1 // SPDX-License-Identifier: GPL-2.0-only
5 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
6 * Doug Evans (dje@spiff.uucp), August 07, 1992
8 * Deadlock detection added.
9 * FIXME: one thing isn't handled yet:
10 * - mandatory locks (requires lots of changes elsewhere)
11 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
13 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
14 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
16 * Converted file_lock_table to a linked list from an array, which eliminates
17 * the limits on how many active file locks are open.
18 * Chad Page (pageone@netcom.com), November 27, 1994
20 * Removed dependency on file descriptors. dup()'ed file descriptors now
21 * get the same locks as the original file descriptors, and a close() on
22 * any file descriptor removes ALL the locks on the file for the current
23 * process. Since locks still depend on the process id, locks are inherited
24 * after an exec() but not after a fork(). This agrees with POSIX, and both
25 * BSD and SVR4 practice.
26 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
28 * Scrapped free list which is redundant now that we allocate locks
29 * dynamically with kmalloc()/kfree().
30 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
32 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
34 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
35 * fcntl() system call. They have the semantics described above.
37 * FL_FLOCK locks are created with calls to flock(), through the flock()
38 * system call, which is new. Old C libraries implement flock() via fcntl()
39 * and will continue to use the old, broken implementation.
41 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
42 * with a file pointer (filp). As a result they can be shared by a parent
43 * process and its children after a fork(). They are removed when the last
44 * file descriptor referring to the file pointer is closed (unless explicitly
47 * FL_FLOCK locks never deadlock, an existing lock is always removed before
48 * upgrading from shared to exclusive (or vice versa). When this happens
49 * any processes blocked by the current lock are woken up and allowed to
50 * run before the new lock is applied.
51 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
53 * Removed some race conditions in flock_lock_file(), marked other possible
54 * races. Just grep for FIXME to see them.
55 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
57 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
58 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
59 * once we've checked for blocking and deadlocking.
60 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
62 * Initial implementation of mandatory locks. SunOS turned out to be
63 * a rotten model, so I implemented the "obvious" semantics.
64 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
65 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
67 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
68 * check if a file has mandatory locks, used by mmap(), open() and creat() to
69 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
71 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
73 * Tidied up block list handling. Added '/proc/locks' interface.
74 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
76 * Fixed deadlock condition for pathological code that mixes calls to
77 * flock() and fcntl().
78 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
80 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
81 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
82 * guarantee sensible behaviour in the case where file system modules might
83 * be compiled with different options than the kernel itself.
84 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
86 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
87 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
88 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
90 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
91 * locks. Changed process synchronisation to avoid dereferencing locks that
92 * have already been freed.
93 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
95 * Made the block list a circular list to minimise searching in the list.
96 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
98 * Made mandatory locking a mount option. Default is not to allow mandatory
100 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
102 * Some adaptations for NFS support.
103 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
105 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
106 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
108 * Use slab allocator instead of kmalloc/kfree.
109 * Use generic list implementation from <linux/list.h>.
110 * Sped up posix_locks_deadlock by only considering blocked locks.
111 * Matthew Wilcox <willy@debian.org>, March, 2000.
113 * Leases and LOCK_MAND
114 * Matthew Wilcox <willy@debian.org>, June, 2000.
115 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
117 * Locking conflicts and dependencies:
118 * If multiple threads attempt to lock the same byte (or flock the same file)
119 * only one can be granted the lock, and other must wait their turn.
120 * The first lock has been "applied" or "granted", the others are "waiting"
121 * and are "blocked" by the "applied" lock..
123 * Waiting and applied locks are all kept in trees whose properties are:
125 * - the root of a tree may be an applied or waiting lock.
126 * - every other node in the tree is a waiting lock that
127 * conflicts with every ancestor of that node.
129 * Every such tree begins life as a waiting singleton which obviously
130 * satisfies the above properties.
132 * The only ways we modify trees preserve these properties:
134 * 1. We may add a new leaf node, but only after first verifying that it
135 * conflicts with all of its ancestors.
136 * 2. We may remove the root of a tree, creating a new singleton
137 * tree from the root and N new trees rooted in the immediate
139 * 3. If the root of a tree is not currently an applied lock, we may
140 * apply it (if possible).
141 * 4. We may upgrade the root of the tree (either extend its range,
142 * or upgrade its entire range from read to write).
144 * When an applied lock is modified in a way that reduces or downgrades any
145 * part of its range, we remove all its children (2 above). This particularly
146 * happens when a lock is unlocked.
148 * For each of those child trees we "wake up" the thread which is
149 * waiting for the lock so it can continue handling as follows: if the
150 * root of the tree applies, we do so (3). If it doesn't, it must
151 * conflict with some applied lock. We remove (wake up) all of its children
152 * (2), and add it is a new leaf to the tree rooted in the applied
153 * lock (1). We then repeat the process recursively with those
158 #include <linux/capability.h>
159 #include <linux/file.h>
160 #include <linux/fdtable.h>
161 #include <linux/fs.h>
162 #include <linux/init.h>
163 #include <linux/security.h>
164 #include <linux/slab.h>
165 #include <linux/syscalls.h>
166 #include <linux/time.h>
167 #include <linux/rcupdate.h>
168 #include <linux/pid_namespace.h>
169 #include <linux/hashtable.h>
170 #include <linux/percpu.h>
172 #define CREATE_TRACE_POINTS
173 #include <trace/events/filelock.h>
175 #include <linux/uaccess.h>
177 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
178 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
179 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
180 #define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
181 #define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
183 static bool lease_breaking(struct file_lock
*fl
)
185 return fl
->fl_flags
& (FL_UNLOCK_PENDING
| FL_DOWNGRADE_PENDING
);
188 static int target_leasetype(struct file_lock
*fl
)
190 if (fl
->fl_flags
& FL_UNLOCK_PENDING
)
192 if (fl
->fl_flags
& FL_DOWNGRADE_PENDING
)
197 int leases_enable
= 1;
198 int lease_break_time
= 45;
201 * The global file_lock_list is only used for displaying /proc/locks, so we
202 * keep a list on each CPU, with each list protected by its own spinlock.
203 * Global serialization is done using file_rwsem.
205 * Note that alterations to the list also require that the relevant flc_lock is
208 struct file_lock_list_struct
{
210 struct hlist_head hlist
;
212 static DEFINE_PER_CPU(struct file_lock_list_struct
, file_lock_list
);
213 DEFINE_STATIC_PERCPU_RWSEM(file_rwsem
);
217 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
218 * It is protected by blocked_lock_lock.
220 * We hash locks by lockowner in order to optimize searching for the lock a
221 * particular lockowner is waiting on.
223 * FIXME: make this value scale via some heuristic? We generally will want more
224 * buckets when we have more lockowners holding locks, but that's a little
225 * difficult to determine without knowing what the workload will look like.
227 #define BLOCKED_HASH_BITS 7
228 static DEFINE_HASHTABLE(blocked_hash
, BLOCKED_HASH_BITS
);
231 * This lock protects the blocked_hash. Generally, if you're accessing it, you
232 * want to be holding this lock.
234 * In addition, it also protects the fl->fl_blocked_requests list, and the
235 * fl->fl_blocker pointer for file_lock structures that are acting as lock
236 * requests (in contrast to those that are acting as records of acquired locks).
238 * Note that when we acquire this lock in order to change the above fields,
239 * we often hold the flc_lock as well. In certain cases, when reading the fields
240 * protected by this lock, we can skip acquiring it iff we already hold the
243 static DEFINE_SPINLOCK(blocked_lock_lock
);
245 static struct kmem_cache
*flctx_cache __read_mostly
;
246 static struct kmem_cache
*filelock_cache __read_mostly
;
248 static struct file_lock_context
*
249 locks_get_lock_context(struct inode
*inode
, int type
)
251 struct file_lock_context
*ctx
;
253 /* paired with cmpxchg() below */
254 ctx
= smp_load_acquire(&inode
->i_flctx
);
255 if (likely(ctx
) || type
== F_UNLCK
)
258 ctx
= kmem_cache_alloc(flctx_cache
, GFP_KERNEL
);
262 spin_lock_init(&ctx
->flc_lock
);
263 INIT_LIST_HEAD(&ctx
->flc_flock
);
264 INIT_LIST_HEAD(&ctx
->flc_posix
);
265 INIT_LIST_HEAD(&ctx
->flc_lease
);
268 * Assign the pointer if it's not already assigned. If it is, then
269 * free the context we just allocated.
