_hash.
1144 if (request->fl_type != F_UNLCK) {
1145 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1146 if (!posix_locks_conflict(request, fl))
1149 locks_copy_conflock(conflock, fl);
1151 if (!(request->fl_flags & FL_SLEEP))
1154 * Deadlock detection and insertion into the blocked
1155 * locks list must be done while holding the same lock!
1158 spin_lock(&blocked_lock_lock);
1160 * Ensure that we don't find any locks blocked on this
1161 * request during deadlock detection.
1163 __locks_wake_up_blocks(request);
1164 if (likely(!posix_locks_deadlock(request, fl))) {
1165 error = FILE_LOCK_DEFERRED;
1166 __locks_insert_block(fl, request,
1167 posix_locks_conflict);
1169 spin_unlock(&blocked_lock_lock);
1174 /* If we're just looking for a conflict, we're done. */
1176 if (request->fl_flags & FL_ACCESS)
1179 /* Find the first old lock with the same owner as the new lock */
1180 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1181 if (posix_same_owner(request, fl))
1185 /* Process locks with this owner. */
1186 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1187 if (!posix_same_owner(request, fl))
1190 /* Detect adjacent or overlapping regions (if same lock type) */
1191 if (request->fl_type == fl->fl_type) {
1192 /* In all comparisons of start vs end, use
1193 * "start - 1" rather than "end + 1". If end
1194 * is OFFSET_MAX, end + 1 will become negative.
1196 if (fl->fl_end < request->fl_start - 1)
1198 /* If the next lock in the list has entirely bigger
1199 * addresses than the new one, insert the lock here.
1201 if (fl->fl_start - 1 > request->fl_end)
1204 /* If we come here, the new and old lock are of the
1205 * same type and adjacent or overlapping. Make one
1206 * lock yielding from the lower start address of both
1207 * locks to the higher end address.
1209 if (fl->fl_start > request->fl_start)
1210 fl->fl_start = request->fl_start;
1212 request->fl_start = fl->fl_start;
1213 if (fl->fl_end < request->fl_end)
1214 fl->fl_end = request->fl_end;
1216 request->fl_end = fl->fl_end;
1218 locks_delete_lock_ctx(fl, &dispose);
1224 /* Processing for different lock types is a bit
1227 if (fl->fl_end < request->fl_start)
1229 if (fl->fl_start > request->fl_end)
1231 if (request->fl_type == F_UNLCK)
1233 if (fl->fl_start < request->fl_start)
1235 /* If the next lock in the list has a higher end
1236 * address than the new one, insert the new one here.
1238 if (fl->fl_end > request->fl_end) {
1242 if (fl->fl_start >= request->fl_start) {
1243 /* The new lock completely replaces an old
1244 * one (This may happen several times).
1247 locks_delete_lock_ctx(fl, &dispose);
1251 * Replace the old lock with new_fl, and
1252 * remove the old one. It's safe to do the
1253 * insert here since we know that we won't be
1254 * using new_fl later, and that the lock is
1255 * just replacing an existing lock.
1260 locks_copy_lock(new_fl, request);
1263 locks_insert_lock_ctx(request, &fl->fl_list);
1264 locks_delete_lock_ctx(fl, &dispose);
1271 * The above code only modifies existing locks in case of merging or
1272 * replacing. If new lock(s) need to be inserted all modifications are
1273 * done below this, so it's safe yet to bail out.
1275 error = -ENOLCK; /* "no luck" */
1276 if (right && left == right && !new_fl2)
1281 if (request->fl_type == F_UNLCK) {
1282 if (request->fl_flags & FL_EXISTS)
1291 locks_copy_lock(new_fl, request);
1292 locks_move_blocks(new_fl, request);
1293 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1298 if (left == right) {
1299 /* The new lock breaks the old one in two pieces,
1300 * so we have to use the second new lock.
1304 locks_copy_lock(left, right);
1305 locks_insert_lock_ctx(left, &fl->fl_list);
1307 right->fl_start = request->fl_end + 1;
1308 locks_wake_up_blocks(right);
1311 left->fl_end = request->fl_start - 1;
1312 locks_wake_up_blocks(left);
1315 spin_unlock(&ctx->flc_lock);
1316 percpu_up_read(&file_rwsem);
1318 * Free any unused locks.
1321 locks_free_lock(new_fl);
1323 locks_free_lock(new_fl2);
1324 locks_dispose_list(&dispose);
1325 trace_posix_lock_inode(inode, request, error);
1331 * posix_lock_file - Apply a POSIX-style lock to a file
1332 * @filp: The file to apply the lock to
1333 * @fl: The lock to be applied
1334 * @conflock: Place to return a copy of the conflicting lock, if found.
1336 * Add a POSIX style lock to a file.
1337 * We merge adjacent & overlapping locks whenever possible.
1338 * POSIX locks are sorted by owner task, then by starting address
1340 * Note that if called with an FL_EXISTS argument, the caller may determine
1341 * whether or not a lock was successfully freed by testing the return
1342 * value for -ENOENT.
