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 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
662 return fl2->fl_lmops == fl1->fl_lmops &&
663 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
664 return fl1->fl_owner == fl2->fl_owner;
667 /* Must be called with the flc_lock held! */
668 static void locks_insert_global_locks(struct file_lock *fl)
670 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
672 percpu_rwsem_assert_held(&file_rwsem);
674 spin_lock(&fll->lock);
675 fl->fl_link_cpu = smp_processor_id();
676 hlist_add_head(&fl->fl_link, &fll->hlist);
677 spin_unlock(&fll->lock);
680 /* Must be called with the flc_lock held! */
681 static void locks_delete_global_locks(struct file_lock *fl)
683 struct file_lock_list_struct *fll;
685 percpu_rwsem_assert_held(&file_rwsem);
688 * Avoid taking lock if already unhashed. This is safe since this check
689 * is done while holding the flc_lock, and new insertions into the list
690 * also require that it be held.
692 if (hlist_unhashed(&fl->fl_link))
695 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
696 spin_lock(&fll->lock);
697 hlist_del_init(&fl->fl_link);
698 spin_unlock(&fll->lock);
702 posix_owner_key(struct file_lock *fl)
704 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
705 return fl->fl_lmops->lm_owner_key(fl);
706 return (unsigned long)fl->fl_owner;
709 static void locks_insert_global_blocked(struct file_lock *waiter)
711 lockdep_assert_held(&blocked_lock_lock);
713 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
716 static void locks_delete_global_blocked(struct file_lock *waiter)
718 lockdep_assert_held(&blocked_lock_lock);
720 hash_del(&waiter->fl_link);
723 /* Remove waiter from blocker's block list.
724 * When blocker ends up pointing to itself then the list is empty.
726 * Must be called with blocked_lock_lock held.
728 static void __locks_delete_block(struct file_lock *waiter)
730 locks_delete_global_blocked(waiter);
731 list_del_init(&waiter->fl_blocked_member);
732 waiter->fl_blocker = NULL;
735 static void __locks_wake_up_blocks(struct file_lock *blocker)
737 while (!list_empty(&blocker->fl_blocked_requests)) {
738 struct file_lock *waiter;
740 waiter = list_first_entry(&blocker->fl_blocked_requests,
741 struct file_lock, fl_blocked_member);
742 __locks_delete_block(waiter);
743 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
744 waiter->fl_lmops->lm_notify(waiter);
746 wake_up(&waiter->fl_wait);
751 * locks_delete_lock - stop waiting for a file lock
752 * @waiter: the lock which was waiting
754 * lockd/nfsd need to disconnect the lock while working on it.
756 int locks_delete_block(struct file_lock *waiter)
758 int status = -ENOENT;
761 * If fl_blocker is NULL, it won't be set again as this thread
762 * "owns" the lock and is the only one that might try to claim
763 * the lock. So it is safe to test fl_blocker locklessly.
764 * Also if fl_blocker is NULL, this waiter is not listed on
765 * fl_blocked_requests for some lock, so no other request can
766 * be added to the list of fl_blocked_requests for this
767 * request. So if fl_blocker is NULL, it is safe to
768 * locklessly check if fl_blocked_requests is empty. If both
769 * of these checks succeed, there is no need to take the lock.
771 if (waiter->fl_blocker == NULL &&
772 list_empty(&waiter->fl_blocked_requests))
774 spin_lock(&blocked_lock_lock);
775 if (waiter->fl_blocker)
777 __locks_wake_up_blocks(waiter);
778 __locks_delete_block(waiter);
779 spin_unlock(&blocked_lock_lock);
782 EXPORT_SYMBOL(locks_delete_block);
784 /* Insert waiter into blocker's block list.
785 * We use a circular list so that processes can be easily woken up in
786 * the order they blocked. The documentation doesn't require this but
787 * it seems like the reasonable thing to do.
789 * Must be called with both the flc_lock and blocked_lock_lock held. The
790 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
791 * but by ensuring that the flc_lock is also held on insertions we can avoid
792 * taking the blocked_lock_lock in some cases when we see that the
793 * fl_blocked_requests list is empty.
795 * Rather than just adding to the list, we check for conflicts with any existing
796 * waiters, and add beneath any waiter that blocks the new waiter.
797 * Thus wakeups don't happen until needed.
799 static void __locks_insert_block(struct file_lock *blocker,
800 struct file_lock *waiter,
801 bool conflict(struct file_lock *,
804 struct file_lock *fl;
805 BUG_ON(!list_empty(&waiter->fl_blocked_member));
808 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
809 if (conflict(fl, waiter)) {
813 waiter->fl_blocker = blocker;
814 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
815 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
816 locks_insert_global_blocked(waiter);
818 /* The requests in waiter->fl_blocked are known to conflict with
819 * waiter, but might not conflict with blocker, or the requests
820 * and lock which block it. So they all need to be woken.
822 __locks_wake_up_blocks(waiter);
825 /* Must be called with flc_lock held. */
826 static void locks_insert_block(struct file_lock *blocker,
827 struct file_lock *waiter,
828 bool conflict(struct file_lock *,
831 spin_lock(&blocked_lock_lock);
832 __locks_insert_block(blocker, waiter, conflict);
833 spin_unlock(&blocked_lock_lock);
837 * Wake up processes blocked waiting for blocker.
839 * Must be called with the inode->flc_lock held!
841 static void locks_wake_up_blocks(struct file_lock *blocker)
844 * Avoid taking global lock if list is empty. This is safe since new
845 * blocked requests are only added to the list under the flc_lock, and
846 * the flc_lock is always held here. Note that removal from the
847 * fl_blocked_requests list does not require the flc_lock, so we must
848 * recheck list_empty() after acquiring the blocked_lock_lock.
850 if (list_empty(&blocker->fl_blocked_requests))
853 spin_lock(&blocked_lock_lock);
854 __locks_wake_up_blocks(blocker);
855 spin_unlock(&blocked_lock_lock);
859 locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
861 list_add_tail(&fl->fl_list, before);
862 locks_insert_global_locks(fl);
866 locks_unlink_lock_ctx(struct file_lock *fl)
868 locks_delete_global_locks(fl);
869 list_del_init(&fl->fl_list);
870 locks_wake_up_blocks(fl);
874 locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
876 locks_unlink_lock_ctx(fl);
878 list_add(&fl->fl_list, dispose);
883 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
884 * checks for shared/exclusive status of overlapping locks.
886 static bool locks_conflict(struct file_lock *caller_fl,
887 struct file_lock *sys_fl)
889 if (sys_fl->fl_type == F_WRLCK)
891 if (caller_fl->fl_type == F_WRLCK)
896 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
897 * checking before calling the locks_conflict().
