1 // SPDX-License-Identifier: GPL-2.0-only
5 * We implement four types of file locks: BSD locks, posix locks, open
6 * file description locks, and leases. For details about BSD locks,
7 * see the flock(2) man page; for details about the other three, see
11 * Locking conflicts and dependencies:
12 * If multiple threads attempt to lock the same byte (or flock the same file)
13 * only one can be granted the lock, and other must wait their turn.
14 * The first lock has been "applied" or "granted", the others are "waiting"
15 * and are "blocked" by the "applied" lock..
17 * Waiting and applied locks are all kept in trees whose properties are:
19 * - the root of a tree may be an applied or waiting lock.
20 * - every other node in the tree is a waiting lock that
21 * conflicts with every ancestor of that node.
23 * Every such tree begins life as a waiting singleton which obviously
24 * satisfies the above properties.
26 * The only ways we modify trees preserve these properties:
28 * 1. We may add a new leaf node, but only after first verifying that it
29 * conflicts with all of its ancestors.
30 * 2. We may remove the root of a tree, creating a new singleton
31 * tree from the root and N new trees rooted in the immediate
33 * 3. If the root of a tree is not currently an applied lock, we may
34 * apply it (if possible).
35 * 4. We may upgrade the root of the tree (either extend its range,
36 * or upgrade its entire range from read to write).
38 * When an applied lock is modified in a way that reduces or downgrades any
39 * part of its range, we remove all its children (2 above). This particularly
40 * happens when a lock is unlocked.
42 * For each of those child trees we "wake up" the thread which is
43 * waiting for the lock so it can continue handling as follows: if the
44 * root of the tree applies, we do so (3). If it doesn't, it must
45 * conflict with some applied lock. We remove (wake up) all of its children
46 * (2), and add it is a new leaf to the tree rooted in the applied
47 * lock (1). We then repeat the process recursively with those
52 #include <linux/capability.h>
53 #include <linux/file.h>
54 #include <linux/fdtable.h>
56 #include <linux/init.h>
57 #include <linux/security.h>
58 #include <linux/slab.h>
59 #include <linux/syscalls.h>
60 #include <linux/time.h>
61 #include <linux/rcupdate.h>
62 #include <linux/pid_namespace.h>
63 #include <linux/hashtable.h>
64 #include <linux/percpu.h>
65 #include <linux/sysctl.h>
67 #define CREATE_TRACE_POINTS
68 #include <trace/events/filelock.h>
70 #include <linux/uaccess.h>
72 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
73 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
74 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
75 #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
76 #define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
78 static bool lease_breaking(struct file_lock *fl)
80 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
83 static int target_leasetype(struct file_lock *fl)
85 if (fl->fl_flags & FL_UNLOCK_PENDING)
87 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
92 static int leases_enable = 1;
93 static int lease_break_time = 45;
96 static struct ctl_table locks_sysctls[] = {
98 .procname = "leases-enable",
99 .data = &leases_enable,
100 .maxlen = sizeof(int),
102 .proc_handler = proc_dointvec,
106 .procname = "lease-break-time",
107 .data = &lease_break_time,
108 .maxlen = sizeof(int),
110 .proc_handler = proc_dointvec,
112 #endif /* CONFIG_MMU */
116 static int __init init_fs_locks_sysctls(void)
118 register_sysctl_init("fs", locks_sysctls);
121 early_initcall(init_fs_locks_sysctls);
122 #endif /* CONFIG_SYSCTL */
125 * The global file_lock_list is only used for displaying /proc/locks, so we
126 * keep a list on each CPU, with each list protected by its own spinlock.
127 * Global serialization is done using file_rwsem.
129 * Note that alterations to the list also require that the relevant flc_lock is
132 struct file_lock_list_struct {
134 struct hlist_head hlist;
136 static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
137 DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
141 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
142 * It is protected by blocked_lock_lock.
144 * We hash locks by lockowner in order to optimize searching for the lock a
145 * particular lockowner is waiting on.
147 * FIXME: make this value scale via some heuristic? We generally will want more
148 * buckets when we have more lockowners holding locks, but that's a little
149 * difficult to determine without knowing what the workload will look like.
151 #define BLOCKED_HASH_BITS 7
152 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
155 * This lock protects the blocked_hash. Generally, if you're accessing it, you
156 * want to be holding this lock.
158 * In addition, it also protects the fl->fl_blocked_requests list, and the
159 * fl->fl_blocker pointer for file_lock structures that are acting as lock
160 * requests (in contrast to those that are acting as records of acquired locks).
162 * Note that when we acquire this lock in order to change the above fields,
163 * we often hold the flc_lock as well. In certain cases, when reading the fields
164 * protected by this lock, we can skip acquiring it iff we already hold the
167 static DEFINE_SPINLOCK(blocked_lock_lock);
169 static struct kmem_cache *flctx_cache __read_mostly;
170 static struct kmem_cache *filelock_cache __read_mostly;
172 static struct file_lock_context *
173 locks_get_lock_context(struct inode *inode, int type)
175 struct file_lock_context *ctx;
177 /* paired with cmpxchg() below */
178 ctx = smp_load_acquire(&inode->i_flctx);
179 if (likely(ctx) || type == F_UNLCK)
182 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
186 spin_lock_init(&ctx->flc_lock);
187 INIT_LIST_HEAD(&ctx->flc_flock);
188 INIT_LIST_HEAD(&ctx->flc_posix);
189 INIT_LIST_HEAD(&ctx->flc_lease);
192 * Assign the pointer if it's not already assigned. If it is, then
193 * free the context we just allocated.
195 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
196 kmem_cache_free(flctx_cache, ctx);
197 ctx = smp_load_acquire(&inode->i_flctx);
200 trace_locks_get_lock_context(inode, type, ctx);
205 locks_dump_ctx_list(struct list_head *list, char *list_type)
207 struct file_lock *fl;
209 list_for_each_entry(fl, list, fl_list) {
210 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);
215 locks_check_ctx_lists(struct inode *inode)
217 struct file_lock_context *ctx = inode->i_flctx;
219 if (unlikely(!list_empty(&ctx->flc_flock) ||
220 !list_empty(&ctx->flc_posix) ||
221 !list_empty(&ctx->flc_lease))) {
222 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
223 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
225 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
226 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
227 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
232 locks_check_ctx_file_list(struct file *filp, struct list_head *list,
235 struct file_lock *fl;
236 struct inode *inode = locks_inode(filp);
238 list_for_each_entry(fl, list, fl_list)
239 if (fl->fl_file == filp)
240 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
241 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
242 list_type, MAJOR(inode->i_sb->s_dev),
243 MINOR(inode->i_sb->s_dev), inode->i_ino,
244 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
248 locks_free_lock_context(struct inode *inode)
250 struct file_lock_context *ctx = inode->i_flctx;
253 locks_check_ctx_lists(inode);
254 kmem_cache_free(flctx_cache, ctx);
258 static void locks_init_lock_heads(struct file_lock *fl)
260 INIT_HLIST_NODE(&fl->fl_link);
261 INIT_LIST_HEAD(&fl->fl_list);
262 INIT_LIST_HEAD(&fl->fl_blocked_requests);
263 INIT_LIST_HEAD(&fl->fl_blocked_member);
264 init_waitqueue_head(&fl->fl_wait);
267 /* Allocate an empty lock structure. */
268 struct file_lock *locks_alloc_lock(void)
270 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
273 locks_init_lock_heads(fl);
277 EXPORT_SYMBOL_GPL(locks_alloc_lock);
279 void locks_release_private(struct file_lock *fl)
281 BUG_ON(waitqueue_active(&fl->fl_wait));
282 BUG_ON(!list_empty(&fl->fl_list));
283 BUG_ON(!list_empty(&fl->fl_blocked_requests));
284 BUG_ON(!list_empty(&fl->fl_blocked_member));
285 BUG_ON(!hlist_unhashed(&fl->fl_link));
288 if (fl->fl_ops->fl_release_private)
289 fl->fl_ops->fl_release_private(fl);
294 if (fl->fl_lmops->lm_put_owner) {
295 fl->fl_lmops->lm_put_owner(fl->fl_owner);
301 EXPORT_SYMBOL_GPL(locks_release_private);
304 * locks_owner_has_blockers - Check for blocking lock requests
305 * @flctx: file lock context
309 * %true: @owner has at least one blocker
310 * %false: @owner has no blockers
312 bool locks_owner_has_blockers(struct file_lock_context *flctx,
315 struct file_lock *fl;
317 spin_lock(&flctx->flc_lock);
318 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
319 if (fl->fl_owner != owner)
321 if (!list_empty(&fl->fl_blocked_requests)) {
322 spin_unlock(&flctx->flc_lock);
326 spin_unlock(&flctx->flc_lock);
329 EXPORT_SYMBOL_GPL(locks_owner_has_blockers);
331 /* Free a lock which is not in use. */
332 void locks_free_lock(struct file_lock *fl)
334 locks_release_private(fl);
335 kmem_cache_free(filelock_cache, fl);
337 EXPORT_SYMBOL(locks_free_lock);
340 locks_dispose_list(struct list_head *dispose)
342 struct file_lock *fl;
344 while (!list_empty(dispose)) {
345 fl = list_first_entry(dispose, struct file_lock, fl_list);
346 list_del_init(&fl->fl_list);
351 void locks_init_lock(struct file_lock *fl)
353 memset(fl, 0, sizeof(struct file_lock));
354 locks_init_lock_heads(fl);
356 EXPORT_SYMBOL(locks_init_lock);
359 * Initialize a new lock from an existing file_lock structure.