271 if (cmpxchg(&inode
->i_flctx
, NULL
, ctx
)) {
272 kmem_cache_free(flctx_cache
, ctx
);
273 ctx
= smp_load_acquire(&inode
->i_flctx
);
276 trace_locks_get_lock_context(inode
, type
, ctx
);
281 locks_dump_ctx_list(struct list_head
*list
, char *list_type
)
283 struct file_lock
*fl
;
285 list_for_each_entry(fl
, list
, fl_list
) {
286 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
);
291 locks_check_ctx_lists(struct inode
*inode
)
293 struct file_lock_context
*ctx
= inode
->i_flctx
;
295 if (unlikely(!list_empty(&ctx
->flc_flock
) ||
296 !list_empty(&ctx
->flc_posix
) ||
297 !list_empty(&ctx
->flc_lease
))) {
298 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
299 MAJOR(inode
->i_sb
->s_dev
), MINOR(inode
->i_sb
->s_dev
),
301 locks_dump_ctx_list(&ctx
->flc_flock
, "FLOCK");
302 locks_dump_ctx_list(&ctx
->flc_posix
, "POSIX");
303 locks_dump_ctx_list(&ctx
->flc_lease
, "LEASE");
308 locks_check_ctx_file_list(struct file
*filp
, struct list_head
*list
,
311 struct file_lock
*fl
;
312 struct inode
*inode
= locks_inode(filp
);
314 list_for_each_entry(fl
, list
, fl_list
)
315 if (fl
->fl_file
== filp
)
316 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
317 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
318 list_type
, MAJOR(inode
->i_sb
->s_dev
),
319 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
,
320 fl
->fl_owner
, fl
->fl_flags
, fl
->fl_type
, fl
->fl_pid
);
324 locks_free_lock_context(struct inode
*inode
)
326 struct file_lock_context
*ctx
= inode
->i_flctx
;
329 locks_check_ctx_lists(inode
);
330 kmem_cache_free(flctx_cache
, ctx
);
334 static void locks_init_lock_heads(struct file_lock
*fl
)
336 INIT_HLIST_NODE(&fl
->fl_link
);
337 INIT_LIST_HEAD(&fl
->fl_list
);
338 INIT_LIST_HEAD(&fl
->fl_blocked_requests
);
339 INIT_LIST_HEAD(&fl
->fl_blocked_member
);
340 init_waitqueue_head(&fl
->fl_wait
);
343 /* Allocate an empty lock structure. */
344 struct file_lock
*locks_alloc_lock(void)
346 struct file_lock
*fl
= kmem_cache_zalloc(filelock_cache
, GFP_KERNEL
);
349 locks_init_lock_heads(fl
);
353 EXPORT_SYMBOL_GPL(locks_alloc_lock
);
355 void locks_release_private(struct file_lock
*fl
)
357 BUG_ON(waitqueue_active(&fl
->fl_wait
));
358 BUG_ON(!list_empty(&fl
->fl_list
));
359 BUG_ON(!list_empty(&fl
->fl_blocked_requests
));
360 BUG_ON(!list_empty(&fl
->fl_blocked_member
));
361 BUG_ON(!hlist_unhashed(&fl
->fl_link
));
364 if (fl
->fl_ops
->fl_release_private
)
365 fl
->fl_ops
->fl_release_private(fl
);
370 if (fl
->fl_lmops
->lm_put_owner
) {
371 fl
->fl_lmops
->lm_put_owner(fl
->fl_owner
);
377 EXPORT_SYMBOL_GPL(locks_release_private
);
379 /* Free a lock which is not in use. */
380 void locks_free_lock(struct file_lock
*fl
)
382 locks_release_private(fl
);
383 kmem_cache_free(filelock_cache
, fl
);
385 EXPORT_SYMBOL(locks_free_lock
);
388 locks_dispose_list(struct list_head
*dispose
)
390 struct file_lock
*fl
;
392 while (!list_empty(dispose
)) {
393 fl
= list_first_entry(dispose
, struct file_lock
, fl_list
);
394 list_del_init(&fl
->fl_list
);
399 void locks_init_lock(struct file_lock
*fl
)
401 memset(fl
, 0, sizeof(struct file_lock
));
402 locks_init_lock_heads(fl
);
404 EXPORT_SYMBOL(locks_init_lock
);
407 * Initialize a new lock from an existing file_lock structure.
409 void locks_copy_conflock(struct file_lock
*new, struct file_lock
*fl
)
411 new->fl_owner
= fl
->fl_owner
;
412 new->fl_pid
= fl
->fl_pid
;
414 new->fl_flags
= fl
->fl_flags
;
415 new->fl_type
= fl
->fl_type
;
416 new->fl_start
= fl
->fl_start
;
417 new->fl_end
= fl
->fl_end
;
418 new->fl_lmops
= fl
->fl_lmops
;
422 if (fl
->fl_lmops
->lm_get_owner
)
423 fl
->fl_lmops
->lm_get_owner(fl
->fl_owner
);
426 EXPORT_SYMBOL(locks_copy_conflock
);
428 void locks_copy_lock(struct file_lock
*new, struct file_lock
*fl
)
430 /* "new" must be a freshly-initialized lock */
431 WARN_ON_ONCE(new->fl_ops
);
433 locks_copy_conflock(new, fl
);
435 new->fl_file
= fl
->fl_file
;
436 new->fl_ops
= fl
->fl_ops
;
439 if (fl
->fl_ops
->fl_copy_lock
)
440 fl
->fl_ops
->fl_copy_lock(new, fl
);
443 EXPORT_SYMBOL(locks_copy_lock
);
445 static void locks_move_blocks(struct file_lock
*new, struct file_lock
*fl
)
450 * As ctx->flc_lock is held, new requests cannot be added to
451 * ->fl_blocked_requests, so we don't need a lock to check if it
454 if (list_empty(&fl
->fl_blocked_requests
))
456 spin_lock(&blocked_lock_lock
);
457 list_splice_init(&fl
->fl_blocked_requests
, &new->fl_blocked_requests
);
458 list_for_each_entry(f
, &new->fl_blocked_requests
, fl_blocked_member
)
460 spin_unlock(&blocked_lock_lock
);
463 static inline int flock_translate_cmd(int cmd
) {
465 return cmd
& (LOCK_MAND
| LOCK_RW
);
477 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
478 static struct file_lock
*
479 flock_make_lock(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
)
481 int type
= flock_translate_cmd(cmd
);
484 return ERR_PTR(type
);
487 fl
= locks_alloc_lock();
489 return ERR_PTR(-ENOMEM
);
496 fl
->fl_pid
= current
->tgid
;
497 fl
->fl_flags
= FL_FLOCK
;
499 fl
->fl_end
= OFFSET_MAX
;
504 static int assign_type(struct file_lock
*fl
, long type
)
518 static int flock64_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
521 switch (l
->l_whence
) {
526 fl
->fl_start
= filp
->f_pos
;
529 fl
->fl_start
= i_size_read(file_inode(filp
));
534 if (l
->l_start
> OFFSET_MAX
- fl
->fl_start
)
536 fl
->fl_start
+= l
->l_start
;
537 if (fl
->fl_start
< 0)
540 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
541 POSIX-2001 defines it. */
543 if (l
->l_len
- 1 > OFFSET_MAX
- fl
->fl_start
)
545 fl
->fl_end
= fl
->fl_start
+ l
->l_len
- 1;
547 } else if (l
->l_len
< 0) {
548 if (fl
->fl_start
+ l
->l_len
< 0)
550 fl
->fl_end
= fl
->fl_start
- 1;
551 fl
->fl_start
+= l
->l_len
;
553 fl
->fl_end
= OFFSET_MAX
;
555 fl
->fl_owner
= current
->files
;
556 fl
->fl_pid
= current
->tgid
;
558 fl
->fl_flags
= FL_POSIX
;
562 return assign_type(fl
, l
->l_type
);
565 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
568 static int flock_to_posix_lock(struct file
*filp
, struct file_lock
*fl
,
571 struct flock64 ll
= {
573 .l_whence
= l
->l_whence
,
574 .l_start
= l
->l_start
,
578 return flock64_to_posix_lock(filp
, fl
, &ll
);
581 /* default lease lock manager operations */
583 lease_break_callback(struct file_lock
*fl
)
585 kill_fasync(&fl
->fl_fasync
, SIGIO
, POLL_MSG
);
590 lease_setup(struct file_lock
*fl
, void **priv
)
592 struct file
*filp
= fl
->fl_file
;
593 struct fasync_struct
*fa
= *priv
;
596 * fasync_insert_entry() returns the old entry if any. If there was no
597 * old entry, then it used "priv" and inserted it into the fasync list.
598 * Clear the pointer to indicate that it shouldn't be freed.
600 if (!fasync_insert_entry(fa
->fa_fd
, filp
, &fl
->fl_fasync
, fa
))
603 __f_setown(filp
, task_pid(current
), PIDTYPE_TGID
, 0);
606 static const struct lock_manager_operations lease_manager_ops
= {
607 .lm_break
= lease_break_callback
,
608 .lm_change
= lease_modify
,
609 .lm_setup
= lease_setup
,
613 * Initialize a lease, use the default lock manager operations
615 static int lease_init(struct file
*filp
, long type
, struct file_lock
*fl
)
617 if (assign_type(fl
, type
) != 0)
621 fl
->fl_pid
= current
->tgid
;
624 fl
->fl_flags
= FL_LEASE
;
626 fl
->fl_end
= OFFSET_MAX
;
628 fl
->fl_lmops
= &lease_manager_ops
;
632 /* Allocate a file_lock initialised to this type of lease */
633 static struct file_lock
*lease_alloc(struct file
*filp
, long type
)
635 struct file_lock
*fl
= locks_alloc_lock();
639 return ERR_PTR(error
);
641 error
= lease_init(filp
, type
, fl
);
644 return ERR_PTR(error
);
649 /* Check if two locks overlap each other.
651 static inline int locks_overlap(struct file_lock
*fl1
, struct file_lock
*fl2
)
653 return ((fl1
->fl_end
>= fl2
->fl_start
) &&
654 (fl2
->fl_end
>= fl1
->fl_start
));
658 * Check whether two locks have the same owner.
660 static int posix_same_owner(struct file_lock
*fl1
, struct file_lock
*fl2
)
662 return fl1
->fl_owner
== fl2
->fl_owner
;
665 /* Must be called with the flc_lock held! */
666 static void locks_insert_global_locks(struct file_lock
*fl
)
668 struct file_lock_list_struct
*fll
= this_cpu_ptr(&file_lock_list
);
670 percpu_rwsem_assert_held(&file_rwsem
);
672 spin_lock(&fll
->lock
);
673 fl
->fl_link_cpu
= smp_processor_id();
674 hlist_add_head(&fl
->fl_link
, &fll
->hlist
);
675 spin_unlock(&fll
->lock
);
678 /* Must be called with the flc_lock held! */
679 static void locks_delete_global_locks(struct file_lock
*fl
)
681 struct file_lock_list_struct
*fll
;
683 percpu_rwsem_assert_held(&file_rwsem
);
686 * Avoid taking lock if already unhashed. This is safe since this check
687 * is done while holding the flc_lock, and new insertions into the list
688 * also require that it be held.
690 if (hlist_unhashed(&fl
->fl_link
))
693 fll
= per_cpu_ptr(&file_lock_list
, fl
->fl_link_cpu
);
694 spin_lock(&fll
->lock
);
695 hlist_del_init(&fl
->fl_link
);
696 spin_unlock(&fll
->lock
);
700 posix_owner_key(struct file_lock
*fl
)
702 return (unsigned long)fl
->fl_owner
;
705 static void locks_insert_global_blocked(struct file_lock
*waiter
)
707 lockdep_assert_held(&blocked_lock_lock
);
709 hash_add(blocked_hash
, &waiter
->fl_link
, posix_owner_key(waiter
));
712 static void locks_delete_global_blocked(struct file_lock
*waiter
)
714 lockdep_assert_held(&blocked_lock_lock
);
716 hash_del(&waiter
->fl_link
);
719 /* Remove waiter from blocker's block list.
