Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/locks.c
4 *
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
8 * fcntl(2).
9 *
10 *
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..
16 *
17 * Waiting and applied locks are all kept in trees whose properties are:
18 *
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.
22 *
23 * Every such tree begins life as a waiting singleton which obviously
24 * satisfies the above properties.
25 *
26 * The only ways we modify trees preserve these properties:
27 *
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
32 * children.
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).
37 *
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.
41 *
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
48 * children.
49 *
50 */
51
52#include <linux/capability.h>
53#include <linux/file.h>
54#include <linux/fdtable.h>
55#include <linux/fs.h>
56#include <linux/init.h>
57#include <linux/security.h>
58#include <linux/slab.h>
59#include <linux/syscalls.h>
60#include <linux/time.h>
61#include <linux/rcupdate.h>
62#include <linux/pid_namespace.h>
63#include <linux/hashtable.h>
64#include <linux/percpu.h>
65#include <linux/sysctl.h>
66
67#define CREATE_TRACE_POINTS
68#include <trace/events/filelock.h>
69
70#include <linux/uaccess.h>
71
72#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
73#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
74#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
75#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
76#define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
77
78static bool lease_breaking(struct file_lock *fl)
79{
80 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
81}
82
83static int target_leasetype(struct file_lock *fl)
84{
85 if (fl->fl_flags & FL_UNLOCK_PENDING)
86 return F_UNLCK;
87 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
88 return F_RDLCK;
89 return fl->fl_type;
90}
91
92static int leases_enable = 1;
93static int lease_break_time = 45;
94
95#ifdef CONFIG_SYSCTL
96static struct ctl_table locks_sysctls[] = {
97 {
98 .procname = "leases-enable",
99 .data = &leases_enable,
100 .maxlen = sizeof(int),
101 .mode = 0644,
102 .proc_handler = proc_dointvec,
103 },
104#ifdef CONFIG_MMU
105 {
106 .procname = "lease-break-time",
107 .data = &lease_break_time,
108 .maxlen = sizeof(int),
109 .mode = 0644,
110 .proc_handler = proc_dointvec,
111 },
112#endif /* CONFIG_MMU */
113 {}
114};
115
116static int __init init_fs_locks_sysctls(void)
117{
118 register_sysctl_init("fs", locks_sysctls);
119 return 0;
120}
121early_initcall(init_fs_locks_sysctls);
122#endif /* CONFIG_SYSCTL */
123
124/*
125 * The global file_lock_list is only used for displaying /proc/locks, so we
126 * keep a list on each CPU, with each list protected by its own spinlock.
127 * Global serialization is done using file_rwsem.
128 *
129 * Note that alterations to the list also require that the relevant flc_lock is
130 * held.
131 */
132struct file_lock_list_struct {
133 spinlock_t lock;
134 struct hlist_head hlist;
135};
136static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
137DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
138
139
140/*
141 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
142 * It is protected by blocked_lock_lock.
143 *
144 * We hash locks by lockowner in order to optimize searching for the lock a
145 * particular lockowner is waiting on.
146 *
147 * FIXME: make this value scale via some heuristic? We generally will want more
148 * buckets when we have more lockowners holding locks, but that's a little
149 * difficult to determine without knowing what the workload will look like.
150 */
151#define BLOCKED_HASH_BITS 7
152static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
153
154/*
155 * This lock protects the blocked_hash. Generally, if you're accessing it, you
156 * want to be holding this lock.
157 *
158 * In addition, it also protects the fl->fl_blocked_requests list, and the
159 * fl->fl_blocker pointer for file_lock structures that are acting as lock
160 * requests (in contrast to those that are acting as records of acquired locks).
161 *
162 * Note that when we acquire this lock in order to change the above fields,
163 * we often hold the flc_lock as well. In certain cases, when reading the fields
164 * protected by this lock, we can skip acquiring it iff we already hold the
165 * flc_lock.
166 */
167static DEFINE_SPINLOCK(blocked_lock_lock);
168
169static struct kmem_cache *flctx_cache __read_mostly;
170static struct kmem_cache *filelock_cache __read_mostly;
171
172static struct file_lock_context *
173locks_get_lock_context(struct inode *inode, int type)
174{
175 struct file_lock_context *ctx;
176
177 /* paired with cmpxchg() below */
178 ctx = locks_inode_context(inode);
179 if (likely(ctx) || type == F_UNLCK)
180 goto out;
181
182 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
183 if (!ctx)
184 goto out;
185
186 spin_lock_init(&ctx->flc_lock);
187 INIT_LIST_HEAD(&ctx->flc_flock);
188 INIT_LIST_HEAD(&ctx->flc_posix);
189 INIT_LIST_HEAD(&ctx->flc_lease);
190
191 /*
192 * Assign the pointer if it's not already assigned. If it is, then
193 * free the context we just allocated.