361 void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
363 new->fl_owner = fl->fl_owner;
364 new->fl_pid = fl->fl_pid;
366 new->fl_flags = fl->fl_flags;
367 new->fl_type = fl->fl_type;
368 new->fl_start = fl->fl_start;
369 new->fl_end = fl->fl_end;
370 new->fl_lmops = fl->fl_lmops;
374 if (fl->fl_lmops->lm_get_owner)
375 fl->fl_lmops->lm_get_owner(fl->fl_owner);
378 EXPORT_SYMBOL(locks_copy_conflock);
380 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
382 /* "new" must be a freshly-initialized lock */
383 WARN_ON_ONCE(new->fl_ops);
385 locks_copy_conflock(new, fl);
387 new->fl_file = fl->fl_file;
388 new->fl_ops = fl->fl_ops;
391 if (fl->fl_ops->fl_copy_lock)
392 fl->fl_ops->fl_copy_lock(new, fl);
395 EXPORT_SYMBOL(locks_copy_lock);
397 static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
402 * As ctx->flc_lock is held, new requests cannot be added to
403 * ->fl_blocked_requests, so we don't need a lock to check if it
406 if (list_empty(&fl->fl_blocked_requests))
408 spin_lock(&blocked_lock_lock);
409 list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
410 list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
412 spin_unlock(&blocked_lock_lock);
415 static inline int flock_translate_cmd(int cmd) {
427 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
428 static struct file_lock *
429 flock_make_lock(struct file *filp, unsigned int cmd, struct file_lock *fl)
431 int type = flock_translate_cmd(cmd);
434 return ERR_PTR(type);
437 fl = locks_alloc_lock();
439 return ERR_PTR(-ENOMEM);
446 fl->fl_pid = current->tgid;
447 fl->fl_flags = FL_FLOCK;
449 fl->fl_end = OFFSET_MAX;
454 static int assign_type(struct file_lock *fl, long type)
468 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
471 switch (l->l_whence) {
476 fl->fl_start = filp->f_pos;
479 fl->fl_start = i_size_read(file_inode(filp));
484 if (l->l_start > OFFSET_MAX - fl->fl_start)
486 fl->fl_start += l->l_start;
487 if (fl->fl_start < 0)
490 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
491 POSIX-2001 defines it. */
493 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
495 fl->fl_end = fl->fl_start + (l->l_len - 1);
497 } else if (l->l_len < 0) {
498 if (fl->fl_start + l->l_len < 0)
500 fl->fl_end = fl->fl_start - 1;
501 fl->fl_start += l->l_len;
503 fl->fl_end = OFFSET_MAX;
505 fl->fl_owner = current->files;
506 fl->fl_pid = current->tgid;
508 fl->fl_flags = FL_POSIX;
512 return assign_type(fl, l->l_type);
515 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
518 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
521 struct flock64 ll = {
523 .l_whence = l->l_whence,
524 .l_start = l->l_start,
528 return flock64_to_posix_lock(filp, fl, &ll);
531 /* default lease lock manager operations */
533 lease_break_callback(struct file_lock *fl)
535 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
540 lease_setup(struct file_lock *fl, void **priv)
542 struct file *filp = fl->fl_file;
543 struct fasync_struct *fa = *priv;
546 * fasync_insert_entry() returns the old entry if any. If there was no
547 * old entry, then it used "priv" and inserted it into the fasync list.
548 * Clear the pointer to indicate that it shouldn't be freed.
550 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
553 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
556 static const struct lock_manager_operations lease_manager_ops = {
557 .lm_break = lease_break_callback,
558 .lm_change = lease_modify,
559 .lm_setup = lease_setup,
563 * Initialize a lease, use the default lock manager operations
565 static int lease_init(struct file *filp, long type, struct file_lock *fl)
567 if (assign_type(fl, type) != 0)
571 fl->fl_pid = current->tgid;
574 fl->fl_flags = FL_LEASE;
576 fl->fl_end = OFFSET_MAX;
578 fl->fl_lmops = &lease_manager_ops;
582 /* Allocate a file_lock initialised to this type of lease */
583 static struct file_lock *lease_alloc(struct file *filp, long type)
585 struct file_lock *fl = locks_alloc_lock();
589 return ERR_PTR(error);
591 error = lease_init(filp, type, fl);
594 return ERR_PTR(error);
599 /* Check if two locks overlap each other.
601 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
603 return ((fl1->fl_end >= fl2->fl_start) &&
604 (fl2->fl_end >= fl1->fl_start));
608 * Check whether two locks have the same owner.
610 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
612 return fl1->fl_owner == fl2->fl_owner;
615 /* Must be called with the flc_lock held! */
616 static void locks_insert_global_locks(struct file_lock *fl)
618 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
620 percpu_rwsem_assert_held(&file_rwsem);
622 spin_lock(&fll->lock);
623 fl->fl_link_cpu = smp_processor_id();
624 hlist_add_head(&fl->fl_link, &fll->hlist);
625 spin_unlock(&fll->lock);
628 /* Must be called with the flc_lock held! */
629 static void locks_delete_global_locks(struct file_lock *fl)
631 struct file_lock_list_struct *fll;
633 percpu_rwsem_assert_held(&file_rwsem);
636 * Avoid taking lock if already unhashed. This is safe since this check
637 * is done while holding the flc_lock, and new insertions into the list
638 * also require that it be held.
640 if (hlist_unhashed(&fl->fl_link))
643 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
644 spin_lock(&fll->lock);
645 hlist_del_init(&fl->fl_link);
646 spin_unlock(&fll->lock);
650 posix_owner_key(struct file_lock *fl)
652 return (unsigned long)fl->fl_owner;
655 static void locks_insert_global_blocked(struct file_lock *waiter)
657 lockdep_assert_held(&blocked_lock_lock);
659 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
662 static void locks_delete_global_blocked(struct file_lock *waiter)
664 lockdep_assert_held(&blocked_lock_lock);
666 hash_del(&waiter->fl_link);
669 /* Remove waiter from blocker's block list.
670 * When blocker ends up pointing to itself then the list is empty.
672 * Must be called with blocked_lock_lock held.
674 static void __locks_delete_block(struct file_lock *waiter)
676 locks_delete_global_blocked(waiter);
677 list_del_init(&waiter->fl_blocked_member);
680 static void __locks_wake_up_blocks(struct file_lock *blocker)
682 while (!list_empty(&blocker->fl_blocked_requests)) {
683 struct file_lock *waiter;
685 waiter = list_first_entry(&blocker->fl_blocked_requests,
686 struct file_lock, fl_blocked_member);
687 __locks_delete_block(waiter);
688 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
689 waiter->fl_lmops->lm_notify(waiter);
691 wake_up(&waiter->fl_wait);
694 * The setting of fl_blocker to NULL marks the "done"
695 * point in deleting a block. Paired with acquire at the top
696 * of locks_delete_block().
698 smp_store_release(&waiter->fl_blocker, NULL);
703 * locks_delete_block - stop waiting for a file lock
704 * @waiter: the lock which was waiting
706 * lockd/nfsd need to disconnect the lock while working on it.
708 int locks_delete_block(struct file_lock *waiter)
710 int status = -ENOENT;
713 * If fl_blocker is NULL, it won't be set again as this thread "owns"
714 * the lock and is the only one that might try to claim the lock.
716 * We use acquire/release to manage fl_blocker so that we can
717 * optimize away taking the blocked_lock_lock in many cases.