899 static bool posix_locks_conflict(struct file_lock *caller_fl,
900 struct file_lock *sys_fl)
902 /* POSIX locks owned by the same process do not conflict with
905 if (posix_same_owner(caller_fl, sys_fl))
908 /* Check whether they overlap */
909 if (!locks_overlap(caller_fl, sys_fl))
912 return locks_conflict(caller_fl, sys_fl);
915 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
916 * checking before calling the locks_conflict().
918 static bool flock_locks_conflict(struct file_lock *caller_fl,
919 struct file_lock *sys_fl)
921 /* FLOCK locks referring to the same filp do not conflict with
924 if (caller_fl->fl_file == sys_fl->fl_file)
926 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
929 return locks_conflict(caller_fl, sys_fl);
933 posix_test_lock(struct file *filp, struct file_lock *fl)
935 struct file_lock *cfl;
936 struct file_lock_context *ctx;
937 struct inode *inode = locks_inode(filp);
939 ctx = smp_load_acquire(&inode->i_flctx);
940 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
941 fl->fl_type = F_UNLCK;
945 spin_lock(&ctx->flc_lock);
946 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
947 if (posix_locks_conflict(fl, cfl)) {
948 locks_copy_conflock(fl, cfl);
952 fl->fl_type = F_UNLCK;
954 spin_unlock(&ctx->flc_lock);
957 EXPORT_SYMBOL(posix_test_lock);
960 * Deadlock detection:
962 * We attempt to detect deadlocks that are due purely to posix file
965 * We assume that a task can be waiting for at most one lock at a time.
966 * So for any acquired lock, the process holding that lock may be
967 * waiting on at most one other lock. That lock in turns may be held by
968 * someone waiting for at most one other lock. Given a requested lock
969 * caller_fl which is about to wait for a conflicting lock block_fl, we
970 * follow this chain of waiters to ensure we are not about to create a
973 * Since we do this before we ever put a process to sleep on a lock, we
974 * are ensured that there is never a cycle; that is what guarantees that
975 * the while() loop in posix_locks_deadlock() eventually completes.
977 * Note: the above assumption may not be true when handling lock
978 * requests from a broken NFS client. It may also fail in the presence
979 * of tasks (such as posix threads) sharing the same open file table.
980 * To handle those cases, we just bail out after a few iterations.
982 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
983 * Because the owner is not even nominally tied to a thread of
984 * execution, the deadlock detection below can't reasonably work well. Just
987 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
988 * locks that just checks for the case where two tasks are attempting to
989 * upgrade from read to write locks on the same inode.
992 #define MAX_DEADLK_ITERATIONS 10
994 /* Find a lock that the owner of the given block_fl is blocking on. */
995 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
997 struct file_lock *fl;
999 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
1000 if (posix_same_owner(fl, block_fl)) {
1001 while (fl->fl_blocker)
1002 fl = fl->fl_blocker;
1009 /* Must be called with the blocked_lock_lock held! */
1010 static int posix_locks_deadlock(struct file_lock *caller_fl,
1011 struct file_lock *block_fl)
1015 lockdep_assert_held(&blocked_lock_lock);
1018 * This deadlock detector can't reasonably detect deadlocks with
1019 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1021 if (IS_OFDLCK(caller_fl))
1024 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
1025 if (i++ > MAX_DEADLK_ITERATIONS)
1027 if (posix_same_owner(caller_fl, block_fl))
1033 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1034 * after any leases, but before any posix locks.
1036 * Note that if called with an FL_EXISTS argument, the caller may determine
1037 * whether or not a lock was successfully freed by testing the return
1038 * value for -ENOENT.
1040 static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1042 struct file_lock *new_fl = NULL;
1043 struct file_lock *fl;
1044 struct file_lock_context *ctx;
1049 ctx = locks_get_lock_context(inode, request->fl_type);
1051 if (request->fl_type != F_UNLCK)
1053 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1056 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1057 new_fl = locks_alloc_lock();
1062 percpu_down_read(&file_rwsem);
1063 spin_lock(&ctx->flc_lock);
1064 if (request->fl_flags & FL_ACCESS)
1067 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1068 if (request->fl_file != fl->fl_file)
1070 if (request->fl_type == fl->fl_type)
1073 locks_delete_lock_ctx(fl, &dispose);
1077 if (request->fl_type == F_UNLCK) {
1078 if ((request->fl_flags & FL_EXISTS) && !found)
1084 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1085 if (!flock_locks_conflict(request, fl))
1088 if (!(request->fl_flags & FL_SLEEP))
1090 error = FILE_LOCK_DEFERRED;
1091 locks_insert_block(fl, request, flock_locks_conflict);
1094 if (request->fl_flags & FL_ACCESS)
1096 locks_copy_lock(new_fl, request);
1097 locks_move_blocks(new_fl, request);
1098 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1103 spin_unlock(&ctx->flc_lock);
1104 percpu_up_read(&file_rwsem);
1106 locks_free_lock(new_fl);
1107 locks_dispose_list(&dispose);
1108 trace_flock_lock_inode(inode, request, error);
1112 static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1113 struct file_lock *conflock)
1115 struct file_lock *fl, *tmp;
1116 struct file_lock *new_fl = NULL;
1117 struct file_lock *new_fl2 = NULL;
1118 struct file_lock *left = NULL;
1119 struct file_lock *right = NULL;
1120 struct file_lock_context *ctx;
1125 ctx = locks_get_lock_context(inode, request->fl_type);
1127 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1130 * We may need two file_lock structures for this operation,
1131 * so we get them in advance to avoid races.
1133 * In some cases we can be sure, that no new locks will be needed
1135 if (!(request->fl_flags & FL_ACCESS) &&
1136 (request->fl_type != F_UNLCK ||
1137 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1138 new_fl = locks_alloc_lock();
1139 new_fl2 = locks_alloc_lock();
1142 percpu_down_read(&file_rwsem);
1143 spin_lock(&ctx->flc_lock);
1145 * New lock request. Walk all POSIX locks and look for conflicts. If
1146 * there are any, either return error or put the request on the
1147 * blocker's list of waiters and the global blocked_hash.
1149 if (request->fl_type != F_UNLCK) {
1150 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1151 if (!posix_locks_conflict(request, fl))
1154 locks_copy_conflock(conflock, fl);
1156 if (!(request->fl_flags & FL_SLEEP))
1159 * Deadlock detection and insertion into the blocked
1160 * locks list must be done while holding the same lock!
1163 spin_lock(&blocked_lock_lock);
1165 * Ensure that we don't find any locks blocked on this
1166 * request during deadlock detection.