194 */
195 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
196 kmem_cache_free(flctx_cache, ctx);
197 ctx = locks_inode_context(inode);
198 }
199out:
200 trace_locks_get_lock_context(inode, type, ctx);
201 return ctx;
202}
203
204static void
205locks_dump_ctx_list(struct list_head *list, char *list_type)
206{
207 struct file_lock *fl;
208
209 list_for_each_entry(fl, list, fl_list) {
210 pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
211 }
212}
213
214static void
215locks_check_ctx_lists(struct inode *inode)
216{
217 struct file_lock_context *ctx = inode->i_flctx;
218
219 if (unlikely(!list_empty(&ctx->flc_flock) ||
220 !list_empty(&ctx->flc_posix) ||
221 !list_empty(&ctx->flc_lease))) {
222 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
223 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
224 inode->i_ino);
225 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
226 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
227 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
228 }
229}
230
231static void
232locks_check_ctx_file_list(struct file *filp, struct list_head *list,
233 char *list_type)
234{
235 struct file_lock *fl;
236 struct inode *inode = locks_inode(filp);
237
238 list_for_each_entry(fl, list, fl_list)
239 if (fl->fl_file == filp)
240 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
241 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
242 list_type, MAJOR(inode->i_sb->s_dev),
243 MINOR(inode->i_sb->s_dev), inode->i_ino,
244 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
245}
246
247void
248locks_free_lock_context(struct inode *inode)
249{
250 struct file_lock_context *ctx = locks_inode_context(inode);
251
252 if (unlikely(ctx)) {
253 locks_check_ctx_lists(inode);
254 kmem_cache_free(flctx_cache, ctx);
255 }
256}
257
258static void locks_init_lock_heads(struct file_lock *fl)
259{
260 INIT_HLIST_NODE(&fl->fl_link);
261 INIT_LIST_HEAD(&fl->fl_list);
262 INIT_LIST_HEAD(&fl->fl_blocked_requests);
263 INIT_LIST_HEAD(&fl->fl_blocked_member);
264 init_waitqueue_head(&fl->fl_wait);
265}
266
267/* Allocate an empty lock structure. */
268struct file_lock *locks_alloc_lock(void)
269{
270 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
271
272 if (fl)
273 locks_init_lock_heads(fl);
274
275 return fl;
276}
277EXPORT_SYMBOL_GPL(locks_alloc_lock);
278
279void locks_release_private(struct file_lock *fl)
280{
281 BUG_ON(waitqueue_active(&fl->fl_wait));
282 BUG_ON(!list_empty(&fl->fl_list));
283 BUG_ON(!list_empty(&fl->fl_blocked_requests));
284 BUG_ON(!list_empty(&fl->fl_blocked_member));
285 BUG_ON(!hlist_unhashed(&fl->fl_link));
286
287 if (fl->fl_ops) {
288 if (fl->fl_ops->fl_release_private)
289 fl->fl_ops->fl_release_private(fl);
290 fl->fl_ops = NULL;
291 }
292
293 if (fl->fl_lmops) {
294 if (fl->fl_lmops->lm_put_owner) {
295 fl->fl_lmops->lm_put_owner(fl->fl_owner);
296 fl->fl_owner = NULL;
297 }
298 fl->fl_lmops = NULL;
299 }
300}
301EXPORT_SYMBOL_GPL(locks_release_private);
302
303/**
304 * locks_owner_has_blockers - Check for blocking lock requests
305 * @flctx: file lock context
306 * @owner: lock owner
307 *
308 * Return values:
309 * %true: @owner has at least one blocker
310 * %false: @owner has no blockers
311 */
312bool locks_owner_has_blockers(struct file_lock_context *flctx,
313 fl_owner_t owner)
314{
315 struct file_lock *fl;
316
317 spin_lock(&flctx->flc_lock);
318 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
319 if (fl->fl_owner != owner)
320 continue;
321 if (!list_empty(&fl->fl_blocked_requests)) {
322 spin_unlock(&flctx->flc_lock);
323 return true;
324 }
325 }
326 spin_unlock(&flctx->flc_lock);
327 return false;
328}
329EXPORT_SYMBOL_GPL(locks_owner_has_blockers);
330
331/* Free a lock which is not in use. */
332void locks_free_lock(struct file_lock *fl)
333{
334 locks_release_private(fl);
335 kmem_cache_free(filelock_cache, fl);
336}
337EXPORT_SYMBOL(locks_free_lock);
338
339static void
340locks_dispose_list(struct list_head *dispose)
341{
342 struct file_lock *fl;
343
344 while (!list_empty(dispose)) {
345 fl = list_first_entry(dispose, struct file_lock, fl_list);
346 list_del_init(&fl->fl_list);
347 locks_free_lock(fl);
348 }
349}
350
351void locks_init_lock(struct file_lock *fl)
352{
353 memset(fl, 0, sizeof(struct file_lock));
354 locks_init_lock_heads(fl);
355}
356EXPORT_SYMBOL(locks_init_lock);
357
358/*
359 * Initialize a new lock from an existing file_lock structure.
360 */
361void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
362{
363 new->fl_owner = fl->fl_owner;
364 new->fl_pid = fl->fl_pid;
365 new->fl_file = NULL;
366 new->fl_flags = fl->fl_flags;
367 new->fl_type = fl->fl_type;
368 new->fl_start = fl->fl_start;
369 new->fl_end = fl->fl_end;
370 new->fl_lmops = fl->fl_lmops;
371 new->fl_ops = NULL;
372
373 if (fl->fl_lmops) {
374 if (fl->fl_lmops->lm_get_owner)
375 fl->fl_lmops->lm_get_owner(fl->fl_owner);
376 }
377}
378EXPORT_SYMBOL(locks_copy_conflock);
379
380void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
381{
382 /* "new" must be a freshly-initialized lock */
383 WARN_ON_ONCE(new->fl_ops);
384
385 locks_copy_conflock(new, fl);
386
387 new->fl_file = fl->fl_file;
388 new->fl_ops = fl->fl_ops;
389
390 if (fl->fl_ops) {
391 if (fl->fl_ops->fl_copy_lock)
392 fl->fl_ops->fl_copy_lock(new, fl);
393 }
394}
395EXPORT_SYMBOL(locks_copy_lock);
396
397static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
398{
399 struct file_lock *f;
400
401 /*
402 * As ctx->flc_lock is held, new requests cannot be added to
403 * ->fl_blocked_requests, so we don't need a lock to check if it
404 * is empty.
405 */
406 if (list_empty(&fl->fl_blocked_requests))
407 return;
408 spin_lock(&blocked_lock_lock);
409 list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
410 list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
411 f->fl_blocker = new;
412 spin_unlock(&blocked_lock_lock);
413}
414
415static inline int flock_translate_cmd(int cmd) {
416 switch (cmd) {
417 case LOCK_SH:
418 return F_RDLCK;
419 case LOCK_EX:
420 return F_WRLCK;
421 case LOCK_UN:
422 return F_UNLCK;
423 }
424 return -EINVAL;
425}
426
427/* Fill in a file_lock structure with an appropriate FLOCK lock. */
428static void flock_make_lock(struct file *filp, struct file_lock *fl, int type)
429{
430 locks_init_lock(fl);
431
432 fl->fl_file = filp;
433 fl->fl_owner = filp;
434 fl->fl_pid = current->tgid;
435 fl->fl_flags = FL_FLOCK;
436 fl->fl_type = type;
437 fl->fl_end = OFFSET_MAX;
438}
439
440static int assign_type(struct file_lock *fl, long type)
441{
442 switch (type) {
443 case F_RDLCK:
444 case F_WRLCK:
445 case F_UNLCK:
446 fl->fl_type = type;
447 break;
448 default:
449 return -EINVAL;
450 }
451 return 0;
452}
453
454static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
455 struct flock64 *l)
456{
457 switch (l->l_whence) {
458 case SEEK_SET:
459 fl->fl_start = 0;
460 break;
461 case SEEK_CUR:
462 fl->fl_start = filp->f_pos;
463 break;
464 case SEEK_END:
465 fl->fl_start = i_size_read(file_inode(filp));
466 break;
467 default:
468 return -EINVAL;
469 }
470 if (l->l_start > OFFSET_MAX - fl->fl_start)
471 return -EOVERFLOW;
472 fl->fl_start += l->l_start;
473 if (fl->fl_start < 0)
474 return -EINVAL;
475
476 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
477 POSIX-2001 defines it. */
478 if (l->l_len > 0) {
479 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
480 return -EOVERFLOW;
481 fl->fl_end = fl->fl_start + (l->l_len - 1);
482
483 } else if (l->l_len < 0) {
484 if (fl->fl_start + l->l_len < 0)
485 return -EINVAL;
486 fl->fl_end = fl->fl_start - 1;
487 fl->fl_start += l->l_len;
488 } else
489 fl->fl_end = OFFSET_MAX;
490
491 fl->fl_owner = current->files;
492 fl->fl_pid = current->tgid;
493 fl->fl_file = filp;
494 fl->fl_flags = FL_POSIX;
495 fl->fl_ops = NULL;
496 fl->fl_lmops = NULL;
497
498 return assign_type(fl, l->l_type);
499}
500
501/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
502 * style lock.