719 * The smp_load_acquire guarantees two things:
721 * 1/ that fl_blocked_requests can be tested locklessly. If something
722 * was recently added to that list it must have been in a locked region
723 * *before* the locked region when fl_blocker was set to NULL.
725 * 2/ that no other thread is accessing 'waiter', so it is safe to free
726 * it. __locks_wake_up_blocks is careful not to touch waiter after
727 * fl_blocker is released.
729 * If a lockless check of fl_blocker shows it to be NULL, we know that
730 * no new locks can be inserted into its fl_blocked_requests list, and
731 * can avoid doing anything further if the list is empty.
733 if (!smp_load_acquire(&waiter->fl_blocker) &&
734 list_empty(&waiter->fl_blocked_requests))
737 spin_lock(&blocked_lock_lock);
738 if (waiter->fl_blocker)
740 __locks_wake_up_blocks(waiter);
741 __locks_delete_block(waiter);
744 * The setting of fl_blocker to NULL marks the "done" point in deleting
745 * a block. Paired with acquire at the top of this function.
747 smp_store_release(&waiter->fl_blocker, NULL);
748 spin_unlock(&blocked_lock_lock);
751 EXPORT_SYMBOL(locks_delete_block);
753 /* Insert waiter into blocker's block list.
754 * We use a circular list so that processes can be easily woken up in
755 * the order they blocked. The documentation doesn't require this but
756 * it seems like the reasonable thing to do.
758 * Must be called with both the flc_lock and blocked_lock_lock held. The
759 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
760 * but by ensuring that the flc_lock is also held on insertions we can avoid
761 * taking the blocked_lock_lock in some cases when we see that the
762 * fl_blocked_requests list is empty.
764 * Rather than just adding to the list, we check for conflicts with any existing
765 * waiters, and add beneath any waiter that blocks the new waiter.
766 * Thus wakeups don't happen until needed.
768 static void __locks_insert_block(struct file_lock *blocker,
769 struct file_lock *waiter,
770 bool conflict(struct file_lock *,
773 struct file_lock *fl;
774 BUG_ON(!list_empty(&waiter->fl_blocked_member));
777 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
778 if (conflict(fl, waiter)) {
782 waiter->fl_blocker = blocker;
783 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
784 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
785 locks_insert_global_blocked(waiter);
787 /* The requests in waiter->fl_blocked are known to conflict with
788 * waiter, but might not conflict with blocker, or the requests
789 * and lock which block it. So they all need to be woken.
791 __locks_wake_up_blocks(waiter);
794 /* Must be called with flc_lock held. */
795 static void locks_insert_block(struct file_lock *blocker,
796 struct file_lock *waiter,
797 bool conflict(struct file_lock *,
800 spin_lock(&blocked_lock_lock);
801 __locks_insert_block(blocker, waiter, conflict);
802 spin_unlock(&blocked_lock_lock);
806 * Wake up processes blocked waiting for blocker.
808 * Must be called with the inode->flc_lock held!
810 static void locks_wake_up_blocks(struct file_lock *blocker)
813 * Avoid taking global lock if list is empty. This is safe since new
814 * blocked requests are only added to the list under the flc_lock, and
815 * the flc_lock is always held here. Note that removal from the
816 * fl_blocked_requests list does not require the flc_lock, so we must
817 * recheck list_empty() after acquiring the blocked_lock_lock.
819 if (list_empty(&blocker->fl_blocked_requests))
822 spin_lock(&blocked_lock_lock);
823 __locks_wake_up_blocks(blocker);
824 spin_unlock(&blocked_lock_lock);
828 locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
830 list_add_tail(&fl->fl_list, before);
831 locks_insert_global_locks(fl);
835 locks_unlink_lock_ctx(struct file_lock *fl)
837 locks_delete_global_locks(fl);
838 list_del_init(&fl->fl_list);
839 locks_wake_up_blocks(fl);
843 locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
845 locks_unlink_lock_ctx(fl);
847 list_add(&fl->fl_list, dispose);
852 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
853 * checks for shared/exclusive status of overlapping locks.
855 static bool locks_conflict(struct file_lock *caller_fl,
856 struct file_lock *sys_fl)
858 if (sys_fl->fl_type == F_WRLCK)
860 if (caller_fl->fl_type == F_WRLCK)
865 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
866 * checking before calling the locks_conflict().
868 static bool posix_locks_conflict(struct file_lock *caller_fl,
869 struct file_lock *sys_fl)
871 /* POSIX locks owned by the same process do not conflict with
874 if (posix_same_owner(caller_fl, sys_fl))
877 /* Check whether they overlap */
878 if (!locks_overlap(caller_fl, sys_fl))
881 return locks_conflict(caller_fl, sys_fl);
884 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
885 * checking before calling the locks_conflict().
887 static bool flock_locks_conflict(struct file_lock *caller_fl,
888 struct file_lock *sys_fl)
890 /* FLOCK locks referring to the same filp do not conflict with
893 if (caller_fl->fl_file == sys_fl->fl_file)
896 return locks_conflict(caller_fl, sys_fl);
900 posix_test_lock(struct file *filp, struct file_lock *fl)
902 struct file_lock *cfl;
903 struct file_lock_context *ctx;
904 struct inode *inode = locks_inode(filp);
906 ctx = smp_load_acquire(&inode->i_flctx);
907 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
908 fl->fl_type = F_UNLCK;
912 spin_lock(&ctx->flc_lock);
913 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
914 if (posix_locks_conflict(fl, cfl)) {
915 locks_copy_conflock(fl, cfl);
919 fl->fl_type = F_UNLCK;
921 spin_unlock(&ctx->flc_lock);
924 EXPORT_SYMBOL(posix_test_lock);
927 * Deadlock detection:
929 * We attempt to detect deadlocks that are due purely to posix file
932 * We assume that a task can be waiting for at most one lock at a time.
933 * So for any acquired lock, the process holding that lock may be
934 * waiting on at most one other lock. That lock in turns may be held by
935 * someone waiting for at most one other lock. Given a requested lock
936 * caller_fl which is about to wait for a conflicting lock block_fl, we
937 * follow this chain of waiters to ensure we are not about to create a
940 * Since we do this before we ever put a process to sleep on a lock, we
941 * are ensured that there is never a cycle; that is what guarantees that
942 * the while() loop in posix_locks_deadlock() eventually completes.
944 * Note: the above assumption may not be true when handling lock
945 * requests from a broken NFS client. It may also fail in the presence
946 * of tasks (such as posix threads) sharing the same open file table.
947 * To handle those cases, we just bail out after a few iterations.
949 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
950 * Because the owner is not even nominally tied to a thread of
951 * execution, the deadlock detection below can't reasonably work well. Just
954 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
955 * locks that just checks for the case where two tasks are attempting to
956 * upgrade from read to write locks on the same inode.
959 #define MAX_DEADLK_ITERATIONS 10
961 /* Find a lock that the owner of the given block_fl is blocking on. */
962 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
964 struct file_lock *fl;
966 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
967 if (posix_same_owner(fl, block_fl)) {
968 while (fl->fl_blocker)
976 /* Must be called with the blocked_lock_lock held! */
977 static int posix_locks_deadlock(struct file_lock *caller_fl,
978 struct file_lock *block_fl)
982 lockdep_assert_held(&blocked_lock_lock);
985 * This deadlock detector can't reasonably detect deadlocks with
986 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
988 if (IS_OFDLCK(caller_fl))
991 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
992 if (i++ > MAX_DEADLK_ITERATIONS)
994 if (posix_same_owner(caller_fl, block_fl))
1000 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1001 * after any leases, but before any posix locks.
1003 * Note that if called with an FL_EXISTS argument, the caller may determine
1004 * whether or not a lock was successfully freed by testing the return
1005 * value for -ENOENT.