1168 __locks_wake_up_blocks(request);
1169 if (likely(!posix_locks_deadlock(request, fl))) {
1170 error = FILE_LOCK_DEFERRED;
1171 __locks_insert_block(fl, request,
1172 posix_locks_conflict);
1174 spin_unlock(&blocked_lock_lock);
1179 /* If we're just looking for a conflict, we're done. */
1181 if (request->fl_flags & FL_ACCESS)
1184 /* Find the first old lock with the same owner as the new lock */
1185 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1186 if (posix_same_owner(request, fl))
1190 /* Process locks with this owner. */
1191 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1192 if (!posix_same_owner(request, fl))
1195 /* Detect adjacent or overlapping regions (if same lock type) */
1196 if (request->fl_type == fl->fl_type) {
1197 /* In all comparisons of start vs end, use
1198 * "start - 1" rather than "end + 1". If end
1199 * is OFFSET_MAX, end + 1 will become negative.
1201 if (fl->fl_end < request->fl_start - 1)
1203 /* If the next lock in the list has entirely bigger
1204 * addresses than the new one, insert the lock here.
1206 if (fl->fl_start - 1 > request->fl_end)
1209 /* If we come here, the new and old lock are of the
1210 * same type and adjacent or overlapping. Make one
1211 * lock yielding from the lower start address of both
1212 * locks to the higher end address.
1214 if (fl->fl_start > request->fl_start)
1215 fl->fl_start = request->fl_start;
1217 request->fl_start = fl->fl_start;
1218 if (fl->fl_end < request->fl_end)
1219 fl->fl_end = request->fl_end;
1221 request->fl_end = fl->fl_end;
1223 locks_delete_lock_ctx(fl, &dispose);
1229 /* Processing for different lock types is a bit
1232 if (fl->fl_end < request->fl_start)
1234 if (fl->fl_start > request->fl_end)
1236 if (request->fl_type == F_UNLCK)
1238 if (fl->fl_start < request->fl_start)
1240 /* If the next lock in the list has a higher end
1241 * address than the new one, insert the new one here.
1243 if (fl->fl_end > request->fl_end) {
1247 if (fl->fl_start >= request->fl_start) {
1248 /* The new lock completely replaces an old
1249 * one (This may happen several times).
1252 locks_delete_lock_ctx(fl, &dispose);
1256 * Replace the old lock with new_fl, and
1257 * remove the old one. It's safe to do the
1258 * insert here since we know that we won't be
1259 * using new_fl later, and that the lock is
1260 * just replacing an existing lock.
1265 locks_copy_lock(new_fl, request);
1268 locks_insert_lock_ctx(request, &fl->fl_list);
1269 locks_delete_lock_ctx(fl, &dispose);
1276 * The above code only modifies existing locks in case of merging or
1277 * replacing. If new lock(s) need to be inserted all modifications are
1278 * done below this, so it's safe yet to bail out.
1280 error = -ENOLCK; /* "no luck" */
1281 if (right && left == right && !new_fl2)
1286 if (request->fl_type == F_UNLCK) {
1287 if (request->fl_flags & FL_EXISTS)
1296 locks_copy_lock(new_fl, request);
1297 locks_move_blocks(new_fl, request);
1298 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1303 if (left == right) {
1304 /* The new lock breaks the old one in two pieces,
1305 * so we have to use the second new lock.
1309 locks_copy_lock(left, right);
1310 locks_insert_lock_ctx(left, &fl->fl_list);
1312 right->fl_start = request->fl_end + 1;
1313 locks_wake_up_blocks(right);
1316 left->fl_end = request->fl_start - 1;
1317 locks_wake_up_blocks(left);
1320 spin_unlock(&ctx->flc_lock);
1321 percpu_up_read(&file_rwsem);
1323 * Free any unused locks.
1326 locks_free_lock(new_fl);
1328 locks_free_lock(new_fl2);
1329 locks_dispose_list(&dispose);
1330 trace_posix_lock_inode(inode, request, error);
1336 * posix_lock_file - Apply a POSIX-style lock to a file
1337 * @filp: The file to apply the lock to
1338 * @fl: The lock to be applied
1339 * @conflock: Place to return a copy of the conflicting lock, if found.
1341 * Add a POSIX style lock to a file.
1342 * We merge adjacent & overlapping locks whenever possible.
1343 * POSIX locks are sorted by owner task, then by starting address
1345 * Note that if called with an FL_EXISTS argument, the caller may determine
1346 * whether or not a lock was successfully freed by testing the return
1347 * value for -ENOENT.
1349 int posix_lock_file(struct file *filp, struct file_lock *fl,
1350 struct file_lock *conflock)
1352 return posix_lock_inode(locks_inode(filp), fl, conflock);
1354 EXPORT_SYMBOL(posix_lock_file);
1357 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1358 * @inode: inode of file to which lock request should be applied
1359 * @fl: The lock to be applied
1361 * Apply a POSIX style lock request to an inode.
1363 static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1368 error = posix_lock_inode(inode, fl, NULL);
1369 if (error != FILE_LOCK_DEFERRED)
1371 error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
1375 locks_delete_block(fl);
1379 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1381 * locks_mandatory_locked - Check for an active lock
1382 * @file: the file to check
1384 * Searches the inode's list of locks to find any POSIX locks which conflict.
1385 * This function is called from locks_verify_locked() only.
1387 int locks_mandatory_locked(struct file *file)
1390 struct inode *inode = locks_inode(file);
1391 struct file_lock_context *ctx;
1392 struct file_lock *fl;
1394 ctx = smp_load_acquire(&inode->i_flctx);
1395 if (!ctx || list_empty_careful(&ctx->flc_posix))
1399 * Search the lock list for this inode for any POSIX locks.
1401 spin_lock(&ctx->flc_lock);
1403 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1404 if (fl->fl_owner != current->files &&
1405 fl->fl_owner != file) {
1410 spin_unlock(&ctx->flc_lock);
1415 * locks_mandatory_area - Check for a conflicting lock
1416 * @inode: the file to check
1417 * @filp: how the file was opened (if it was)
1418 * @start: first byte in the file to check
1419 * @end: lastbyte in the file to check
1420 * @type: %F_WRLCK for a write lock, else %F_RDLCK
1422 * Searches the inode's list of locks to find any POSIX locks which conflict.
1424 int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1425 loff_t end, unsigned char type)
1427 struct file_lock fl;
1431 locks_init_lock(&fl);
1432 fl.fl_pid = current->tgid;
1434 fl.fl_flags = FL_POSIX | FL_ACCESS;
1435 if (filp && !(filp->f_flags & O_NONBLOCK))
1438 fl.fl_start = start;
1444 fl.fl_flags &= ~FL_SLEEP;
1445 error = posix_lock_inode(inode, &fl, NULL);
1451 fl.fl_flags |= FL_SLEEP;
1452 fl.fl_owner = current->files;
1453 error = posix_lock_inode(inode, &fl, NULL);
1454 if (error != FILE_LOCK_DEFERRED)
1456 error = wait_event_interruptible(fl.fl_wait, !fl.fl_blocker);
1459 * If we've been sleeping someone might have
1460 * changed the permissions behind our back.