720 * When blocker ends up pointing to itself then the list is empty.
722 * Must be called with blocked_lock_lock held.
724 static void __locks_delete_block(struct file_lock
*waiter
)
726 locks_delete_global_blocked(waiter
);
727 list_del_init(&waiter
->fl_blocked_member
);
728 waiter
->fl_blocker
= NULL
;
731 static void __locks_wake_up_blocks(struct file_lock
*blocker
)
733 while (!list_empty(&blocker
->fl_blocked_requests
)) {
734 struct file_lock
*waiter
;
736 waiter
= list_first_entry(&blocker
->fl_blocked_requests
,
737 struct file_lock
, fl_blocked_member
);
738 __locks_delete_block(waiter
);
739 if (waiter
->fl_lmops
&& waiter
->fl_lmops
->lm_notify
)
740 waiter
->fl_lmops
->lm_notify(waiter
);
742 wake_up(&waiter
->fl_wait
);
747 * locks_delete_lock - stop waiting for a file lock
748 * @waiter: the lock which was waiting
750 * lockd/nfsd need to disconnect the lock while working on it.
752 int locks_delete_block(struct file_lock
*waiter
)
754 int status
= -ENOENT
;
756 spin_lock(&blocked_lock_lock
);
757 if (waiter
->fl_blocker
)
759 __locks_wake_up_blocks(waiter
);
760 __locks_delete_block(waiter
);
761 spin_unlock(&blocked_lock_lock
);
764 EXPORT_SYMBOL(locks_delete_block
);
766 /* Insert waiter into blocker's block list.
767 * We use a circular list so that processes can be easily woken up in
768 * the order they blocked. The documentation doesn't require this but
769 * it seems like the reasonable thing to do.
771 * Must be called with both the flc_lock and blocked_lock_lock held. The
772 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
773 * but by ensuring that the flc_lock is also held on insertions we can avoid
774 * taking the blocked_lock_lock in some cases when we see that the
775 * fl_blocked_requests list is empty.
777 * Rather than just adding to the list, we check for conflicts with any existing
778 * waiters, and add beneath any waiter that blocks the new waiter.
779 * Thus wakeups don't happen until needed.
781 static void __locks_insert_block(struct file_lock
*blocker
,
782 struct file_lock
*waiter
,
783 bool conflict(struct file_lock
*,
786 struct file_lock
*fl
;
787 BUG_ON(!list_empty(&waiter
->fl_blocked_member
));
790 list_for_each_entry(fl
, &blocker
->fl_blocked_requests
, fl_blocked_member
)
791 if (conflict(fl
, waiter
)) {
795 waiter
->fl_blocker
= blocker
;
796 list_add_tail(&waiter
->fl_blocked_member
, &blocker
->fl_blocked_requests
);
797 if (IS_POSIX(blocker
) && !IS_OFDLCK(blocker
))
798 locks_insert_global_blocked(waiter
);
800 /* The requests in waiter->fl_blocked are known to conflict with
801 * waiter, but might not conflict with blocker, or the requests
802 * and lock which block it. So they all need to be woken.
804 __locks_wake_up_blocks(waiter
);
807 /* Must be called with flc_lock held. */
808 static void locks_insert_block(struct file_lock
*blocker
,
809 struct file_lock
*waiter
,
810 bool conflict(struct file_lock
*,
813 spin_lock(&blocked_lock_lock
);
814 __locks_insert_block(blocker
, waiter
, conflict
);
815 spin_unlock(&blocked_lock_lock
);
819 * Wake up processes blocked waiting for blocker.
821 * Must be called with the inode->flc_lock held!
823 static void locks_wake_up_blocks(struct file_lock
*blocker
)
826 * Avoid taking global lock if list is empty. This is safe since new
827 * blocked requests are only added to the list under the flc_lock, and
828 * the flc_lock is always held here. Note that removal from the
829 * fl_blocked_requests list does not require the flc_lock, so we must
830 * recheck list_empty() after acquiring the blocked_lock_lock.
832 if (list_empty(&blocker
->fl_blocked_requests
))
835 spin_lock(&blocked_lock_lock
);
836 __locks_wake_up_blocks(blocker
);
837 spin_unlock(&blocked_lock_lock
);
841 locks_insert_lock_ctx(struct file_lock
*fl
, struct list_head
*before
)
843 list_add_tail(&fl
->fl_list
, before
);
844 locks_insert_global_locks(fl
);
848 locks_unlink_lock_ctx(struct file_lock
*fl
)
850 locks_delete_global_locks(fl
);
851 list_del_init(&fl
->fl_list
);
852 locks_wake_up_blocks(fl
);
856 locks_delete_lock_ctx(struct file_lock
*fl
, struct list_head
*dispose
)
858 locks_unlink_lock_ctx(fl
);
860 list_add(&fl
->fl_list
, dispose
);
865 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
866 * checks for shared/exclusive status of overlapping locks.
868 static bool locks_conflict(struct file_lock
*caller_fl
,
869 struct file_lock
*sys_fl
)
871 if (sys_fl
->fl_type
== F_WRLCK
)
873 if (caller_fl
->fl_type
== F_WRLCK
)
878 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
879 * checking before calling the locks_conflict().
881 static bool posix_locks_conflict(struct file_lock
*caller_fl
,
882 struct file_lock
*sys_fl
)
884 /* POSIX locks owned by the same process do not conflict with
887 if (posix_same_owner(caller_fl
, sys_fl
))
890 /* Check whether they overlap */
891 if (!locks_overlap(caller_fl
, sys_fl
))
894 return locks_conflict(caller_fl
, sys_fl
);
897 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
898 * checking before calling the locks_conflict().
900 static bool flock_locks_conflict(struct file_lock
*caller_fl
,
901 struct file_lock
*sys_fl
)
903 /* FLOCK locks referring to the same filp do not conflict with
906 if (caller_fl
->fl_file
== sys_fl
->fl_file
)
908 if ((caller_fl
->fl_type
& LOCK_MAND
) || (sys_fl
->fl_type
& LOCK_MAND
))
911 return locks_conflict(caller_fl
, sys_fl
);
915 posix_test_lock(struct file
*filp
, struct file_lock
*fl
)
917 struct file_lock
*cfl
;
918 struct file_lock_context
*ctx
;
919 struct inode
*inode
= locks_inode(filp
);
921 ctx
= smp_load_acquire(&inode
->i_flctx
);
922 if (!ctx
|| list_empty_careful(&ctx
->flc_posix
)) {
923 fl
->fl_type
= F_UNLCK
;
927 spin_lock(&ctx
->flc_lock
);
928 list_for_each_entry(cfl
, &ctx
->flc_posix
, fl_list
) {
929 if (posix_locks_conflict(fl
, cfl
)) {
930 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
;
1107 ctx
= locks_get_lock_context(inode
, request
->fl_type
);
1109 return (request
->fl_type
== F_UNLCK
) ? 0 : -ENOMEM
;
1112 * We may need two file_lock structures for this operation,
1113 * so we get them in advance to avoid races.
1115 * In some cases we can be sure, that no new locks will be needed
1117 if (!(request
->fl_flags
& FL_ACCESS
) &&
1118 (request
->fl_type
!= F_UNLCK
||
1119 request
->fl_start
!= 0 || request
->fl_end
!= OFFSET_MAX
)) {
1120 new_fl
= locks_alloc_lock();
1121 new_fl2
= locks_alloc_lock();
1124 percpu_down_read(&file_rwsem
);
1125 spin_lock(&ctx
->flc_lock
);
1127 * New lock request. Walk all POSIX locks and look for conflicts. If
1128 * there are any, either return error or put the request on the
1129 * blocker's list of waiters and the global blocked_hash.
1131 if (request
->fl_type
!= F_UNLCK
) {
1132 list_for_each_entry(fl
, &ctx
->flc_posix
, fl_list
) {
1133 if (!posix_locks_conflict(request
, fl
))
1136 locks_copy_conflock(conflock
, fl
);
1138 if (!(request
->fl_flags
& FL_SLEEP
))
1141 * Deadlock detection and insertion into the blocked
1142 * locks list must be done while holding the same lock!
1145 spin_lock(&blocked_lock_lock
);
1147 * Ensure that we don't find any locks blocked on this
1148 * request during deadlock detection.
1150 __locks_wake_up_blocks(request
);
1151 if (likely(!posix_locks_deadlock(request
, fl
))) {
1152 error
= FILE_LOCK_DEFERRED
;
1153 __locks_insert_block(fl
, request
,
1154 posix_locks_conflict
);
1156 spin_unlock(&blocked_lock_lock
);
1161 /* If we're just looking for a conflict, we're done. */
1163 if (request
->fl_flags
& FL_ACCESS
)
1166 /* Find the first old lock with the same owner as the new lock */
1167 list_for_each_entry(fl
, &ctx
->flc_posix
, fl_list
) {
1168 if (posix_same_owner(request
, fl
))
1172 /* Process locks with this owner. */
1173 list_for_each_entry_safe_from(fl
, tmp
, &ctx
->flc_posix
, fl_list
) {
1174 if (!posix_same_owner(request
, fl
))
1177 /* Detect adjacent or overlapping regions (if same lock type) */
1178 if (request
->fl_type
== fl
->fl_type
) {
1179 /* In all comparisons of start vs end, use
1180 * "start - 1" rather than "end + 1". If end
1181 * is OFFSET_MAX, end + 1 will become negative.
1183 if (fl
->fl_end
< request
->fl_start
- 1)
1185 /* If the next lock in the list has entirely bigger
1186 * addresses than the new one, insert the lock here.
1188 if (fl
->fl_start
- 1 > request
->fl_end
)
1191 /* If we come here, the new and old lock are of the
1192 * same type and adjacent or overlapping. Make one
1193 * lock yielding from the lower start address of both
1194 * locks to the higher end address.
1196 if (fl
->fl_start
> request
->fl_start
)
1197 fl
->fl_start
= request
->fl_start
;
1199 request
->fl_start
= fl
->fl_start
;
1200 if (fl
->fl_end
< request
->fl_end
)
1201 fl
->fl_end
= request
->fl_end
;
1203 request
->fl_end
= fl
->fl_end
;
1205 locks_delete_lock_ctx(fl
, &dispose
);
1211 /* Processing for different lock types is a bit
1214 if (fl
->fl_end
< request
->fl_start
)
1216 if (fl
->fl_start
> request
->fl_end
)
1218 if (request
->fl_type
== F_UNLCK
)
1220 if (fl
->fl_start
< request
->fl_start
)
1222 /* If the next lock in the list has a higher end
1223 * address than the new one, insert the new one here.
