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>
55 #include <linux/filelock.h>
57 #include <linux/init.h>
58 #include <linux/security.h>
59 #include <linux/slab.h>
60 #include <linux/syscalls.h>
61 #include <linux/time.h>
62 #include <linux/rcupdate.h>
63 #include <linux/pid_namespace.h>
64 #include <linux/hashtable.h>
65 #include <linux/percpu.h>
66 #include <linux/sysctl.h>
68 #define CREATE_TRACE_POINTS
69 #include <trace/events/filelock.h>
71 #include <linux/uaccess.h>
73 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
74 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
75 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
76 #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
77 #define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
79 static bool lease_breaking(struct file_lock
*fl
)
81 return fl
->fl_flags
& (FL_UNLOCK_PENDING
| FL_DOWNGRADE_PENDING
);
84 static int target_leasetype(struct file_lock
*fl
)
86 if (fl
->fl_flags
& FL_UNLOCK_PENDING
)
88 if (fl
->fl_flags
& FL_DOWNGRADE_PENDING
)
93 static int leases_enable
= 1;
94 static int lease_break_time
= 45;
97 static struct ctl_table locks_sysctls
[] = {
99 .procname
= "leases-enable",
100 .data
= &leases_enable
,
101 .maxlen
= sizeof(int),
103 .proc_handler
= proc_dointvec
,
107 .procname
= "lease-break-time",
108 .data
= &lease_break_time
,
109 .maxlen
= sizeof(int),
111 .proc_handler
= proc_dointvec
,
113 #endif /* CONFIG_MMU */
117 static int __init
init_fs_locks_sysctls(void)
119 register_sysctl_init("fs", locks_sysctls
);
122 early_initcall(init_fs_locks_sysctls
);
123 #endif /* CONFIG_SYSCTL */
126 * The global file_lock_list is only used for displaying /proc/locks, so we
127 * keep a list on each CPU, with each list protected by its own spinlock.
128 * Global serialization is done using file_rwsem.
130 * Note that alterations to the list also require that the relevant flc_lock is
133 struct file_lock_list_struct
{
135 struct hlist_head hlist
;
137 static DEFINE_PER_CPU(struct file_lock_list_struct
, file_lock_list
);
138 DEFINE_STATIC_PERCPU_RWSEM(file_rwsem
);
142 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
143 * It is protected by blocked_lock_lock.
145 * We hash locks by lockowner in order to optimize searching for the lock a
146 * particular lockowner is waiting on.
148 * FIXME: make this value scale via some heuristic? We generally will want more
149 * buckets when we have more lockowners holding locks, but that's a little
150 * difficult to determine without knowing what the workload will look like.
152 #define BLOCKED_HASH_BITS 7
153 static DEFINE_HASHTABLE(blocked_hash
, BLOCKED_HASH_BITS
);
156 * This lock protects the blocked_hash. Generally, if you're accessing it, you
157 * want to be holding this lock.
159 * In addition, it also protects the fl->fl_blocked_requests list, and the
160 * fl->fl_blocker pointer for file_lock structures that are acting as lock
161 * requests (in contrast to those that are acting as records of acquired locks).
163 * Note that when we acquire this lock in order to change the above fields,
164 * we often hold the flc_lock as well. In certain cases, when reading the fields
165 * protected by this lock, we can skip acquiring it iff we already hold the
168 static DEFINE_SPINLOCK(blocked_lock_lock
);
170 static struct kmem_cache
*flctx_cache __ro_after_init
;
171 static struct kmem_cache
*filelock_cache __ro_after_init
;
173 static struct file_lock_context
*
174 locks_get_lock_context(struct inode
*inode
, int type
)
176 struct file_lock_context
*ctx
;
178 /* paired with cmpxchg() below */
179 ctx
= locks_inode_context(inode
);
180 if (likely(ctx
) || type
== F_UNLCK
)
183 ctx
= kmem_cache_alloc(flctx_cache
, GFP_KERNEL
);
187 spin_lock_init(&ctx
->flc_lock
);
188 INIT_LIST_HEAD(&ctx
->flc_flock
);
189 INIT_LIST_HEAD(&ctx
->flc_posix
);
190 INIT_LIST_HEAD(&ctx
->flc_lease
);
193 * Assign the pointer if it's not already assigned. If it is, then
194 * free the context we just allocated.
196 if (cmpxchg(&inode
->i_flctx
, NULL
, ctx
)) {
197 kmem_cache_free(flctx_cache
, ctx
);
198 ctx
= locks_inode_context(inode
);
201 trace_locks_get_lock_context(inode
, type
, ctx
);
206 locks_dump_ctx_list(struct list_head
*list
, char *list_type
)
208 struct file_lock
*fl
;
210 list_for_each_entry(fl
, list
, fl_list
) {
211 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
);
216 locks_check_ctx_lists(struct inode
*inode
)
218 struct file_lock_context
*ctx
= inode
->i_flctx
;
220 if (unlikely(!list_empty(&ctx
->flc_flock
) ||
221 !list_empty(&ctx
->flc_posix
) ||
222 !list_empty(&ctx
->flc_lease
))) {
223 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
224 MAJOR(inode
->i_sb
->s_dev
), MINOR(inode
->i_sb
->s_dev
),
226 locks_dump_ctx_list(&ctx
->flc_flock
, "FLOCK");
227 locks_dump_ctx_list(&ctx
->flc_posix
, "POSIX");
228 locks_dump_ctx_list(&ctx
->flc_lease
, "LEASE");
233 locks_check_ctx_file_list(struct file
*filp
, struct list_head
*list
,
236 struct file_lock
*fl
;
237 struct inode
*inode
= file_inode(filp
);
239 list_for_each_entry(fl
, list
, fl_list
)
240 if (fl
->fl_file
== filp
)
241 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
242 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
243 list_type
, MAJOR(inode
->i_sb
->s_dev
),
244 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
,
245 fl
->fl_owner
, fl
->fl_flags
, fl
->fl_type
, fl
->fl_pid
);
249 locks_free_lock_context(struct inode
*inode
)
251 struct file_lock_context
*ctx
= locks_inode_context(inode
);
254 locks_check_ctx_lists(inode
);
255 kmem_cache_free(flctx_cache
, ctx
);
259 static void locks_init_lock_heads(struct file_lock
*fl
)
261 INIT_HLIST_NODE(&fl
->fl_link
);
262 INIT_LIST_HEAD(&fl
->fl_list
);
263 INIT_LIST_HEAD(&fl
->fl_blocked_requests
);
264 INIT_LIST_HEAD(&fl
->fl_blocked_member
);
265 init_waitqueue_head(&fl
->fl_wait
);
268 /* Allocate an empty lock structure. */
269 struct file_lock
*locks_alloc_lock(void)
271 struct file_lock
*fl
= kmem_cache_zalloc(filelock_cache
, GFP_KERNEL
);
274 locks_init_lock_heads(fl
);
278 EXPORT_SYMBOL_GPL(locks_alloc_lock
);
280 void locks_release_private(struct file_lock
*fl
)
282 BUG_ON(waitqueue_active(&fl
->fl_wait
));
283 BUG_ON(!list_empty(&fl
->fl_list
));
284 BUG_ON(!list_empty(&fl
->fl_blocked_requests
));
285 BUG_ON(!list_empty(&fl
->fl_blocked_member
));
286 BUG_ON(!hlist_unhashed(&fl
->fl_link
));
289 if (fl
->fl_ops
->fl_release_private
)
290 fl
->fl_ops
->fl_release_private(fl
);
295 if (fl
->fl_lmops
->lm_put_owner
) {
296 fl
->fl_lmops
->lm_put_owner(fl
->fl_owner
);
302 EXPORT_SYMBOL_GPL(locks_release_private
);
305 * locks_owner_has_blockers - Check for blocking lock requests
306 * @flctx: file lock context
310 * %true: @owner has at least one blocker
311 * %false: @owner has no blockers
313 bool locks_owner_has_blockers(struct file_lock_context
*flctx
,
316 struct file_lock
*fl
;
318 spin_lock(&flctx
->flc_lock
);
319 list_for_each_entry(fl
, &flctx
->flc_posix
, fl_list
) {
320 if (fl
->fl_owner
!= owner
)
322 if (!list_empty(&fl
->fl_blocked_requests
)) {
323 spin_unlock(&flctx
->flc_lock
);
327 spin_unlock(&flctx
->flc_lock
);
330 EXPORT_SYMBOL_GPL(locks_owner_has_blockers
);
332 /* Free a lock which is not in use. */
333 void locks_free_lock(struct file_lock
*fl
)
335 locks_release_private(fl
);
336 kmem_cache_free(filelock_cache
, fl
);
338 EXPORT_SYMBOL(locks_free_lock
);
341 locks_dispose_list(struct list_head
*dispose
)
343 struct file_lock
*fl
;
345 while (!list_empty(dispose
)) {
346 fl
= list_first_entry(dispose
, struct file_lock
, fl_list
);
347 list_del_init(&fl
->fl_list
);
352 void locks_init_lock(struct file_lock
*fl
)
354 memset(fl
, 0, sizeof(struct file_lock
));
355 locks_init_lock_heads(fl
);
357 EXPORT_SYMBOL(locks_init_lock
);
360 * Initialize a new lock from an existing file_lock structure.