1462 if (__mandatory_lock(inode))
1468 locks_delete_block(&fl);
1472 EXPORT_SYMBOL(locks_mandatory_area);
1473 #endif /* CONFIG_MANDATORY_FILE_LOCKING */
1475 static void lease_clear_pending(struct file_lock *fl, int arg)
1479 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1482 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1486 /* We already had a lease on this file; just change its type */
1487 int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1489 int error = assign_type(fl, arg);
1493 lease_clear_pending(fl, arg);
1494 locks_wake_up_blocks(fl);
1495 if (arg == F_UNLCK) {
1496 struct file *filp = fl->fl_file;
1499 filp->f_owner.signum = 0;
1500 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1501 if (fl->fl_fasync != NULL) {
1502 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1503 fl->fl_fasync = NULL;
1505 locks_delete_lock_ctx(fl, dispose);
1509 EXPORT_SYMBOL(lease_modify);
1511 static bool past_time(unsigned long then)
1514 /* 0 is a special value meaning "this never expires": */
1516 return time_after(jiffies, then);
1519 static void time_out_leases(struct inode *inode, struct list_head *dispose)
1521 struct file_lock_context *ctx = inode->i_flctx;
1522 struct file_lock *fl, *tmp;
1524 lockdep_assert_held(&ctx->flc_lock);
1526 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1527 trace_time_out_leases(inode, fl);
1528 if (past_time(fl->fl_downgrade_time))
1529 lease_modify(fl, F_RDLCK, dispose);
1530 if (past_time(fl->fl_break_time))
1531 lease_modify(fl, F_UNLCK, dispose);
1535 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1539 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1543 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1548 rc = locks_conflict(breaker, lease);
1550 trace_leases_conflict(rc, lease, breaker);
1555 any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1557 struct file_lock_context *ctx = inode->i_flctx;
1558 struct file_lock *fl;
1560 lockdep_assert_held(&ctx->flc_lock);
1562 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1563 if (leases_conflict(fl, breaker))
1570 * __break_lease - revoke all outstanding leases on file
1571 * @inode: the inode of the file to return
1572 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1574 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1577 * break_lease (inlined for speed) has checked there already is at least
1578 * some kind of lock (maybe a lease) on this file. Leases are broken on
1579 * a call to open() or truncate(). This function can sleep unless you
1580 * specified %O_NONBLOCK to your open().
1582 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1585 struct file_lock_context *ctx;
1586 struct file_lock *new_fl, *fl, *tmp;
1587 unsigned long break_time;
1588 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1591 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1593 return PTR_ERR(new_fl);
1594 new_fl->fl_flags = type;
1596 /* typically we will check that ctx is non-NULL before calling */
1597 ctx = smp_load_acquire(&inode->i_flctx);
1603 percpu_down_read(&file_rwsem);
1604 spin_lock(&ctx->flc_lock);
1606 time_out_leases(inode, &dispose);
1608 if (!any_leases_conflict(inode, new_fl))
1612 if (lease_break_time > 0) {
1613 break_time = jiffies + lease_break_time * HZ;
1614 if (break_time == 0)
1615 break_time++; /* so that 0 means no break time */
1618 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1619 if (!leases_conflict(fl, new_fl))
1622 if (fl->fl_flags & FL_UNLOCK_PENDING)
1624 fl->fl_flags |= FL_UNLOCK_PENDING;
1625 fl->fl_break_time = break_time;
1627 if (lease_breaking(fl))
1629 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1630 fl->fl_downgrade_time = break_time;
1632 if (fl->fl_lmops->lm_break(fl))
1633 locks_delete_lock_ctx(fl, &dispose);
1636 if (list_empty(&ctx->flc_lease))
1639 if (mode & O_NONBLOCK) {
1640 trace_break_lease_noblock(inode, new_fl);
1641 error = -EWOULDBLOCK;
1646 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1647 break_time = fl->fl_break_time;
1648 if (break_time != 0)
1649 break_time -= jiffies;
1650 if (break_time == 0)
1652 locks_insert_block(fl, new_fl, leases_conflict);
1653 trace_break_lease_block(inode, new_fl);
1654 spin_unlock(&ctx->flc_lock);
1655 percpu_up_read(&file_rwsem);
1657 locks_dispose_list(&dispose);
1658 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1659 !new_fl->fl_blocker, break_time);
1661 percpu_down_read(&file_rwsem);
1662 spin_lock(&ctx->flc_lock);
1663 trace_break_lease_unblock(inode, new_fl);
1664 locks_delete_block(new_fl);
1667 * Wait for the next conflicting lease that has not been
1671 time_out_leases(inode, &dispose);
1672 if (any_leases_conflict(inode, new_fl))
1677 spin_unlock(&ctx->flc_lock);
1678 percpu_up_read(&file_rwsem);
1679 locks_dispose_list(&dispose);
1680 locks_free_lock(new_fl);
1683 EXPORT_SYMBOL(__break_lease);
1686 * lease_get_mtime - update modified time of an inode with exclusive lease
1688 * @time: pointer to a timespec which contains the last modified time
1690 * This is to force NFS clients to flush their caches for files with
1691 * exclusive leases. The justification is that if someone has an
1692 * exclusive lease, then they could be modifying it.
1694 void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1696 bool has_lease = false;
1697 struct file_lock_context *ctx;
1698 struct file_lock *fl;
1700 ctx = smp_load_acquire(&inode->i_flctx);
1701 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1702 spin_lock(&ctx->flc_lock);
1703 fl = list_first_entry_or_null(&ctx->flc_lease,
1704 struct file_lock, fl_list);
1705 if (fl && (fl->fl_type == F_WRLCK))
1707 spin_unlock(&ctx->flc_lock);
1711 *time = current_time(inode);
1713 EXPORT_SYMBOL(lease_get_mtime);
1716 * fcntl_getlease - Enquire what lease is currently active
1719 * The value returned by this function will be one of
1720 * (if no lease break is pending):
1722 * %F_RDLCK to indicate a shared lease is held.
1724 * %F_WRLCK to indicate an exclusive lease is held.
1726 * %F_UNLCK to indicate no lease is held.
1728 * (if a lease break is pending):
1730 * %F_RDLCK to indicate an exclusive lease needs to be
1731 * changed to a shared lease (or removed).
1733 * %F_UNLCK to indicate the lease needs to be removed.
1735 * XXX: sfr & willy disagree over whether F_INPROGRESS
1736 * should be returned to userspace.