1225 if (fl
->fl_end
> request
->fl_end
) {
1229 if (fl
->fl_start
>= request
->fl_start
) {
1230 /* The new lock completely replaces an old
1231 * one (This may happen several times).
1234 locks_delete_lock_ctx(fl
, &dispose
);
1238 * Replace the old lock with new_fl, and
1239 * remove the old one. It's safe to do the
1240 * insert here since we know that we won't be
1241 * using new_fl later, and that the lock is
1242 * just replacing an existing lock.
1247 locks_copy_lock(new_fl
, request
);
1250 locks_insert_lock_ctx(request
, &fl
->fl_list
);
1251 locks_delete_lock_ctx(fl
, &dispose
);
1258 * The above code only modifies existing locks in case of merging or
1259 * replacing. If new lock(s) need to be inserted all modifications are
1260 * done below this, so it's safe yet to bail out.
1262 error
= -ENOLCK
; /* "no luck" */
1263 if (right
&& left
== right
&& !new_fl2
)
1268 if (request
->fl_type
== F_UNLCK
) {
1269 if (request
->fl_flags
& FL_EXISTS
)
1278 locks_copy_lock(new_fl
, request
);
1279 locks_move_blocks(new_fl
, request
);
1280 locks_insert_lock_ctx(new_fl
, &fl
->fl_list
);
1285 if (left
== right
) {
1286 /* The new lock breaks the old one in two pieces,
1287 * so we have to use the second new lock.
1291 locks_copy_lock(left
, right
);
1292 locks_insert_lock_ctx(left
, &fl
->fl_list
);
1294 right
->fl_start
= request
->fl_end
+ 1;
1295 locks_wake_up_blocks(right
);
1298 left
->fl_end
= request
->fl_start
- 1;
1299 locks_wake_up_blocks(left
);
1302 spin_unlock(&ctx
->flc_lock
);
1303 percpu_up_read(&file_rwsem
);
1305 * Free any unused locks.
1308 locks_free_lock(new_fl
);
1310 locks_free_lock(new_fl2
);
1311 locks_dispose_list(&dispose
);
1312 trace_posix_lock_inode(inode
, request
, error
);
1318 * posix_lock_file - Apply a POSIX-style lock to a file
1319 * @filp: The file to apply the lock to
1320 * @fl: The lock to be applied
1321 * @conflock: Place to return a copy of the conflicting lock, if found.
1323 * Add a POSIX style lock to a file.
1324 * We merge adjacent & overlapping locks whenever possible.
1325 * POSIX locks are sorted by owner task, then by starting address
1327 * Note that if called with an FL_EXISTS argument, the caller may determine
1328 * whether or not a lock was successfully freed by testing the return
1329 * value for -ENOENT.
1331 int posix_lock_file(struct file
*filp
, struct file_lock
*fl
,
1332 struct file_lock
*conflock
)
1334 return posix_lock_inode(locks_inode(filp
), fl
, conflock
);
1336 EXPORT_SYMBOL(posix_lock_file
);
1339 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1340 * @inode: inode of file to which lock request should be applied
1341 * @fl: The lock to be applied
1343 * Apply a POSIX style lock request to an inode.
1345 static int posix_lock_inode_wait(struct inode
*inode
, struct file_lock
*fl
)
1350 error
= posix_lock_inode(inode
, fl
, NULL
);
1351 if (error
!= FILE_LOCK_DEFERRED
)
1353 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_blocker
);
1357 locks_delete_block(fl
);
1361 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1363 * locks_mandatory_locked - Check for an active lock
1364 * @file: the file to check
1366 * Searches the inode's list of locks to find any POSIX locks which conflict.
1367 * This function is called from locks_verify_locked() only.
1369 int locks_mandatory_locked(struct file
*file
)
1372 struct inode
*inode
= locks_inode(file
);
1373 struct file_lock_context
*ctx
;
1374 struct file_lock
*fl
;
1376 ctx
= smp_load_acquire(&inode
->i_flctx
);
1377 if (!ctx
|| list_empty_careful(&ctx
->flc_posix
))
1381 * Search the lock list for this inode for any POSIX locks.
1383 spin_lock(&ctx
->flc_lock
);
1385 list_for_each_entry(fl
, &ctx
->flc_posix
, fl_list
) {
1386 if (fl
->fl_owner
!= current
->files
&&
1387 fl
->fl_owner
!= file
) {
1392 spin_unlock(&ctx
->flc_lock
);
1397 * locks_mandatory_area - Check for a conflicting lock
1398 * @inode: the file to check
1399 * @filp: how the file was opened (if it was)
1400 * @start: first byte in the file to check
1401 * @end: lastbyte in the file to check
1402 * @type: %F_WRLCK for a write lock, else %F_RDLCK
1404 * Searches the inode's list of locks to find any POSIX locks which conflict.
1406 int locks_mandatory_area(struct inode
*inode
, struct file
*filp
, loff_t start
,
1407 loff_t end
, unsigned char type
)
1409 struct file_lock fl
;
1413 locks_init_lock(&fl
);
1414 fl
.fl_pid
= current
->tgid
;
1416 fl
.fl_flags
= FL_POSIX
| FL_ACCESS
;
1417 if (filp
&& !(filp
->f_flags
& O_NONBLOCK
))
1420 fl
.fl_start
= start
;
1426 fl
.fl_flags
&= ~FL_SLEEP
;
1427 error
= posix_lock_inode(inode
, &fl
, NULL
);
1433 fl
.fl_flags
|= FL_SLEEP
;
1434 fl
.fl_owner
= current
->files
;
1435 error
= posix_lock_inode(inode
, &fl
, NULL
);
1436 if (error
!= FILE_LOCK_DEFERRED
)
1438 error
= wait_event_interruptible(fl
.fl_wait
, !fl
.fl_blocker
);
1441 * If we've been sleeping someone might have
1442 * changed the permissions behind our back.
1444 if (__mandatory_lock(inode
))
1450 locks_delete_block(&fl
);
1454 EXPORT_SYMBOL(locks_mandatory_area
);
1455 #endif /* CONFIG_MANDATORY_FILE_LOCKING */
1457 static void lease_clear_pending(struct file_lock
*fl
, int arg
)
1461 fl
->fl_flags
&= ~FL_UNLOCK_PENDING
;
1464 fl
->fl_flags
&= ~FL_DOWNGRADE_PENDING
;
1468 /* We already had a lease on this file; just change its type */
1469 int lease_modify(struct file_lock
*fl
, int arg
, struct list_head
*dispose
)
1471 int error
= assign_type(fl
, arg
);
1475 lease_clear_pending(fl
, arg
);
1476 locks_wake_up_blocks(fl
);
1477 if (arg
== F_UNLCK
) {
1478 struct file
*filp
= fl
->fl_file
;
1481 filp
->f_owner
.signum
= 0;
1482 fasync_helper(0, fl
->fl_file
, 0, &fl
->fl_fasync
);
1483 if (fl
->fl_fasync
!= NULL
) {
1484 printk(KERN_ERR
"locks_delete_lock: fasync == %p\n", fl
->fl_fasync
);
1485 fl
->fl_fasync
= NULL
;
1487 locks_delete_lock_ctx(fl
, dispose
);
1491 EXPORT_SYMBOL(lease_modify
);
1493 static bool past_time(unsigned long then
)
1496 /* 0 is a special value meaning "this never expires": */
1498 return time_after(jiffies
, then
);
1501 static void time_out_leases(struct inode
*inode
, struct list_head
*dispose
)
1503 struct file_lock_context
*ctx
= inode
->i_flctx
;
1504 struct file_lock
*fl
, *tmp
;
1506 lockdep_assert_held(&ctx
->flc_lock
);
1508 list_for_each_entry_safe(fl
, tmp
, &ctx
->flc_lease
, fl_list
) {
1509 trace_time_out_leases(inode
, fl
);
1510 if (past_time(fl
->fl_downgrade_time
))
1511 lease_modify(fl
, F_RDLCK
, dispose
);
1512 if (past_time(fl
->fl_break_time
))
1513 lease_modify(fl
, F_UNLCK
, dispose
);
1517 static bool leases_conflict(struct file_lock
*lease
, struct file_lock
*breaker
)
1521 if ((breaker
->fl_flags
& FL_LAYOUT
) != (lease
->fl_flags
& FL_LAYOUT
)) {
1525 if ((breaker
->fl_flags
& FL_DELEG
) && (lease
->fl_flags
& FL_LEASE
)) {
1530 rc
= locks_conflict(breaker
, lease
);
1532 trace_leases_conflict(rc
, lease
, breaker
);
1537 any_leases_conflict(struct inode
*inode
, struct file_lock
*breaker
)
1539 struct file_lock_context
*ctx
= inode
->i_flctx
;
1540 struct file_lock
*fl
;
1542 lockdep_assert_held(&ctx
->flc_lock
);
1544 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1545 if (leases_conflict(fl
, breaker
))
1552 * __break_lease - revoke all outstanding leases on file
1553 * @inode: the inode of the file to return
1554 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1556 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1559 * break_lease (inlined for speed) has checked there already is at least
1560 * some kind of lock (maybe a lease) on this file. Leases are broken on
1561 * a call to open() or truncate(). This function can sleep unless you
1562 * specified %O_NONBLOCK to your open().