1007 static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1009 struct file_lock *new_fl = NULL;
1010 struct file_lock *fl;
1011 struct file_lock_context *ctx;
1016 ctx = locks_get_lock_context(inode, request->fl_type);
1018 if (request->fl_type != F_UNLCK)
1020 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1023 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1024 new_fl = locks_alloc_lock();
1029 percpu_down_read(&file_rwsem);
1030 spin_lock(&ctx->flc_lock);
1031 if (request->fl_flags & FL_ACCESS)
1034 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1035 if (request->fl_file != fl->fl_file)
1037 if (request->fl_type == fl->fl_type)
1040 locks_delete_lock_ctx(fl, &dispose);
1044 if (request->fl_type == F_UNLCK) {
1045 if ((request->fl_flags & FL_EXISTS) && !found)
1051 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1052 if (!flock_locks_conflict(request, fl))
1055 if (!(request->fl_flags & FL_SLEEP))
1057 error = FILE_LOCK_DEFERRED;
1058 locks_insert_block(fl, request, flock_locks_conflict);
1061 if (request->fl_flags & FL_ACCESS)
1063 locks_copy_lock(new_fl, request);
1064 locks_move_blocks(new_fl, request);
1065 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1070 spin_unlock(&ctx->flc_lock);
1071 percpu_up_read(&file_rwsem);
1073 locks_free_lock(new_fl);
1074 locks_dispose_list(&dispose);
1075 trace_flock_lock_inode(inode, request, error);
1079 static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1080 struct file_lock *conflock)
1082 struct file_lock *fl, *tmp;
1083 struct file_lock *new_fl = NULL;
1084 struct file_lock *new_fl2 = NULL;
1085 struct file_lock *left = NULL;
1086 struct file_lock *right = NULL;
1087 struct file_lock_context *ctx;
1092 ctx = locks_get_lock_context(inode, request->fl_type);
1094 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1097 * We may need two file_lock structures for this operation,
1098 * so we get them in advance to avoid races.
1100 * In some cases we can be sure, that no new locks will be needed
1102 if (!(request->fl_flags & FL_ACCESS) &&
1103 (request->fl_type != F_UNLCK ||
1104 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1105 new_fl = locks_alloc_lock();
1106 new_fl2 = locks_alloc_lock();
1109 percpu_down_read(&file_rwsem);
1110 spin_lock(&ctx->flc_lock);
1112 * New lock request. Walk all POSIX locks and look for conflicts. If
1113 * there are any, either return error or put the request on the
1114 * blocker's list of waiters and the global blocked_hash.
1116 if (request->fl_type != F_UNLCK) {
1117 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1118 if (!posix_locks_conflict(request, fl))
1121 locks_copy_conflock(conflock, fl);
1123 if (!(request->fl_flags & FL_SLEEP))
1126 * Deadlock detection and insertion into the blocked
1127 * locks list must be done while holding the same lock!
1130 spin_lock(&blocked_lock_lock);
1132 * Ensure that we don't find any locks blocked on this
1133 * request during deadlock detection.
1135 __locks_wake_up_blocks(request);
1136 if (likely(!posix_locks_deadlock(request, fl))) {
1137 error = FILE_LOCK_DEFERRED;
1138 __locks_insert_block(fl, request,
1139 posix_locks_conflict);
1141 spin_unlock(&blocked_lock_lock);
1146 /* If we're just looking for a conflict, we're done. */
1148 if (request->fl_flags & FL_ACCESS)
1151 /* Find the first old lock with the same owner as the new lock */
1152 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1153 if (posix_same_owner(request, fl))
1157 /* Process locks with this owner. */
1158 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1159 if (!posix_same_owner(request, fl))
1162 /* Detect adjacent or overlapping regions (if same lock type) */
1163 if (request->fl_type == fl->fl_type) {
1164 /* In all comparisons of start vs end, use
1165 * "start - 1" rather than "end + 1". If end
1166 * is OFFSET_MAX, end + 1 will become negative.
1168 if (fl->fl_end < request->fl_start - 1)
1170 /* If the next lock in the list has entirely bigger
1171 * addresses than the new one, insert the lock here.
1173 if (fl->fl_start - 1 > request->fl_end)
1176 /* If we come here, the new and old lock are of the
1177 * same type and adjacent or overlapping. Make one
1178 * lock yielding from the lower start address of both
1179 * locks to the higher end address.
1181 if (fl->fl_start > request->fl_start)
1182 fl->fl_start = request->fl_start;
1184 request->fl_start = fl->fl_start;
1185 if (fl->fl_end < request->fl_end)
1186 fl->fl_end = request->fl_end;
1188 request->fl_end = fl->fl_end;
1190 locks_delete_lock_ctx(fl, &dispose);
1196 /* Processing for different lock types is a bit
1199 if (fl->fl_end < request->fl_start)
1201 if (fl->fl_start > request->fl_end)
1203 if (request->fl_type == F_UNLCK)
1205 if (fl->fl_start < request->fl_start)
1207 /* If the next lock in the list has a higher end
1208 * address than the new one, insert the new one here.
1210 if (fl->fl_end > request->fl_end) {
1214 if (fl->fl_start >= request->fl_start) {
1215 /* The new lock completely replaces an old
1216 * one (This may happen several times).
1219 locks_delete_lock_ctx(fl, &dispose);
1223 * Replace the old lock with new_fl, and
1224 * remove the old one. It's safe to do the
1225 * insert here since we know that we won't be
1226 * using new_fl later, and that the lock is
1227 * just replacing an existing lock.
1232 locks_copy_lock(new_fl, request);
1233 locks_move_blocks(new_fl, request);
1236 locks_insert_lock_ctx(request, &fl->fl_list);
1237 locks_delete_lock_ctx(fl, &dispose);
1244 * The above code only modifies existing locks in case of merging or
1245 * replacing. If new lock(s) need to be inserted all modifications are
1246 * done below this, so it's safe yet to bail out.
1248 error = -ENOLCK; /* "no luck" */
1249 if (right && left == right && !new_fl2)
1254 if (request->fl_type == F_UNLCK) {
1255 if (request->fl_flags & FL_EXISTS)
1264 locks_copy_lock(new_fl, request);
1265 locks_move_blocks(new_fl, request);
1266 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1271 if (left == right) {
1272 /* The new lock breaks the old one in two pieces,
1273 * so we have to use the second new lock.
1277 locks_copy_lock(left, right);
1278 locks_insert_lock_ctx(left, &fl->fl_list);
1280 right->fl_start = request->fl_end + 1;
1281 locks_wake_up_blocks(right);
1284 left->fl_end = request->fl_start - 1;
1285 locks_wake_up_blocks(left);
1288 spin_unlock(&ctx->flc_lock);
1289 percpu_up_read(&file_rwsem);
1291 * Free any unused locks.
1294 locks_free_lock(new_fl);
1296 locks_free_lock(new_fl2);
1297 locks_dispose_list(&dispose);
1298 trace_posix_lock_inode(inode, request, error);
1304 * posix_lock_file - Apply a POSIX-style lock to a file
1305 * @filp: The file to apply the lock to
1306 * @fl: The lock to be applied
1307 * @conflock: Place to return a copy of the conflicting lock, if found.
1309 * Add a POSIX style lock to a file.
1310 * We merge adjacent & overlapping locks whenever possible.
1311 * POSIX locks are sorted by owner task, then by starting address
1313 * Note that if called with an FL_EXISTS argument, the caller may determine
1314 * whether or not a lock was successfully freed by testing the return
1315 * value for -ENOENT.
1317 int posix_lock_file(struct file *filp, struct file_lock *fl,
1318 struct file_lock *conflock)
1320 return posix_lock_inode(locks_inode(filp), fl, conflock);
1322 EXPORT_SYMBOL(posix_lock_file);
1325 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1326 * @inode: inode of file to which lock request should be applied
1327 * @fl: The lock to be applied
1329 * Apply a POSIX style lock request to an inode.
1331 static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1336 error = posix_lock_inode(inode, fl, NULL);
1337 if (error != FILE_LOCK_DEFERRED)
1339 error = wait_event_interruptible(fl->fl_wait,
1340 list_empty(&fl->fl_blocked_member));
1344 locks_delete_block(fl);
1348 static void lease_clear_pending(struct file_lock *fl, int arg)
1352 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1355 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1359 /* We already had a lease on this file; just change its type */
1360 int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1362 int error = assign_type(fl, arg);
1366 lease_clear_pending(fl, arg);
1367 locks_wake_up_blocks(fl);
1368 if (arg == F_UNLCK) {
1369 struct file *filp = fl->fl_file;
1372 filp->f_owner.signum = 0;
1373 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1374 if (fl->fl_fasync != NULL) {
1375 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1376 fl->fl_fasync = NULL;
1378 locks_delete_lock_ctx(fl, dispose);
1382 EXPORT_SYMBOL(lease_modify);
1384 static bool past_time(unsigned long then)
1387 /* 0 is a special value meaning "this never expires": */
1389 return time_after(jiffies, then);
1392 static void time_out_leases(struct inode *inode, struct list_head *dispose)
1394 struct file_lock_context *ctx = inode->i_flctx;
1395 struct file_lock *fl, *tmp;
1397 lockdep_assert_held(&ctx->flc_lock);
1399 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1400 trace_time_out_leases(inode, fl);
1401 if (past_time(fl->fl_downgrade_time))
1402 lease_modify(fl, F_RDLCK, dispose);
1403 if (past_time(fl->fl_break_time))
1404 lease_modify(fl, F_UNLCK, dispose);
1408 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1412 if (lease->fl_lmops->lm_breaker_owns_lease
1413 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1415 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1419 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1424 rc = locks_conflict(breaker, lease);
1426 trace_leases_conflict(rc, lease, breaker);
1431 any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1433 struct file_lock_context *ctx = inode->i_flctx;
1434 struct file_lock *fl;
1436 lockdep_assert_held(&ctx->flc_lock);
1438 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1439 if (leases_conflict(fl, breaker))
1446 * __break_lease - revoke all outstanding leases on file
1447 * @inode: the inode of the file to return
1448 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1450 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1453 * break_lease (inlined for speed) has checked there already is at least
1454 * some kind of lock (maybe a lease) on this file. Leases are broken on
1455 * a call to open() or truncate(). This function can sleep unless you
1456 * specified %O_NONBLOCK to your open().