1738 int fcntl_getlease(struct file *filp)
1740 struct file_lock *fl;
1741 struct inode *inode = locks_inode(filp);
1742 struct file_lock_context *ctx;
1746 ctx = smp_load_acquire(&inode->i_flctx);
1747 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1748 percpu_down_read(&file_rwsem);
1749 spin_lock(&ctx->flc_lock);
1750 time_out_leases(inode, &dispose);
1751 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1752 if (fl->fl_file != filp)
1754 type = target_leasetype(fl);
1757 spin_unlock(&ctx->flc_lock);
1758 percpu_up_read(&file_rwsem);
1760 locks_dispose_list(&dispose);
1766 * check_conflicting_open - see if the given file points to an inode that has
1767 * an existing open that would conflict with the
1769 * @filp: file to check
1770 * @arg: type of lease that we're trying to acquire
1771 * @flags: current lock flags
1773 * Check to see if there's an existing open fd on this file that would
1774 * conflict with the lease we're trying to set.
1777 check_conflicting_open(struct file *filp, const long arg, int flags)
1779 struct inode *inode = locks_inode(filp);
1780 int self_wcount = 0, self_rcount = 0;
1782 if (flags & FL_LAYOUT)
1786 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1787 else if (arg != F_WRLCK)
1791 * Make sure that only read/write count is from lease requestor.
1792 * Note that this will result in denying write leases when i_writecount
1793 * is negative, which is what we want. (We shouldn't grant write leases
1794 * on files open for execution.)
1796 if (filp->f_mode & FMODE_WRITE)
1798 else if (filp->f_mode & FMODE_READ)
1801 if (atomic_read(&inode->i_writecount) != self_wcount ||
1802 atomic_read(&inode->i_readcount) != self_rcount)
1809 generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1811 struct file_lock *fl, *my_fl = NULL, *lease;
1812 struct inode *inode = locks_inode(filp);
1813 struct file_lock_context *ctx;
1814 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1819 trace_generic_add_lease(inode, lease);
1821 /* Note that arg is never F_UNLCK here */
1822 ctx = locks_get_lock_context(inode, arg);
1827 * In the delegation case we need mutual exclusion with
1828 * a number of operations that take the i_mutex. We trylock
1829 * because delegations are an optional optimization, and if
1830 * there's some chance of a conflict--we'd rather not
1831 * bother, maybe that's a sign this just isn't a good file to
1832 * hand out a delegation on.
1834 if (is_deleg && !inode_trylock(inode))
1837 if (is_deleg && arg == F_WRLCK) {
1838 /* Write delegations are not currently supported: */
1839 inode_unlock(inode);
1844 percpu_down_read(&file_rwsem);
1845 spin_lock(&ctx->flc_lock);
1846 time_out_leases(inode, &dispose);
1847 error = check_conflicting_open(filp, arg, lease->fl_flags);
1852 * At this point, we know that if there is an exclusive
1853 * lease on this file, then we hold it on this filp
1854 * (otherwise our open of this file would have blocked).
1855 * And if we are trying to acquire an exclusive lease,
1856 * then the file is not open by anyone (including us)
1857 * except for this filp.
1860 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1861 if (fl->fl_file == filp &&
1862 fl->fl_owner == lease->fl_owner) {
1868 * No exclusive leases if someone else has a lease on
1874 * Modifying our existing lease is OK, but no getting a
1875 * new lease if someone else is opening for write:
1877 if (fl->fl_flags & FL_UNLOCK_PENDING)
1881 if (my_fl != NULL) {
1883 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1893 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1895 * The check in break_lease() is lockless. It's possible for another
1896 * open to race in after we did the earlier check for a conflicting
1897 * open but before the lease was inserted. Check again for a
1898 * conflicting open and cancel the lease if there is one.
1900 * We also add a barrier here to ensure that the insertion of the lock
1901 * precedes these checks.
1904 error = check_conflicting_open(filp, arg, lease->fl_flags);
1906 locks_unlink_lock_ctx(lease);
1911 if (lease->fl_lmops->lm_setup)
1912 lease->fl_lmops->lm_setup(lease, priv);
1914 spin_unlock(&ctx->flc_lock);
1915 percpu_up_read(&file_rwsem);
1916 locks_dispose_list(&dispose);
1918 inode_unlock(inode);
1919 if (!error && !my_fl)
1924 static int generic_delete_lease(struct file *filp, void *owner)
1926 int error = -EAGAIN;
1927 struct file_lock *fl, *victim = NULL;
1928 struct inode *inode = locks_inode(filp);
1929 struct file_lock_context *ctx;
1932 ctx = smp_load_acquire(&inode->i_flctx);
1934 trace_generic_delete_lease(inode, NULL);
1938 percpu_down_read(&file_rwsem);
1939 spin_lock(&ctx->flc_lock);
1940 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1941 if (fl->fl_file == filp &&
1942 fl->fl_owner == owner) {
1947 trace_generic_delete_lease(inode, victim);
1949 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1950 spin_unlock(&ctx->flc_lock);
1951 percpu_up_read(&file_rwsem);
1952 locks_dispose_list(&dispose);
1957 * generic_setlease - sets a lease on an open file
1958 * @filp: file pointer
1959 * @arg: type of lease to obtain
1960 * @flp: input - file_lock to use, output - file_lock inserted
1961 * @priv: private data for lm_setup (may be NULL if lm_setup
1962 * doesn't require it)
1964 * The (input) flp->fl_lmops->lm_break function is required
1967 int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1970 struct inode *inode = locks_inode(filp);
1973 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1975 if (!S_ISREG(inode->i_mode))
1977 error = security_file_lock(filp, arg);
1983 return generic_delete_lease(filp, *priv);
1986 if (!(*flp)->fl_lmops->lm_break) {
1991 return generic_add_lease(filp, arg, flp, priv);
1996 EXPORT_SYMBOL(generic_setlease);
1999 * vfs_setlease - sets a lease on an open file
2000 * @filp: file pointer
2001 * @arg: type of lease to obtain
2002 * @lease: file_lock to use when adding a lease
2003 * @priv: private info for lm_setup when adding a lease (may be
2004 * NULL if lm_setup doesn't require it)
2006 * Call this to establish a lease on the file. The "lease" argument is not
2007 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2008 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2009 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2012 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2013 * may be NULL if the lm_setup operation doesn't require it.
2016 vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
2018 if (filp->f_op->setlease)
2019 return filp->f_op->setlease(filp, arg, lease, priv);
2021 return generic_setlease(filp, arg, lease, priv);
2023 EXPORT_SYMBOL_GPL(vfs_setlease);
2025 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
2027 struct file_lock *fl;
2028 struct fasync_struct *new;
2031 fl = lease_alloc(filp, arg);
2035 new = fasync_alloc();
2037 locks_free_lock(fl);
2042 error = vfs_setlease(filp, arg, &fl, (void **)&new);
2044 locks_free_lock(fl);
2051 * fcntl_setlease - sets a lease on an open file
2052 * @fd: open file descriptor
2053 * @filp: file pointer
2054 * @arg: type of lease to obtain
2056 * Call this fcntl to establish a lease on the file.