1564 int __break_lease(struct inode
*inode
, unsigned int mode
, unsigned int type
)
1567 struct file_lock_context
*ctx
;
1568 struct file_lock
*new_fl
, *fl
, *tmp
;
1569 unsigned long break_time
;
1570 int want_write
= (mode
& O_ACCMODE
) != O_RDONLY
;
1573 new_fl
= lease_alloc(NULL
, want_write
? F_WRLCK
: F_RDLCK
);
1575 return PTR_ERR(new_fl
);
1576 new_fl
->fl_flags
= type
;
1578 /* typically we will check that ctx is non-NULL before calling */
1579 ctx
= smp_load_acquire(&inode
->i_flctx
);
1585 percpu_down_read(&file_rwsem
);
1586 spin_lock(&ctx
->flc_lock
);
1588 time_out_leases(inode
, &dispose
);
1590 if (!any_leases_conflict(inode
, new_fl
))
1594 if (lease_break_time
> 0) {
1595 break_time
= jiffies
+ lease_break_time
* HZ
;
1596 if (break_time
== 0)
1597 break_time
++; /* so that 0 means no break time */
1600 list_for_each_entry_safe(fl
, tmp
, &ctx
->flc_lease
, fl_list
) {
1601 if (!leases_conflict(fl
, new_fl
))
1604 if (fl
->fl_flags
& FL_UNLOCK_PENDING
)
1606 fl
->fl_flags
|= FL_UNLOCK_PENDING
;
1607 fl
->fl_break_time
= break_time
;
1609 if (lease_breaking(fl
))
1611 fl
->fl_flags
|= FL_DOWNGRADE_PENDING
;
1612 fl
->fl_downgrade_time
= break_time
;
1614 if (fl
->fl_lmops
->lm_break(fl
))
1615 locks_delete_lock_ctx(fl
, &dispose
);
1618 if (list_empty(&ctx
->flc_lease
))
1621 if (mode
& O_NONBLOCK
) {
1622 trace_break_lease_noblock(inode
, new_fl
);
1623 error
= -EWOULDBLOCK
;
1628 fl
= list_first_entry(&ctx
->flc_lease
, struct file_lock
, fl_list
);
1629 break_time
= fl
->fl_break_time
;
1630 if (break_time
!= 0)
1631 break_time
-= jiffies
;
1632 if (break_time
== 0)
1634 locks_insert_block(fl
, new_fl
, leases_conflict
);
1635 trace_break_lease_block(inode
, new_fl
);
1636 spin_unlock(&ctx
->flc_lock
);
1637 percpu_up_read(&file_rwsem
);
1639 locks_dispose_list(&dispose
);
1640 error
= wait_event_interruptible_timeout(new_fl
->fl_wait
,
1641 !new_fl
->fl_blocker
, break_time
);
1643 percpu_down_read(&file_rwsem
);
1644 spin_lock(&ctx
->flc_lock
);
1645 trace_break_lease_unblock(inode
, new_fl
);
1646 locks_delete_block(new_fl
);
1649 * Wait for the next conflicting lease that has not been
1653 time_out_leases(inode
, &dispose
);
1654 if (any_leases_conflict(inode
, new_fl
))
1659 spin_unlock(&ctx
->flc_lock
);
1660 percpu_up_read(&file_rwsem
);
1661 locks_dispose_list(&dispose
);
1663 locks_free_lock(new_fl
);
1666 EXPORT_SYMBOL(__break_lease
);
1669 * lease_get_mtime - update modified time of an inode with exclusive lease
1671 * @time: pointer to a timespec which contains the last modified time
1673 * This is to force NFS clients to flush their caches for files with
1674 * exclusive leases. The justification is that if someone has an
1675 * exclusive lease, then they could be modifying it.
1677 void lease_get_mtime(struct inode
*inode
, struct timespec64
*time
)
1679 bool has_lease
= false;
1680 struct file_lock_context
*ctx
;
1681 struct file_lock
*fl
;
1683 ctx
= smp_load_acquire(&inode
->i_flctx
);
1684 if (ctx
&& !list_empty_careful(&ctx
->flc_lease
)) {
1685 spin_lock(&ctx
->flc_lock
);
1686 fl
= list_first_entry_or_null(&ctx
->flc_lease
,
1687 struct file_lock
, fl_list
);
1688 if (fl
&& (fl
->fl_type
== F_WRLCK
))
1690 spin_unlock(&ctx
->flc_lock
);
1694 *time
= current_time(inode
);
1696 EXPORT_SYMBOL(lease_get_mtime
);
1699 * fcntl_getlease - Enquire what lease is currently active
1702 * The value returned by this function will be one of
1703 * (if no lease break is pending):
1705 * %F_RDLCK to indicate a shared lease is held.
1707 * %F_WRLCK to indicate an exclusive lease is held.
1709 * %F_UNLCK to indicate no lease is held.
1711 * (if a lease break is pending):
1713 * %F_RDLCK to indicate an exclusive lease needs to be
1714 * changed to a shared lease (or removed).
1716 * %F_UNLCK to indicate the lease needs to be removed.
1718 * XXX: sfr & willy disagree over whether F_INPROGRESS
1719 * should be returned to userspace.
1721 int fcntl_getlease(struct file
*filp
)
1723 struct file_lock
*fl
;
1724 struct inode
*inode
= locks_inode(filp
);
1725 struct file_lock_context
*ctx
;
1729 ctx
= smp_load_acquire(&inode
->i_flctx
);
1730 if (ctx
&& !list_empty_careful(&ctx
->flc_lease
)) {
1731 percpu_down_read(&file_rwsem
);
1732 spin_lock(&ctx
->flc_lock
);
1733 time_out_leases(inode
, &dispose
);
1734 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1735 if (fl
->fl_file
!= filp
)
1737 type
= target_leasetype(fl
);
1740 spin_unlock(&ctx
->flc_lock
);
1741 percpu_up_read(&file_rwsem
);
1743 locks_dispose_list(&dispose
);
1749 * check_conflicting_open - see if the given file points to an inode that has
1750 * an existing open that would conflict with the
1752 * @filp: file to check
1753 * @arg: type of lease that we're trying to acquire
1754 * @flags: current lock flags
1756 * Check to see if there's an existing open fd on this file that would
1757 * conflict with the lease we're trying to set.
1760 check_conflicting_open(struct file
*filp
, const long arg
, int flags
)
1762 struct inode
*inode
= locks_inode(filp
);
1763 int self_wcount
= 0, self_rcount
= 0;
1765 if (flags
& FL_LAYOUT
)
1769 return inode_is_open_for_write(inode
) ? -EAGAIN
: 0;
1770 else if (arg
!= F_WRLCK
)
1774 * Make sure that only read/write count is from lease requestor.
1775 * Note that this will result in denying write leases when i_writecount
1776 * is negative, which is what we want. (We shouldn't grant write leases
1777 * on files open for execution.)
1779 if (filp
->f_mode
& FMODE_WRITE
)
1781 else if (filp
->f_mode
& FMODE_READ
)
1784 if (atomic_read(&inode
->i_writecount
) != self_wcount
||
1785 atomic_read(&inode
->i_readcount
) != self_rcount
)
1792 generic_add_lease(struct file
*filp
, long arg
, struct file_lock
**flp
, void **priv
)
1794 struct file_lock
*fl
, *my_fl
= NULL
, *lease
;
1795 struct inode
*inode
= locks_inode(filp
);
1796 struct file_lock_context
*ctx
;
1797 bool is_deleg
= (*flp
)->fl_flags
& FL_DELEG
;
1802 trace_generic_add_lease(inode
, lease
);
1804 /* Note that arg is never F_UNLCK here */
1805 ctx
= locks_get_lock_context(inode
, arg
);
1810 * In the delegation case we need mutual exclusion with
1811 * a number of operations that take the i_mutex. We trylock
1812 * because delegations are an optional optimization, and if
1813 * there's some chance of a conflict--we'd rather not
1814 * bother, maybe that's a sign this just isn't a good file to
1815 * hand out a delegation on.
1817 if (is_deleg
&& !inode_trylock(inode
))
1820 if (is_deleg
&& arg
== F_WRLCK
) {
1821 /* Write delegations are not currently supported: */
1822 inode_unlock(inode
);
1827 percpu_down_read(&file_rwsem
);
1828 spin_lock(&ctx
->flc_lock
);
1829 time_out_leases(inode
, &dispose
);
1830 error
= check_conflicting_open(filp
, arg
, lease
->fl_flags
);
1835 * At this point, we know that if there is an exclusive
1836 * lease on this file, then we hold it on this filp
1837 * (otherwise our open of this file would have blocked).
1838 * And if we are trying to acquire an exclusive lease,
1839 * then the file is not open by anyone (including us)
1840 * except for this filp.
1843 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1844 if (fl
->fl_file
== filp
&&
1845 fl
->fl_owner
== lease
->fl_owner
) {
1851 * No exclusive leases if someone else has a lease on
1857 * Modifying our existing lease is OK, but no getting a
1858 * new lease if someone else is opening for write:
1860 if (fl
->fl_flags
& FL_UNLOCK_PENDING
)
1864 if (my_fl
!= NULL
) {
1866 error
= lease
->fl_lmops
->lm_change(lease
, arg
, &dispose
);
1876 locks_insert_lock_ctx(lease
, &ctx
->flc_lease
);
1878 * The check in break_lease() is lockless. It's possible for another
1879 * open to race in after we did the earlier check for a conflicting
1880 * open but before the lease was inserted. Check again for a
1881 * conflicting open and cancel the lease if there is one.
1883 * We also add a barrier here to ensure that the insertion of the lock
1884 * precedes these checks.