1344 int posix_lock_file(struct file *filp, struct file_lock *fl,
1345 struct file_lock *conflock)
1347 return posix_lock_inode(locks_inode(filp), fl, conflock);
1349 EXPORT_SYMBOL(posix_lock_file);
1352 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1353 * @inode: inode of file to which lock request should be applied
1354 * @fl: The lock to be applied
1356 * Apply a POSIX style lock request to an inode.
1358 static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1363 error = posix_lock_inode(inode, fl, NULL);
1364 if (error != FILE_LOCK_DEFERRED)
1366 error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
1370 locks_delete_block(fl);
1374 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1376 * locks_mandatory_locked - Check for an active lock
1377 * @file: the file to check
1379 * Searches the inode's list of locks to find any POSIX locks which conflict.
1380 * This function is called from locks_verify_locked() only.
1382 int locks_mandatory_locked(struct file *file)
1385 struct inode *inode = locks_inode(file);
1386 struct file_lock_context *ctx;
1387 struct file_lock *fl;
1389 ctx = smp_load_acquire(&inode->i_flctx);
1390 if (!ctx || list_empty_careful(&ctx->flc_posix))
1394 * Search the lock list for this inode for any POSIX locks.
1396 spin_lock(&ctx->flc_lock);
1398 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1399 if (fl->fl_owner != current->files &&
1400 fl->fl_owner != file) {
1405 spin_unlock(&ctx->flc_lock);
1410 * locks_mandatory_area - Check for a conflicting lock
1411 * @inode: the file to check
1412 * @filp: how the file was opened (if it was)
1413 * @start: first byte in the file to check
1414 * @end: lastbyte in the file to check
1415 * @type: %F_WRLCK for a write lock, else %F_RDLCK
1417 * Searches the inode's list of locks to find any POSIX locks which conflict.
1419 int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1420 loff_t end, unsigned char type)
1422 struct file_lock fl;
1426 locks_init_lock(&fl);
1427 fl.fl_pid = current->tgid;
1429 fl.fl_flags = FL_POSIX | FL_ACCESS;
1430 if (filp && !(filp->f_flags & O_NONBLOCK))
1433 fl.fl_start = start;
1439 fl.fl_flags &= ~FL_SLEEP;
1440 error = posix_lock_inode(inode, &fl, NULL);
1446 fl.fl_flags |= FL_SLEEP;
1447 fl.fl_owner = current->files;
1448 error = posix_lock_inode(inode, &fl, NULL);
1449 if (error != FILE_LOCK_DEFERRED)
1451 error = wait_event_interruptible(fl.fl_wait, !fl.fl_blocker);
1454 * If we've been sleeping someone might have
1455 * changed the permissions behind our back.
1457 if (__mandatory_lock(inode))
1463 locks_delete_block(&fl);
1467 EXPORT_SYMBOL(locks_mandatory_area);
1468 #endif /* CONFIG_MANDATORY_FILE_LOCKING */
1470 static void lease_clear_pending(struct file_lock *fl, int arg)
1474 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1477 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1481 /* We already had a lease on this file; just change its type */
1482 int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1484 int error = assign_type(fl, arg);
1488 lease_clear_pending(fl, arg);
1489 locks_wake_up_blocks(fl);
1490 if (arg == F_UNLCK) {
1491 struct file *filp = fl->fl_file;
1494 filp->f_owner.signum = 0;
1495 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1496 if (fl->fl_fasync != NULL) {
1497 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1498 fl->fl_fasync = NULL;
1500 locks_delete_lock_ctx(fl, dispose);
1504 EXPORT_SYMBOL(lease_modify);
1506 static bool past_time(unsigned long then)
1509 /* 0 is a special value meaning "this never expires": */
1511 return time_after(jiffies, then);
1514 static void time_out_leases(struct inode *inode, struct list_head *dispose)
1516 struct file_lock_context *ctx = inode->i_flctx;
1517 struct file_lock *fl, *tmp;
1519 lockdep_assert_held(&ctx->flc_lock);
1521 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1522 trace_time_out_leases(inode, fl);
1523 if (past_time(fl->fl_downgrade_time))
1524 lease_modify(fl, F_RDLCK, dispose);
1525 if (past_time(fl->fl_break_time))
1526 lease_modify(fl, F_UNLCK, dispose);
1530 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1534 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1538 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1543 rc = locks_conflict(breaker, lease);
1545 trace_leases_conflict(rc, lease, breaker);
1550 any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1552 struct file_lock_context *ctx = inode->i_flctx;
1553 struct file_lock *fl;
1555 lockdep_assert_held(&ctx->flc_lock);
1557 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1558 if (leases_conflict(fl, breaker))
1565 * __break_lease - revoke all outstanding leases on file
1566 * @inode: the inode of the file to return
1567 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1569 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1572 * break_lease (inlined for speed) has checked there already is at least
1573 * some kind of lock (maybe a lease) on this file. Leases are broken on
1574 * a call to open() or truncate(). This function can sleep unless you
1575 * specified %O_NONBLOCK to your open().