503 */
504static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
505 struct flock *l)
506{
507 struct flock64 ll = {
508 .l_type = l->l_type,
509 .l_whence = l->l_whence,
510 .l_start = l->l_start,
511 .l_len = l->l_len,
512 };
513
514 return flock64_to_posix_lock(filp, fl, &ll);
515}
516
517/* default lease lock manager operations */
518static bool
519lease_break_callback(struct file_lock *fl)
520{
521 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
522 return false;
523}
524
525static void
526lease_setup(struct file_lock *fl, void **priv)
527{
528 struct file *filp = fl->fl_file;
529 struct fasync_struct *fa = *priv;
530
531 /*
532 * fasync_insert_entry() returns the old entry if any. If there was no
533 * old entry, then it used "priv" and inserted it into the fasync list.
534 * Clear the pointer to indicate that it shouldn't be freed.
535 */
536 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
537 *priv = NULL;
538
539 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
540}
541
542static const struct lock_manager_operations lease_manager_ops = {
543 .lm_break = lease_break_callback,
544 .lm_change = lease_modify,
545 .lm_setup = lease_setup,
546};
547
548/*
549 * Initialize a lease, use the default lock manager operations
550 */
551static int lease_init(struct file *filp, long type, struct file_lock *fl)
552{
553 if (assign_type(fl, type) != 0)
554 return -EINVAL;
555
556 fl->fl_owner = filp;
557 fl->fl_pid = current->tgid;
558
559 fl->fl_file = filp;
560 fl->fl_flags = FL_LEASE;
561 fl->fl_start = 0;
562 fl->fl_end = OFFSET_MAX;
563 fl->fl_ops = NULL;
564 fl->fl_lmops = &lease_manager_ops;
565 return 0;
566}
567
568/* Allocate a file_lock initialised to this type of lease */
569static struct file_lock *lease_alloc(struct file *filp, long type)
570{
571 struct file_lock *fl = locks_alloc_lock();
572 int error = -ENOMEM;
573
574 if (fl == NULL)
575 return ERR_PTR(error);
576
577 error = lease_init(filp, type, fl);
578 if (error) {
579 locks_free_lock(fl);
580 return ERR_PTR(error);
581 }
582 return fl;
583}
584
585/* Check if two locks overlap each other.
586 */
587static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
588{
589 return ((fl1->fl_end >= fl2->fl_start) &&
590 (fl2->fl_end >= fl1->fl_start));
591}
592
593/*
594 * Check whether two locks have the same owner.
595 */
596static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
597{
598 return fl1->fl_owner == fl2->fl_owner;
599}
600
601/* Must be called with the flc_lock held! */
602static void locks_insert_global_locks(struct file_lock *fl)
603{
604 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
605
606 percpu_rwsem_assert_held(&file_rwsem);
607
608 spin_lock(&fll->lock);
609 fl->fl_link_cpu = smp_processor_id();
610 hlist_add_head(&fl->fl_link, &fll->hlist);
611 spin_unlock(&fll->lock);
612}
613
614/* Must be called with the flc_lock held! */
615static void locks_delete_global_locks(struct file_lock *fl)
616{
617 struct file_lock_list_struct *fll;
618
619 percpu_rwsem_assert_held(&file_rwsem);
620
621 /*
622 * Avoid taking lock if already unhashed. This is safe since this check
623 * is done while holding the flc_lock, and new insertions into the list
624 * also require that it be held.
625 */
626 if (hlist_unhashed(&fl->fl_link))
627 return;
628
629 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
630 spin_lock(&fll->lock);
631 hlist_del_init(&fl->fl_link);
632 spin_unlock(&fll->lock);
633}
634
635static unsigned long
636posix_owner_key(struct file_lock *fl)
637{
638 return (unsigned long)fl->fl_owner;
639}
640
641static void locks_insert_global_blocked(struct file_lock *waiter)
642{
643 lockdep_assert_held(&blocked_lock_lock);
644
645 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
646}
647
648static void locks_delete_global_blocked(struct file_lock *waiter)
649{
650 lockdep_assert_held(&blocked_lock_lock);
651
652 hash_del(&waiter->fl_link);
653}
654
655/* Remove waiter from blocker's block list.
656 * When blocker ends up pointing to itself then the list is empty.
657 *
658 * Must be called with blocked_lock_lock held.
659 */
660static void __locks_delete_block(struct file_lock *waiter)
661{
662 locks_delete_global_blocked(waiter);
663 list_del_init(&waiter->fl_blocked_member);
664}
665
666static void __locks_wake_up_blocks(struct file_lock *blocker)
667{
668 while (!list_empty(&blocker->fl_blocked_requests)) {
669 struct file_lock *waiter;
670
671 waiter = list_first_entry(&blocker->fl_blocked_requests,
672 struct file_lock, fl_blocked_member);
673 __locks_delete_block(waiter);
674 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
675 waiter->fl_lmops->lm_notify(waiter);
676 else
677 wake_up(&waiter->fl_wait);
678
679 /*
680 * The setting of fl_blocker to NULL marks the "done"
681 * point in deleting a block. Paired with acquire at the top
682 * of locks_delete_block().
683 */
684 smp_store_release(&waiter->fl_blocker, NULL);
685 }
686}
687
688/**
689 * locks_delete_block - stop waiting for a file lock
690 * @waiter: the lock which was waiting
691 *
692 * lockd/nfsd need to disconnect the lock while working on it.