1887 error
= check_conflicting_open(filp
, arg
, lease
->fl_flags
);
1889 locks_unlink_lock_ctx(lease
);
1894 if (lease
->fl_lmops
->lm_setup
)
1895 lease
->fl_lmops
->lm_setup(lease
, priv
);
1897 spin_unlock(&ctx
->flc_lock
);
1898 percpu_up_read(&file_rwsem
);
1899 locks_dispose_list(&dispose
);
1901 inode_unlock(inode
);
1902 if (!error
&& !my_fl
)
1907 static int generic_delete_lease(struct file
*filp
, void *owner
)
1909 int error
= -EAGAIN
;
1910 struct file_lock
*fl
, *victim
= NULL
;
1911 struct inode
*inode
= locks_inode(filp
);
1912 struct file_lock_context
*ctx
;
1915 ctx
= smp_load_acquire(&inode
->i_flctx
);
1917 trace_generic_delete_lease(inode
, NULL
);
1921 percpu_down_read(&file_rwsem
);
1922 spin_lock(&ctx
->flc_lock
);
1923 list_for_each_entry(fl
, &ctx
->flc_lease
, fl_list
) {
1924 if (fl
->fl_file
== filp
&&
1925 fl
->fl_owner
== owner
) {
1930 trace_generic_delete_lease(inode
, victim
);
1932 error
= fl
->fl_lmops
->lm_change(victim
, F_UNLCK
, &dispose
);
1933 spin_unlock(&ctx
->flc_lock
);
1934 percpu_up_read(&file_rwsem
);
1935 locks_dispose_list(&dispose
);
1940 * generic_setlease - sets a lease on an open file
1941 * @filp: file pointer
1942 * @arg: type of lease to obtain
1943 * @flp: input - file_lock to use, output - file_lock inserted
1944 * @priv: private data for lm_setup (may be NULL if lm_setup
1945 * doesn't require it)
1947 * The (input) flp->fl_lmops->lm_break function is required
1950 int generic_setlease(struct file
*filp
, long arg
, struct file_lock
**flp
,
1953 struct inode
*inode
= locks_inode(filp
);
1956 if ((!uid_eq(current_fsuid(), inode
->i_uid
)) && !capable(CAP_LEASE
))
1958 if (!S_ISREG(inode
->i_mode
))
1960 error
= security_file_lock(filp
, arg
);
1966 return generic_delete_lease(filp
, *priv
);
1969 if (!(*flp
)->fl_lmops
->lm_break
) {
1974 return generic_add_lease(filp
, arg
, flp
, priv
);
1979 EXPORT_SYMBOL(generic_setlease
);
1981 #if IS_ENABLED(CONFIG_SRCU)
1983 * Kernel subsystems can register to be notified on any attempt to set
1984 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1985 * to close files that it may have cached when there is an attempt to set a
1986 * conflicting lease.
1988 static struct srcu_notifier_head lease_notifier_chain
;
1991 lease_notifier_chain_init(void)
1993 srcu_init_notifier_head(&lease_notifier_chain
);
1997 setlease_notifier(long arg
, struct file_lock
*lease
)
2000 srcu_notifier_call_chain(&lease_notifier_chain
, arg
, lease
);
2003 int lease_register_notifier(struct notifier_block
*nb
)
2005 return srcu_notifier_chain_register(&lease_notifier_chain
, nb
);
2007 EXPORT_SYMBOL_GPL(lease_register_notifier
);
2009 void lease_unregister_notifier(struct notifier_block
*nb
)
2011 srcu_notifier_chain_unregister(&lease_notifier_chain
, nb
);
2013 EXPORT_SYMBOL_GPL(lease_unregister_notifier
);
2015 #else /* !IS_ENABLED(CONFIG_SRCU) */
2017 lease_notifier_chain_init(void)
2022 setlease_notifier(long arg
, struct file_lock
*lease
)
2026 int lease_register_notifier(struct notifier_block
*nb
)
2030 EXPORT_SYMBOL_GPL(lease_register_notifier
);
2032 void lease_unregister_notifier(struct notifier_block
*nb
)
2035 EXPORT_SYMBOL_GPL(lease_unregister_notifier
);
2037 #endif /* IS_ENABLED(CONFIG_SRCU) */
2040 * vfs_setlease - sets a lease on an open file
2041 * @filp: file pointer
2042 * @arg: type of lease to obtain
2043 * @lease: file_lock to use when adding a lease
2044 * @priv: private info for lm_setup when adding a lease (may be
2045 * NULL if lm_setup doesn't require it)
2047 * Call this to establish a lease on the file. The "lease" argument is not
2048 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2049 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2050 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2053 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2054 * may be NULL if the lm_setup operation doesn't require it.
2057 vfs_setlease(struct file
*filp
, long arg
, struct file_lock
**lease
, void **priv
)
2060 setlease_notifier(arg
, *lease
);
2061 if (filp
->f_op
->setlease
)
2062 return filp
->f_op
->setlease(filp
, arg
, lease
, priv
);
2064 return generic_setlease(filp
, arg
, lease
, priv
);
2066 EXPORT_SYMBOL_GPL(vfs_setlease
);
2068 static int do_fcntl_add_lease(unsigned int fd
, struct file
*filp
, long arg
)
2070 struct file_lock
*fl
;
2071 struct fasync_struct
*new;
2074 fl
= lease_alloc(filp
, arg
);
2078 new = fasync_alloc();
2080 locks_free_lock(fl
);
2085 error
= vfs_setlease(filp
, arg
, &fl
, (void **)&new);
2087 locks_free_lock(fl
);
2094 * fcntl_setlease - sets a lease on an open file
2095 * @fd: open file descriptor
2096 * @filp: file pointer
2097 * @arg: type of lease to obtain
2099 * Call this fcntl to establish a lease on the file.
2100 * Note that you also need to call %F_SETSIG to
2101 * receive a signal when the lease is broken.
2103 int fcntl_setlease(unsigned int fd
, struct file
*filp
, long arg
)
2106 return vfs_setlease(filp
, F_UNLCK
, NULL
, (void **)&filp
);
2107 return do_fcntl_add_lease(fd
, filp
, arg
);
2111 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2112 * @inode: inode of the file to apply to
2113 * @fl: The lock to be applied
2115 * Apply a FLOCK style lock request to an inode.
2117 static int flock_lock_inode_wait(struct inode
*inode
, struct file_lock
*fl
)
2122 error
= flock_lock_inode(inode
, fl
);
2123 if (error
!= FILE_LOCK_DEFERRED
)
2125 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_blocker
);
2129 locks_delete_block(fl
);
2134 * locks_lock_inode_wait - Apply a lock to an inode
2135 * @inode: inode of the file to apply to
2136 * @fl: The lock to be applied
2138 * Apply a POSIX or FLOCK style lock request to an inode.
2140 int locks_lock_inode_wait(struct inode
*inode
, struct file_lock
*fl
)
2143 switch (fl
->fl_flags
& (FL_POSIX
|FL_FLOCK
)) {
2145 res
= posix_lock_inode_wait(inode
, fl
);
2148 res
= flock_lock_inode_wait(inode
, fl
);
2155 EXPORT_SYMBOL(locks_lock_inode_wait
);
2158 * sys_flock: - flock() system call.
2159 * @fd: the file descriptor to lock.
2160 * @cmd: the type of lock to apply.
2162 * Apply a %FL_FLOCK style lock to an open file descriptor.
2163 * The @cmd can be one of:
2165 * - %LOCK_SH -- a shared lock.
2166 * - %LOCK_EX -- an exclusive lock.
2167 * - %LOCK_UN -- remove an existing lock.
2168 * - %LOCK_MAND -- a 'mandatory' flock.
2169 * This exists to emulate Windows Share Modes.
2171 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
2172 * processes read and write access respectively.
2174 SYSCALL_DEFINE2(flock
, unsigned int, fd
, unsigned int, cmd
)
2176 struct fd f
= fdget(fd
);
2177 struct file_lock
*lock
;
2178 int can_sleep
, unlock
;
2185 can_sleep
= !(cmd
& LOCK_NB
);
2187 unlock
= (cmd
== LOCK_UN
);
2189 if (!unlock
&& !(cmd
& LOCK_MAND
) &&
2190 !(f
.file
->f_mode
& (FMODE_READ
|FMODE_WRITE
)))
2193 lock
= flock_make_lock(f
.file
, cmd
, NULL
);
2195 error
= PTR_ERR(lock
);
2200 lock
->fl_flags
|= FL_SLEEP
;
2202 error
= security_file_lock(f
.file
, lock
->fl_type
);
2206 if (f
.file
->f_op
->flock
)
2207 error
= f
.file
->f_op
->flock(f
.file
,
2208 (can_sleep
) ? F_SETLKW
: F_SETLK
,
2211 error
= locks_lock_file_wait(f
.file
, lock
);
2214 locks_free_lock(lock
);
2223 * vfs_test_lock - test file byte range lock
2224 * @filp: The file to test lock for
2225 * @fl: The lock to test; also used to hold result
2227 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2228 * setting conf->fl_type to something other than F_UNLCK.
2230 int vfs_test_lock(struct file
*filp
, struct file_lock
*fl
)
2232 if (filp
->f_op
->lock
)
2233 return filp
->f_op
->lock(filp
, F_GETLK
, fl
);
2234 posix_test_lock(filp
, fl
);
2237 EXPORT_SYMBOL_GPL(vfs_test_lock
);
2240 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2241 * @fl: The file_lock who's fl_pid should be translated
2242 * @ns: The namespace into which the pid should be translated
2244 * Used to tranlate a fl_pid into a namespace virtual pid number
2246 static pid_t
locks_translate_pid(struct file_lock
*fl
, struct pid_namespace
*ns
)
2253 if (IS_REMOTELCK(fl
))
2256 * If the flock owner process is dead and its pid has been already
2257 * freed, the translation below won't work, but we still want to show
2258 * flock owner pid number in init pidns.
2260 if (ns
== &init_pid_ns
)
2261 return (pid_t
)fl
->fl_pid
;
2264 pid
= find_pid_ns(fl
->fl_pid
, &init_pid_ns
);
2265 vnr
= pid_nr_ns(pid
, ns
);
2270 static int posix_lock_to_flock(struct flock
*flock
, struct file_lock
*fl
)
2272 flock
->l_pid
= locks_translate_pid(fl
, task_active_pid_ns(current
));
2273 #if BITS_PER_LONG == 32
2275 * Make sure we can represent the posix lock via
2276 * legacy 32bit flock.
2278 if (fl
->fl_start
> OFFT_OFFSET_MAX
)
2280 if (fl
->fl_end
!= OFFSET_MAX
&& fl
->fl_end
> OFFT_OFFSET_MAX
)
2283 flock
->l_start
= fl
->fl_start
;
2284 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
2285 fl
->fl_end
- fl
->fl_start
+ 1;
2286 flock
->l_whence
= 0;
2287 flock
->l_type
= fl
->fl_type
;
2291 #if BITS_PER_LONG == 32
2292 static void posix_lock_to_flock64(struct flock64
*flock
, struct file_lock
*fl
)
2294 flock
->l_pid
= locks_translate_pid(fl
, task_active_pid_ns(current
));
2295 flock
->l_start
= fl
->fl_start
;
2296 flock
->l_len
= fl
->fl_end
== OFFSET_MAX
? 0 :
2297 fl
->fl_end
- fl
->fl_start
+ 1;
2298 flock
->l_whence
= 0;
2299 flock
->l_type
= fl
->fl_type
;
2303 /* Report the first existing lock that would conflict with l.