1577 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1580 struct file_lock_context *ctx;
1581 struct file_lock *new_fl, *fl, *tmp;
1582 unsigned long break_time;
1583 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1586 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1588 return PTR_ERR(new_fl);
1589 new_fl->fl_flags = type;
1591 /* typically we will check that ctx is non-NULL before calling */
1592 ctx = smp_load_acquire(&inode->i_flctx);
1598 percpu_down_read(&file_rwsem);
1599 spin_lock(&ctx->flc_lock);
1601 time_out_leases(inode, &dispose);
1603 if (!any_leases_conflict(inode, new_fl))
1607 if (lease_break_time > 0) {
1608 break_time = jiffies + lease_break_time * HZ;
1609 if (break_time == 0)
1610 break_time++; /* so that 0 means no break time */
1613 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1614 if (!leases_conflict(fl, new_fl))
1617 if (fl->fl_flags & FL_UNLOCK_PENDING)
1619 fl->fl_flags |= FL_UNLOCK_PENDING;
1620 fl->fl_break_time = break_time;
1622 if (lease_breaking(fl))
1624 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1625 fl->fl_downgrade_time = break_time;
1627 if (fl->fl_lmops->lm_break(fl))
1628 locks_delete_lock_ctx(fl, &dispose);
1631 if (list_empty(&ctx->flc_lease))
1634 if (mode & O_NONBLOCK) {
1635 trace_break_lease_noblock(inode, new_fl);
1636 error = -EWOULDBLOCK;
1641 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1642 break_time = fl->fl_break_time;
1643 if (break_time != 0)
1644 break_time -= jiffies;
1645 if (break_time == 0)
1647 locks_insert_block(fl, new_fl, leases_conflict);
1648 trace_break_lease_block(inode, new_fl);
1649 spin_unlock(&ctx->flc_lock);
1650 percpu_up_read(&file_rwsem);
1652 locks_dispose_list(&dispose);
1653 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1654 !new_fl->fl_blocker, break_time);
1656 percpu_down_read(&file_rwsem);
1657 spin_lock(&ctx->flc_lock);
1658 trace_break_lease_unblock(inode, new_fl);
1659 locks_delete_block(new_fl);
1662 * Wait for the next conflicting lease that has not been
1666 time_out_leases(inode, &dispose);
1667 if (any_leases_conflict(inode, new_fl))
1672 spin_unlock(&ctx->flc_lock);
1673 percpu_up_read(&file_rwsem);
1674 locks_dispose_list(&dispose);
1675 locks_free_lock(new_fl);
1678 EXPORT_SYMBOL(__break_lease);
1681 * lease_get_mtime - update modified time of an inode with exclusive lease
1683 * @time: pointer to a timespec which contains the last modified time
1685 * This is to force NFS clients to flush their caches for files with
1686 * exclusive leases. The justification is that if someone has an
1687 * exclusive lease, then they could be modifying it.
1689 void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1691 bool has_lease = false;
1692 struct file_lock_context *ctx;
1693 struct file_lock *fl;
1695 ctx = smp_load_acquire(&inode->i_flctx);
1696 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1697 spin_lock(&ctx->flc_lock);
1698 fl = list_first_entry_or_null(&ctx->flc_lease,
1699 struct file_lock, fl_list);
1700 if (fl && (fl->fl_type == F_WRLCK))
1702 spin_unlock(&ctx->flc_lock);
1706 *time = current_time(inode);
1708 EXPORT_SYMBOL(lease_get_mtime);
1711 * fcntl_getlease - Enquire what lease is currently active
1714 * The value returned by this function will be one of
1715 * (if no lease break is pending):
1717 * %F_RDLCK to indicate a shared lease is held.
1719 * %F_WRLCK to indicate an exclusive lease is held.
1721 * %F_UNLCK to indicate no lease is held.
1723 * (if a lease break is pending):
1725 * %F_RDLCK to indicate an exclusive lease needs to be
1726 * changed to a shared lease (or removed).
1728 * %F_UNLCK to indicate the lease needs to be removed.
1730 * XXX: sfr & willy disagree over whether F_INPROGRESS
1731 * should be returned to userspace.
1733 int fcntl_getlease(struct file *filp)
1735 struct file_lock *fl;
1736 struct inode *inode = locks_inode(filp);
1737 struct file_lock_context *ctx;
1741 ctx = smp_load_acquire(&inode->i_flctx);
1742 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1743 percpu_down_read(&file_rwsem);
1744 spin_lock(&ctx->flc_lock);
1745 time_out_leases(inode, &dispose);
1746 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1747 if (fl->fl_file != filp)
1749 type = target_leasetype(fl);
1752 spin_unlock(&ctx->flc_lock);
1753 percpu_up_read(&file_rwsem);
1755 locks_dispose_list(&dispose);
1761 * check_conflicting_open - see if the given file points to an inode that has
1762 * an existing open that would conflict with the
1764 * @filp: file to check
1765 * @arg: type of lease that we're trying to acquire
1766 * @flags: current lock flags
1768 * Check to see if there's an existing open fd on this file that would
1769 * conflict with the lease we're trying to set.
1772 check_conflicting_open(struct file *filp, const long arg, int flags)
1774 struct inode *inode = locks_inode(filp);
1775 int self_wcount = 0, self_rcount = 0;
1777 if (flags & FL_LAYOUT)
1781 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1782 else if (arg != F_WRLCK)
1786 * Make sure that only read/write count is from lease requestor.