2057 * Note that you also need to call %F_SETSIG to
2058 * receive a signal when the lease is broken.
2060 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2063 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2064 return do_fcntl_add_lease(fd, filp, arg);
2068 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2069 * @inode: inode of the file to apply to
2070 * @fl: The lock to be applied
2072 * Apply a FLOCK style lock request to an inode.
2074 static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2079 error = flock_lock_inode(inode, fl);
2080 if (error != FILE_LOCK_DEFERRED)
2082 error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
2086 locks_delete_block(fl);
2091 * locks_lock_inode_wait - Apply a lock to an inode
2092 * @inode: inode of the file to apply to
2093 * @fl: The lock to be applied
2095 * Apply a POSIX or FLOCK style lock request to an inode.
2097 int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2100 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2102 res = posix_lock_inode_wait(inode, fl);
2105 res = flock_lock_inode_wait(inode, fl);
2112 EXPORT_SYMBOL(locks_lock_inode_wait);
2115 * sys_flock: - flock() system call.
2116 * @fd: the file descriptor to lock.
2117 * @cmd: the type of lock to apply.
2119 * Apply a %FL_FLOCK style lock to an open file descriptor.
2120 * The @cmd can be one of:
2122 * - %LOCK_SH -- a shared lock.
2123 * - %LOCK_EX -- an exclusive lock.
2124 * - %LOCK_UN -- remove an existing lock.
2125 * - %LOCK_MAND -- a 'mandatory' flock.
2126 * This exists to emulate Windows Share Modes.
2128 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
2129 * processes read and write access respectively.
2131 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2133 struct fd f = fdget(fd);
2134 struct file_lock *lock;
2135 int can_sleep, unlock;
2142 can_sleep = !(cmd & LOCK_NB);
2144 unlock = (cmd == LOCK_UN);
2146 if (!unlock && !(cmd & LOCK_MAND) &&
2147 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2150 lock = flock_make_lock(f.file, cmd, NULL);
2152 error = PTR_ERR(lock);
2157 lock->fl_flags |= FL_SLEEP;
2159 error = security_file_lock(f.file, lock->fl_type);
2163 if (f.file->f_op->flock)
2164 error = f.file->f_op->flock(f.file,
2165 (can_sleep) ? F_SETLKW : F_SETLK,
2168 error = locks_lock_file_wait(f.file, lock);
2171 locks_free_lock(lock);
2180 * vfs_test_lock - test file byte range lock
2181 * @filp: The file to test lock for
2182 * @fl: The lock to test; also used to hold result
2184 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2185 * setting conf->fl_type to something other than F_UNLCK.
2187 int vfs_test_lock(struct file *filp, struct file_lock *fl)
2189 if (filp->f_op->lock)
2190 return filp->f_op->lock(filp, F_GETLK, fl);
2191 posix_test_lock(filp, fl);
2194 EXPORT_SYMBOL_GPL(vfs_test_lock);
2197 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2198 * @fl: The file_lock who's fl_pid should be translated
2199 * @ns: The namespace into which the pid should be translated
2201 * Used to tranlate a fl_pid into a namespace virtual pid number
2203 static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2210 if (IS_REMOTELCK(fl))
2213 * If the flock owner process is dead and its pid has been already
2214 * freed, the translation below won't work, but we still want to show
2215 * flock owner pid number in init pidns.
2217 if (ns == &init_pid_ns)
2218 return (pid_t)fl->fl_pid;
2221 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2222 vnr = pid_nr_ns(pid, ns);
2227 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2229 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2230 #if BITS_PER_LONG == 32
2232 * Make sure we can represent the posix lock via
2233 * legacy 32bit flock.
2235 if (fl->fl_start > OFFT_OFFSET_MAX)
2237 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2240 flock->l_start = fl->fl_start;
2241 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2242 fl->fl_end - fl->fl_start + 1;
2243 flock->l_whence = 0;
2244 flock->l_type = fl->fl_type;
2248 #if BITS_PER_LONG == 32
2249 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2251 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2252 flock->l_start = fl->fl_start;
2253 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2254 fl->fl_end - fl->fl_start + 1;
2255 flock->l_whence = 0;
2256 flock->l_type = fl->fl_type;
2260 /* Report the first existing lock that would conflict with l.
2261 * This implements the F_GETLK command of fcntl().
2263 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2265 struct file_lock *fl;
2268 fl = locks_alloc_lock();
2272 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2275 error = flock_to_posix_lock(filp, fl, flock);
2279 if (cmd == F_OFD_GETLK) {
2281 if (flock->l_pid != 0)
2285 fl->fl_flags |= FL_OFDLCK;
2286 fl->fl_owner = filp;
2289 error = vfs_test_lock(filp, fl);
2293 flock->l_type = fl->fl_type;
2294 if (fl->fl_type != F_UNLCK) {
2295 error = posix_lock_to_flock(flock, fl);
2300 locks_free_lock(fl);
2305 * vfs_lock_file - file byte range lock
2306 * @filp: The file to apply the lock to
2307 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2308 * @fl: The lock to be applied
2309 * @conf: Place to return a copy of the conflicting lock, if found.
2311 * A caller that doesn't care about the conflicting lock may pass NULL
2312 * as the final argument.
2314 * If the filesystem defines a private ->lock() method, then @conf will
2315 * be left unchanged; so a caller that cares should initialize it to
2316 * some acceptable default.
2318 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2319 * locks, the ->lock() interface may return asynchronously, before the lock has
2320 * been granted or denied by the underlying filesystem, if (and only if)
2321 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2322 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2323 * the request is for a blocking lock. When ->lock() does return asynchronously,
2324 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2325 * request completes.
2326 * If the request is for non-blocking lock the file system should return
2327 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2328 * with the result. If the request timed out the callback routine will return a
2329 * nonzero return code and the file system should release the lock. The file
2330 * system is also responsible to keep a corresponding posix lock when it
2331 * grants a lock so the VFS can find out which locks are locally held and do
2332 * the correct lock cleanup when required.