693 */
694int locks_delete_block(struct file_lock *waiter)
695{
696 int status = -ENOENT;
697
698 /*
699 * If fl_blocker is NULL, it won't be set again as this thread "owns"
700 * the lock and is the only one that might try to claim the lock.
701 *
702 * We use acquire/release to manage fl_blocker so that we can
703 * optimize away taking the blocked_lock_lock in many cases.
704 *
705 * The smp_load_acquire guarantees two things:
706 *
707 * 1/ that fl_blocked_requests can be tested locklessly. If something
708 * was recently added to that list it must have been in a locked region
709 * *before* the locked region when fl_blocker was set to NULL.
710 *
711 * 2/ that no other thread is accessing 'waiter', so it is safe to free
712 * it. __locks_wake_up_blocks is careful not to touch waiter after
713 * fl_blocker is released.
714 *
715 * If a lockless check of fl_blocker shows it to be NULL, we know that
716 * no new locks can be inserted into its fl_blocked_requests list, and
717 * can avoid doing anything further if the list is empty.
718 */
719 if (!smp_load_acquire(&waiter->fl_blocker) &&
720 list_empty(&waiter->fl_blocked_requests))
721 return status;
722
723 spin_lock(&blocked_lock_lock);
724 if (waiter->fl_blocker)
725 status = 0;
726 __locks_wake_up_blocks(waiter);
727 __locks_delete_block(waiter);
728
729 /*
730 * The setting of fl_blocker to NULL marks the "done" point in deleting
731 * a block. Paired with acquire at the top of this function.
732 */
733 smp_store_release(&waiter->fl_blocker, NULL);
734 spin_unlock(&blocked_lock_lock);
735 return status;
736}
737EXPORT_SYMBOL(locks_delete_block);
738
739/* Insert waiter into blocker's block list.
740 * We use a circular list so that processes can be easily woken up in
741 * the order they blocked. The documentation doesn't require this but
742 * it seems like the reasonable thing to do.
743 *
744 * Must be called with both the flc_lock and blocked_lock_lock held. The
745 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
746 * but by ensuring that the flc_lock is also held on insertions we can avoid
747 * taking the blocked_lock_lock in some cases when we see that the
748 * fl_blocked_requests list is empty.
749 *
750 * Rather than just adding to the list, we check for conflicts with any existing
751 * waiters, and add beneath any waiter that blocks the new waiter.
752 * Thus wakeups don't happen until needed.
753 */
754static void __locks_insert_block(struct file_lock *blocker,
755 struct file_lock *waiter,
756 bool conflict(struct file_lock *,
757 struct file_lock *))
758{
759 struct file_lock *fl;
760 BUG_ON(!list_empty(&waiter->fl_blocked_member));
761
762new_blocker:
763 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
764 if (conflict(fl, waiter)) {
765 blocker = fl;
766 goto new_blocker;
767 }
768 waiter->fl_blocker = blocker;
769 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
770 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
771 locks_insert_global_blocked(waiter);
772
773 /* The requests in waiter->fl_blocked are known to conflict with
774 * waiter, but might not conflict with blocker, or the requests
775 * and lock which block it. So they all need to be woken.
776 */
777 __locks_wake_up_blocks(waiter);
778}
779
780/* Must be called with flc_lock held. */
781static void locks_insert_block(struct file_lock *blocker,
782 struct file_lock *waiter,
783 bool conflict(struct file_lock *,
784 struct file_lock *))
785{
786 spin_lock(&blocked_lock_lock);
787 __locks_insert_block(blocker, waiter, conflict);
788 spin_unlock(&blocked_lock_lock);
789}
790
791/*
792 * Wake up processes blocked waiting for blocker.
793 *
794 * Must be called with the inode->flc_lock held!
795 */
796static void locks_wake_up_blocks(struct file_lock *blocker)
797{
798 /*
799 * Avoid taking global lock if list is empty. This is safe since new
800 * blocked requests are only added to the list under the flc_lock, and
801 * the flc_lock is always held here. Note that removal from the
802 * fl_blocked_requests list does not require the flc_lock, so we must
803 * recheck list_empty() after acquiring the blocked_lock_lock.
804 */
805 if (list_empty(&blocker->fl_blocked_requests))
806 return;
807
808 spin_lock(&blocked_lock_lock);
809 __locks_wake_up_blocks(blocker);
810 spin_unlock(&blocked_lock_lock);
811}
812
813static void
814locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
815{
816 list_add_tail(&fl->fl_list, before);
817 locks_insert_global_locks(fl);
818}
819
820static void
821locks_unlink_lock_ctx(struct file_lock *fl)
822{
823 locks_delete_global_locks(fl);
824 list_del_init(&fl->fl_list);
825 locks_wake_up_blocks(fl);
826}
827
828static void
829locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
830{
831 locks_unlink_lock_ctx(fl);
832 if (dispose)
833 list_add(&fl->fl_list, dispose);
834 else
835 locks_free_lock(fl);
836}
837
838/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
839 * checks for shared/exclusive status of overlapping locks.
840 */
841static bool locks_conflict(struct file_lock *caller_fl,
842 struct file_lock *sys_fl)
843{
844 if (sys_fl->fl_type == F_WRLCK)
845 return true;
846 if (caller_fl->fl_type == F_WRLCK)
847 return true;
848 return false;
849}
850
851/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
852 * checking before calling the locks_conflict().
853 */
854static bool posix_locks_conflict(struct file_lock *caller_fl,
855 struct file_lock *sys_fl)
856{
857 /* POSIX locks owned by the same process do not conflict with
858 * each other.
859 */
860 if (posix_same_owner(caller_fl, sys_fl))
861 return false;
862
863 /* Check whether they overlap */
864 if (!locks_overlap(caller_fl, sys_fl))
865 return false;
866
867 return locks_conflict(caller_fl, sys_fl);
868}
869
870/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
871 * checking before calling the locks_conflict().
872 */
873static bool flock_locks_conflict(struct file_lock *caller_fl,
874 struct file_lock *sys_fl)
875{
876 /* FLOCK locks referring to the same filp do not conflict with
877 * each other.