1787 * Note that this will result in denying write leases when i_writecount
1788 * is negative, which is what we want. (We shouldn't grant write leases
1789 * on files open for execution.)
1791 if (filp->f_mode & FMODE_WRITE)
1793 else if (filp->f_mode & FMODE_READ)
1796 if (atomic_read(&inode->i_writecount) != self_wcount ||
1797 atomic_read(&inode->i_readcount) != self_rcount)
1804 generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1806 struct file_lock *fl, *my_fl = NULL, *lease;
1807 struct inode *inode = locks_inode(filp);
1808 struct file_lock_context *ctx;
1809 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1814 trace_generic_add_lease(inode, lease);
1816 /* Note that arg is never F_UNLCK here */
1817 ctx = locks_get_lock_context(inode, arg);
1822 * In the delegation case we need mutual exclusion with
1823 * a number of operations that take the i_mutex. We trylock
1824 * because delegations are an optional optimization, and if
1825 * there's some chance of a conflict--we'd rather not
1826 * bother, maybe that's a sign this just isn't a good file to
1827 * hand out a delegation on.
1829 if (is_deleg && !inode_trylock(inode))
1832 if (is_deleg && arg == F_WRLCK) {
1833 /* Write delegations are not currently supported: */
1834 inode_unlock(inode);
1839 percpu_down_read(&file_rwsem);
1840 spin_lock(&ctx->flc_lock);
1841 time_out_leases(inode, &dispose);
1842 error = check_conflicting_open(filp, arg, lease->fl_flags);
1847 * At this point, we know that if there is an exclusive
1848 * lease on this file, then we hold it on this filp
1849 * (otherwise our open of this file would have blocked).
1850 * And if we are trying to acquire an exclusive lease,
1851 * then the file is not open by anyone (including us)
1852 * except for this filp.
1855 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1856 if (fl->fl_file == filp &&
1857 fl->fl_owner == lease->fl_owner) {
1863 * No exclusive leases if someone else has a lease on
1869 * Modifying our existing lease is OK, but no getting a
1870 * new lease if someone else is opening for write:
1872 if (fl->fl_flags & FL_UNLOCK_PENDING)
1876 if (my_fl != NULL) {
1878 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1888 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1890 * The check in break_lease() is lockless. It's possible for another
1891 * open to race in after we did the earlier check for a conflicting
1892 * open but before the lease was inserted. Check again for a
1893 * conflicting open and cancel the lease if there is one.
1895 * We also add a barrier here to ensure that the insertion of the lock
1896 * precedes these checks.
1899 error = check_conflicting_open(filp, arg, lease->fl_flags);
1901 locks_unlink_lock_ctx(lease);
1906 if (lease->fl_lmops->lm_setup)
1907 lease->fl_lmops->lm_setup(lease, priv);
1909 spin_unlock(&ctx->flc_lock);
1910 percpu_up_read(&file_rwsem);
1911 locks_dispose_list(&dispose);
1913 inode_unlock(inode);
1914 if (!error && !my_fl)
1919 static int generic_delete_lease(struct file *filp, void *owner)
1921 int error = -EAGAIN;
1922 struct file_lock *fl, *victim = NULL;
1923 struct inode *inode = locks_inode(filp);
1924 struct file_lock_context *ctx;
1927 ctx = smp_load_acquire(&inode->i_flctx);
1929 trace_generic_delete_lease(inode, NULL);
1933 percpu_down_read(&file_rwsem);
1934 spin_lock(&ctx->flc_lock);
1935 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1936 if (fl->fl_file == filp &&
1937 fl->fl_owner == owner) {
1942 trace_generic_delete_lease(inode, victim);
1944 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1945 spin_unlock(&ctx->flc_lock);
1946 percpu_up_read(&file_rwsem);
1947 locks_dispose_list(&dispose);
1952 * generic_setlease - sets a lease on an open file
1953 * @filp: file pointer
1954 * @arg: type of lease to obtain
1955 * @flp: input - file_lock to use, output - file_lock inserted
1956 * @priv: private data for lm_setup (may be NULL if lm_setup
1957 * doesn't require it)
1959 * The (input) flp->fl_lmops->lm_break function is required
1962 int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1965 struct inode *inode = locks_inode(filp);
1968 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1970 if (!S_ISREG(inode->i_mode))
1972 error = security_file_lock(filp, arg);
1978 return generic_delete_lease(filp, *priv);
1981 if (!(*flp)->fl_lmops->lm_break) {
1986 return generic_add_lease(filp, arg, flp, priv);
1991 EXPORT_SYMBOL(generic_setlease);
1994 * vfs_setlease - sets a lease on an open file
1995 * @filp: file pointer
1996 * @arg: type of lease to obtain
1997 * @lease: file_lock to use when adding a lease
1998 * @priv: private info for lm_setup when adding a lease (may be
1999 * NULL if lm_setup doesn't require it)
2001 * Call this to establish a lease on the file. The "lease" argument is not
2002 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2003 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2004 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2007 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2008 * may be NULL if the lm_setup operation doesn't require it.