1458 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1461 struct file_lock_context *ctx;
1462 struct file_lock *new_fl, *fl, *tmp;
1463 unsigned long break_time;
1464 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1467 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1469 return PTR_ERR(new_fl);
1470 new_fl->fl_flags = type;
1472 /* typically we will check that ctx is non-NULL before calling */
1473 ctx = smp_load_acquire(&inode->i_flctx);
1479 percpu_down_read(&file_rwsem);
1480 spin_lock(&ctx->flc_lock);
1482 time_out_leases(inode, &dispose);
1484 if (!any_leases_conflict(inode, new_fl))
1488 if (lease_break_time > 0) {
1489 break_time = jiffies + lease_break_time * HZ;
1490 if (break_time == 0)
1491 break_time++; /* so that 0 means no break time */
1494 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1495 if (!leases_conflict(fl, new_fl))
1498 if (fl->fl_flags & FL_UNLOCK_PENDING)
1500 fl->fl_flags |= FL_UNLOCK_PENDING;
1501 fl->fl_break_time = break_time;
1503 if (lease_breaking(fl))
1505 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1506 fl->fl_downgrade_time = break_time;
1508 if (fl->fl_lmops->lm_break(fl))
1509 locks_delete_lock_ctx(fl, &dispose);
1512 if (list_empty(&ctx->flc_lease))
1515 if (mode & O_NONBLOCK) {
1516 trace_break_lease_noblock(inode, new_fl);
1517 error = -EWOULDBLOCK;
1522 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1523 break_time = fl->fl_break_time;
1524 if (break_time != 0)
1525 break_time -= jiffies;
1526 if (break_time == 0)
1528 locks_insert_block(fl, new_fl, leases_conflict);
1529 trace_break_lease_block(inode, new_fl);
1530 spin_unlock(&ctx->flc_lock);
1531 percpu_up_read(&file_rwsem);
1533 locks_dispose_list(&dispose);
1534 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1535 list_empty(&new_fl->fl_blocked_member),
1538 percpu_down_read(&file_rwsem);
1539 spin_lock(&ctx->flc_lock);
1540 trace_break_lease_unblock(inode, new_fl);
1541 locks_delete_block(new_fl);
1544 * Wait for the next conflicting lease that has not been
1548 time_out_leases(inode, &dispose);
1549 if (any_leases_conflict(inode, new_fl))
1554 spin_unlock(&ctx->flc_lock);
1555 percpu_up_read(&file_rwsem);
1556 locks_dispose_list(&dispose);
1558 locks_free_lock(new_fl);
1561 EXPORT_SYMBOL(__break_lease);
1564 * lease_get_mtime - update modified time of an inode with exclusive lease
1566 * @time: pointer to a timespec which contains the last modified time
1568 * This is to force NFS clients to flush their caches for files with
1569 * exclusive leases. The justification is that if someone has an
1570 * exclusive lease, then they could be modifying it.
1572 void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1574 bool has_lease = false;
1575 struct file_lock_context *ctx;
1576 struct file_lock *fl;
1578 ctx = smp_load_acquire(&inode->i_flctx);
1579 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1580 spin_lock(&ctx->flc_lock);
1581 fl = list_first_entry_or_null(&ctx->flc_lease,
1582 struct file_lock, fl_list);
1583 if (fl && (fl->fl_type == F_WRLCK))
1585 spin_unlock(&ctx->flc_lock);
1589 *time = current_time(inode);
1591 EXPORT_SYMBOL(lease_get_mtime);
1594 * fcntl_getlease - Enquire what lease is currently active
1597 * The value returned by this function will be one of
1598 * (if no lease break is pending):
1600 * %F_RDLCK to indicate a shared lease is held.
1602 * %F_WRLCK to indicate an exclusive lease is held.
1604 * %F_UNLCK to indicate no lease is held.
1606 * (if a lease break is pending):
1608 * %F_RDLCK to indicate an exclusive lease needs to be
1609 * changed to a shared lease (or removed).
1611 * %F_UNLCK to indicate the lease needs to be removed.
1613 * XXX: sfr & willy disagree over whether F_INPROGRESS
1614 * should be returned to userspace.
1616 int fcntl_getlease(struct file *filp)
1618 struct file_lock *fl;
1619 struct inode *inode = locks_inode(filp);
1620 struct file_lock_context *ctx;
1624 ctx = smp_load_acquire(&inode->i_flctx);
1625 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1626 percpu_down_read(&file_rwsem);
1627 spin_lock(&ctx->flc_lock);
1628 time_out_leases(inode, &dispose);
1629 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1630 if (fl->fl_file != filp)
1632 type = target_leasetype(fl);
1635 spin_unlock(&ctx->flc_lock);
1636 percpu_up_read(&file_rwsem);
1638 locks_dispose_list(&dispose);
1644 * check_conflicting_open - see if the given file points to an inode that has
1645 * an existing open that would conflict with the
1647 * @filp: file to check
1648 * @arg: type of lease that we're trying to acquire
1649 * @flags: current lock flags
1651 * Check to see if there's an existing open fd on this file that would
1652 * conflict with the lease we're trying to set.
1655 check_conflicting_open(struct file *filp, const long arg, int flags)
1657 struct inode *inode = locks_inode(filp);
1658 int self_wcount = 0, self_rcount = 0;
1660 if (flags & FL_LAYOUT)
1662 if (flags & FL_DELEG)
1663 /* We leave these checks to the caller */
1667 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1668 else if (arg != F_WRLCK)
1672 * Make sure that only read/write count is from lease requestor.
1673 * Note that this will result in denying write leases when i_writecount
1674 * is negative, which is what we want. (We shouldn't grant write leases
1675 * on files open for execution.)
1677 if (filp->f_mode & FMODE_WRITE)
1679 else if (filp->f_mode & FMODE_READ)
1682 if (atomic_read(&inode->i_writecount) != self_wcount ||
1683 atomic_read(&inode->i_readcount) != self_rcount)
1690 generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1692 struct file_lock *fl, *my_fl = NULL, *lease;
1693 struct inode *inode = locks_inode(filp);
1694 struct file_lock_context *ctx;
1695 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1700 trace_generic_add_lease(inode, lease);
1702 /* Note that arg is never F_UNLCK here */
1703 ctx = locks_get_lock_context(inode, arg);
1708 * In the delegation case we need mutual exclusion with
1709 * a number of operations that take the i_mutex. We trylock
1710 * because delegations are an optional optimization, and if
1711 * there's some chance of a conflict--we'd rather not
1712 * bother, maybe that's a sign this just isn't a good file to
1713 * hand out a delegation on.
1715 if (is_deleg && !inode_trylock(inode))
1718 if (is_deleg && arg == F_WRLCK) {
1719 /* Write delegations are not currently supported: */
1720 inode_unlock(inode);
1725 percpu_down_read(&file_rwsem);
1726 spin_lock(&ctx->flc_lock);
1727 time_out_leases(inode, &dispose);
1728 error = check_conflicting_open(filp, arg, lease->fl_flags);
1733 * At this point, we know that if there is an exclusive
1734 * lease on this file, then we hold it on this filp
1735 * (otherwise our open of this file would have blocked).
1736 * And if we are trying to acquire an exclusive lease,
1737 * then the file is not open by anyone (including us)
1738 * except for this filp.