878 */
879 if (caller_fl->fl_file == sys_fl->fl_file)
880 return false;
881
882 return locks_conflict(caller_fl, sys_fl);
883}
884
885void
886posix_test_lock(struct file *filp, struct file_lock *fl)
887{
888 struct file_lock *cfl;
889 struct file_lock_context *ctx;
890 struct inode *inode = locks_inode(filp);
891 void *owner;
892 void (*func)(void);
893
894 ctx = locks_inode_context(inode);
895 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
896 fl->fl_type = F_UNLCK;
897 return;
898 }
899
900retry:
901 spin_lock(&ctx->flc_lock);
902 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
903 if (!posix_locks_conflict(fl, cfl))
904 continue;
905 if (cfl->fl_lmops && cfl->fl_lmops->lm_lock_expirable
906 && (*cfl->fl_lmops->lm_lock_expirable)(cfl)) {
907 owner = cfl->fl_lmops->lm_mod_owner;
908 func = cfl->fl_lmops->lm_expire_lock;
909 __module_get(owner);
910 spin_unlock(&ctx->flc_lock);
911 (*func)();
912 module_put(owner);
913 goto retry;
914 }
915 locks_copy_conflock(fl, cfl);
916 goto out;
917 }
918 fl->fl_type = F_UNLCK;
919out:
920 spin_unlock(&ctx->flc_lock);
921 return;
922}
923EXPORT_SYMBOL(posix_test_lock);
924
925/*
926 * Deadlock detection:
927 *
928 * We attempt to detect deadlocks that are due purely to posix file
929 * locks.
930 *
931 * We assume that a task can be waiting for at most one lock at a time.
932 * So for any acquired lock, the process holding that lock may be
933 * waiting on at most one other lock. That lock in turns may be held by
934 * someone waiting for at most one other lock. Given a requested lock
935 * caller_fl which is about to wait for a conflicting lock block_fl, we
936 * follow this chain of waiters to ensure we are not about to create a
937 * cycle.
938 *
939 * Since we do this before we ever put a process to sleep on a lock, we
940 * are ensured that there is never a cycle; that is what guarantees that
941 * the while() loop in posix_locks_deadlock() eventually completes.
942 *
943 * Note: the above assumption may not be true when handling lock
944 * requests from a broken NFS client. It may also fail in the presence
945 * of tasks (such as posix threads) sharing the same open file table.
946 * To handle those cases, we just bail out after a few iterations.
947 *
948 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
949 * Because the owner is not even nominally tied to a thread of
950 * execution, the deadlock detection below can't reasonably work well. Just
951 * skip it for those.
952 *
953 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
954 * locks that just checks for the case where two tasks are attempting to
955 * upgrade from read to write locks on the same inode.
956 */
957
958#define MAX_DEADLK_ITERATIONS 10
959
960/* Find a lock that the owner of the given block_fl is blocking on. */
961static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
962{
963 struct file_lock *fl;
964
965 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
966 if (posix_same_owner(fl, block_fl)) {
967 while (fl->fl_blocker)
968 fl = fl->fl_blocker;
969 return fl;
970 }
971 }
972 return NULL;
973}
974
975/* Must be called with the blocked_lock_lock held! */
976static int posix_locks_deadlock(struct file_lock *caller_fl,
977 struct file_lock *block_fl)
978{
979 int i = 0;
980
981 lockdep_assert_held(&blocked_lock_lock);
982
983 /*
984 * This deadlock detector can't reasonably detect deadlocks with
985 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
986 */
987 if (IS_OFDLCK(caller_fl))
988 return 0;
989
990 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
991 if (i++ > MAX_DEADLK_ITERATIONS)
992 return 0;
993 if (posix_same_owner(caller_fl, block_fl))
994 return 1;
995 }
996 return 0;
997}
998
999/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1000 * after any leases, but before any posix locks.
1001 *
1002 * Note that if called with an FL_EXISTS argument, the caller may determine
1003 * whether or not a lock was successfully freed by testing the return
1004 * value for -ENOENT.
1005 */
1006static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1007{
1008 struct file_lock *new_fl = NULL;
1009 struct file_lock *fl;
1010 struct file_lock_context *ctx;
1011 int error = 0;
1012 bool found = false;
1013 LIST_HEAD(dispose);
1014
1015 ctx = locks_get_lock_context(inode, request->fl_type);
1016 if (!ctx) {
1017 if (request->fl_type != F_UNLCK)
1018 return -ENOMEM;
1019 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1020 }
1021
1022 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1023 new_fl = locks_alloc_lock();
1024 if (!new_fl)
1025 return -ENOMEM;
1026 }
1027
1028 percpu_down_read(&file_rwsem);
1029 spin_lock(&ctx->flc_lock);
1030 if (request->fl_flags & FL_ACCESS)
1031 goto find_conflict;
1032
1033 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1034 if (request->fl_file != fl->fl_file)
1035 continue;
1036 if (request->fl_type == fl->fl_type)
1037 goto out;
1038 found = true;
1039 locks_delete_lock_ctx(fl, &dispose);
1040 break;
1041 }
1042
1043 if (request->fl_type == F_UNLCK) {
1044 if ((request->fl_flags & FL_EXISTS) && !found)
1045 error = -ENOENT;
1046 goto out;
1047 }
1048
1049find_conflict:
1050 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1051 if (!flock_locks_conflict(request, fl))
1052 continue;
1053 error = -EAGAIN;
1054 if (!(request->fl_flags & FL_SLEEP))
1055 goto out;
1056 error = FILE_LOCK_DEFERRED;
1057 locks_insert_block(fl, request, flock_locks_conflict);
1058 goto out;
1059 }
1060 if (request->fl_flags & FL_ACCESS)
1061 goto out;
1062 locks_copy_lock(new_fl, request);
1063 locks_move_blocks(new_fl, request);
1064 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1065 new_fl = NULL;
1066 error = 0;
1067
1068out:
1069 spin_unlock(&ctx->flc_lock);
1070 percpu_up_read(&file_rwsem);
1071 if (new_fl)
1072 locks_free_lock(new_fl);
1073 locks_dispose_list(&dispose);
1074 trace_flock_lock_inode(inode, request, error);
1075 return error;
1076}
1077
1078static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1079 struct file_lock *conflock)
1080{
1081 struct file_lock *fl, *tmp;
1082 struct file_lock *new_fl = NULL;
1083 struct file_lock *new_fl2 = NULL;
1084 struct file_lock *left = NULL;
1085 struct file_lock *right = NULL;
1086 struct file_lock_context *ctx;
1087 int error;
1088 bool added = false;
1089 LIST_HEAD(dispose);
1090 void *owner;
1091 void (*func)(void);
1092
1093 ctx = locks_get_lock_context(inode, request->fl_type);
1094 if (!ctx)
1095 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1096
1097 /*
1098 * We may need two file_lock structures for this operation,
1099 * so we get them in advance to avoid races.