2011 vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
2013 if (filp->f_op->setlease)
2014 return filp->f_op->setlease(filp, arg, lease, priv);
2016 return generic_setlease(filp, arg, lease, priv);
2018 EXPORT_SYMBOL_GPL(vfs_setlease);
2020 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
2022 struct file_lock *fl;
2023 struct fasync_struct *new;
2026 fl = lease_alloc(filp, arg);
2030 new = fasync_alloc();
2032 locks_free_lock(fl);
2037 error = vfs_setlease(filp, arg, &fl, (void **)&new);
2039 locks_free_lock(fl);
2046 * fcntl_setlease - sets a lease on an open file
2047 * @fd: open file descriptor
2048 * @filp: file pointer
2049 * @arg: type of lease to obtain
2051 * Call this fcntl to establish a lease on the file.
2052 * Note that you also need to call %F_SETSIG to
2053 * receive a signal when the lease is broken.
2055 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2058 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2059 return do_fcntl_add_lease(fd, filp, arg);
2063 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2064 * @inode: inode of the file to apply to
2065 * @fl: The lock to be applied
2067 * Apply a FLOCK style lock request to an inode.
2069 static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2074 error = flock_lock_inode(inode, fl);
2075 if (error != FILE_LOCK_DEFERRED)
2077 error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
2081 locks_delete_block(fl);
2086 * locks_lock_inode_wait - Apply a lock to an inode
2087 * @inode: inode of the file to apply to
2088 * @fl: The lock to be applied
2090 * Apply a POSIX or FLOCK style lock request to an inode.
2092 int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2095 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2097 res = posix_lock_inode_wait(inode, fl);
2100 res = flock_lock_inode_wait(inode, fl);
2107 EXPORT_SYMBOL(locks_lock_inode_wait);
2110 * sys_flock: - flock() system call.
2111 * @fd: the file descriptor to lock.
2112 * @cmd: the type of lock to apply.
2114 * Apply a %FL_FLOCK style lock to an open file descriptor.
2115 * The @cmd can be one of:
2117 * - %LOCK_SH -- a shared lock.
2118 * - %LOCK_EX -- an exclusive lock.
2119 * - %LOCK_UN -- remove an existing lock.
2120 * - %LOCK_MAND -- a 'mandatory' flock.
2121 * This exists to emulate Windows Share Modes.
2123 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
2124 * processes read and write access respectively.
2126 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2128 struct fd f = fdget(fd);
2129 struct file_lock *lock;
2130 int can_sleep, unlock;
2137 can_sleep = !(cmd & LOCK_NB);
2139 unlock = (cmd == LOCK_UN);
2141 if (!unlock && !(cmd & LOCK_MAND) &&
2142 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2145 lock = flock_make_lock(f.file, cmd, NULL);
2147 error = PTR_ERR(lock);
2152 lock->fl_flags |= FL_SLEEP;
2154 error = security_file_lock(f.file, lock->fl_type);
2158 if (f.file->f_op->flock)
2159 error = f.file->f_op->flock(f.file,
2160 (can_sleep) ? F_SETLKW : F_SETLK,
2163 error = locks_lock_file_wait(f.file, lock);
2166 locks_free_lock(lock);
2175 * vfs_test_lock - test file byte range lock
2176 * @filp: The file to test lock for
2177 * @fl: The lock to test; also used to hold result
2179 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2180 * setting conf->fl_type to something other than F_UNLCK.
2182 int vfs_test_lock(struct file *filp, struct file_lock *fl)
2184 if (filp->f_op->lock)
2185 return filp->f_op->lock(filp, F_GETLK, fl);
2186 posix_test_lock(filp, fl);
2189 EXPORT_SYMBOL_GPL(vfs_test_lock);
2192 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2193 * @fl: The file_lock who's fl_pid should be translated
2194 * @ns: The namespace into which the pid should be translated
2196 * Used to tranlate a fl_pid into a namespace virtual pid number
2198 static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2205 if (IS_REMOTELCK(fl))
2208 * If the flock owner process is dead and its pid has been already
2209 * freed, the translation below won't work, but we still want to show
2210 * flock owner pid number in init pidns.
2212 if (ns == &init_pid_ns)
2213 return (pid_t)fl->fl_pid;
2216 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2217 vnr = pid_nr_ns(pid, ns);
2222 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2224 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2225 #if BITS_PER_LONG == 32
2227 * Make sure we can represent the posix lock via
2228 * legacy 32bit flock.
2230 if (fl->fl_start > OFFT_OFFSET_MAX)
2232 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2235 flock->l_start = fl->fl_start;
2236 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2237 fl->fl_end - fl->fl_start + 1;
2238 flock->l_whence = 0;
2239 flock->l_type = fl->fl_type;
2243 #if BITS_PER_LONG == 32
2244 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2246 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2247 flock->l_start = fl->fl_start;
2248 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2249 fl->fl_end - fl->fl_start + 1;
2250 flock->l_whence = 0;
2251 flock->l_type = fl->fl_type;
2255 /* Report the first existing lock that would conflict with l.
2256 * This implements the F_GETLK command of fcntl().