2304 * This implements the F_GETLK command of fcntl().
2306 int fcntl_getlk(struct file
*filp
, unsigned int cmd
, struct flock
*flock
)
2308 struct file_lock
*fl
;
2311 fl
= locks_alloc_lock();
2315 if (flock
->l_type
!= F_RDLCK
&& flock
->l_type
!= F_WRLCK
)
2318 error
= flock_to_posix_lock(filp
, fl
, flock
);
2322 if (cmd
== F_OFD_GETLK
) {
2324 if (flock
->l_pid
!= 0)
2328 fl
->fl_flags
|= FL_OFDLCK
;
2329 fl
->fl_owner
= filp
;
2332 error
= vfs_test_lock(filp
, fl
);
2336 flock
->l_type
= fl
->fl_type
;
2337 if (fl
->fl_type
!= F_UNLCK
) {
2338 error
= posix_lock_to_flock(flock
, fl
);
2343 locks_free_lock(fl
);
2348 * vfs_lock_file - file byte range lock
2349 * @filp: The file to apply the lock to
2350 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2351 * @fl: The lock to be applied
2352 * @conf: Place to return a copy of the conflicting lock, if found.
2354 * A caller that doesn't care about the conflicting lock may pass NULL
2355 * as the final argument.
2357 * If the filesystem defines a private ->lock() method, then @conf will
2358 * be left unchanged; so a caller that cares should initialize it to
2359 * some acceptable default.
2361 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2362 * locks, the ->lock() interface may return asynchronously, before the lock has
2363 * been granted or denied by the underlying filesystem, if (and only if)
2364 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2365 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2366 * the request is for a blocking lock. When ->lock() does return asynchronously,
2367 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2368 * request completes.
2369 * If the request is for non-blocking lock the file system should return
2370 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2371 * with the result. If the request timed out the callback routine will return a
2372 * nonzero return code and the file system should release the lock. The file
2373 * system is also responsible to keep a corresponding posix lock when it
2374 * grants a lock so the VFS can find out which locks are locally held and do
2375 * the correct lock cleanup when required.
2376 * The underlying filesystem must not drop the kernel lock or call
2377 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2380 int vfs_lock_file(struct file
*filp
, unsigned int cmd
, struct file_lock
*fl
, struct file_lock
*conf
)
2382 if (filp
->f_op
->lock
)
2383 return filp
->f_op
->lock(filp
, cmd
, fl
);
2385 return posix_lock_file(filp
, fl
, conf
);
2387 EXPORT_SYMBOL_GPL(vfs_lock_file
);
2389 static int do_lock_file_wait(struct file
*filp
, unsigned int cmd
,
2390 struct file_lock
*fl
)
2394 error
= security_file_lock(filp
, fl
->fl_type
);
2399 error
= vfs_lock_file(filp
, cmd
, fl
, NULL
);
2400 if (error
!= FILE_LOCK_DEFERRED
)
2402 error
= wait_event_interruptible(fl
->fl_wait
, !fl
->fl_blocker
);
2406 locks_delete_block(fl
);
2411 /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2413 check_fmode_for_setlk(struct file_lock
*fl
)
2415 switch (fl
->fl_type
) {
2417 if (!(fl
->fl_file
->f_mode
& FMODE_READ
))
2421 if (!(fl
->fl_file
->f_mode
& FMODE_WRITE
))
2427 /* Apply the lock described by l to an open file descriptor.
2428 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2430 int fcntl_setlk(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
2431 struct flock
*flock
)
2433 struct file_lock
*file_lock
= locks_alloc_lock();
2434 struct inode
*inode
= locks_inode(filp
);
2438 if (file_lock
== NULL
)
2441 /* Don't allow mandatory locks on files that may be memory mapped
2444 if (mandatory_lock(inode
) && mapping_writably_mapped(filp
->f_mapping
)) {
2449 error
= flock_to_posix_lock(filp
, file_lock
, flock
);
2453 error
= check_fmode_for_setlk(file_lock
);
2458 * If the cmd is requesting file-private locks, then set the
2459 * FL_OFDLCK flag and override the owner.
2464 if (flock
->l_pid
!= 0)
2468 file_lock
->fl_flags
|= FL_OFDLCK
;
2469 file_lock
->fl_owner
= filp
;
2473 if (flock
->l_pid
!= 0)
2477 file_lock
->fl_flags
|= FL_OFDLCK
;
2478 file_lock
->fl_owner
= filp
;
2481 file_lock
->fl_flags
|= FL_SLEEP
;
2484 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2487 * Attempt to detect a close/fcntl race and recover by releasing the
2488 * lock that was just acquired. There is no need to do that when we're
2489 * unlocking though, or for OFD locks.
2491 if (!error
&& file_lock
->fl_type
!= F_UNLCK
&&
2492 !(file_lock
->fl_flags
& FL_OFDLCK
)) {
2494 * We need that spin_lock here - it prevents reordering between
2495 * update of i_flctx->flc_posix and check for it done in
2496 * close(). rcu_read_lock() wouldn't do.
2498 spin_lock(¤t
->files
->file_lock
);
2500 spin_unlock(¤t
->files
->file_lock
);
2502 file_lock
->fl_type
= F_UNLCK
;
2503 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2504 WARN_ON_ONCE(error
);
2509 trace_fcntl_setlk(inode
, file_lock
, error
);
2510 locks_free_lock(file_lock
);
2514 #if BITS_PER_LONG == 32
2515 /* Report the first existing lock that would conflict with l.
2516 * This implements the F_GETLK command of fcntl().
2518 int fcntl_getlk64(struct file
*filp
, unsigned int cmd
, struct flock64
*flock
)
2520 struct file_lock
*fl
;
2523 fl
= locks_alloc_lock();
2528 if (flock
->l_type
!= F_RDLCK
&& flock
->l_type
!= F_WRLCK
)
2531 error
= flock64_to_posix_lock(filp
, fl
, flock
);
2535 if (cmd
== F_OFD_GETLK
) {
2537 if (flock
->l_pid
!= 0)
2541 fl
->fl_flags
|= FL_OFDLCK
;
2542 fl
->fl_owner
= filp
;
2545 error
= vfs_test_lock(filp
, fl
);
2549 flock
->l_type
= fl
->fl_type
;
2550 if (fl
->fl_type
!= F_UNLCK
)
2551 posix_lock_to_flock64(flock
, fl
);
2554 locks_free_lock(fl
);
2558 /* Apply the lock described by l to an open file descriptor.
2559 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2561 int fcntl_setlk64(unsigned int fd
, struct file
*filp
, unsigned int cmd
,
2562 struct flock64
*flock
)
2564 struct file_lock
*file_lock
= locks_alloc_lock();
2565 struct inode
*inode
= locks_inode(filp
);
2569 if (file_lock
== NULL
)
2572 /* Don't allow mandatory locks on files that may be memory mapped
2575 if (mandatory_lock(inode
) && mapping_writably_mapped(filp
->f_mapping
)) {
2580 error
= flock64_to_posix_lock(filp
, file_lock
, flock
);
2584 error
= check_fmode_for_setlk(file_lock
);
2589 * If the cmd is requesting file-private locks, then set the
2590 * FL_OFDLCK flag and override the owner.
2595 if (flock
->l_pid
!= 0)
2599 file_lock
->fl_flags
|= FL_OFDLCK
;
2600 file_lock
->fl_owner
= filp
;
2604 if (flock
->l_pid
!= 0)
2608 file_lock
->fl_flags
|= FL_OFDLCK
;
2609 file_lock
->fl_owner
= filp
;
2612 file_lock
->fl_flags
|= FL_SLEEP
;
2615 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2618 * Attempt to detect a close/fcntl race and recover by releasing the
2619 * lock that was just acquired. There is no need to do that when we're
2620 * unlocking though, or for OFD locks.
2622 if (!error
&& file_lock
->fl_type
!= F_UNLCK
&&
2623 !(file_lock
->fl_flags
& FL_OFDLCK
)) {
2625 * We need that spin_lock here - it prevents reordering between
2626 * update of i_flctx->flc_posix and check for it done in
2627 * close(). rcu_read_lock() wouldn't do.
2629 spin_lock(¤t
->files
->file_lock
);
2631 spin_unlock(¤t
->files
->file_lock
);
2633 file_lock
->fl_type
= F_UNLCK
;
2634 error
= do_lock_file_wait(filp
, cmd
, file_lock
);
2635 WARN_ON_ONCE(error
);
2640 locks_free_lock(file_lock
);
2643 #endif /* BITS_PER_LONG == 32 */
2646 * This function is called when the file is being removed
2647 * from the task's fd array. POSIX locks belonging to this task
2648 * are deleted at this time.
2650 void locks_remove_posix(struct file
*filp
, fl_owner_t owner
)
2653 struct inode
*inode
= locks_inode(filp
);
2654 struct file_lock lock
;
2655 struct file_lock_context
*ctx
;
2658 * If there are no locks held on this file, we don't need to call
2659 * posix_lock_file(). Another process could be setting a lock on this
2660 * file at the same time, but we wouldn't remove that lock anyway.