1100 *
1101 * In some cases we can be sure, that no new locks will be needed
1102 */
1103 if (!(request->fl_flags & FL_ACCESS) &&
1104 (request->fl_type != F_UNLCK ||
1105 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1106 new_fl = locks_alloc_lock();
1107 new_fl2 = locks_alloc_lock();
1108 }
1109
1110retry:
1111 percpu_down_read(&file_rwsem);
1112 spin_lock(&ctx->flc_lock);
1113 /*
1114 * New lock request. Walk all POSIX locks and look for conflicts. If
1115 * there are any, either return error or put the request on the
1116 * blocker's list of waiters and the global blocked_hash.
1117 */
1118 if (request->fl_type != F_UNLCK) {
1119 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1120 if (!posix_locks_conflict(request, fl))
1121 continue;
1122 if (fl->fl_lmops && fl->fl_lmops->lm_lock_expirable
1123 && (*fl->fl_lmops->lm_lock_expirable)(fl)) {
1124 owner = fl->fl_lmops->lm_mod_owner;
1125 func = fl->fl_lmops->lm_expire_lock;
1126 __module_get(owner);
1127 spin_unlock(&ctx->flc_lock);
1128 percpu_up_read(&file_rwsem);
1129 (*func)();
1130 module_put(owner);
1131 goto retry;
1132 }
1133 if (conflock)
1134 locks_copy_conflock(conflock, fl);
1135 error = -EAGAIN;
1136 if (!(request->fl_flags & FL_SLEEP))
1137 goto out;
1138 /*
1139 * Deadlock detection and insertion into the blocked
1140 * locks list must be done while holding the same lock!
1141 */
1142 error = -EDEADLK;
1143 spin_lock(&blocked_lock_lock);
1144 /*
1145 * Ensure that we don't find any locks blocked on this
1146 * request during deadlock detection.
1147 */
1148 __locks_wake_up_blocks(request);
1149 if (likely(!posix_locks_deadlock(request, fl))) {
1150 error = FILE_LOCK_DEFERRED;
1151 __locks_insert_block(fl, request,
1152 posix_locks_conflict);
1153 }
1154 spin_unlock(&blocked_lock_lock);
1155 goto out;
1156 }
1157 }
1158
1159 /* If we're just looking for a conflict, we're done. */
1160 error = 0;
1161 if (request->fl_flags & FL_ACCESS)
1162 goto out;
1163
1164 /* Find the first old lock with the same owner as the new lock */
1165 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1166 if (posix_same_owner(request, fl))
1167 break;
1168 }
1169
1170 /* Process locks with this owner. */
1171 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1172 if (!posix_same_owner(request, fl))
1173 break;
1174
1175 /* Detect adjacent or overlapping regions (if same lock type) */
1176 if (request->fl_type == fl->fl_type) {
1177 /* In all comparisons of start vs end, use
1178 * "start - 1" rather than "end + 1". If end
1179 * is OFFSET_MAX, end + 1 will become negative.
1180 */
1181 if (fl->fl_end < request->fl_start - 1)
1182 continue;
1183 /* If the next lock in the list has entirely bigger
1184 * addresses than the new one, insert the lock here.
1185 */
1186 if (fl->fl_start - 1 > request->fl_end)
1187 break;
1188
1189 /* If we come here, the new and old lock are of the
1190 * same type and adjacent or overlapping. Make one
1191 * lock yielding from the lower start address of both
1192 * locks to the higher end address.
1193 */
1194 if (fl->fl_start > request->fl_start)
1195 fl->fl_start = request->fl_start;
1196 else
1197 request->fl_start = fl->fl_start;
1198 if (fl->fl_end < request->fl_end)
1199 fl->fl_end = request->fl_end;
1200 else
1201 request->fl_end = fl->fl_end;
1202 if (added) {
1203 locks_delete_lock_ctx(fl, &dispose);
1204 continue;
1205 }
1206 request = fl;
1207 added = true;
1208 } else {
1209 /* Processing for different lock types is a bit
1210 * more complex.
1211 */
1212 if (fl->fl_end < request->fl_start)
1213 continue;
1214 if (fl->fl_start > request->fl_end)
1215 break;
1216 if (request->fl_type == F_UNLCK)
1217 added = true;
1218 if (fl->fl_start < request->fl_start)
1219 left = fl;
1220 /* If the next lock in the list has a higher end
1221 * address than the new one, insert the new one here.
1222 */
1223 if (fl->fl_end > request->fl_end) {
1224 right = fl;
1225 break;
1226 }
1227 if (fl->fl_start >= request->fl_start) {
1228 /* The new lock completely replaces an old
1229 * one (This may happen several times).
1230 */
1231 if (added) {
1232 locks_delete_lock_ctx(fl, &dispose);
1233 continue;
1234 }
1235 /*
1236 * Replace the old lock with new_fl, and
1237 * remove the old one. It's safe to do the
1238 * insert here since we know that we won't be
1239 * using new_fl later, and that the lock is
1240 * just replacing an existing lock.
1241 */
1242 error = -ENOLCK;
1243 if (!new_fl)
1244 goto out;
1245 locks_copy_lock(new_fl, request);
1246 locks_move_blocks(new_fl, request);
1247 request = new_fl;
1248 new_fl = NULL;
1249 locks_insert_lock_ctx(request, &fl->fl_list);
1250 locks_delete_lock_ctx(fl, &dispose);
1251 added = true;
1252 }
1253 }
1254 }
1255
1256 /*
1257 * The above code only modifies existing locks in case of merging or
1258 * replacing. If new lock(s) need to be inserted all modifications are
1259 * done below this, so it's safe yet to bail out.
1260 */
1261 error = -ENOLCK; /* "no luck" */
1262 if (right && left == right && !new_fl2)
1263 goto out;
1264
1265 error = 0;
1266 if (!added) {
1267 if (request->fl_type == F_UNLCK) {
1268 if (request->fl_flags & FL_EXISTS)
1269 error = -ENOENT;
1270 goto out;
1271 }
1272
1273 if (!new_fl) {
1274 error = -ENOLCK;
1275 goto out;
1276 }
1277 locks_copy_lock(new_fl, request);
1278 locks_move_blocks(new_fl, request);
1279 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1280 fl = new_fl;
1281 new_fl = NULL;
1282 }
1283 if (right) {
1284 if (left == right) {
1285 /* The new lock breaks the old one in two pieces,
1286 * so we have to use the second new lock.