2258 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2260 struct file_lock *fl;
2263 fl = locks_alloc_lock();
2267 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2270 error = flock_to_posix_lock(filp, fl, flock);
2274 if (cmd == F_OFD_GETLK) {
2276 if (flock->l_pid != 0)
2280 fl->fl_flags |= FL_OFDLCK;
2281 fl->fl_owner = filp;
2284 error = vfs_test_lock(filp, fl);
2288 flock->l_type = fl->fl_type;
2289 if (fl->fl_type != F_UNLCK) {
2290 error = posix_lock_to_flock(flock, fl);
2295 locks_free_lock(fl);
2300 * vfs_lock_file - file byte range lock
2301 * @filp: The file to apply the lock to
2302 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2303 * @fl: The lock to be applied
2304 * @conf: Place to return a copy of the conflicting lock, if found.
2306 * A caller that doesn't care about the conflicting lock may pass NULL
2307 * as the final argument.
2309 * If the filesystem defines a private ->lock() method, then @conf will
2310 * be left unchanged; so a caller that cares should initialize it to
2311 * some acceptable default.
2313 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2314 * locks, the ->lock() interface may return asynchronously, before the lock has
2315 * been granted or denied by the underlying filesystem, if (and only if)
2316 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2317 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2318 * the request is for a blocking lock. When ->lock() does return asynchronously,
2319 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2320 * request completes.
2321 * If the request is for non-blocking lock the file system should return
2322 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2323 * with the result. If the request timed out the callback routine will return a
2324 * nonzero return code and the file system should release the lock. The file
2325 * system is also responsible to keep a corresponding posix lock when it
2326 * grants a lock so the VFS can find out which locks are locally held and do
2327 * the correct lock cleanup when required.
2328 * The underlying filesystem must not drop the kernel lock or call
2329 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2332 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2334 if (filp->f_op->lock)
2335 return filp->f_op->lock(filp, cmd, fl);
2337 return posix_lock_file(filp, fl, conf);
2339 EXPORT_SYMBOL_GPL(vfs_lock_file);
2341 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2342 struct file_lock *fl)
2346 error = security_file_lock(filp, fl->fl_type);
2351 error = vfs_lock_file(filp, cmd, fl, NULL);
2352 if (error != FILE_LOCK_DEFERRED)
2354 error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
2358 locks_delete_block(fl);
2363 /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2365 check_fmode_for_setlk(struct file_lock *fl)
2367 switch (fl->fl_type) {
2369 if (!(fl->fl_file->f_mode & FMODE_READ))
2373 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2379 /* Apply the lock described by l to an open file descriptor.
2380 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2382 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2383 struct flock *flock)
2385 struct file_lock *file_lock = locks_alloc_lock();
2386 struct inode *inode = locks_inode(filp);
2390 if (file_lock == NULL)
2393 /* Don't allow mandatory locks on files that may be memory mapped
2396 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2401 error = flock_to_posix_lock(filp, file_lock, flock);
2405 error = check_fmode_for_setlk(file_lock);
2410 * If the cmd is requesting file-private locks, then set the
2411 * FL_OFDLCK flag and override the owner.
2416 if (flock->l_pid != 0)
2420 file_lock->fl_flags |= FL_OFDLCK;
2421 file_lock->fl_owner = filp;
2425 if (flock->l_pid != 0)
2429 file_lock->fl_flags |= FL_OFDLCK;
2430 file_lock->fl_owner = filp;
2433 file_lock->fl_flags |= FL_SLEEP;
2436 error = do_lock_file_wait(filp, cmd, file_lock);
2439 * Attempt to detect a close/fcntl race and recover by releasing the
2440 * lock that was just acquired. There is no need to do that when we're
2441 * unlocking though, or for OFD locks.
2443 if (!error && file_lock->fl_type != F_UNLCK &&
2444 !(file_lock->fl_flags & FL_OFDLCK)) {
2446 * We need that spin_lock here - it prevents reordering between
2447 * update of i_flctx->flc_posix and check for it done in
2448 * close(). rcu_read_lock() wouldn't do.
2450 spin_lock(¤t->files->file_lock);
2452 spin_unlock(¤t->files->file_lock);
2454 file_lock->fl_type = F_UNLCK;
2455 error = do_lock_file_wait(filp, cmd, file_lock);
2456 WARN_ON_ONCE(error);
2461 trace_fcntl_setlk(inode, file_lock, error);
2462 locks_free_lock(file_lock);
2466 #if BITS_PER_LONG == 32
2467 /* Report the first existing lock that would conflict with l.
2468 * This implements the F_GETLK command of fcntl().
2470 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2472 struct file_lock *fl;
2475 fl = locks_alloc_lock();
2480 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2483 error = flock64_to_posix_lock(filp, fl, flock);
2487 if (cmd == F_OFD_GETLK) {
2489 if (flock->l_pid != 0)
2493 fl->fl_flags |= FL_OFDLCK;
2494 fl->fl_owner = filp;
2497 error = vfs_test_lock(filp, fl);
2501 flock->l_type = fl->fl_type;
2502 if (fl->fl_type != F_UNLCK)
2503 posix_lock_to_flock64(flock, fl);
2506 locks_free_lock(fl);
2510 /* Apply the lock described by l to an open file descriptor.