362 void locks_copy_conflock(struct file_lock
*new, struct file_lock
*fl
)
364 new->fl_owner
= fl
->fl_owner
;
365 new->fl_pid
= fl
->fl_pid
;
367 new->fl_flags
= fl
->fl_flags
;
368 new->fl_type
= fl
->fl_type
;
369 new->fl_start
= fl
->fl_start
;
370 new->fl_end
= fl
->fl_end
;
371 new->fl_lmops
= fl
->fl_lmops
;
375 if (fl
->fl_lmops
->lm_get_owner
)
376 fl
->fl_lmops
->lm_get_owner(fl
->fl_owner
);
379 EXPORT_SYMBOL(locks_copy_conflock
);
381 void locks_copy_lock(struct file_lock
*new, struct file_lock
*fl
)
383 /* "new" must be a freshly-initialized lock */
384 WARN_ON_ONCE(new->fl_ops
);
386 locks_copy_conflock(new, fl
);
388 new->fl_file
= fl
->fl_file
;
389 new->fl_ops
= fl
->fl_ops
;
392 if (fl
->fl_ops
->fl_copy_lock
)
393 fl
->fl_ops
->fl_copy_lock(new, fl
);
396 EXPORT_SYMBOL(locks_copy_lock
);
398 static void locks_move_blocks(struct file_lock
*new, struct file_lock
*fl
)
403 * As ctx->flc_lock is held, new requests cannot be added to
404 * ->fl_blocked_requests, so we don't need a lock to check if it
407 if (list_empty(&fl
->fl_blocked_requests
))
409 spin_lock(&blocked_lock_lock
);
410 list_splice_init(&fl
->fl_blocked_requests
, &new->fl_blocked_requests
);
411 list_for_each_entry(f
, &new->fl_blocked_requests
, fl_blocked_member
)
413 spin_unlock(&blocked_lock_lock
);
416 static inline int flock_translate_cmd(int cmd
) {
428 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
429 static void flock_make_lock(struct file
*filp
, struct file_lock
*fl
, int type
)
435 fl
->fl_pid
= current
->tgid
;
436 fl
->fl_flags
= FL_FLOCK
;
438 fl
->fl_end
= OFFSET_MAX
;
441 static int assign_type(struct file_lock
*fl
, int type
)
455 static int flock64_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
458 switch (l
->l_whence
) {
463 fl
->fl_start
= filp
->f_pos
;
466 fl
->fl_start
= i_size_read(file_inode(filp
));
471 if (l
->l_start
> OFFSET_MAX
- fl
->fl_start
)
473 fl
->fl_start
+= l
->l_start
;
474 if (fl
->fl_start
< 0)
477 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
478 POSIX-2001 defines it. */
480 if (l
->l_len
- 1 > OFFSET_MAX
- fl
->fl_start
)
482 fl
->fl_end
= fl
->fl_start
+ (l
->l_len
- 1);
484 } else if (l
->l_len
< 0) {
485 if (fl
->fl_start
+ l
->l_len
< 0)
487 fl
->fl_end
= fl
->fl_start
- 1;
488 fl
->fl_start
+= l
->l_len
;
490 fl
->fl_end
= OFFSET_MAX
;
492 fl
->fl_owner
= current
->files
;
493 fl
->fl_pid
= current
->tgid
;
495 fl
->fl_flags
= FL_POSIX
;
499 return assign_type(fl
, l
->l_type
);
502 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
505 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
508 struct flock64 ll
= {
510 .l_whence
= l
->l_whence
,
511 .l_start
= l
->l_start
,
515 return flock64_to_posix_lock(filp
, fl
, &ll
);
518 /* default lease lock manager operations */
520 lease_break_callback(struct file_lock
*fl
)
522 kill_fasync(&fl
->fl_fasync
, SIGIO
, POLL_MSG
);
527 lease_setup(struct file_lock
*fl
, void **priv
)
529 struct file
*filp
= fl
->fl_file
;
530 struct fasync_struct
*fa
= *priv
;
533 * fasync_insert_entry() returns the old entry if any. If there was no
534 * old entry, then it used "priv" and inserted it into the fasync list.
535 * Clear the pointer to indicate that it shouldn't be freed.
537 if (!fasync_insert_entry(fa
->fa_fd
, filp
, &fl
->fl_fasync
, fa
))
540 __f_setown(filp
, task_pid(current
), PIDTYPE_TGID
, 0);
543 static const struct lock_manager_operations lease_manager_ops
= {
544 .lm_break
= lease_break_callback
,
545 .lm_change
= lease_modify
,
546 .lm_setup
= lease_setup
,
550 * Initialize a lease, use the default lock manager operations
552 static int lease_init(struct file
*filp
, int type
, struct file_lock
*fl
)
554 if (assign_type(fl
, type
) != 0)
558 fl
->fl_pid
= current
->tgid
;
561 fl
->fl_flags
= FL_LEASE
;
563 fl
->fl_end
= OFFSET_MAX
;
565 fl
->fl_lmops
= &lease_manager_ops
;
569 /* Allocate a file_lock initialised to this type of lease */
570 static struct file_lock
*lease_alloc(struct file
*filp
, int type
)
572 struct file_lock
*fl
= locks_alloc_lock();
576 return ERR_PTR(error
);
578 error
= lease_init(filp
, type
, fl
);
581 return ERR_PTR(error
);
586 /* Check if two locks overlap each other.
588 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
590 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
591 (fl2
->fl_end
>= fl1
->fl_start
));
595 * Check whether two locks have the same owner.
597 static int posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
599 return fl1
->fl_owner
== fl2
->fl_owner
;
602 /* Must be called with the flc_lock held! */
603 static void locks_insert_global_locks(struct file_lock
*fl
)
605 struct file_lock_list_struct
*fll
= this_cpu_ptr(&file_lock_list
);
607 percpu_rwsem_assert_held(&file_rwsem
);
609 spin_lock(&fll
->lock
);
610 fl
->fl_link_cpu
= smp_processor_id();
611 hlist_add_head(&fl
->fl_link
, &fll
->hlist
);
612 spin_unlock(&fll
->lock
);
615 /* Must be called with the flc_lock held! */
616 static void locks_delete_global_locks(struct file_lock
*fl
)
618 struct file_lock_list_struct
*fll
;
620 percpu_rwsem_assert_held(&file_rwsem
);
623 * Avoid taking lock if already unhashed. This is safe since this check
624 * is done while holding the flc_lock, and new insertions into the list
625 * also require that it be held.
627 if (hlist_unhashed(&fl
->fl_link
))
630 fll
= per_cpu_ptr(&file_lock_list
, fl
->fl_link_cpu
);
631 spin_lock(&fll
->lock
);
632 hlist_del_init(&fl
->fl_link
);
633 spin_unlock(&fll
->lock
);
637 posix_owner_key(struct file_lock
*fl
)
639 return (unsigned long)fl
->fl_owner
;
642 static void locks_insert_global_blocked(struct file_lock
*waiter
)
644 lockdep_assert_held(&blocked_lock_lock
);
646 hash_add(blocked_hash
, &waiter
->fl_link
, posix_owner_key(waiter
));
649 static void locks_delete_global_blocked(struct file_lock
*waiter
)
651 lockdep_assert_held(&blocked_lock_lock
);
653 hash_del(&waiter
->fl_link
);
656 /* Remove waiter from blocker's block list.
657 * When blocker ends up pointing to itself then the list is empty.
659 * Must be called with blocked_lock_lock held.
661 static void __locks_delete_block(struct file_lock
*waiter
)
663 locks_delete_global_blocked(waiter
);
664 list_del_init(&waiter
->fl_blocked_member
);
667 static void __locks_wake_up_blocks(struct file_lock
*blocker
)
669 while (!list_empty(&blocker
->fl_blocked_requests
)) {
670 struct file_lock
*waiter
;
672 waiter
= list_first_entry(&blocker
->fl_blocked_requests
,
673 struct file_lock
, fl_blocked_member
);
674 __locks_delete_block(waiter
);
675 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->lm_notify
)
676 waiter
->fl_lmops
->lm_notify(waiter
);
678 wake_up(&waiter
->fl_wait
);
681 * The setting of fl_blocker to NULL marks the "done"
682 * point in deleting a block. Paired with acquire at the top
683 * of locks_delete_block().
685 smp_store_release(&waiter
->fl_blocker
, NULL
);
690 * locks_delete_block - stop waiting for a file lock
691 * @waiter: the lock which was waiting
693 * lockd/nfsd need to disconnect the lock while working on it.
695 int locks_delete_block(struct file_lock
*waiter
)
697 int status
= -ENOENT
;
700 * If fl_blocker is NULL, it won't be set again as this thread "owns"
701 * the lock and is the only one that might try to claim the lock.
703 * We use acquire/release to manage fl_blocker so that we can
704 * optimize away taking the blocked_lock_lock in many cases.
706 * The smp_load_acquire guarantees two things:
708 * 1/ that fl_blocked_requests can be tested locklessly. If something
709 * was recently added to that list it must have been in a locked region
710 * *before* the locked region when fl_blocker was set to NULL.
712 * 2/ that no other thread is accessing 'waiter', so it is safe to free
713 * it. __locks_wake_up_blocks is careful not to touch waiter after
714 * fl_blocker is released.
716 * If a lockless check of fl_blocker shows it to be NULL, we know that
717 * no new locks can be inserted into its fl_blocked_requests list, and
718 * can avoid doing anything further if the list is empty.
720 if (!smp_load_acquire(&waiter
->fl_blocker
) &&
721 list_empty(&waiter
->fl_blocked_requests
))
724 spin_lock(&blocked_lock_lock
);
725 if (waiter
->fl_blocker
)
727 __locks_wake_up_blocks(waiter
);
728 __locks_delete_block(waiter
);
731 * The setting of fl_blocker to NULL marks the "done" point in deleting
732 * a block. Paired with acquire at the top of this function.
734 smp_store_release(&waiter
->fl_blocker
, NULL
);
735 spin_unlock(&blocked_lock_lock
);
738 EXPORT_SYMBOL(locks_delete_block
);
740 /* Insert waiter into blocker's block list.
741 * We use a circular list so that processes can be easily woken up in
742 * the order they blocked. The documentation doesn't require this but
743 * it seems like the reasonable thing to do.
745 * Must be called with both the flc_lock and blocked_lock_lock held. The
746 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
747 * but by ensuring that the flc_lock is also held on insertions we can avoid
748 * taking the blocked_lock_lock in some cases when we see that the
749 * fl_blocked_requests list is empty.
751 * Rather than just adding to the list, we check for conflicts with any existing
752 * waiters, and add beneath any waiter that blocks the new waiter.
753 * Thus wakeups don't happen until needed.
755 static void __locks_insert_block(struct file_lock
*blocker
,
756 struct file_lock
*waiter
,
757 bool conflict(struct file_lock
*,
760 struct file_lock
*fl
;
761 BUG_ON(!list_empty(&waiter
->fl_blocked_member
));
764 list_for_each_entry(fl
, &blocker
->fl_blocked_requests
, fl_blocked_member
)
765 if (conflict(fl
, waiter
)) {
769 waiter
->fl_blocker
= blocker
;
770 list_add_tail(&waiter
->fl_blocked_member
, &blocker
->fl_blocked_requests
);
771 if (IS_POSIX(blocker
) && !IS_OFDLCK(blocker
))
772 locks_insert_global_blocked(waiter
);
774 /* The requests in waiter->fl_blocked are known to conflict with
775 * waiter, but might not conflict with blocker, or the requests
776 * and lock which block it. So they all need to be woken.