1287 */
1288 left = new_fl2;
1289 new_fl2 = NULL;
1290 locks_copy_lock(left, right);
1291 locks_insert_lock_ctx(left, &fl->fl_list);
1292 }
1293 right->fl_start = request->fl_end + 1;
1294 locks_wake_up_blocks(right);
1295 }
1296 if (left) {
1297 left->fl_end = request->fl_start - 1;
1298 locks_wake_up_blocks(left);
1299 }
1300 out:
1301 spin_unlock(&ctx->flc_lock);
1302 percpu_up_read(&file_rwsem);
1303 /*
1304 * Free any unused locks.
1305 */
1306 if (new_fl)
1307 locks_free_lock(new_fl);
1308 if (new_fl2)
1309 locks_free_lock(new_fl2);
1310 locks_dispose_list(&dispose);
1311 trace_posix_lock_inode(inode, request, error);
1312
1313 return error;
1314}
1315
1316/**
1317 * posix_lock_file - Apply a POSIX-style lock to a file
1318 * @filp: The file to apply the lock to
1319 * @fl: The lock to be applied
1320 * @conflock: Place to return a copy of the conflicting lock, if found.
1321 *
1322 * Add a POSIX style lock to a file.
1323 * We merge adjacent & overlapping locks whenever possible.
1324 * POSIX locks are sorted by owner task, then by starting address
1325 *
1326 * Note that if called with an FL_EXISTS argument, the caller may determine
1327 * whether or not a lock was successfully freed by testing the return
1328 * value for -ENOENT.
1329 */
1330int posix_lock_file(struct file *filp, struct file_lock *fl,
1331 struct file_lock *conflock)
1332{
1333 return posix_lock_inode(locks_inode(filp), fl, conflock);
1334}
1335EXPORT_SYMBOL(posix_lock_file);
1336
1337/**
1338 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1339 * @inode: inode of file to which lock request should be applied
1340 * @fl: The lock to be applied
1341 *
1342 * Apply a POSIX style lock request to an inode.
1343 */
1344static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1345{
1346 int error;
1347 might_sleep ();
1348 for (;;) {
1349 error = posix_lock_inode(inode, fl, NULL);
1350 if (error != FILE_LOCK_DEFERRED)
1351 break;
1352 error = wait_event_interruptible(fl->fl_wait,
1353 list_empty(&fl->fl_blocked_member));
1354 if (error)
1355 break;
1356 }
1357 locks_delete_block(fl);
1358 return error;
1359}
1360
1361static void lease_clear_pending(struct file_lock *fl, int arg)
1362{
1363 switch (arg) {
1364 case F_UNLCK:
1365 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1366 fallthrough;
1367 case F_RDLCK:
1368 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1369 }
1370}
1371
1372/* We already had a lease on this file; just change its type */
1373int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1374{
1375 int error = assign_type(fl, arg);
1376
1377 if (error)
1378 return error;
1379 lease_clear_pending(fl, arg);
1380 locks_wake_up_blocks(fl);
1381 if (arg == F_UNLCK) {
1382 struct file *filp = fl->fl_file;
1383
1384 f_delown(filp);
1385 filp->f_owner.signum = 0;
1386 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1387 if (fl->fl_fasync != NULL) {
1388 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1389 fl->fl_fasync = NULL;
1390 }
1391 locks_delete_lock_ctx(fl, dispose);
1392 }
1393 return 0;
1394}
1395EXPORT_SYMBOL(lease_modify);
1396
1397static bool past_time(unsigned long then)
1398{
1399 if (!then)
1400 /* 0 is a special value meaning "this never expires": */
1401 return false;
1402 return time_after(jiffies, then);
1403}
1404
1405static void time_out_leases(struct inode *inode, struct list_head *dispose)
1406{
1407 struct file_lock_context *ctx = inode->i_flctx;
1408 struct file_lock *fl, *tmp;
1409
1410 lockdep_assert_held(&ctx->flc_lock);
1411
1412 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1413 trace_time_out_leases(inode, fl);
1414 if (past_time(fl->fl_downgrade_time))
1415 lease_modify(fl, F_RDLCK, dispose);
1416 if (past_time(fl->fl_break_time))
1417 lease_modify(fl, F_UNLCK, dispose);
1418 }
1419}
1420
1421static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1422{
1423 bool rc;
1424
1425 if (lease->fl_lmops->lm_breaker_owns_lease
1426 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1427 return false;
1428 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1429 rc = false;
1430 goto trace;
1431 }
1432 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1433 rc = false;
1434 goto trace;
1435 }
1436
1437 rc = locks_conflict(breaker, lease);
1438trace:
1439 trace_leases_conflict(rc, lease, breaker);
1440 return rc;
1441}
1442
1443static bool
1444any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1445{
1446 struct file_lock_context *ctx = inode->i_flctx;
1447 struct file_lock *fl;
1448
1449 lockdep_assert_held(&ctx->flc_lock);
1450
1451 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1452 if (leases_conflict(fl, breaker))
1453 return true;
1454 }
1455 return false;
1456}
1457
1458/**
1459 * __break_lease - revoke all outstanding leases on file
1460 * @inode: the inode of the file to return
1461 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1462 * break all leases
1463 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1464 * only delegations
1465 *
1466 * break_lease (inlined for speed) has checked there already is at least
1467 * some kind of lock (maybe a lease) on this file. Leases are broken on
1468 * a call to open() or truncate(). This function can sleep unless you
1469 * specified %O_NONBLOCK to your open().