1741 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1742 if (fl->fl_file == filp &&
1743 fl->fl_owner == lease->fl_owner) {
1749 * No exclusive leases if someone else has a lease on
1755 * Modifying our existing lease is OK, but no getting a
1756 * new lease if someone else is opening for write:
1758 if (fl->fl_flags & FL_UNLOCK_PENDING)
1762 if (my_fl != NULL) {
1764 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1774 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1776 * The check in break_lease() is lockless. It's possible for another
1777 * open to race in after we did the earlier check for a conflicting
1778 * open but before the lease was inserted. Check again for a
1779 * conflicting open and cancel the lease if there is one.
1781 * We also add a barrier here to ensure that the insertion of the lock
1782 * precedes these checks.
1785 error = check_conflicting_open(filp, arg, lease->fl_flags);
1787 locks_unlink_lock_ctx(lease);
1792 if (lease->fl_lmops->lm_setup)
1793 lease->fl_lmops->lm_setup(lease, priv);
1795 spin_unlock(&ctx->flc_lock);
1796 percpu_up_read(&file_rwsem);
1797 locks_dispose_list(&dispose);
1799 inode_unlock(inode);
1800 if (!error && !my_fl)
1805 static int generic_delete_lease(struct file *filp, void *owner)
1807 int error = -EAGAIN;
1808 struct file_lock *fl, *victim = NULL;
1809 struct inode *inode = locks_inode(filp);
1810 struct file_lock_context *ctx;
1813 ctx = smp_load_acquire(&inode->i_flctx);
1815 trace_generic_delete_lease(inode, NULL);
1819 percpu_down_read(&file_rwsem);
1820 spin_lock(&ctx->flc_lock);
1821 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1822 if (fl->fl_file == filp &&
1823 fl->fl_owner == owner) {
1828 trace_generic_delete_lease(inode, victim);
1830 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1831 spin_unlock(&ctx->flc_lock);
1832 percpu_up_read(&file_rwsem);
1833 locks_dispose_list(&dispose);
1838 * generic_setlease - sets a lease on an open file
1839 * @filp: file pointer
1840 * @arg: type of lease to obtain
1841 * @flp: input - file_lock to use, output - file_lock inserted
1842 * @priv: private data for lm_setup (may be NULL if lm_setup
1843 * doesn't require it)
1845 * The (input) flp->fl_lmops->lm_break function is required
1848 int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1851 struct inode *inode = locks_inode(filp);
1854 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1856 if (!S_ISREG(inode->i_mode))
1858 error = security_file_lock(filp, arg);
1864 return generic_delete_lease(filp, *priv);
1867 if (!(*flp)->fl_lmops->lm_break) {
1872 return generic_add_lease(filp, arg, flp, priv);
1877 EXPORT_SYMBOL(generic_setlease);
1879 #if IS_ENABLED(CONFIG_SRCU)
1881 * Kernel subsystems can register to be notified on any attempt to set
1882 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1883 * to close files that it may have cached when there is an attempt to set a
1884 * conflicting lease.
1886 static struct srcu_notifier_head lease_notifier_chain;
1889 lease_notifier_chain_init(void)
1891 srcu_init_notifier_head(&lease_notifier_chain);
1895 setlease_notifier(long arg, struct file_lock *lease)
1898 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1901 int lease_register_notifier(struct notifier_block *nb)
1903 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1905 EXPORT_SYMBOL_GPL(lease_register_notifier);
1907 void lease_unregister_notifier(struct notifier_block *nb)
1909 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1911 EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1913 #else /* !IS_ENABLED(CONFIG_SRCU) */
1915 lease_notifier_chain_init(void)
1920 setlease_notifier(long arg, struct file_lock *lease)
1924 int lease_register_notifier(struct notifier_block *nb)
1928 EXPORT_SYMBOL_GPL(lease_register_notifier);
1930 void lease_unregister_notifier(struct notifier_block *nb)
1933 EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1935 #endif /* IS_ENABLED(CONFIG_SRCU) */
1938 * vfs_setlease - sets a lease on an open file
1939 * @filp: file pointer
1940 * @arg: type of lease to obtain
1941 * @lease: file_lock to use when adding a lease
1942 * @priv: private info for lm_setup when adding a lease (may be
1943 * NULL if lm_setup doesn't require it)
1945 * Call this to establish a lease on the file. The "lease" argument is not
1946 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1947 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1948 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1951 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1952 * may be NULL if the lm_setup operation doesn't require it.
1955 vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1958 setlease_notifier(arg, *lease);
1959 if (filp->f_op->setlease)
1960 return filp->f_op->setlease(filp, arg, lease, priv);
1962 return generic_setlease(filp, arg, lease, priv);
1964 EXPORT_SYMBOL_GPL(vfs_setlease);
1966 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1968 struct file_lock *fl;
1969 struct fasync_struct *new;
1972 fl = lease_alloc(filp, arg);
1976 new = fasync_alloc();
1978 locks_free_lock(fl);
1983 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1985 locks_free_lock(fl);
1992 * fcntl_setlease - sets a lease on an open file
1993 * @fd: open file descriptor
1994 * @filp: file pointer
1995 * @arg: type of lease to obtain
1997 * Call this fcntl to establish a lease on the file.
1998 * Note that you also need to call %F_SETSIG to
1999 * receive a signal when the lease is broken.
2001 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2004 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2005 return do_fcntl_add_lease(fd, filp, arg);
2009 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2010 * @inode: inode of the file to apply to
2011 * @fl: The lock to be applied
2013 * Apply a FLOCK style lock request to an inode.
2015 static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2020 error = flock_lock_inode(inode, fl);
2021 if (error != FILE_LOCK_DEFERRED)
2023 error = wait_event_interruptible(fl->fl_wait,
2024 list_empty(&fl->fl_blocked_member));
2028 locks_delete_block(fl);
2033 * locks_lock_inode_wait - Apply a lock to an inode
2034 * @inode: inode of the file to apply to
2035 * @fl: The lock to be applied
2037 * Apply a POSIX or FLOCK style lock request to an inode.
2039 int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2042 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2044 res = posix_lock_inode_wait(inode, fl);
2047 res = flock_lock_inode_wait(inode, fl);
2054 EXPORT_SYMBOL(locks_lock_inode_wait);
2057 * sys_flock: - flock() system call.
2058 * @fd: the file descriptor to lock.
2059 * @cmd: the type of lock to apply.
2061 * Apply a %FL_FLOCK style lock to an open file descriptor.
2062 * The @cmd can be one of:
2064 * - %LOCK_SH -- a shared lock.
2065 * - %LOCK_EX -- an exclusive lock.
2066 * - %LOCK_UN -- remove an existing lock.
2067 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2069 * %LOCK_MAND support has been removed from the kernel.
2071 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2073 struct fd f = fdget(fd);
2074 struct file_lock *lock;
2075 int can_sleep, unlock;
2082 can_sleep = !(cmd & LOCK_NB);
2084 unlock = (cmd == LOCK_UN);
2086 if (!unlock && !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2090 * LOCK_MAND locks were broken for a long time in that they never
2091 * conflicted with one another and didn't prevent any sort of open,
2092 * read or write activity.
2094 * Just ignore these requests now, to preserve legacy behavior, but
2095 * throw a warning to let people know that they don't actually work.
2097 if (cmd & LOCK_MAND) {
2098 pr_warn_once("Attempt to set a LOCK_MAND lock via flock(2). This support has been removed and the request ignored.\n");
2103 lock = flock_make_lock(f.file, cmd, NULL);
2105 error = PTR_ERR(lock);
2110 lock->fl_flags |= FL_SLEEP;
2112 error = security_file_lock(f.file, lock->fl_type);
2116 if (f.file->f_op->flock)
2117 error = f.file->f_op->flock(f.file,
2118 (can_sleep) ? F_SETLKW : F_SETLK,
2121 error = locks_lock_file_wait(f.file, lock);
2124 locks_free_lock(lock);
2133 * vfs_test_lock - test file byte range lock
2134 * @filp: The file to test lock for
2135 * @fl: The lock to test; also used to hold result
2137 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2138 * setting conf->fl_type to something other than F_UNLCK.
2140 int vfs_test_lock(struct file *filp, struct file_lock *fl)
2142 if (filp->f_op->lock)
2143 return filp->f_op->lock(filp, F_GETLK, fl);
2144 posix_test_lock(filp, fl);
2147 EXPORT_SYMBOL_GPL(vfs_test_lock);
2150 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2151 * @fl: The file_lock who's fl_pid should be translated
2152 * @ns: The namespace into which the pid should be translated
2154 * Used to tranlate a fl_pid into a namespace virtual pid number
2156 static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2163 if (IS_REMOTELCK(fl))
2166 * If the flock owner process is dead and its pid has been already
2167 * freed, the translation below won't work, but we still want to show
2168 * flock owner pid number in init pidns.