778 __locks_wake_up_blocks(waiter
);
781 /* Must be called with flc_lock held. */
782 static void locks_insert_block(struct file_lock
*blocker
,
783 struct file_lock
*waiter
,
784 bool conflict(struct file_lock
*,
787 spin_lock(&blocked_lock_lock
);
788 __locks_insert_block(blocker
, waiter
, conflict
);
789 spin_unlock(&blocked_lock_lock
);
793 * Wake up processes blocked waiting for blocker.
795 * Must be called with the inode->flc_lock held!
797 static void locks_wake_up_blocks(struct file_lock
*blocker
)
800 * Avoid taking global lock if list is empty. This is safe since new
801 * blocked requests are only added to the list under the flc_lock, and
802 * the flc_lock is always held here. Note that removal from the
803 * fl_blocked_requests list does not require the flc_lock, so we must
804 * recheck list_empty() after acquiring the blocked_lock_lock.
806 if (list_empty(&blocker
->fl_blocked_requests
))
809 spin_lock(&blocked_lock_lock
);
810 __locks_wake_up_blocks(blocker
);
811 spin_unlock(&blocked_lock_lock
);
815 locks_insert_lock_ctx(struct file_lock
*fl
, struct list_head
*before
)
817 list_add_tail(&fl
->fl_list
, before
);
818 locks_insert_global_locks(fl
);
822 locks_unlink_lock_ctx(struct file_lock
*fl
)
824 locks_delete_global_locks(fl
);
825 list_del_init(&fl
->fl_list
);
826 locks_wake_up_blocks(fl
);
830 locks_delete_lock_ctx(struct file_lock
*fl
, struct list_head
*dispose
)
832 locks_unlink_lock_ctx(fl
);
834 list_add(&fl
->fl_list
, dispose
);
839 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
840 * checks for shared/exclusive status of overlapping locks.
842 static bool locks_conflict(struct file_lock
*caller_fl
,
843 struct file_lock
*sys_fl
)
845 if (sys_fl
->fl_type
== F_WRLCK
)
847 if (caller_fl
->fl_type
== F_WRLCK
)
852 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
853 * checking before calling the locks_conflict().
855 static bool posix_locks_conflict(struct file_lock
*caller_fl
,
856 struct file_lock
*sys_fl
)
858 /* POSIX locks owned by the same process do not conflict with
861 if (posix_same_owner(caller_fl
, sys_fl
))
864 /* Check whether they overlap */
865 if (!locks_overlap(caller_fl
, sys_fl
))
868 return locks_conflict(caller_fl
, sys_fl
);
871 /* Determine if lock sys_fl blocks lock caller_fl. Used on xx_GETLK
872 * path so checks for additional GETLK-specific things like F_UNLCK.
874 static bool posix_test_locks_conflict(struct file_lock
*caller_fl
,
875 struct file_lock
*sys_fl
)
877 /* F_UNLCK checks any locks on the same fd. */
878 if (caller_fl
->fl_type
== F_UNLCK
) {
879 if (!posix_same_owner(caller_fl
, sys_fl
))
881 return locks_overlap(caller_fl
, sys_fl
);
883 return posix_locks_conflict(caller_fl
, sys_fl
);
886 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
887 * checking before calling the locks_conflict().
889 static bool flock_locks_conflict(struct file_lock
*caller_fl
,
890 struct file_lock
*sys_fl
)
892 /* FLOCK locks referring to the same filp do not conflict with
895 if (caller_fl
->fl_file
== sys_fl
->fl_file
)
898 return locks_conflict(caller_fl
, sys_fl
);
902 posix_test_lock(struct file
*filp
, struct file_lock
*fl
)
904 struct file_lock
*cfl
;
905 struct file_lock_context
*ctx
;
906 struct inode
*inode
= file_inode(filp
);
910 ctx
= locks_inode_context(inode
);
911 if (!ctx
|| list_empty_careful(&ctx
->flc_posix
)) {
912 fl
->fl_type
= F_UNLCK
;
917 spin_lock(&ctx
->flc_lock
);
918 list_for_each_entry(cfl
, &ctx
->flc_posix
, fl_list
) {
919 if (!posix_test_locks_conflict(fl
, cfl
))
921 if (cfl
->fl_lmops
&& cfl
->fl_lmops
->lm_lock_expirable
922 && (*cfl
->fl_lmops
->lm_lock_expirable
)(cfl
)) {
923 owner
= cfl
->fl_lmops
->lm_mod_owner
;
924 func
= cfl
->fl_lmops
->lm_expire_lock
;
926 spin_unlock(&ctx
->flc_lock
);
931 locks_copy_conflock(fl
, cfl
);
934 fl
->fl_type
= F_UNLCK
;
936 spin_unlock(&ctx
->flc_lock
);
939 EXPORT_SYMBOL(posix_test_lock
);
942 * Deadlock detection:
944 * We attempt to detect deadlocks that are due purely to posix file
947 * We assume that a task can be waiting for at most one lock at a time.
948 * So for any acquired lock, the process holding that lock may be
949 * waiting on at most one other lock. That lock in turns may be held by
950 * someone waiting for at most one other lock. Given a requested lock
951 * caller_fl which is about to wait for a conflicting lock block_fl, we
952 * follow this chain of waiters to ensure we are not about to create a
955 * Since we do this before we ever put a process to sleep on a lock, we
956 * are ensured that there is never a cycle; that is what guarantees that
957 * the while() loop in posix_locks_deadlock() eventually completes.
959 * Note: the above assumption may not be true when handling lock
960 * requests from a broken NFS client. It may also fail in the presence
961 * of tasks (such as posix threads) sharing the same open file table.
962 * To handle those cases, we just bail out after a few iterations.
964 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
965 * Because the owner is not even nominally tied to a thread of
966 * execution, the deadlock detection below can't reasonably work well. Just
969 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
970 * locks that just checks for the case where two tasks are attempting to
971 * upgrade from read to write locks on the same inode.
974 #define MAX_DEADLK_ITERATIONS 10
976 /* Find a lock that the owner of the given block_fl is blocking on. */
977 static struct file_lock
*what_owner_is_waiting_for(struct file_lock
*block_fl
)
979 struct file_lock
*fl
;
981 hash_for_each_possible(blocked_hash
, fl
, fl_link
, posix_owner_key(block_fl
)) {
982 if (posix_same_owner(fl
, block_fl
)) {
983 while (fl
->fl_blocker
)
991 /* Must be called with the blocked_lock_lock held! */
992 static int posix_locks_deadlock(struct file_lock
*caller_fl
,
993 struct file_lock
*block_fl
)
997 lockdep_assert_held(&blocked_lock_lock
);
1000 * This deadlock detector can't reasonably detect deadlocks with
1001 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1003 if (IS_OFDLCK(caller_fl
))
1006 while ((block_fl
= what_owner_is_waiting_for(block_fl
))) {
1007 if (i
++ > MAX_DEADLK_ITERATIONS
)
1009 if (posix_same_owner(caller_fl
, block_fl
))
1015 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1016 * after any leases, but before any posix locks.
1018 * Note that if called with an FL_EXISTS argument, the caller may determine
1019 * whether or not a lock was successfully freed by testing the return
1020 * value for -ENOENT.
1022 static int flock_lock_inode(struct inode
*inode
, struct file_lock
*request
)
1024 struct file_lock
*new_fl
= NULL
;
1025 struct file_lock
*fl
;
1026 struct file_lock_context
*ctx
;
1031 ctx
= locks_get_lock_context(inode
, request
->fl_type
);
1033 if (request
->fl_type
!= F_UNLCK
)
1035 return (request
->fl_flags
& FL_EXISTS
) ? -ENOENT
: 0;
1038 if (!(request
->fl_flags
& FL_ACCESS
) && (request
->fl_type
!= F_UNLCK
)) {
1039 new_fl
= locks_alloc_lock();
1044 percpu_down_read(&file_rwsem
);
1045 spin_lock(&ctx
->flc_lock
);
1046 if (request
->fl_flags
& FL_ACCESS
)
1049 list_for_each_entry(fl
, &ctx
->flc_flock
, fl_list
) {
1050 if (request
->fl_file
!= fl
->fl_file
)
1052 if (request
->fl_type
== fl
->fl_type
)
1055 locks_delete_lock_ctx(fl
, &dispose
);
1059 if (request
->fl_type
== F_UNLCK
) {
1060 if ((request
->fl_flags
& FL_EXISTS
) && !found
)
1066 list_for_each_entry(fl
, &ctx
->flc_flock
, fl_list
) {
1067 if (!flock_locks_conflict(request
, fl
))
1070 if (!(request
->fl_flags
& FL_SLEEP
))
1072 error
= FILE_LOCK_DEFERRED
;
1073 locks_insert_block(fl
, request
, flock_locks_conflict
);
1076 if (request
->fl_flags
& FL_ACCESS
)
1078 locks_copy_lock(new_fl
, request
);
1079 locks_move_blocks(new_fl
, request
);
1080 locks_insert_lock_ctx(new_fl
, &ctx
->flc_flock
);
1085 spin_unlock(&ctx
->flc_lock
);
1086 percpu_up_read(&file_rwsem
);
1088 locks_free_lock(new_fl
);
1089 locks_dispose_list(&dispose
);
1090 trace_flock_lock_inode(inode
, request
, error
);
1094 static int posix_lock_inode(struct inode
*inode
, struct file_lock
*request
,
1095 struct file_lock
*conflock
)
1097 struct file_lock
*fl
, *tmp
;
1098 struct file_lock
*new_fl
= NULL
;
1099 struct file_lock
*new_fl2
= NULL
;
1100 struct file_lock
*left
= NULL
;
1101 struct file_lock
*right
= NULL
;
1102 struct file_lock_context
*ctx
;
1109 ctx
= locks_get_lock_context(inode
, request
->fl_type
);
1111 return (request
->fl_type
== F_UNLCK
) ? 0 : -ENOMEM
;
1114 * We may need two file_lock structures for this operation,
1115 * so we get them in advance to avoid races.
1117 * In some cases we can be sure, that no new locks will be needed
1119 if (!(request
->fl_flags
& FL_ACCESS
) &&
1120 (request
->fl_type
!= F_UNLCK
||
1121 request
->fl_start
!= 0 || request
->fl_end
!= OFFSET_MAX
)) {
1122 new_fl
= locks_alloc_lock();
1123 new_fl2
= locks_alloc_lock();
1127 percpu_down_read(&file_rwsem
);
1128 spin_lock(&ctx
->flc_lock
);
1130 * New lock request. Walk all POSIX locks and look for conflicts. If
1131 * there are any, either return error or put the request on the
1132 * blocker's list of waiters and the global blocked_hash.