2511 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2513 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2514 struct flock64 *flock)
2516 struct file_lock *file_lock = locks_alloc_lock();
2517 struct inode *inode = locks_inode(filp);
2521 if (file_lock == NULL)
2524 /* Don't allow mandatory locks on files that may be memory mapped
2527 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2532 error = flock64_to_posix_lock(filp, file_lock, flock);
2536 error = check_fmode_for_setlk(file_lock);
2541 * If the cmd is requesting file-private locks, then set the
2542 * FL_OFDLCK flag and override the owner.
2547 if (flock->l_pid != 0)
2551 file_lock->fl_flags |= FL_OFDLCK;
2552 file_lock->fl_owner = filp;
2556 if (flock->l_pid != 0)
2560 file_lock->fl_flags |= FL_OFDLCK;
2561 file_lock->fl_owner = filp;
2564 file_lock->fl_flags |= FL_SLEEP;
2567 error = do_lock_file_wait(filp, cmd, file_lock);
2570 * Attempt to detect a close/fcntl race and recover by releasing the
2571 * lock that was just acquired. There is no need to do that when we're
2572 * unlocking though, or for OFD locks.
2574 if (!error && file_lock->fl_type != F_UNLCK &&
2575 !(file_lock->fl_flags & FL_OFDLCK)) {
2577 * We need that spin_lock here - it prevents reordering between
2578 * update of i_flctx->flc_posix and check for it done in
2579 * close(). rcu_read_lock() wouldn't do.
2581 spin_lock(¤t->files->file_lock);
2583 spin_unlock(¤t->files->file_lock);
2585 file_lock->fl_type = F_UNLCK;
2586 error = do_lock_file_wait(filp, cmd, file_lock);
2587 WARN_ON_ONCE(error);
2592 locks_free_lock(file_lock);
2595 #endif /* BITS_PER_LONG == 32 */
2598 * This function is called when the file is being removed
2599 * from the task's fd array. POSIX locks belonging to this task
2600 * are deleted at this time.
2602 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2605 struct inode *inode = locks_inode(filp);
2606 struct file_lock lock;
2607 struct file_lock_context *ctx;
2610 * If there are no locks held on this file, we don't need to call
2611 * posix_lock_file(). Another process could be setting a lock on this
2612 * file at the same time, but we wouldn't remove that lock anyway.
2614 ctx = smp_load_acquire(&inode->i_flctx);
2615 if (!ctx || list_empty(&ctx->flc_posix))
2618 locks_init_lock(&lock);
2619 lock.fl_type = F_UNLCK;
2620 lock.fl_flags = FL_POSIX | FL_CLOSE;
2622 lock.fl_end = OFFSET_MAX;
2623 lock.fl_owner = owner;
2624 lock.fl_pid = current->tgid;
2625 lock.fl_file = filp;
2627 lock.fl_lmops = NULL;
2629 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2631 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2632 lock.fl_ops->fl_release_private(&lock);
2633 trace_locks_remove_posix(inode, &lock, error);
2635 EXPORT_SYMBOL(locks_remove_posix);
2637 /* The i_flctx must be valid when calling into here */
2639 locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2641 struct file_lock fl;
2642 struct inode *inode = locks_inode(filp);
2644 if (list_empty(&flctx->flc_flock))
2647 flock_make_lock(filp, LOCK_UN, &fl);
2648 fl.fl_flags |= FL_CLOSE;
2650 if (filp->f_op->flock)
2651 filp->f_op->flock(filp, F_SETLKW, &fl);
2653 flock_lock_inode(inode, &fl);
2655 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2656 fl.fl_ops->fl_release_private(&fl);
2659 /* The i_flctx must be valid when calling into here */
2661 locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2663 struct file_lock *fl, *tmp;
2666 if (list_empty(&ctx->flc_lease))
2669 percpu_down_read(&file_rwsem);
2670 spin_lock(&ctx->flc_lock);
2671 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2672 if (filp == fl->fl_file)
2673 lease_modify(fl, F_UNLCK, &dispose);
2674 spin_unlock(&ctx->flc_lock);
2675 percpu_up_read(&file_rwsem);
2677 locks_dispose_list(&dispose);
2681 * This function is called on the last close of an open file.
2683 void locks_remove_file(struct file *filp)
2685 struct file_lock_context *ctx;
2687 ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2691 /* remove any OFD locks */
2692 locks_remove_posix(filp, filp);
2694 /* remove flock locks */
2695 locks_remove_flock(filp, ctx);
2697 /* remove any leases */
2698 locks_remove_lease(filp, ctx);
2700 spin_lock(&ctx->flc_lock);
2701 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2702 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2703 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2704 spin_unlock(&ctx->flc_lock);
2708 * vfs_cancel_lock - file byte range unblock lock
2709 * @filp: The file to apply the unblock to
2710 * @fl: The lock to be unblocked
2712 * Used by lock managers to cancel blocked requests
2714 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2716 if (filp->f_op->lock)
2717 return filp->f_op->lock(filp, F_CANCELLK, fl);
2720 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2722 #ifdef CONFIG_PROC_FS
2723 #include <linux/proc_fs.h>
2724 #include <linux/seq_file.h>
2726 struct locks_iterator {
2731 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2732 loff_t id, char *pfx)
2734 struct inode *inode = NULL;
2735 unsigned int fl_pid;
2736 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2738 fl_pid = locks_translate_pid(fl, proc_pidns);
2740 * If lock owner is dead (and pid is freed) or not visible in current
2741 * pidns, zero is shown as a pid value. Check lock info from
2742 * init_pid_ns to get saved lock pid value.