2662 ctx
= smp_load_acquire(&inode
->i_flctx
);
2663 if (!ctx
|| list_empty(&ctx
->flc_posix
))
2666 locks_init_lock(&lock
);
2667 lock
.fl_type
= F_UNLCK
;
2668 lock
.fl_flags
= FL_POSIX
| FL_CLOSE
;
2670 lock
.fl_end
= OFFSET_MAX
;
2671 lock
.fl_owner
= owner
;
2672 lock
.fl_pid
= current
->tgid
;
2673 lock
.fl_file
= filp
;
2675 lock
.fl_lmops
= NULL
;
2677 error
= vfs_lock_file(filp
, F_SETLK
, &lock
, NULL
);
2679 if (lock
.fl_ops
&& lock
.fl_ops
->fl_release_private
)
2680 lock
.fl_ops
->fl_release_private(&lock
);
2681 trace_locks_remove_posix(inode
, &lock
, error
);
2683 EXPORT_SYMBOL(locks_remove_posix
);
2685 /* The i_flctx must be valid when calling into here */
2687 locks_remove_flock(struct file
*filp
, struct file_lock_context
*flctx
)
2689 struct file_lock fl
;
2690 struct inode
*inode
= locks_inode(filp
);
2692 if (list_empty(&flctx
->flc_flock
))
2695 flock_make_lock(filp
, LOCK_UN
, &fl
);
2696 fl
.fl_flags
|= FL_CLOSE
;
2698 if (filp
->f_op
->flock
)
2699 filp
->f_op
->flock(filp
, F_SETLKW
, &fl
);
2701 flock_lock_inode(inode
, &fl
);
2703 if (fl
.fl_ops
&& fl
.fl_ops
->fl_release_private
)
2704 fl
.fl_ops
->fl_release_private(&fl
);
2707 /* The i_flctx must be valid when calling into here */
2709 locks_remove_lease(struct file
*filp
, struct file_lock_context
*ctx
)
2711 struct file_lock
*fl
, *tmp
;
2714 if (list_empty(&ctx
->flc_lease
))
2717 percpu_down_read(&file_rwsem
);
2718 spin_lock(&ctx
->flc_lock
);
2719 list_for_each_entry_safe(fl
, tmp
, &ctx
->flc_lease
, fl_list
)
2720 if (filp
== fl
->fl_file
)
2721 lease_modify(fl
, F_UNLCK
, &dispose
);
2722 spin_unlock(&ctx
->flc_lock
);
2723 percpu_up_read(&file_rwsem
);
2725 locks_dispose_list(&dispose
);
2729 * This function is called on the last close of an open file.
2731 void locks_remove_file(struct file
*filp
)
2733 struct file_lock_context
*ctx
;
2735 ctx
= smp_load_acquire(&locks_inode(filp
)->i_flctx
);
2739 /* remove any OFD locks */
2740 locks_remove_posix(filp
, filp
);
2742 /* remove flock locks */
2743 locks_remove_flock(filp
, ctx
);
2745 /* remove any leases */
2746 locks_remove_lease(filp
, ctx
);
2748 spin_lock(&ctx
->flc_lock
);
2749 locks_check_ctx_file_list(filp
, &ctx
->flc_posix
, "POSIX");
2750 locks_check_ctx_file_list(filp
, &ctx
->flc_flock
, "FLOCK");
2751 locks_check_ctx_file_list(filp
, &ctx
->flc_lease
, "LEASE");
2752 spin_unlock(&ctx
->flc_lock
);
2756 * vfs_cancel_lock - file byte range unblock lock
2757 * @filp: The file to apply the unblock to
2758 * @fl: The lock to be unblocked
2760 * Used by lock managers to cancel blocked requests
2762 int vfs_cancel_lock(struct file
*filp
, struct file_lock
*fl
)
2764 if (filp
->f_op
->lock
)
2765 return filp
->f_op
->lock(filp
, F_CANCELLK
, fl
);
2768 EXPORT_SYMBOL_GPL(vfs_cancel_lock
);
2770 #ifdef CONFIG_PROC_FS
2771 #include <linux/proc_fs.h>
2772 #include <linux/seq_file.h>
2774 struct locks_iterator
{
2779 static void lock_get_status(struct seq_file
*f
, struct file_lock
*fl
,
2780 loff_t id
, char *pfx
)
2782 struct inode
*inode
= NULL
;
2783 unsigned int fl_pid
;
2784 struct pid_namespace
*proc_pidns
= file_inode(f
->file
)->i_sb
->s_fs_info
;
2786 fl_pid
= locks_translate_pid(fl
, proc_pidns
);
2788 * If lock owner is dead (and pid is freed) or not visible in current
2789 * pidns, zero is shown as a pid value. Check lock info from
2790 * init_pid_ns to get saved lock pid value.
2793 if (fl
->fl_file
!= NULL
)
2794 inode
= locks_inode(fl
->fl_file
);
2796 seq_printf(f
, "%lld:%s ", id
, pfx
);
2798 if (fl
->fl_flags
& FL_ACCESS
)
2799 seq_puts(f
, "ACCESS");
2800 else if (IS_OFDLCK(fl
))
2801 seq_puts(f
, "OFDLCK");
2803 seq_puts(f
, "POSIX ");
2805 seq_printf(f
, " %s ",
2806 (inode
== NULL
) ? "*NOINODE*" :
2807 mandatory_lock(inode
) ? "MANDATORY" : "ADVISORY ");
2808 } else if (IS_FLOCK(fl
)) {
2809 if (fl
->fl_type
& LOCK_MAND
) {
2810 seq_puts(f
, "FLOCK MSNFS ");
2812 seq_puts(f
, "FLOCK ADVISORY ");
2814 } else if (IS_LEASE(fl
)) {
2815 if (fl
->fl_flags
& FL_DELEG
)
2816 seq_puts(f
, "DELEG ");
2818 seq_puts(f
, "LEASE ");
2820 if (lease_breaking(fl
))
2821 seq_puts(f
, "BREAKING ");
2822 else if (fl
->fl_file
)
2823 seq_puts(f
, "ACTIVE ");
2825 seq_puts(f
, "BREAKER ");
2827 seq_puts(f
, "UNKNOWN UNKNOWN ");
2829 if (fl
->fl_type
& LOCK_MAND
) {
2830 seq_printf(f
, "%s ",
2831 (fl
->fl_type
& LOCK_READ
)
2832 ? (fl
->fl_type
& LOCK_WRITE
) ? "RW " : "READ "
2833 : (fl
->fl_type
& LOCK_WRITE
) ? "WRITE" : "NONE ");
2835 int type
= IS_LEASE(fl
) ? target_leasetype(fl
) : fl
->fl_type
;
2837 seq_printf(f
, "%s ", (type
== F_WRLCK
) ? "WRITE" :
2838 (type
== F_RDLCK
) ? "READ" : "UNLCK");
2841 /* userspace relies on this representation of dev_t */
2842 seq_printf(f
, "%d %02x:%02x:%lu ", fl_pid
,
2843 MAJOR(inode
->i_sb
->s_dev
),
2844 MINOR(inode
->i_sb
->s_dev
), inode
->i_ino
);
2846 seq_printf(f
, "%d <none>:0 ", fl_pid
);
2849 if (fl
->fl_end
== OFFSET_MAX
)
2850 seq_printf(f
, "%Ld EOF\n", fl
->fl_start
);
2852 seq_printf(f
, "%Ld %Ld\n", fl
->fl_start
, fl
->fl_end
);
2854 seq_puts(f
, "0 EOF\n");
2858 static int locks_show(struct seq_file
*f
, void *v
)
2860 struct locks_iterator
*iter
= f
->private;
2861 struct file_lock
*fl
, *bfl
;
2862 struct pid_namespace
*proc_pidns
= file_inode(f
->file
)->i_sb
->s_fs_info
;
2864 fl
= hlist_entry(v
, struct file_lock
, fl_link
);
2866 if (locks_translate_pid(fl
, proc_pidns
) == 0)
2869 lock_get_status(f
, fl
, iter
->li_pos
, "");
2871 list_for_each_entry(bfl
, &fl
->fl_blocked_requests
, fl_blocked_member
)
2872 lock_get_status(f
, bfl
, iter
->li_pos
, " ->");
2877 static void __show_fd_locks(struct seq_file
*f
,
2878 struct list_head
*head
, int *id
,
2879 struct file
*filp
, struct files_struct
*files
)
2881 struct file_lock
*fl
;
2883 list_for_each_entry(fl
, head
, fl_list
) {
2885 if (filp
!= fl
->fl_file
)
2887 if (fl
->fl_owner
!= files
&&
2888 fl
->fl_owner
!= filp
)
2892 seq_puts(f
, "lock:\t");
2893 lock_get_status(f
, fl
, *id
, "");
2897 void show_fd_locks(struct seq_file
*f
,
2898 struct file
*filp
, struct files_struct
*files
)
2900 struct inode
*inode
= locks_inode(filp
);
2901 struct file_lock_context
*ctx
;
2904 ctx
= smp_load_acquire(&inode
->i_flctx
);
2908 spin_lock(&ctx
->flc_lock
);
2909 __show_fd_locks(f
, &ctx
->flc_flock
, &id
, filp
, files
);
2910 __show_fd_locks(f
, &ctx
->flc_posix
, &id
, filp
, files
);
2911 __show_fd_locks(f
, &ctx
->flc_lease
, &id
, filp
, files
);
2912 spin_unlock(&ctx
->flc_lock
);
2915 static void *locks_start(struct seq_file
*f
, loff_t
*pos
)
2916 __acquires(&blocked_lock_lock
)
2918 struct locks_iterator
*iter
= f
->private;
2920 iter
->li_pos
= *pos
+ 1;
2921 percpu_down_write(&file_rwsem
);
2922 spin_lock(&blocked_lock_lock
);
2923 return seq_hlist_start_percpu(&file_lock_list
.hlist
, &iter
->li_cpu
, *pos
);
2926 static void *locks_next(struct seq_file
*f
, void *v
, loff_t
*pos
)
2928 struct locks_iterator
*iter
= f
->private;
2931 return seq_hlist_next_percpu(v
, &file_lock_list
.hlist
, &iter
->li_cpu
, pos
);
2934 static void locks_stop(struct seq_file
*f
, void *v
)
2935 __releases(&blocked_lock_lock
)
2937 spin_unlock(&blocked_lock_lock
);
2938 percpu_up_write(&file_rwsem
);
2941 static const struct seq_operations locks_seq_operations
= {
2942 .start
= locks_start
,
2948 static int __init
proc_locks_init(void)
2950 proc_create_seq_private("locks", 0, NULL
, &locks_seq_operations
,
2951 sizeof(struct locks_iterator
), NULL
);
2954 fs_initcall(proc_locks_init
);
2957 static int __init
filelock_init(void)
2961 flctx_cache
= kmem_cache_create("file_lock_ctx",
2962 sizeof(struct file_lock_context
), 0, SLAB_PANIC
, NULL
);
2964 filelock_cache
= kmem_cache_create("file_lock_cache",
2965 sizeof(struct file_lock
), 0, SLAB_PANIC
, NULL
);
2967 for_each_possible_cpu(i
) {
2968 struct file_lock_list_struct
*fll
= per_cpu_ptr(&file_lock_list
, i
);
2970 spin_lock_init(&fll
->lock
);
2971 INIT_HLIST_HEAD(&fll
->hlist
);
2974 lease_notifier_chain_init();
2977 core_initcall(filelock_init
);