1134 if (request
->fl_type
!= F_UNLCK
) {
1135 list_for_each_entry(fl
, &ctx
->flc_posix
, fl_list
) {
1136 if (!posix_locks_conflict(request
, fl
))
1138 if (fl
->fl_lmops
&& fl
->fl_lmops
->lm_lock_expirable
1139 && (*fl
->fl_lmops
->lm_lock_expirable
)(fl
)) {
1140 owner
= fl
->fl_lmops
->lm_mod_owner
;
1141 func
= fl
->fl_lmops
->lm_expire_lock
;
1142 __module_get(owner
);
1143 spin_unlock(&ctx
->flc_lock
);
1144 percpu_up_read(&file_rwsem
);
1150 locks_copy_conflock(conflock
, fl
);
1152 if (!(request
->fl_flags
& FL_SLEEP
))
1155 * Deadlock detection and insertion into the blocked
1156 * locks list must be done while holding the same lock!
1159 spin_lock(&blocked_lock_lock
);
1161 * Ensure that we don't find any locks blocked on this
1162 * request during deadlock detection.
1164 __locks_wake_up_blocks(request
);
1165 if (likely(!posix_locks_deadlock(request
, fl
))) {
1166 error
= FILE_LOCK_DEFERRED
;
1167 __locks_insert_block(fl
, request
,
1168 posix_locks_conflict
);
1170 spin_unlock(&blocked_lock_lock
);
1175 /* If we're just looking for a conflict, we're done. */
1177 if (request
->fl_flags
& FL_ACCESS
)
1180 /* Find the first old lock with the same owner as the new lock */
1181 list_for_each_entry(fl
, &ctx
->flc_posix
, fl_list
) {
1182 if (posix_same_owner(request
, fl
))
1186 /* Process locks with this owner. */
1187 list_for_each_entry_safe_from(fl
, tmp
, &ctx
->flc_posix
, fl_list
) {
1188 if (!posix_same_owner(request
, fl
))
1191 /* Detect adjacent or overlapping regions (if same lock type) */
1192 if (request
->fl_type
== fl
->fl_type
) {
1193 /* In all comparisons of start vs end, use
1194 * "start - 1" rather than "end + 1". If end
1195 * is OFFSET_MAX, end + 1 will become negative.
1197 if (fl
->fl_end
< request
->fl_start
- 1)
1199 /* If the next lock in the list has entirely bigger
1200 * addresses than the new one, insert the lock here.
1202 if (fl
->fl_start
- 1 > request
->fl_end
)
1205 /* If we come here, the new and old lock are of the
1206 * same type and adjacent or overlapping. Make one
1207 * lock yielding from the lower start address of both
1208 * locks to the higher end address.
1210 if (fl
->fl_start
> request
->fl_start
)
1211 fl
->fl_start
= request
->fl_start
;
1213 request
->fl_start
= fl
->fl_start
;
1214 if (fl
->fl_end
< request
->fl_end
)
1215 fl
->fl_end
= request
->fl_end
;
1217 request
->fl_end
= fl
->fl_end
;
1219 locks_delete_lock_ctx(fl
, &dispose
);
1225 /* Processing for different lock types is a bit
1228 if (fl
->fl_end
< request
->fl_start
)
1230 if (fl
->fl_start
> request
->fl_end
)
1232 if (request
->fl_type
== F_UNLCK
)
1234 if (fl
->fl_start
< request
->fl_start
)
1236 /* If the next lock in the list has a higher end
1237 * address than the new one, insert the new one here.
1239 if (fl
->fl_end
> request
->fl_end
) {
1243 if (fl
->fl_start
>= request
->fl_start
) {
1244 /* The new lock completely replaces an old
1245 * one (This may happen several times).
1248 locks_delete_lock_ctx(fl
, &dispose
);
1252 * Replace the old lock with new_fl, and
1253 * remove the old one. It's safe to do the
1254 * insert here since we know that we won't be
1255 * using new_fl later, and that the lock is
1256 * just replacing an existing lock.
1261 locks_copy_lock(new_fl
, request
);
1262 locks_move_blocks(new_fl
, request
);
1265 locks_insert_lock_ctx(request
, &fl
->fl_list
);
1266 locks_delete_lock_ctx(fl
, &dispose
);
1273 * The above code only modifies existing locks in case of merging or
1274 * replacing. If new lock(s) need to be inserted all modifications are
1275 * done below this, so it's safe yet to bail out.
1277 error
= -ENOLCK
; /* "no luck" */
1278 if (right
&& left
== right
&& !new_fl2
)
1283 if (request
->fl_type
== F_UNLCK
) {
1284 if (request
->fl_flags
& FL_EXISTS
)
1293 locks_copy_lock(new_fl
, request
);
1294 locks_move_blocks(new_fl
, request
);
1295 locks_insert_lock_ctx(new_fl
, &fl
->fl_list
);
1300 if (left
== right
) {
1301 /* The new lock breaks the old one in two pieces,
1302 * so we have to use the second new lock.
1306 locks_copy_lock(left
, right
);
1307 locks_insert_lock_ctx(left
, &fl
->fl_list
);
1309 right
->fl_start
= request
->fl_end
+ 1;
1310 locks_wake_up_blocks(right
);
1313 left
->fl_end
= request
->fl_start
- 1;
1314 locks_wake_up_blocks(left
);
1317 spin_unlock(&ctx
->flc_lock
);
1318 percpu_up_read(&file_rwsem
);
1319 trace_posix_lock_inode(inode
, request
, error
);
1321 * Free any unused locks.
1324 locks_free_lock(new_fl
);
1326 locks_free_lock(new_fl2
);
1327 locks_dispose_list(&dispose
);
1333 * posix_lock_file - Apply a POSIX-style lock to a file
1334 * @filp: The file to apply the lock to
1335 * @fl: The lock to be applied
1336 * @conflock: Place to return a copy of the conflicting lock, if found.
1338 * Add a POSIX style lock to a file.
1339 * We merge adjacent & overlapping locks whenever possible.
1340 * POSIX locks are sorted by owner task, then by starting address
1342 * Note that if called with an FL_EXISTS argument, the caller may determine
1343 * whether or not a lock was successfully freed by testing the return
1344 * value for -ENOENT.
1346 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
,
1347 struct file_lock
*conflock
)
1349 return posix_lock_inode(file_inode(filp
), fl
, conflock
);
1351 EXPORT_SYMBOL(posix_lock_file
);
1354 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1355 * @inode: inode of file to which lock request should be applied
1356 * @fl: The lock to be applied
1358 * Apply a POSIX style lock request to an inode.
1360 static int posix_lock_inode_wait(struct inode
*inode
, struct file_lock
*fl
)
1365 error
= posix_lock_inode(inode
, fl
, NULL
);
1366 if (error
!= FILE_LOCK_DEFERRED
)
1368 error
= wait_event_interruptible(fl
->fl_wait
,
1369 list_empty(&fl
->fl_blocked_member
));
1373 locks_delete_block(fl
);
1377 static void lease_clear_pending(struct file_lock
*fl
, int arg
)
1381 fl
->fl_flags
&= ~FL_UNLOCK_PENDING
;
1384 fl
->fl_flags
&= ~FL_DOWNGRADE_PENDING
;
1388 /* We already had a lease on this file; just change its type */
1389 int lease_modify(struct file_lock
*fl
, int arg
, struct list_head
*dispose
)
1391 int error
= assign_type(fl
, arg
);
1395 lease_clear_pending(fl
, arg
);
1396 locks_wake_up_blocks(fl
);
1397 if (arg
== F_UNLCK
) {
1398 struct file
*filp
= fl
->fl_file
;
1401 filp
->f_owner
.signum
= 0;
1402 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
1403 if (fl
->fl_fasync
!= NULL
) {
1404 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
1405 fl
->fl_fasync
= NULL
;
1407 locks_delete_lock_ctx(fl
, dispose
);
1411 EXPORT_SYMBOL(lease_modify
);
1413 static bool past_time(unsigned long then
)
1416 /* 0 is a special value meaning "this never expires": */
1418 return time_after(jiffies
, then
);
1421 static void time_out_leases(struct inode
*inode
, struct list_head
*dispose
)
1423 struct file_lock_context
*ctx
= inode
->i_flctx
;
1424 struct file_lock
*fl
, *tmp
;
1426 lockdep_assert_held(&ctx
->flc_lock
);
1428 list_for_each_entry_safe(fl
, tmp
, &ctx
->flc_lease
, fl_list
) {
1429 trace_time_out_leases(inode
, fl
);
1430 if (past_time(fl
->fl_downgrade_time
))
1431 lease_modify(fl
, F_RDLCK
, dispose
);
1432 if (past_time(fl
->fl_break_time
))
1433 lease_modify(fl
, F_UNLCK
, dispose
);
1437 static bool leases_conflict(struct file_lock
*lease
, struct file_lock
*breaker
)
1441 if (lease
->fl_lmops
->lm_breaker_owns_lease
1442 && lease
->fl_lmops
->lm_breaker_owns_lease(lease
))
1444 if ((breaker
->fl_flags
& FL_LAYOUT
) != (lease
->fl_flags
& FL_LAYOUT
)) {
1448 if ((breaker
->fl_flags
& FL_DELEG
) && (lease
->fl_flags
& FL_LEASE
)) {
1453 rc
= locks_conflict(breaker
, lease
);
1455 trace_leases_conflict(rc
, lease
, breaker
);
1460 any_leases_conflict(struct inode
*inode
, struct file_lock
*breaker
)
1462 struct file_lock_context
*ctx
= inode
->i_flctx
;
1463 struct file_lock
*fl
;
1465 lockdep_assert_held(&ctx
->flc_lock
);
1467 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1468 if (leases_conflict(fl
, breaker
))
1475 * __break_lease - revoke all outstanding leases on file
1476 * @inode: the inode of the file to return
1477 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1479 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1482 * break_lease (inlined for speed) has checked there already is at least
1483 * some kind of lock (maybe a lease) on this file. Leases are broken on
1484 * a call to open() or truncate(). This function can sleep unless you
1485 * specified %O_NONBLOCK to your open().