2745 if (fl->fl_file != NULL)
2746 inode = locks_inode(fl->fl_file);
2748 seq_printf(f, "%lld:%s ", id, pfx);
2750 if (fl->fl_flags & FL_ACCESS)
2751 seq_puts(f, "ACCESS");
2752 else if (IS_OFDLCK(fl))
2753 seq_puts(f, "OFDLCK");
2755 seq_puts(f, "POSIX ");
2757 seq_printf(f, " %s ",
2758 (inode == NULL) ? "*NOINODE*" :
2759 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2760 } else if (IS_FLOCK(fl)) {
2761 if (fl->fl_type & LOCK_MAND) {
2762 seq_puts(f, "FLOCK MSNFS ");
2764 seq_puts(f, "FLOCK ADVISORY ");
2766 } else if (IS_LEASE(fl)) {
2767 if (fl->fl_flags & FL_DELEG)
2768 seq_puts(f, "DELEG ");
2770 seq_puts(f, "LEASE ");
2772 if (lease_breaking(fl))
2773 seq_puts(f, "BREAKING ");
2774 else if (fl->fl_file)
2775 seq_puts(f, "ACTIVE ");
2777 seq_puts(f, "BREAKER ");
2779 seq_puts(f, "UNKNOWN UNKNOWN ");
2781 if (fl->fl_type & LOCK_MAND) {
2782 seq_printf(f, "%s ",
2783 (fl->fl_type & LOCK_READ)
2784 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2785 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2787 seq_printf(f, "%s ",
2788 (lease_breaking(fl))
2789 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2790 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2793 /* userspace relies on this representation of dev_t */
2794 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2795 MAJOR(inode->i_sb->s_dev),
2796 MINOR(inode->i_sb->s_dev), inode->i_ino);
2798 seq_printf(f, "%d <none>:0 ", fl_pid);
2801 if (fl->fl_end == OFFSET_MAX)
2802 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2804 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2806 seq_puts(f, "0 EOF\n");
2810 static int locks_show(struct seq_file *f, void *v)
2812 struct locks_iterator *iter = f->private;
2813 struct file_lock *fl, *bfl;
2814 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2816 fl = hlist_entry(v, struct file_lock, fl_link);
2818 if (locks_translate_pid(fl, proc_pidns) == 0)
2821 lock_get_status(f, fl, iter->li_pos, "");
2823 list_for_each_entry(bfl, &fl->fl_blocked_requests, fl_blocked_member)
2824 lock_get_status(f, bfl, iter->li_pos, " ->");
2829 static void __show_fd_locks(struct seq_file *f,
2830 struct list_head *head, int *id,
2831 struct file *filp, struct files_struct *files)
2833 struct file_lock *fl;
2835 list_for_each_entry(fl, head, fl_list) {
2837 if (filp != fl->fl_file)
2839 if (fl->fl_owner != files &&
2840 fl->fl_owner != filp)
2844 seq_puts(f, "lock:\t");
2845 lock_get_status(f, fl, *id, "");
2849 void show_fd_locks(struct seq_file *f,
2850 struct file *filp, struct files_struct *files)
2852 struct inode *inode = locks_inode(filp);
2853 struct file_lock_context *ctx;
2856 ctx = smp_load_acquire(&inode->i_flctx);
2860 spin_lock(&ctx->flc_lock);
2861 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2862 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2863 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2864 spin_unlock(&ctx->flc_lock);
2867 static void *locks_start(struct seq_file *f, loff_t *pos)
2868 __acquires(&blocked_lock_lock)
2870 struct locks_iterator *iter = f->private;
2872 iter->li_pos = *pos + 1;
2873 percpu_down_write(&file_rwsem);
2874 spin_lock(&blocked_lock_lock);
2875 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2878 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2880 struct locks_iterator *iter = f->private;
2883 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2886 static void locks_stop(struct seq_file *f, void *v)
2887 __releases(&blocked_lock_lock)
2889 spin_unlock(&blocked_lock_lock);
2890 percpu_up_write(&file_rwsem);
2893 static const struct seq_operations locks_seq_operations = {
2894 .start = locks_start,
2900 static int __init proc_locks_init(void)
2902 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2903 sizeof(struct locks_iterator), NULL);
2906 fs_initcall(proc_locks_init);
2909 static int __init filelock_init(void)
2913 flctx_cache = kmem_cache_create("file_lock_ctx",
2914 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2916 filelock_cache = kmem_cache_create("file_lock_cache",
2917 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2919 for_each_possible_cpu(i) {
2920 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2922 spin_lock_init(&fll->lock);
2923 INIT_HLIST_HEAD(&fll->hlist);
2928 core_initcall(filelock_init);