2333 * The underlying filesystem must not drop the kernel lock or call
2334 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2337 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2339 if (filp->f_op->lock)
2340 return filp->f_op->lock(filp, cmd, fl);
2342 return posix_lock_file(filp, fl, conf);
2344 EXPORT_SYMBOL_GPL(vfs_lock_file);
2346 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2347 struct file_lock *fl)
2351 error = security_file_lock(filp, fl->fl_type);
2356 error = vfs_lock_file(filp, cmd, fl, NULL);
2357 if (error != FILE_LOCK_DEFERRED)
2359 error = wait_event_interruptible(fl->fl_wait, !fl->fl_blocker);
2363 locks_delete_block(fl);
2368 /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2370 check_fmode_for_setlk(struct file_lock *fl)
2372 switch (fl->fl_type) {
2374 if (!(fl->fl_file->f_mode & FMODE_READ))
2378 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2384 /* Apply the lock described by l to an open file descriptor.
2385 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2387 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2388 struct flock *flock)
2390 struct file_lock *file_lock = locks_alloc_lock();
2391 struct inode *inode = locks_inode(filp);
2395 if (file_lock == NULL)
2398 /* Don't allow mandatory locks on files that may be memory mapped
2401 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2406 error = flock_to_posix_lock(filp, file_lock, flock);
2410 error = check_fmode_for_setlk(file_lock);
2415 * If the cmd is requesting file-private locks, then set the
2416 * FL_OFDLCK flag and override the owner.
2421 if (flock->l_pid != 0)
2425 file_lock->fl_flags |= FL_OFDLCK;
2426 file_lock->fl_owner = filp;
2430 if (flock->l_pid != 0)
2434 file_lock->fl_flags |= FL_OFDLCK;
2435 file_lock->fl_owner = filp;
2438 file_lock->fl_flags |= FL_SLEEP;
2441 error = do_lock_file_wait(filp, cmd, file_lock);
2444 * Attempt to detect a close/fcntl race and recover by releasing the
2445 * lock that was just acquired. There is no need to do that when we're
2446 * unlocking though, or for OFD locks.
2448 if (!error && file_lock->fl_type != F_UNLCK &&
2449 !(file_lock->fl_flags & FL_OFDLCK)) {
2451 * We need that spin_lock here - it prevents reordering between
2452 * update of i_flctx->flc_posix and check for it done in
2453 * close(). rcu_read_lock() wouldn't do.
2455 spin_lock(¤t->files->file_lock);
2457 spin_unlock(¤t->files->file_lock);
2459 file_lock->fl_type = F_UNLCK;
2460 error = do_lock_file_wait(filp, cmd, file_lock);
2461 WARN_ON_ONCE(error);
2466 trace_fcntl_setlk(inode, file_lock, error);
2467 locks_free_lock(file_lock);
2471 #if BITS_PER_LONG == 32
2472 /* Report the first existing lock that would conflict with l.
2473 * This implements the F_GETLK command of fcntl().
2475 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2477 struct file_lock *fl;
2480 fl = locks_alloc_lock();
2485 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2488 error = flock64_to_posix_lock(filp, fl, flock);
2492 if (cmd == F_OFD_GETLK) {
2494 if (flock->l_pid != 0)
2498 fl->fl_flags |= FL_OFDLCK;
2499 fl->fl_owner = filp;
2502 error = vfs_test_lock(filp, fl);
2506 flock->l_type = fl->fl_type;
2507 if (fl->fl_type != F_UNLCK)
2508 posix_lock_to_flock64(flock, fl);
2511 locks_free_lock(fl);
2515 /* Apply the lock described by l to an open file descriptor.
2516 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2518 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2519 struct flock64 *flock)
2521 struct file_lock *file_lock = locks_alloc_lock();
2522 struct inode *inode = locks_inode(filp);
2526 if (file_lock == NULL)
2529 /* Don't allow mandatory locks on files that may be memory mapped
2532 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2537 error = flock64_to_posix_lock(filp, file_lock, flock);
2541 error = check_fmode_for_setlk(file_lock);
2546 * If the cmd is requesting file-private locks, then set the
2547 * FL_OFDLCK flag and override the owner.
2552 if (flock->l_pid != 0)
2556 file_lock->fl_flags |= FL_OFDLCK;
2557 file_lock->fl_owner = filp;
2561 if (flock->l_pid != 0)
2565 file_lock->fl_flags |= FL_OFDLCK;
2566 file_lock->fl_owner = filp;
2569 file_lock->fl_flags |= FL_SLEEP;
2572 error = do_lock_file_wait(filp, cmd, file_lock);
2575 * Attempt to detect a close/fcntl race and recover by releasing the
2576 * lock that was just acquired. There is no need to do that when we're
2577 * unlocking though, or for OFD locks.
2579 if (!error && file_lock->fl_type != F_UNLCK &&
2580 !(file_lock->fl_flags & FL_OFDLCK)) {
2582 * We need that spin_lock here - it prevents reordering between
2583 * update of i_flctx->flc_posix and check for it done in
2584 * close(). rcu_read_lock() wouldn't do.
2586 spin_lock(¤t->files->file_lock);
2588 spin_unlock(¤t->files->file_lock);
2590 file_lock->fl_type = F_UNLCK;
2591 error = do_lock_file_wait(filp, cmd, file_lock);
2592 WARN_ON_ONCE(error);
2597 locks_free_lock(file_lock);
2600 #endif /* BITS_PER_LONG == 32 */
2603 * This function is called when the file is being removed
2604 * from the task's fd array. POSIX locks belonging to this task
2605 * are deleted at this time.
2607 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2610 struct inode *inode = locks_inode(filp);
2611 struct file_lock lock;
2612 struct file_lock_context *ctx;
2615 * If there are no locks held on this file, we don't need to call
2616 * posix_lock_file(). Another process could be setting a lock on this
2617 * file at the same time, but we wouldn't remove that lock anyway.
2619 ctx = smp_load_acquire(&inode->i_flctx);
2620 if (!ctx || list_empty(&ctx->flc_posix))
2623 locks_init_lock(&lock);
2624 lock.fl_type = F_UNLCK;
2625 lock.fl_flags = FL_POSIX | FL_CLOSE;
2627 lock.fl_end = OFFSET_MAX;
2628 lock.fl_owner = owner;
2629 lock.fl_pid = current->tgid;
2630 lock.fl_file = filp;
2632 lock.fl_lmops = NULL;
2634 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2636 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2637 lock.fl_ops->fl_release_private(&lock);
2638 trace_locks_remove_posix(inode, &lock, error);
2640 EXPORT_SYMBOL(locks_remove_posix);
2642 /* The i_flctx must be valid when calling into here */
2644 locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2646 struct file_lock fl;
2647 struct inode *inode = locks_inode(filp);
2649 if (list_empty(&flctx->flc_flock))
2652 flock_make_lock(filp, LOCK_UN, &fl);
2653 fl.fl_flags |= FL_CLOSE;
2655 if (filp->f_op->flock)
2656 filp->f_op->flock(filp, F_SETLKW, &fl);
2658 flock_lock_inode(inode, &fl);
2660 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2661 fl.fl_ops->fl_release_private(&fl);
2664 /* The i_flctx must be valid when calling into here */
2666 locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2668 struct file_lock *fl, *tmp;
2671 if (list_empty(&ctx->flc_lease))
2674 percpu_down_read(&file_rwsem);
2675 spin_lock(&ctx->flc_lock);
2676 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2677 if (filp == fl->fl_file)
2678 lease_modify(fl, F_UNLCK, &dispose);
2679 spin_unlock(&ctx->flc_lock);
2680 percpu_up_read(&file_rwsem);
2682 locks_dispose_list(&dispose);
2686 * This function is called on the last close of an open file.