1487 int __break_lease(struct inode
*inode
, unsigned int mode
, unsigned int type
)
1490 struct file_lock_context
*ctx
;
1491 struct file_lock
*new_fl
, *fl
, *tmp
;
1492 unsigned long break_time
;
1493 int want_write
= (mode
& O_ACCMODE
) != O_RDONLY
;
1496 new_fl
= lease_alloc(NULL
, want_write
? F_WRLCK
: F_RDLCK
);
1498 return PTR_ERR(new_fl
);
1499 new_fl
->fl_flags
= type
;
1501 /* typically we will check that ctx is non-NULL before calling */
1502 ctx
= locks_inode_context(inode
);
1508 percpu_down_read(&file_rwsem
);
1509 spin_lock(&ctx
->flc_lock
);
1511 time_out_leases(inode
, &dispose
);
1513 if (!any_leases_conflict(inode
, new_fl
))
1517 if (lease_break_time
> 0) {
1518 break_time
= jiffies
+ lease_break_time
* HZ
;
1519 if (break_time
== 0)
1520 break_time
++; /* so that 0 means no break time */
1523 list_for_each_entry_safe(fl
, tmp
, &ctx
->flc_lease
, fl_list
) {
1524 if (!leases_conflict(fl
, new_fl
))
1527 if (fl
->fl_flags
& FL_UNLOCK_PENDING
)
1529 fl
->fl_flags
|= FL_UNLOCK_PENDING
;
1530 fl
->fl_break_time
= break_time
;
1532 if (lease_breaking(fl
))
1534 fl
->fl_flags
|= FL_DOWNGRADE_PENDING
;
1535 fl
->fl_downgrade_time
= break_time
;
1537 if (fl
->fl_lmops
->lm_break(fl
))
1538 locks_delete_lock_ctx(fl
, &dispose
);
1541 if (list_empty(&ctx
->flc_lease
))
1544 if (mode
& O_NONBLOCK
) {
1545 trace_break_lease_noblock(inode
, new_fl
);
1546 error
= -EWOULDBLOCK
;
1551 fl
= list_first_entry(&ctx
->flc_lease
, struct file_lock
, fl_list
);
1552 break_time
= fl
->fl_break_time
;
1553 if (break_time
!= 0)
1554 break_time
-= jiffies
;
1555 if (break_time
== 0)
1557 locks_insert_block(fl
, new_fl
, leases_conflict
);
1558 trace_break_lease_block(inode
, new_fl
);
1559 spin_unlock(&ctx
->flc_lock
);
1560 percpu_up_read(&file_rwsem
);
1562 locks_dispose_list(&dispose
);
1563 error
= wait_event_interruptible_timeout(new_fl
->fl_wait
,
1564 list_empty(&new_fl
->fl_blocked_member
),
1567 percpu_down_read(&file_rwsem
);
1568 spin_lock(&ctx
->flc_lock
);
1569 trace_break_lease_unblock(inode
, new_fl
);
1570 locks_delete_block(new_fl
);
1573 * Wait for the next conflicting lease that has not been
1577 time_out_leases(inode
, &dispose
);
1578 if (any_leases_conflict(inode
, new_fl
))
1583 spin_unlock(&ctx
->flc_lock
);
1584 percpu_up_read(&file_rwsem
);
1585 locks_dispose_list(&dispose
);
1587 locks_free_lock(new_fl
);
1590 EXPORT_SYMBOL(__break_lease
);
1593 * lease_get_mtime - update modified time of an inode with exclusive lease
1595 * @time: pointer to a timespec which contains the last modified time
1597 * This is to force NFS clients to flush their caches for files with
1598 * exclusive leases. The justification is that if someone has an
1599 * exclusive lease, then they could be modifying it.
1601 void lease_get_mtime(struct inode
*inode
, struct timespec64
*time
)
1603 bool has_lease
= false;
1604 struct file_lock_context
*ctx
;
1605 struct file_lock
*fl
;
1607 ctx
= locks_inode_context(inode
);
1608 if (ctx
&& !list_empty_careful(&ctx
->flc_lease
)) {
1609 spin_lock(&ctx
->flc_lock
);
1610 fl
= list_first_entry_or_null(&ctx
->flc_lease
,
1611 struct file_lock
, fl_list
);
1612 if (fl
&& (fl
->fl_type
== F_WRLCK
))
1614 spin_unlock(&ctx
->flc_lock
);
1618 *time
= current_time(inode
);
1620 EXPORT_SYMBOL(lease_get_mtime
);
1623 * fcntl_getlease - Enquire what lease is currently active
1626 * The value returned by this function will be one of
1627 * (if no lease break is pending):
1629 * %F_RDLCK to indicate a shared lease is held.
1631 * %F_WRLCK to indicate an exclusive lease is held.
1633 * %F_UNLCK to indicate no lease is held.
1635 * (if a lease break is pending):
1637 * %F_RDLCK to indicate an exclusive lease needs to be
1638 * changed to a shared lease (or removed).
1640 * %F_UNLCK to indicate the lease needs to be removed.
1642 * XXX: sfr & willy disagree over whether F_INPROGRESS
1643 * should be returned to userspace.
1645 int fcntl_getlease(struct file
*filp
)
1647 struct file_lock
*fl
;
1648 struct inode
*inode
= file_inode(filp
);
1649 struct file_lock_context
*ctx
;
1653 ctx
= locks_inode_context(inode
);
1654 if (ctx
&& !list_empty_careful(&ctx
->flc_lease
)) {
1655 percpu_down_read(&file_rwsem
);
1656 spin_lock(&ctx
->flc_lock
);
1657 time_out_leases(inode
, &dispose
);
1658 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1659 if (fl
->fl_file
!= filp
)
1661 type
= target_leasetype(fl
);
1664 spin_unlock(&ctx
->flc_lock
);
1665 percpu_up_read(&file_rwsem
);
1667 locks_dispose_list(&dispose
);
1673 * check_conflicting_open - see if the given file points to an inode that has
1674 * an existing open that would conflict with the
1676 * @filp: file to check
1677 * @arg: type of lease that we're trying to acquire
1678 * @flags: current lock flags
1680 * Check to see if there's an existing open fd on this file that would
1681 * conflict with the lease we're trying to set.
1684 check_conflicting_open(struct file
*filp
, const int arg
, int flags
)
1686 struct inode
*inode
= file_inode(filp
);
1687 int self_wcount
= 0, self_rcount
= 0;
1689 if (flags
& FL_LAYOUT
)
1691 if (flags
& FL_DELEG
)
1692 /* We leave these checks to the caller */
1696 return inode_is_open_for_write(inode
) ? -EAGAIN
: 0;
1697 else if (arg
!= F_WRLCK
)
1701 * Make sure that only read/write count is from lease requestor.
1702 * Note that this will result in denying write leases when i_writecount
1703 * is negative, which is what we want. (We shouldn't grant write leases
1704 * on files open for execution.)
1706 if (filp
->f_mode
& FMODE_WRITE
)
1708 else if (filp
->f_mode
& FMODE_READ
)
1711 if (atomic_read(&inode
->i_writecount
) != self_wcount
||
1712 atomic_read(&inode
->i_readcount
) != self_rcount
)
1719 generic_add_lease(struct file
*filp
, int arg
, struct file_lock
**flp
, void **priv
)
1721 struct file_lock
*fl
, *my_fl
= NULL
, *lease
;
1722 struct inode
*inode
= file_inode(filp
);
1723 struct file_lock_context
*ctx
;
1724 bool is_deleg
= (*flp
)->fl_flags
& FL_DELEG
;
1729 trace_generic_add_lease(inode
, lease
);
1731 /* Note that arg is never F_UNLCK here */
1732 ctx
= locks_get_lock_context(inode
, arg
);
1737 * In the delegation case we need mutual exclusion with
1738 * a number of operations that take the i_mutex. We trylock
1739 * because delegations are an optional optimization, and if
1740 * there's some chance of a conflict--we'd rather not
1741 * bother, maybe that's a sign this just isn't a good file to
1742 * hand out a delegation on.
1744 if (is_deleg
&& !inode_trylock(inode
))
1747 percpu_down_read(&file_rwsem
);
1748 spin_lock(&ctx
->flc_lock
);
1749 time_out_leases(inode
, &dispose
);
1750 error
= check_conflicting_open(filp
, arg
, lease
->fl_flags
);
1755 * At this point, we know that if there is an exclusive
1756 * lease on this file, then we hold it on this filp
1757 * (otherwise our open of this file would have blocked).
1758 * And if we are trying to acquire an exclusive lease,
1759 * then the file is not open by anyone (including us)
1760 * except for this filp.
1763 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1764 if (fl
->fl_file
== filp
&&
1765 fl
->fl_owner
== lease
->fl_owner
) {
1771 * No exclusive leases if someone else has a lease on
1777 * Modifying our existing lease is OK, but no getting a
1778 * new lease if someone else is opening for write:
1780 if (fl
->fl_flags
& FL_UNLOCK_PENDING
)
1784 if (my_fl
!= NULL
) {
1786 error
= lease
->fl_lmops
->lm_change(lease
, arg
, &dispose
);
1796 locks_insert_lock_ctx(lease
, &ctx
->flc_lease
);
1798 * The check in break_lease() is lockless. It's possible for another
1799 * open to race in after we did the earlier check for a conflicting
1800 * open but before the lease was inserted. Check again for a
1801 * conflicting open and cancel the lease if there is one.
1803 * We also add a barrier here to ensure that the insertion of the lock
1804 * precedes these checks.