2170 if (ns == &init_pid_ns)
2171 return (pid_t)fl->fl_pid;
2174 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2175 vnr = pid_nr_ns(pid, ns);
2180 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2182 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2183 #if BITS_PER_LONG == 32
2185 * Make sure we can represent the posix lock via
2186 * legacy 32bit flock.
2188 if (fl->fl_start > OFFT_OFFSET_MAX)
2190 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2193 flock->l_start = fl->fl_start;
2194 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2195 fl->fl_end - fl->fl_start + 1;
2196 flock->l_whence = 0;
2197 flock->l_type = fl->fl_type;
2201 #if BITS_PER_LONG == 32
2202 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2204 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2205 flock->l_start = fl->fl_start;
2206 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2207 fl->fl_end - fl->fl_start + 1;
2208 flock->l_whence = 0;
2209 flock->l_type = fl->fl_type;
2213 /* Report the first existing lock that would conflict with l.
2214 * This implements the F_GETLK command of fcntl().
2216 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2218 struct file_lock *fl;
2221 fl = locks_alloc_lock();
2225 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2228 error = flock_to_posix_lock(filp, fl, flock);
2232 if (cmd == F_OFD_GETLK) {
2234 if (flock->l_pid != 0)
2237 fl->fl_flags |= FL_OFDLCK;
2238 fl->fl_owner = filp;
2241 error = vfs_test_lock(filp, fl);
2245 flock->l_type = fl->fl_type;
2246 if (fl->fl_type != F_UNLCK) {
2247 error = posix_lock_to_flock(flock, fl);
2252 locks_free_lock(fl);
2257 * vfs_lock_file - file byte range lock
2258 * @filp: The file to apply the lock to
2259 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2260 * @fl: The lock to be applied
2261 * @conf: Place to return a copy of the conflicting lock, if found.
2263 * A caller that doesn't care about the conflicting lock may pass NULL
2264 * as the final argument.
2266 * If the filesystem defines a private ->lock() method, then @conf will
2267 * be left unchanged; so a caller that cares should initialize it to
2268 * some acceptable default.
2270 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2271 * locks, the ->lock() interface may return asynchronously, before the lock has
2272 * been granted or denied by the underlying filesystem, if (and only if)
2273 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2274 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2275 * the request is for a blocking lock. When ->lock() does return asynchronously,
2276 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2277 * request completes.
2278 * If the request is for non-blocking lock the file system should return
2279 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2280 * with the result. If the request timed out the callback routine will return a
2281 * nonzero return code and the file system should release the lock. The file
2282 * system is also responsible to keep a corresponding posix lock when it
2283 * grants a lock so the VFS can find out which locks are locally held and do
2284 * the correct lock cleanup when required.
2285 * The underlying filesystem must not drop the kernel lock or call
2286 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2289 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2291 if (filp->f_op->lock)
2292 return filp->f_op->lock(filp, cmd, fl);
2294 return posix_lock_file(filp, fl, conf);
2296 EXPORT_SYMBOL_GPL(vfs_lock_file);
2298 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2299 struct file_lock *fl)
2303 error = security_file_lock(filp, fl->fl_type);
2308 error = vfs_lock_file(filp, cmd, fl, NULL);
2309 if (error != FILE_LOCK_DEFERRED)
2311 error = wait_event_interruptible(fl->fl_wait,
2312 list_empty(&fl->fl_blocked_member));
2316 locks_delete_block(fl);
2321 /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2323 check_fmode_for_setlk(struct file_lock *fl)
2325 switch (fl->fl_type) {
2327 if (!(fl->fl_file->f_mode & FMODE_READ))
2331 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2337 /* Apply the lock described by l to an open file descriptor.
2338 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2340 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2341 struct flock *flock)
2343 struct file_lock *file_lock = locks_alloc_lock();
2344 struct inode *inode = locks_inode(filp);
2348 if (file_lock == NULL)
2351 error = flock_to_posix_lock(filp, file_lock, flock);
2355 error = check_fmode_for_setlk(file_lock);
2360 * If the cmd is requesting file-private locks, then set the
2361 * FL_OFDLCK flag and override the owner.
2366 if (flock->l_pid != 0)
2370 file_lock->fl_flags |= FL_OFDLCK;
2371 file_lock->fl_owner = filp;
2375 if (flock->l_pid != 0)
2379 file_lock->fl_flags |= FL_OFDLCK;
2380 file_lock->fl_owner = filp;
2383 file_lock->fl_flags |= FL_SLEEP;
2386 error = do_lock_file_wait(filp, cmd, file_lock);
2389 * Attempt to detect a close/fcntl race and recover by releasing the
2390 * lock that was just acquired. There is no need to do that when we're
2391 * unlocking though, or for OFD locks.
2393 if (!error && file_lock->fl_type != F_UNLCK &&
2394 !(file_lock->fl_flags & FL_OFDLCK)) {
2395 struct files_struct *files = current->files;
2397 * We need that spin_lock here - it prevents reordering between
2398 * update of i_flctx->flc_posix and check for it done in
2399 * close(). rcu_read_lock() wouldn't do.
2401 spin_lock(&files->file_lock);
2402 f = files_lookup_fd_locked(files, fd);
2403 spin_unlock(&files->file_lock);
2405 file_lock->fl_type = F_UNLCK;
2406 error = do_lock_file_wait(filp, cmd, file_lock);
2407 WARN_ON_ONCE(error);
2412 trace_fcntl_setlk(inode, file_lock, error);
2413 locks_free_lock(file_lock);
2417 #if BITS_PER_LONG == 32
2418 /* Report the first existing lock that would conflict with l.
2419 * This implements the F_GETLK command of fcntl().
2421 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2423 struct file_lock *fl;
2426 fl = locks_alloc_lock();
2431 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2434 error = flock64_to_posix_lock(filp, fl, flock);
2438 if (cmd == F_OFD_GETLK) {
2440 if (flock->l_pid != 0)
2444 fl->fl_flags |= FL_OFDLCK;
2445 fl->fl_owner = filp;
2448 error = vfs_test_lock(filp, fl);
2452 flock->l_type = fl->fl_type;
2453 if (fl->fl_type != F_UNLCK)
2454 posix_lock_to_flock64(flock, fl);
2457 locks_free_lock(fl);
2461 /* Apply the lock described by l to an open file descriptor.
2462 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2464 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2465 struct flock64 *flock)
2467 struct file_lock *file_lock = locks_alloc_lock();
2471 if (file_lock == NULL)
2474 error = flock64_to_posix_lock(filp, file_lock, flock);
2478 error = check_fmode_for_setlk(file_lock);
2483 * If the cmd is requesting file-private locks, then set the
2484 * FL_OFDLCK flag and override the owner.
2489 if (flock->l_pid != 0)
2493 file_lock->fl_flags |= FL_OFDLCK;
2494 file_lock->fl_owner = filp;
2498 if (flock->l_pid != 0)
2502 file_lock->fl_flags |= FL_OFDLCK;
2503 file_lock->fl_owner = filp;
2506 file_lock->fl_flags |= FL_SLEEP;
2509 error = do_lock_file_wait(filp, cmd, file_lock);
2512 * Attempt to detect a close/fcntl race and recover by releasing the
2513 * lock that was just acquired. There is no need to do that when we're
2514 * unlocking though, or for OFD locks.
2516 if (!error && file_lock->fl_type != F_UNLCK &&
2517 !(file_lock->fl_flags & FL_OFDLCK)) {
2518 struct files_struct *files = current->files;
2520 * We need that spin_lock here - it prevents reordering between
2521 * update of i_flctx->flc_posix and check for it done in
2522 * close(). rcu_read_lock() wouldn't do.
2524 spin_lock(&files->file_lock);
2525 f = files_lookup_fd_locked(files, fd);
2526 spin_unlock(&files->file_lock);
2528 file_lock->fl_type = F_UNLCK;
2529 error = do_lock_file_wait(filp, cmd, file_lock);
2530 WARN_ON_ONCE(error);
2535 locks_free_lock(file_lock);
2538 #endif /* BITS_PER_LONG == 32 */
2541 * This function is called when the file is being removed
2542 * from the task's fd array. POSIX locks belonging to this task
2543 * are deleted at this time.
2545 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2548 struct inode *inode = locks_inode(filp);
2549 struct file_lock lock;
2550 struct file_lock_context *ctx;
2553 * If there are no locks held on this file, we don't need to call
2554 * posix_lock_file(). Another process could be setting a lock on this
2555 * file at the same time, but we wouldn't remove that lock anyway.