2688 void locks_remove_file(struct file *filp)
2690 struct file_lock_context *ctx;
2692 ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2696 /* remove any OFD locks */
2697 locks_remove_posix(filp, filp);
2699 /* remove flock locks */
2700 locks_remove_flock(filp, ctx);
2702 /* remove any leases */
2703 locks_remove_lease(filp, ctx);
2705 spin_lock(&ctx->flc_lock);
2706 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2707 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2708 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2709 spin_unlock(&ctx->flc_lock);
2713 * vfs_cancel_lock - file byte range unblock lock
2714 * @filp: The file to apply the unblock to
2715 * @fl: The lock to be unblocked
2717 * Used by lock managers to cancel blocked requests
2719 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2721 if (filp->f_op->lock)
2722 return filp->f_op->lock(filp, F_CANCELLK, fl);
2725 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2727 #ifdef CONFIG_PROC_FS
2728 #include <linux/proc_fs.h>
2729 #include <linux/seq_file.h>
2731 struct locks_iterator {
2736 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2737 loff_t id, char *pfx)
2739 struct inode *inode = NULL;
2740 unsigned int fl_pid;
2741 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2743 fl_pid = locks_translate_pid(fl, proc_pidns);
2745 * If lock owner is dead (and pid is freed) or not visible in current
2746 * pidns, zero is shown as a pid value. Check lock info from
2747 * init_pid_ns to get saved lock pid value.
2750 if (fl->fl_file != NULL)
2751 inode = locks_inode(fl->fl_file);
2753 seq_printf(f, "%lld:%s ", id, pfx);
2755 if (fl->fl_flags & FL_ACCESS)
2756 seq_puts(f, "ACCESS");
2757 else if (IS_OFDLCK(fl))
2758 seq_puts(f, "OFDLCK");
2760 seq_puts(f, "POSIX ");
2762 seq_printf(f, " %s ",
2763 (inode == NULL) ? "*NOINODE*" :
2764 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2765 } else if (IS_FLOCK(fl)) {
2766 if (fl->fl_type & LOCK_MAND) {
2767 seq_puts(f, "FLOCK MSNFS ");
2769 seq_puts(f, "FLOCK ADVISORY ");
2771 } else if (IS_LEASE(fl)) {
2772 if (fl->fl_flags & FL_DELEG)
2773 seq_puts(f, "DELEG ");
2775 seq_puts(f, "LEASE ");
2777 if (lease_breaking(fl))
2778 seq_puts(f, "BREAKING ");
2779 else if (fl->fl_file)
2780 seq_puts(f, "ACTIVE ");
2782 seq_puts(f, "BREAKER ");
2784 seq_puts(f, "UNKNOWN UNKNOWN ");
2786 if (fl->fl_type & LOCK_MAND) {
2787 seq_printf(f, "%s ",
2788 (fl->fl_type & LOCK_READ)
2789 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2790 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2792 seq_printf(f, "%s ",
2793 (lease_breaking(fl))
2794 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2795 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2798 /* userspace relies on this representation of dev_t */
2799 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2800 MAJOR(inode->i_sb->s_dev),
2801 MINOR(inode->i_sb->s_dev), inode->i_ino);
2803 seq_printf(f, "%d <none>:0 ", fl_pid);
2806 if (fl->fl_end == OFFSET_MAX)
2807 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2809 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2811 seq_puts(f, "0 EOF\n");
2815 static int locks_show(struct seq_file *f, void *v)
2817 struct locks_iterator *iter = f->private;
2818 struct file_lock *fl, *bfl;
2819 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2821 fl = hlist_entry(v, struct file_lock, fl_link);
2823 if (locks_translate_pid(fl, proc_pidns) == 0)
2826 lock_get_status(f, fl, iter->li_pos, "");
2828 list_for_each_entry(bfl, &fl->fl_blocked_requests, fl_blocked_member)
2829 lock_get_status(f, bfl, iter->li_pos, " ->");
2834 static void __show_fd_locks(struct seq_file *f,
2835 struct list_head *head, int *id,
2836 struct file *filp, struct files_struct *files)
2838 struct file_lock *fl;
2840 list_for_each_entry(fl, head, fl_list) {
2842 if (filp != fl->fl_file)
2844 if (fl->fl_owner != files &&
2845 fl->fl_owner != filp)
2849 seq_puts(f, "lock:\t");
2850 lock_get_status(f, fl, *id, "");
2854 void show_fd_locks(struct seq_file *f,
2855 struct file *filp, struct files_struct *files)
2857 struct inode *inode = locks_inode(filp);
2858 struct file_lock_context *ctx;
2861 ctx = smp_load_acquire(&inode->i_flctx);
2865 spin_lock(&ctx->flc_lock);
2866 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2867 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2868 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2869 spin_unlock(&ctx->flc_lock);
2872 static void *locks_start(struct seq_file *f, loff_t *pos)
2873 __acquires(&blocked_lock_lock)
2875 struct locks_iterator *iter = f->private;
2877 iter->li_pos = *pos + 1;
2878 percpu_down_write(&file_rwsem);
2879 spin_lock(&blocked_lock_lock);
2880 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2883 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2885 struct locks_iterator *iter = f->private;
2888 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2891 static void locks_stop(struct seq_file *f, void *v)
2892 __releases(&blocked_lock_lock)
2894 spin_unlock(&blocked_lock_lock);
2895 percpu_up_write(&file_rwsem);
2898 static const struct seq_operations locks_seq_operations = {
2899 .start = locks_start,
2905 static int __init proc_locks_init(void)
2907 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2908 sizeof(struct locks_iterator), NULL);
2911 fs_initcall(proc_locks_init);
2914 static int __init filelock_init(void)
2918 flctx_cache = kmem_cache_create("file_lock_ctx",
2919 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2921 filelock_cache = kmem_cache_create("file_lock_cache",
2922 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2924 for_each_possible_cpu(i) {
2925 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2927 spin_lock_init(&fll->lock);
2928 INIT_HLIST_HEAD(&fll->hlist);
2933 core_initcall(filelock_init);