1807 error
= check_conflicting_open(filp
, arg
, lease
->fl_flags
);
1809 locks_unlink_lock_ctx(lease
);
1814 if (lease
->fl_lmops
->lm_setup
)
1815 lease
->fl_lmops
->lm_setup(lease
, priv
);
1817 spin_unlock(&ctx
->flc_lock
);
1818 percpu_up_read(&file_rwsem
);
1819 locks_dispose_list(&dispose
);
1821 inode_unlock(inode
);
1822 if (!error
&& !my_fl
)
1827 static int generic_delete_lease(struct file
*filp
, void *owner
)
1829 int error
= -EAGAIN
;
1830 struct file_lock
*fl
, *victim
= NULL
;
1831 struct inode
*inode
= file_inode(filp
);
1832 struct file_lock_context
*ctx
;
1835 ctx
= locks_inode_context(inode
);
1837 trace_generic_delete_lease(inode
, NULL
);
1841 percpu_down_read(&file_rwsem
);
1842 spin_lock(&ctx
->flc_lock
);
1843 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1844 if (fl
->fl_file
== filp
&&
1845 fl
->fl_owner
== owner
) {
1850 trace_generic_delete_lease(inode
, victim
);
1852 error
= fl
->fl_lmops
->lm_change(victim
, F_UNLCK
, &dispose
);
1853 spin_unlock(&ctx
->flc_lock
);
1854 percpu_up_read(&file_rwsem
);
1855 locks_dispose_list(&dispose
);
1860 * generic_setlease - sets a lease on an open file
1861 * @filp: file pointer
1862 * @arg: type of lease to obtain
1863 * @flp: input - file_lock to use, output - file_lock inserted
1864 * @priv: private data for lm_setup (may be NULL if lm_setup
1865 * doesn't require it)
1867 * The (input) flp->fl_lmops->lm_break function is required
1870 int generic_setlease(struct file
*filp
, int arg
, struct file_lock
**flp
,
1873 struct inode
*inode
= file_inode(filp
);
1874 vfsuid_t vfsuid
= i_uid_into_vfsuid(file_mnt_idmap(filp
), inode
);
1877 if ((!vfsuid_eq_kuid(vfsuid
, current_fsuid())) && !capable(CAP_LEASE
))
1879 if (!S_ISREG(inode
->i_mode
))
1881 error
= security_file_lock(filp
, arg
);
1887 return generic_delete_lease(filp
, *priv
);
1890 if (!(*flp
)->fl_lmops
->lm_break
) {
1895 return generic_add_lease(filp
, arg
, flp
, priv
);
1900 EXPORT_SYMBOL(generic_setlease
);
1903 * Kernel subsystems can register to be notified on any attempt to set
1904 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1905 * to close files that it may have cached when there is an attempt to set a
1906 * conflicting lease.
1908 static struct srcu_notifier_head lease_notifier_chain
;
1911 lease_notifier_chain_init(void)
1913 srcu_init_notifier_head(&lease_notifier_chain
);
1917 setlease_notifier(int arg
, struct file_lock
*lease
)
1920 srcu_notifier_call_chain(&lease_notifier_chain
, arg
, lease
);
1923 int lease_register_notifier(struct notifier_block
*nb
)
1925 return srcu_notifier_chain_register(&lease_notifier_chain
, nb
);
1927 EXPORT_SYMBOL_GPL(lease_register_notifier
);
1929 void lease_unregister_notifier(struct notifier_block
*nb
)
1931 srcu_notifier_chain_unregister(&lease_notifier_chain
, nb
);
1933 EXPORT_SYMBOL_GPL(lease_unregister_notifier
);
1936 * vfs_setlease - sets a lease on an open file
1937 * @filp: file pointer
1938 * @arg: type of lease to obtain
1939 * @lease: file_lock to use when adding a lease
1940 * @priv: private info for lm_setup when adding a lease (may be
1941 * NULL if lm_setup doesn't require it)
1943 * Call this to establish a lease on the file. The "lease" argument is not
1944 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1945 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1946 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1949 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1950 * may be NULL if the lm_setup operation doesn't require it.
1953 vfs_setlease(struct file
*filp
, int arg
, struct file_lock
**lease
, void **priv
)
1956 setlease_notifier(arg
, *lease
);
1957 if (filp
->f_op
->setlease
)
1958 return filp
->f_op
->setlease(filp
, arg
, lease
, priv
);
1960 return generic_setlease(filp
, arg
, lease
, priv
);
1962 EXPORT_SYMBOL_GPL(vfs_setlease
);
1964 static int do_fcntl_add_lease(unsigned int fd
, struct file
*filp
, int arg
)
1966 struct file_lock
*fl
;
1967 struct fasync_struct
*new;
1970 fl
= lease_alloc(filp
, arg
);
1974 new = fasync_alloc();
1976 locks_free_lock(fl
);
1981 error
= vfs_setlease(filp
, arg
, &fl
, (void **)&new);
1983 locks_free_lock(fl
);
1990 * fcntl_setlease - sets a lease on an open file
1991 * @fd: open file descriptor
1992 * @filp: file pointer
1993 * @arg: type of lease to obtain
1995 * Call this fcntl to establish a lease on the file.
1996 * Note that you also need to call %F_SETSIG to
1997 * receive a signal when the lease is broken.
1999 int fcntl_setlease(unsigned int fd
, struct file
*filp
, int arg
)
2002 return vfs_setlease(filp
, F_UNLCK
, NULL
, (void **)&filp
);
2003 return do_fcntl_add_lease(fd
, filp
, arg
);
2007 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2008 * @inode: inode of the file to apply to
2009 * @fl: The lock to be applied
2011 * Apply a FLOCK style lock request to an inode.
2013 static int flock_lock_inode_wait(struct inode
*inode
, struct file_lock
*fl
)
2018 error
= flock_lock_inode(inode
, fl
);
2019 if (error
!= FILE_LOCK_DEFERRED
)
2021 error
= wait_event_interruptible(fl
->fl_wait
,
2022 list_empty(&fl
->fl_blocked_member
));
2026 locks_delete_block(fl
);
2031 * locks_lock_inode_wait - Apply a lock to an inode
2032 * @inode: inode of the file to apply to
2033 * @fl: The lock to be applied
2035 * Apply a POSIX or FLOCK style lock request to an inode.
2037 int locks_lock_inode_wait(struct inode
*inode
, struct file_lock
*fl
)
2040 switch (fl
->fl_flags
& (FL_POSIX
|FL_FLOCK
)) {
2042 res
= posix_lock_inode_wait(inode
, fl
);
2045 res
= flock_lock_inode_wait(inode
, fl
);
2052 EXPORT_SYMBOL(locks_lock_inode_wait
);
2055 * sys_flock: - flock() system call.
2056 * @fd: the file descriptor to lock.
2057 * @cmd: the type of lock to apply.
2059 * Apply a %FL_FLOCK style lock to an open file descriptor.
2060 * The @cmd can be one of:
2062 * - %LOCK_SH -- a shared lock.
2063 * - %LOCK_EX -- an exclusive lock.
2064 * - %LOCK_UN -- remove an existing lock.
2065 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2067 * %LOCK_MAND support has been removed from the kernel.
2069 SYSCALL_DEFINE2(flock
, unsigned int, fd
, unsigned int, cmd
)
2071 int can_sleep
, error
, type
;
2072 struct file_lock fl
;
2076 * LOCK_MAND locks were broken for a long time in that they never
2077 * conflicted with one another and didn't prevent any sort of open,
2078 * read or write activity.
2080 * Just ignore these requests now, to preserve legacy behavior, but
2081 * throw a warning to let people know that they don't actually work.
2083 if (cmd
& LOCK_MAND
) {
2084 pr_warn_once("%s(%d): Attempt to set a LOCK_MAND lock via flock(2). This support has been removed and the request ignored.\n", current
->comm
, current
->pid
);
2088 type
= flock_translate_cmd(cmd
& ~LOCK_NB
);
2097 if (type
!= F_UNLCK
&& !(f
.file
->f_mode
& (FMODE_READ
| FMODE_WRITE
)))
2100 flock_make_lock(f
.file
, &fl
, type
);
2102 error
= security_file_lock(f
.file
, fl
.fl_type
);
2106 can_sleep
= !(cmd
& LOCK_NB
);
2108 fl
.fl_flags
|= FL_SLEEP
;
2110 if (f
.file
->f_op
->flock
)
2111 error
= f
.file
->f_op
->flock(f
.file
,
2112 (can_sleep
) ? F_SETLKW
: F_SETLK
,
2115 error
= locks_lock_file_wait(f
.file
, &fl
);
2117 locks_release_private(&fl
);
2125 * vfs_test_lock - test file byte range lock
2126 * @filp: The file to test lock for
2127 * @fl: The lock to test; also used to hold result
2129 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2130 * setting conf->fl_type to something other than F_UNLCK.
2132 int vfs_test_lock(struct file
*filp
, struct file_lock
*fl
)
2134 WARN_ON_ONCE(filp
!= fl
->fl_file
);
2135 if (filp
->f_op
->lock
)
2136 return filp
->f_op
->lock(filp
, F_GETLK
, fl
);
2137 posix_test_lock(filp
, fl
);
2140 EXPORT_SYMBOL_GPL(vfs_test_lock
);
2143 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2144 * @fl: The file_lock who's fl_pid should be translated
2145 * @ns: The namespace into which the pid should be translated
2147 * Used to translate a fl_pid into a namespace virtual pid number
2149 static pid_t
locks_translate_pid(struct file_lock
*fl
, struct pid_namespace
*ns
)
2156 if (IS_REMOTELCK(fl
))
2159 * If the flock owner process is dead and its pid has been already
2160 * freed, the translation below won't work, but we still want to show
2161 * flock owner pid number in init pidns.
2163 if (ns
== &init_pid_ns
)
2164 return (pid_t
)fl
->fl_pid
;
2167 pid
= find_pid_ns(fl
->fl_pid
, &init_pid_ns
);
2168 vnr
= pid_nr_ns(pid
, ns
);
2173 static int posix_lock_to_flock(struct flock
*flock
, struct file_lock
*fl
)
2175 flock
->l_pid
= locks_translate_pid(fl
, task_active_pid_ns(current
));
2176 #if BITS_PER_LONG == 32
2178 * Make sure we can represent the posix lock via
2179 * legacy 32bit flock.
2181 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
2183 if (fl
->fl_end
!= OFFSET_MAX
&& fl
->fl_end
> OFFT_OFFSET_MAX
)
2186 flock
->l_start
= fl
->fl_start
;
2187 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
2188 fl
->fl_end
- fl
->fl_start
+ 1;
2189 flock
->l_whence
= 0;
2190 flock
->l_type
= fl
->fl_type
;
2194 #if BITS_PER_LONG == 32
2195 static void posix_lock_to_flock64(struct flock64
*flock
, struct file_lock
*fl
)
2197 flock
->l_pid
= locks_translate_pid(fl
, task_active_pid_ns(current
));
2198 flock
->l_start
= fl
->fl_start
;
2199 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
2200 fl
->fl_end
- fl
->fl_start
+ 1;
2201 flock
->l_whence
= 0;
2202 flock
->l_type
= fl
->fl_type
;
2206 /* Report the first existing lock that would conflict with l.
2207 * This implements the F_GETLK command of fcntl().