2557 ctx = smp_load_acquire(&inode->i_flctx);
2558 if (!ctx || list_empty(&ctx->flc_posix))
2561 locks_init_lock(&lock);
2562 lock.fl_type = F_UNLCK;
2563 lock.fl_flags = FL_POSIX | FL_CLOSE;
2565 lock.fl_end = OFFSET_MAX;
2566 lock.fl_owner = owner;
2567 lock.fl_pid = current->tgid;
2568 lock.fl_file = filp;
2570 lock.fl_lmops = NULL;
2572 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2574 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2575 lock.fl_ops->fl_release_private(&lock);
2576 trace_locks_remove_posix(inode, &lock, error);
2578 EXPORT_SYMBOL(locks_remove_posix);
2580 /* The i_flctx must be valid when calling into here */
2582 locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2584 struct file_lock fl;
2585 struct inode *inode = locks_inode(filp);
2587 if (list_empty(&flctx->flc_flock))
2590 flock_make_lock(filp, LOCK_UN, &fl);
2591 fl.fl_flags |= FL_CLOSE;
2593 if (filp->f_op->flock)
2594 filp->f_op->flock(filp, F_SETLKW, &fl);
2596 flock_lock_inode(inode, &fl);
2598 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2599 fl.fl_ops->fl_release_private(&fl);
2602 /* The i_flctx must be valid when calling into here */
2604 locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2606 struct file_lock *fl, *tmp;
2609 if (list_empty(&ctx->flc_lease))
2612 percpu_down_read(&file_rwsem);
2613 spin_lock(&ctx->flc_lock);
2614 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2615 if (filp == fl->fl_file)
2616 lease_modify(fl, F_UNLCK, &dispose);
2617 spin_unlock(&ctx->flc_lock);
2618 percpu_up_read(&file_rwsem);
2620 locks_dispose_list(&dispose);
2624 * This function is called on the last close of an open file.
2626 void locks_remove_file(struct file *filp)
2628 struct file_lock_context *ctx;
2630 ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2634 /* remove any OFD locks */
2635 locks_remove_posix(filp, filp);
2637 /* remove flock locks */
2638 locks_remove_flock(filp, ctx);
2640 /* remove any leases */
2641 locks_remove_lease(filp, ctx);
2643 spin_lock(&ctx->flc_lock);
2644 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2645 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2646 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2647 spin_unlock(&ctx->flc_lock);
2651 * vfs_cancel_lock - file byte range unblock lock
2652 * @filp: The file to apply the unblock to
2653 * @fl: The lock to be unblocked
2655 * Used by lock managers to cancel blocked requests
2657 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2659 if (filp->f_op->lock)
2660 return filp->f_op->lock(filp, F_CANCELLK, fl);
2663 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2665 #ifdef CONFIG_PROC_FS
2666 #include <linux/proc_fs.h>
2667 #include <linux/seq_file.h>
2669 struct locks_iterator {
2674 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2675 loff_t id, char *pfx, int repeat)
2677 struct inode *inode = NULL;
2678 unsigned int fl_pid;
2679 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2682 fl_pid = locks_translate_pid(fl, proc_pidns);
2684 * If lock owner is dead (and pid is freed) or not visible in current
2685 * pidns, zero is shown as a pid value. Check lock info from
2686 * init_pid_ns to get saved lock pid value.
2689 if (fl->fl_file != NULL)
2690 inode = locks_inode(fl->fl_file);
2692 seq_printf(f, "%lld: ", id);
2695 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2698 if (fl->fl_flags & FL_ACCESS)
2699 seq_puts(f, "ACCESS");
2700 else if (IS_OFDLCK(fl))
2701 seq_puts(f, "OFDLCK");
2703 seq_puts(f, "POSIX ");
2705 seq_printf(f, " %s ",
2706 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2707 } else if (IS_FLOCK(fl)) {
2708 seq_puts(f, "FLOCK ADVISORY ");
2709 } else if (IS_LEASE(fl)) {
2710 if (fl->fl_flags & FL_DELEG)
2711 seq_puts(f, "DELEG ");
2713 seq_puts(f, "LEASE ");
2715 if (lease_breaking(fl))
2716 seq_puts(f, "BREAKING ");
2717 else if (fl->fl_file)
2718 seq_puts(f, "ACTIVE ");
2720 seq_puts(f, "BREAKER ");
2722 seq_puts(f, "UNKNOWN UNKNOWN ");
2724 type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2726 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2727 (type == F_RDLCK) ? "READ" : "UNLCK");
2729 /* userspace relies on this representation of dev_t */
2730 seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2731 MAJOR(inode->i_sb->s_dev),
2732 MINOR(inode->i_sb->s_dev), inode->i_ino);
2734 seq_printf(f, "%d <none>:0 ", fl_pid);
2737 if (fl->fl_end == OFFSET_MAX)
2738 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2740 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2742 seq_puts(f, "0 EOF\n");
2746 static struct file_lock *get_next_blocked_member(struct file_lock *node)
2748 struct file_lock *tmp;
2750 /* NULL node or root node */
2751 if (node == NULL || node->fl_blocker == NULL)
2754 /* Next member in the linked list could be itself */
2755 tmp = list_next_entry(node, fl_blocked_member);
2756 if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2764 static int locks_show(struct seq_file *f, void *v)
2766 struct locks_iterator *iter = f->private;
2767 struct file_lock *cur, *tmp;
2768 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2771 cur = hlist_entry(v, struct file_lock, fl_link);
2773 if (locks_translate_pid(cur, proc_pidns) == 0)
2776 /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2777 * is the left child of current node, the next silibing in fl_blocked_member is the
2778 * right child, we can alse get the parent of current node from fl_blocker, so this
2779 * question becomes traversal of a binary tree
2781 while (cur != NULL) {
2783 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2785 lock_get_status(f, cur, iter->li_pos, "", level);
2787 if (!list_empty(&cur->fl_blocked_requests)) {
2789 cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2790 struct file_lock, fl_blocked_member);
2794 tmp = get_next_blocked_member(cur);
2795 /* Fall back to parent node */
2796 while (tmp == NULL && cur->fl_blocker != NULL) {
2797 cur = cur->fl_blocker;
2799 tmp = get_next_blocked_member(cur);
2808 static void __show_fd_locks(struct seq_file *f,
2809 struct list_head *head, int *id,
2810 struct file *filp, struct files_struct *files)
2812 struct file_lock *fl;
2814 list_for_each_entry(fl, head, fl_list) {
2816 if (filp != fl->fl_file)
2818 if (fl->fl_owner != files &&
2819 fl->fl_owner != filp)
2823 seq_puts(f, "lock:\t");
2824 lock_get_status(f, fl, *id, "", 0);
2828 void show_fd_locks(struct seq_file *f,
2829 struct file *filp, struct files_struct *files)
2831 struct inode *inode = locks_inode(filp);
2832 struct file_lock_context *ctx;
2835 ctx = smp_load_acquire(&inode->i_flctx);
2839 spin_lock(&ctx->flc_lock);
2840 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2841 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2842 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2843 spin_unlock(&ctx->flc_lock);
2846 static void *locks_start(struct seq_file *f, loff_t *pos)
2847 __acquires(&blocked_lock_lock)
2849 struct locks_iterator *iter = f->private;
2851 iter->li_pos = *pos + 1;
2852 percpu_down_write(&file_rwsem);
2853 spin_lock(&blocked_lock_lock);
2854 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2857 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2859 struct locks_iterator *iter = f->private;
2862 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2865 static void locks_stop(struct seq_file *f, void *v)
2866 __releases(&blocked_lock_lock)
2868 spin_unlock(&blocked_lock_lock);
2869 percpu_up_write(&file_rwsem);
2872 static const struct seq_operations locks_seq_operations = {
2873 .start = locks_start,
2879 static int __init proc_locks_init(void)
2881 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2882 sizeof(struct locks_iterator), NULL);
2885 fs_initcall(proc_locks_init);
2888 static int __init filelock_init(void)
2892 flctx_cache = kmem_cache_create("file_lock_ctx",
2893 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2895 filelock_cache = kmem_cache_create("file_lock_cache",
2896 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2898 for_each_possible_cpu(i) {
2899 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2901 spin_lock_init(&fll->lock);
2902 INIT_HLIST_HEAD(&fll->hlist);
2905 lease_notifier_chain_init();
2908 core_initcall(filelock_init);