2209 int fcntl_getlk(struct file
*filp
, unsigned int cmd
, struct flock
*flock
)
2211 struct file_lock
*fl
;
2214 fl
= locks_alloc_lock();
2218 if (cmd
!= F_OFD_GETLK
&& flock
->l_type
!= F_RDLCK
2219 && flock
->l_type
!= F_WRLCK
)
2222 error
= flock_to_posix_lock(filp
, fl
, flock
);
2226 if (cmd
== F_OFD_GETLK
) {
2228 if (flock
->l_pid
!= 0)
2231 fl
->fl_flags
|= FL_OFDLCK
;
2232 fl
->fl_owner
= filp
;
2235 error
= vfs_test_lock(filp
, fl
);
2239 flock
->l_type
= fl
->fl_type
;
2240 if (fl
->fl_type
!= F_UNLCK
) {
2241 error
= posix_lock_to_flock(flock
, fl
);
2246 locks_free_lock(fl
);
2251 * vfs_lock_file - file byte range lock
2252 * @filp: The file to apply the lock to
2253 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2254 * @fl: The lock to be applied
2255 * @conf: Place to return a copy of the conflicting lock, if found.
2257 * A caller that doesn't care about the conflicting lock may pass NULL
2258 * as the final argument.
2260 * If the filesystem defines a private ->lock() method, then @conf will
2261 * be left unchanged; so a caller that cares should initialize it to
2262 * some acceptable default.
2264 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2265 * locks, the ->lock() interface may return asynchronously, before the lock has
2266 * been granted or denied by the underlying filesystem, if (and only if)
2267 * lm_grant is set. Additionally EXPORT_OP_ASYNC_LOCK in export_operations
2268 * flags need to be set.
2270 * Callers expecting ->lock() to return asynchronously will only use F_SETLK,
2271 * not F_SETLKW; they will set FL_SLEEP if (and only if) the request is for a
2272 * blocking lock. When ->lock() does return asynchronously, it must return
2273 * FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock request completes.
2274 * If the request is for non-blocking lock the file system should return
2275 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2276 * with the result. If the request timed out the callback routine will return a
2277 * nonzero return code and the file system should release the lock. The file
2278 * system is also responsible to keep a corresponding posix lock when it
2279 * grants a lock so the VFS can find out which locks are locally held and do
2280 * the correct lock cleanup when required.
2281 * The underlying filesystem must not drop the kernel lock or call
2282 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2285 int vfs_lock_file(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
, struct file_lock
*conf
)
2287 WARN_ON_ONCE(filp
!= fl
->fl_file
);
2288 if (filp
->f_op
->lock
)
2289 return filp
->f_op
->lock(filp
, cmd
, fl
);
2291 return posix_lock_file(filp
, fl
, conf
);
2293 EXPORT_SYMBOL_GPL(vfs_lock_file
);
2295 static int do_lock_file_wait(struct file
*filp
, unsigned int cmd
,
2296 struct file_lock
*fl
)
2300 error
= security_file_lock(filp
, fl
->fl_type
);
2305 error
= vfs_lock_file(filp
, cmd
, fl
, NULL
);
2306 if (error
!= FILE_LOCK_DEFERRED
)
2308 error
= wait_event_interruptible(fl
->fl_wait
,
2309 list_empty(&fl
->fl_blocked_member
));
2313 locks_delete_block(fl
);
2318 /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2320 check_fmode_for_setlk(struct file_lock
*fl
)
2322 switch (fl
->fl_type
) {
2324 if (!(fl
->fl_file
->f_mode
& FMODE_READ
))
2328 if (!(fl
->fl_file
->f_mode
& FMODE_WRITE
))
2334 /* Apply the lock described by l to an open file descriptor.
2335 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2337 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
2338 struct flock
*flock
)
2340 struct file_lock
*file_lock
= locks_alloc_lock();
2341 struct inode
*inode
= file_inode(filp
);
2345 if (file_lock
== NULL
)
2348 error
= flock_to_posix_lock(filp
, file_lock
, flock
);
2352 error
= check_fmode_for_setlk(file_lock
);
2357 * If the cmd is requesting file-private locks, then set the
2358 * FL_OFDLCK flag and override the owner.
2363 if (flock
->l_pid
!= 0)
2367 file_lock
->fl_flags
|= FL_OFDLCK
;
2368 file_lock
->fl_owner
= filp
;
2372 if (flock
->l_pid
!= 0)
2376 file_lock
->fl_flags
|= FL_OFDLCK
;
2377 file_lock
->fl_owner
= filp
;
2380 file_lock
->fl_flags
|= FL_SLEEP
;
2383 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2386 * Attempt to detect a close/fcntl race and recover by releasing the
2387 * lock that was just acquired. There is no need to do that when we're
2388 * unlocking though, or for OFD locks.
2390 if (!error
&& file_lock
->fl_type
!= F_UNLCK
&&
2391 !(file_lock
->fl_flags
& FL_OFDLCK
)) {
2392 struct files_struct
*files
= current
->files
;
2394 * We need that spin_lock here - it prevents reordering between
2395 * update of i_flctx->flc_posix and check for it done in
2396 * close(). rcu_read_lock() wouldn't do.
2398 spin_lock(&files
->file_lock
);
2399 f
= files_lookup_fd_locked(files
, fd
);
2400 spin_unlock(&files
->file_lock
);
2402 file_lock
->fl_type
= F_UNLCK
;
2403 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2404 WARN_ON_ONCE(error
);
2409 trace_fcntl_setlk(inode
, file_lock
, error
);
2410 locks_free_lock(file_lock
);
2414 #if BITS_PER_LONG == 32
2415 /* Report the first existing lock that would conflict with l.
2416 * This implements the F_GETLK command of fcntl().
2418 int fcntl_getlk64(struct file
*filp
, unsigned int cmd
, struct flock64
*flock
)
2420 struct file_lock
*fl
;
2423 fl
= locks_alloc_lock();
2428 if (cmd
!= F_OFD_GETLK
&& flock
->l_type
!= F_RDLCK
2429 && flock
->l_type
!= F_WRLCK
)
2432 error
= flock64_to_posix_lock(filp
, fl
, flock
);
2436 if (cmd
== F_OFD_GETLK
) {
2438 if (flock
->l_pid
!= 0)
2441 fl
->fl_flags
|= FL_OFDLCK
;
2442 fl
->fl_owner
= filp
;
2445 error
= vfs_test_lock(filp
, fl
);
2449 flock
->l_type
= fl
->fl_type
;
2450 if (fl
->fl_type
!= F_UNLCK
)
2451 posix_lock_to_flock64(flock
, fl
);
2454 locks_free_lock(fl
);
2458 /* Apply the lock described by l to an open file descriptor.
2459 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2461 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
2462 struct flock64
*flock
)
2464 struct file_lock
*file_lock
= locks_alloc_lock();
2468 if (file_lock
== NULL
)
2471 error
= flock64_to_posix_lock(filp
, file_lock
, flock
);
2475 error
= check_fmode_for_setlk(file_lock
);
2480 * If the cmd is requesting file-private locks, then set the
2481 * FL_OFDLCK flag and override the owner.
2486 if (flock
->l_pid
!= 0)
2490 file_lock
->fl_flags
|= FL_OFDLCK
;
2491 file_lock
->fl_owner
= filp
;
2495 if (flock
->l_pid
!= 0)
2499 file_lock
->fl_flags
|= FL_OFDLCK
;
2500 file_lock
->fl_owner
= filp
;
2503 file_lock
->fl_flags
|= FL_SLEEP
;
2506 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2509 * Attempt to detect a close/fcntl race and recover by releasing the
2510 * lock that was just acquired. There is no need to do that when we're
2511 * unlocking though, or for OFD locks.
2513 if (!error
&& file_lock
->fl_type
!= F_UNLCK
&&
2514 !(file_lock
->fl_flags
& FL_OFDLCK
)) {
2515 struct files_struct
*files
= current
->files
;
2517 * We need that spin_lock here - it prevents reordering between
2518 * update of i_flctx->flc_posix and check for it done in
2519 * close(). rcu_read_lock() wouldn't do.
2521 spin_lock(&files
->file_lock
);
2522 f
= files_lookup_fd_locked(files
, fd
);
2523 spin_unlock(&files
->file_lock
);
2525 file_lock
->fl_type
= F_UNLCK
;
2526 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2527 WARN_ON_ONCE(error
);
2532 locks_free_lock(file_lock
);
2535 #endif /* BITS_PER_LONG == 32 */
2538 * This function is called when the file is being removed
2539 * from the task's fd array. POSIX locks belonging to this task
2540 * are deleted at this time.
2542 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
2545 struct inode
*inode
= file_inode(filp
);
2546 struct file_lock lock
;
2547 struct file_lock_context
*ctx
;
2550 * If there are no locks held on this file, we don't need to call
2551 * posix_lock_file(). Another process could be setting a lock on this
2552 * file at the same time, but we wouldn't remove that lock anyway.
2554 ctx
= locks_inode_context(inode
);
2555 if (!ctx
|| list_empty(&ctx
->flc_posix
))
2558 locks_init_lock(&lock
);
2559 lock
.fl_type
= F_UNLCK
;
2560 lock
.fl_flags
= FL_POSIX
| FL_CLOSE
;
2562 lock
.fl_end
= OFFSET_MAX
;
2563 lock
.fl_owner
= owner
;
2564 lock
.fl_pid
= current
->tgid
;
2565 lock
.fl_file
= filp
;
2567 lock
.fl_lmops
= NULL
;
2569 error
= vfs_lock_file(filp
, F_SETLK
, &lock
, NULL
);
2571 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
2572 lock
.fl_ops
->fl_release_private(&lock
);
2573 trace_locks_remove_posix(inode
, &lock
, error
);
2575 EXPORT_SYMBOL(locks_remove_posix
);
2577 /* The i_flctx must be valid when calling into here */
2579 locks_remove_flock(struct file
*filp
, struct file_lock_context
*flctx
)
2581 struct file_lock fl
;
2582 struct inode
*inode
= file_inode(filp
);
2584 if (list_empty(&flctx
->flc_flock
))
2587 flock_make_lock(filp
, &fl
, F_UNLCK
);
2588 fl
.fl_flags
|= FL_CLOSE
;
2590 if (filp
->f_op
->flock
)
2591 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
2593 flock_lock_inode(inode
, &fl
);
2595 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
2596 fl
.fl_ops
->fl_release_private(&fl
);
2599 /* The i_flctx must be valid when calling into here */
2601 locks_remove_lease(struct file
*filp
, struct file_lock_context
*ctx
)
2603 struct file_lock
*fl
, *tmp
;
2606 if (list_empty(&ctx
->flc_lease
))
2609 percpu_down_read(&file_rwsem
);
2610 spin_lock(&ctx
->flc_lock
);
2611 list_for_each_entry_safe(fl
, tmp
, &ctx
->flc_lease
, fl_list
)
2612 if (filp
== fl
->fl_file
)
2613 lease_modify(fl
, F_UNLCK
, &dispose
);
2614 spin_unlock(&ctx
->flc_lock
);
2615 percpu_up_read(&file_rwsem
);
2617 locks_dispose_list(&dispose
);
2621 * This function is called on the last close of an open file.
2623 void locks_remove_file(struct file
*filp
)
2625 struct file_lock_context
*ctx
;
2627 ctx
= locks_inode_context(file_inode(filp
));
2631 /* remove any OFD locks */
2632 locks_remove_posix(filp
, filp
);
2634 /* remove flock locks */
2635 locks_remove_flock(filp
, ctx
);
2637 /* remove any leases */
2638 locks_remove_lease(filp
, ctx
);
2640 spin_lock(&ctx
->flc_lock
);
2641 locks_check_ctx_file_list(filp
, &ctx
->flc_posix
, "POSIX");
2642 locks_check_ctx_file_list(filp
, &ctx
->flc_flock
, "FLOCK");
2643 locks_check_ctx_file_list(filp
, &ctx
->flc_lease
, "LEASE");
2644 spin_unlock(&ctx
->flc_lock
);
2648 * vfs_cancel_lock - file byte range unblock lock
2649 * @filp: The file to apply the unblock to
2650 * @fl: The lock to be unblocked
2652 * Used by lock managers to cancel blocked requests
2654 int vfs_cancel_lock(struct file
*filp
, struct file_lock
*fl
)
2656 WARN_ON_ONCE(filp
!= fl
->fl_file
);
2657 if (filp
->f_op
->lock
)
2658 return filp
->f_op
->lock(filp
, F_CANCELLK
, fl
);
2661 EXPORT_SYMBOL_GPL(vfs_cancel_lock
);
2664 * vfs_inode_has_locks - are any file locks held on @inode?
2665 * @inode: inode to check for locks
2667 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2670 bool vfs_inode_has_locks(struct inode
*inode
)
2672 struct file_lock_context
*ctx
;
2675 ctx
= locks_inode_context(inode
);
2679 spin_lock(&ctx
->flc_lock
);
2680 ret
= !list_empty(&ctx
->flc_posix
) || !list_empty(&ctx
->flc_flock
);
2681 spin_unlock(&ctx
->flc_lock
);
2684 EXPORT_SYMBOL_GPL(vfs_inode_has_locks
);
2686 #ifdef CONFIG_PROC_FS
2687 #include <linux/proc_fs.h>
2688 #include <linux/seq_file.h>
2690 struct locks_iterator
{
2695 static void lock_get_status(struct seq_file
*f
, struct file_lock
*fl
,
2696 loff_t id
, char *pfx
, int repeat
)
2698 struct inode
*inode
= NULL
;
2699 unsigned int fl_pid
;
2700 struct pid_namespace
*proc_pidns
= proc_pid_ns(file_inode(f
->file
)->i_sb
);
2703 fl_pid
= locks_translate_pid(fl
, proc_pidns
);
2705 * If lock owner is dead (and pid is freed) or not visible in current
2706 * pidns, zero is shown as a pid value. Check lock info from
2707 * init_pid_ns to get saved lock pid value.
2710 if (fl
->fl_file
!= NULL
)
2711 inode
= file_inode(fl
->fl_file
);
2713 seq_printf(f
, "%lld: ", id
);
2716 seq_printf(f
, "%*s", repeat
- 1 + (int)strlen(pfx
), pfx
);
2719 if (fl
->fl_flags
& FL_ACCESS
)
2720 seq_puts(f
, "ACCESS");
2721 else if (IS_OFDLCK(fl
))
2722 seq_puts(f
, "OFDLCK");
2724 seq_puts(f
, "POSIX ");
2726 seq_printf(f
, " %s ",
2727 (inode
== NULL
) ? "*NOINODE*" : "ADVISORY ");
2728 } else if (IS_FLOCK(fl
)) {
2729 seq_puts(f
, "FLOCK ADVISORY ");
2730 } else if (IS_LEASE(fl
)) {
2731 if (fl
->fl_flags
& FL_DELEG
)
2732 seq_puts(f
, "DELEG ");
2734 seq_puts(f
, "LEASE ");
2736 if (lease_breaking(fl
))
2737 seq_puts(f
, "BREAKING ");
2738 else if (fl
->fl_file
)
2739 seq_puts(f
, "ACTIVE ");
2741 seq_puts(f
, "BREAKER ");
2743 seq_puts(f
, "UNKNOWN UNKNOWN ");
2745 type
= IS_LEASE(fl
) ? target_leasetype(fl
) : fl
->fl_type
;
2747 seq_printf(f
, "%s ", (type
== F_WRLCK
) ? "WRITE" :
2748 (type
== F_RDLCK
) ? "READ" : "UNLCK");
2750 /* userspace relies on this representation of dev_t */
2751 seq_printf(f
, "%d %02x:%02x:%lu ", fl_pid
,
2752 MAJOR(inode
->i_sb
->s_dev
),
2753 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2755 seq_printf(f
, "%d <none>:0 ", fl_pid
);
2758 if (fl
->fl_end
== OFFSET_MAX
)
2759 seq_printf(f
, "%Ld EOF\n", fl
->fl_start
);
2761 seq_printf(f
, "%Ld %Ld\n", fl
->fl_start
, fl
->fl_end
);
2763 seq_puts(f
, "0 EOF\n");
2767 static struct file_lock
*get_next_blocked_member(struct file_lock
*node
)
2769 struct file_lock
*tmp
;
2771 /* NULL node or root node */
2772 if (node
== NULL
|| node
->fl_blocker
== NULL
)
2775 /* Next member in the linked list could be itself */
2776 tmp
= list_next_entry(node
, fl_blocked_member
);
2777 if (list_entry_is_head(tmp
, &node
->fl_blocker
->fl_blocked_requests
, fl_blocked_member
)
2785 static int locks_show(struct seq_file
*f
, void *v
)
2787 struct locks_iterator
*iter
= f
->private;
2788 struct file_lock
*cur
, *tmp
;
2789 struct pid_namespace
*proc_pidns
= proc_pid_ns(file_inode(f
->file
)->i_sb
);
2792 cur
= hlist_entry(v
, struct file_lock
, fl_link
);
2794 if (locks_translate_pid(cur
, proc_pidns
) == 0)
2797 /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2798 * is the left child of current node, the next silibing in fl_blocked_member is the
2799 * right child, we can alse get the parent of current node from fl_blocker, so this
2800 * question becomes traversal of a binary tree
2802 while (cur
!= NULL
) {
2804 lock_get_status(f
, cur
, iter
->li_pos
, "-> ", level
);
2806 lock_get_status(f
, cur
, iter
->li_pos
, "", level
);
2808 if (!list_empty(&cur
->fl_blocked_requests
)) {
2810 cur
= list_first_entry_or_null(&cur
->fl_blocked_requests
,
2811 struct file_lock
, fl_blocked_member
);
2815 tmp
= get_next_blocked_member(cur
);
2816 /* Fall back to parent node */
2817 while (tmp
== NULL
&& cur
->fl_blocker
!= NULL
) {
2818 cur
= cur
->fl_blocker
;
2820 tmp
= get_next_blocked_member(cur
);
2829 static void __show_fd_locks(struct seq_file
*f
,
2830 struct list_head
*head
, int *id
,
2831 struct file
*filp
, struct files_struct
*files
)
2833 struct file_lock
*fl
;
2835 list_for_each_entry(fl
, head
, fl_list
) {
2837 if (filp
!= fl
->fl_file
)
2839 if (fl
->fl_owner
!= files
&&
2840 fl
->fl_owner
!= filp
)
2844 seq_puts(f
, "lock:\t");
2845 lock_get_status(f
, fl
, *id
, "", 0);
2849 void show_fd_locks(struct seq_file
*f
,
2850 struct file
*filp
, struct files_struct
*files
)
2852 struct inode
*inode
= file_inode(filp
);
2853 struct file_lock_context
*ctx
;
2856 ctx
= locks_inode_context(inode
);
2860 spin_lock(&ctx
->flc_lock
);
2861 __show_fd_locks(f
, &ctx
->flc_flock
, &id
, filp
, files
);
2862 __show_fd_locks(f
, &ctx
->flc_posix
, &id
, filp
, files
);
2863 __show_fd_locks(f
, &ctx
->flc_lease
, &id
, filp
, files
);
2864 spin_unlock(&ctx
->flc_lock
);
2867 static void *locks_start(struct seq_file
*f
, loff_t
*pos
)
2868 __acquires(&blocked_lock_lock
)
2870 struct locks_iterator
*iter
= f
->private;
2872 iter
->li_pos
= *pos
+ 1;
2873 percpu_down_write(&file_rwsem
);
2874 spin_lock(&blocked_lock_lock
);
2875 return seq_hlist_start_percpu(&file_lock_list
.hlist
, &iter
->li_cpu
, *pos
);
2878 static void *locks_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2880 struct locks_iterator
*iter
= f
->private;
2883 return seq_hlist_next_percpu(v
, &file_lock_list
.hlist
, &iter
->li_cpu
, pos
);
2886 static void locks_stop(struct seq_file
*f
, void *v
)
2887 __releases(&blocked_lock_lock
)
2889 spin_unlock(&blocked_lock_lock
);
2890 percpu_up_write(&file_rwsem
);
2893 static const struct seq_operations locks_seq_operations
= {
2894 .start
= locks_start
,
2900 static int __init
proc_locks_init(void)
2902 proc_create_seq_private("locks", 0, NULL
, &locks_seq_operations
,
2903 sizeof(struct locks_iterator
), NULL
);
2906 fs_initcall(proc_locks_init
);
2909 static int __init
filelock_init(void)
2913 flctx_cache
= kmem_cache_create("file_lock_ctx",
2914 sizeof(struct file_lock_context
), 0, SLAB_PANIC
, NULL
);
2916 filelock_cache
= kmem_cache_create("file_lock_cache",
2917 sizeof(struct file_lock
), 0, SLAB_PANIC
, NULL
);
2919 for_each_possible_cpu(i
) {
2920 struct file_lock_list_struct
*fll
= per_cpu_ptr(&file_lock_list
, i
);
2922 spin_lock_init(&fll
->lock
);
2923 INIT_HLIST_HEAD(&fll
->hlist
);
2926 lease_notifier_chain_init();
2929 core_initcall(filelock_init
);