1470 */
1471int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1472{
1473 int error = 0;
1474 struct file_lock_context *ctx;
1475 struct file_lock *new_fl, *fl, *tmp;
1476 unsigned long break_time;
1477 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1478 LIST_HEAD(dispose);
1479
1480 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1481 if (IS_ERR(new_fl))
1482 return PTR_ERR(new_fl);
1483 new_fl->fl_flags = type;
1484
1485 /* typically we will check that ctx is non-NULL before calling */
1486 ctx = locks_inode_context(inode);
1487 if (!ctx) {
1488 WARN_ON_ONCE(1);
1489 goto free_lock;
1490 }
1491
1492 percpu_down_read(&file_rwsem);
1493 spin_lock(&ctx->flc_lock);
1494
1495 time_out_leases(inode, &dispose);
1496
1497 if (!any_leases_conflict(inode, new_fl))
1498 goto out;
1499
1500 break_time = 0;
1501 if (lease_break_time > 0) {
1502 break_time = jiffies + lease_break_time * HZ;
1503 if (break_time == 0)
1504 break_time++; /* so that 0 means no break time */
1505 }
1506
1507 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1508 if (!leases_conflict(fl, new_fl))
1509 continue;
1510 if (want_write) {
1511 if (fl->fl_flags & FL_UNLOCK_PENDING)
1512 continue;
1513 fl->fl_flags |= FL_UNLOCK_PENDING;
1514 fl->fl_break_time = break_time;
1515 } else {
1516 if (lease_breaking(fl))
1517 continue;
1518 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1519 fl->fl_downgrade_time = break_time;
1520 }
1521 if (fl->fl_lmops->lm_break(fl))
1522 locks_delete_lock_ctx(fl, &dispose);
1523 }
1524
1525 if (list_empty(&ctx->flc_lease))
1526 goto out;
1527
1528 if (mode & O_NONBLOCK) {
1529 trace_break_lease_noblock(inode, new_fl);
1530 error = -EWOULDBLOCK;
1531 goto out;
1532 }
1533
1534restart:
1535 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1536 break_time = fl->fl_break_time;
1537 if (break_time != 0)
1538 break_time -= jiffies;
1539 if (break_time == 0)
1540 break_time++;
1541 locks_insert_block(fl, new_fl, leases_conflict);
1542 trace_break_lease_block(inode, new_fl);
1543 spin_unlock(&ctx->flc_lock);
1544 percpu_up_read(&file_rwsem);
1545
1546 locks_dispose_list(&dispose);
1547 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1548 list_empty(&new_fl->fl_blocked_member),
1549 break_time);
1550
1551 percpu_down_read(&file_rwsem);
1552 spin_lock(&ctx->flc_lock);
1553 trace_break_lease_unblock(inode, new_fl);
1554 locks_delete_block(new_fl);
1555 if (error >= 0) {
1556 /*
1557 * Wait for the next conflicting lease that has not been
1558 * broken yet
1559 */
1560 if (error == 0)
1561 time_out_leases(inode, &dispose);
1562 if (any_leases_conflict(inode, new_fl))
1563 goto restart;
1564 error = 0;
1565 }
1566out:
1567 spin_unlock(&ctx->flc_lock);
1568 percpu_up_read(&file_rwsem);
1569 locks_dispose_list(&dispose);
1570free_lock:
1571 locks_free_lock(new_fl);
1572 return error;
1573}
1574EXPORT_SYMBOL(__break_lease);
1575
1576/**
1577 * lease_get_mtime - update modified time of an inode with exclusive lease
1578 * @inode: the inode
1579 * @time: pointer to a timespec which contains the last modified time
1580 *
1581 * This is to force NFS clients to flush their caches for files with
1582 * exclusive leases. The justification is that if someone has an
1583 * exclusive lease, then they could be modifying it.
1584 */
1585void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1586{
1587 bool has_lease = false;
1588 struct file_lock_context *ctx;
1589 struct file_lock *fl;
1590
1591 ctx = locks_inode_context(inode);
1592 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1593 spin_lock(&ctx->flc_lock);
1594 fl = list_first_entry_or_null(&ctx->flc_lease,
1595 struct file_lock, fl_list);
1596 if (fl && (fl->fl_type == F_WRLCK))
1597 has_lease = true;
1598 spin_unlock(&ctx->flc_lock);
1599 }
1600
1601 if (has_lease)
1602 *time = current_time(inode);
1603}
1604EXPORT_SYMBOL(lease_get_mtime);
1605
1606/**
1607 * fcntl_getlease - Enquire what lease is currently active
1608 * @filp: the file
1609 *
1610 * The value returned by this function will be one of
1611 * (if no lease break is pending):
1612 *
1613 * %F_RDLCK to indicate a shared lease is held.
1614 *
1615 * %F_WRLCK to indicate an exclusive lease is held.
1616 *
1617 * %F_UNLCK to indicate no lease is held.
1618 *
1619 * (if a lease break is pending):
1620 *
1621 * %F_RDLCK to indicate an exclusive lease needs to be
1622 * changed to a shared lease (or removed).
1623 *
1624 * %F_UNLCK to indicate the lease needs to be removed.
1625 *
1626 * XXX: sfr & willy disagree over whether F_INPROGRESS
1627 * should be returned to userspace.
1628 */
1629int fcntl_getlease(struct file *filp)
1630{
1631 struct file_lock *fl;
1632 struct inode *inode = locks_inode(filp);
1633 struct file_lock_context *ctx;
1634 int type = F_UNLCK;
1635 LIST_HEAD(dispose);
1636
1637 ctx = locks_inode_context(inode);
1638 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1639 percpu_down_read(&file_rwsem);
1640 spin_lock(&ctx->flc_lock);
1641 time_out_leases(inode, &dispose);
1642 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1643 if (fl->fl_file != filp)
1644 continue;
1645 type = target_leasetype(fl);
1646 break;
1647 }
1648 spin_unlock(&ctx->flc_lock);
1649 percpu_up_read(&file_rwsem);
1650
1651 locks_dispose_list(&dispose);
1652 }
1653 return type;
1654}
1655
1656/**
1657 * check_conflicting_open - see if the given file points to an inode that has
1658 * an existing open that would conflict with the
1659 * desired lease.
1660 * @filp: file to check
1661 * @arg: type of lease that we're trying to acquire
1662 * @flags: current lock flags
1663 *
1664 * Check to see if there's an existing open fd on this file that would
1665 * conflict with the lease we're trying to set.
1666 */
1667static int
1668check_conflicting_open(struct file *filp, const long arg, int flags)
1669{
1670 struct inode *inode = locks_inode(filp);
1671 int self_wcount = 0, self_rcount = 0;
1672
1673 if (flags & FL_LAYOUT)
1674 return 0;
1675 if (flags & FL_DELEG)
1676 /* We leave these checks to the caller */
1677 return 0;
1678
1679 if (arg == F_RDLCK)
1680 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1681 else if (arg != F_WRLCK)
1682 return 0;
1683
1684 /*
1685 * Make sure that only read/write count is from lease requestor.
1686 * Note that this will result in denying write leases when i_writecount
1687 * is negative, which is what we want. (We shouldn't grant write leases
1688 * on files open for execution.)
1689 */
1690 if (filp->f_mode & FMODE_WRITE)
1691 self_wcount = 1;
1692 else if (filp->f_mode & FMODE_READ)
1693 self_rcount = 1;
1694
1695 if (atomic_read(&inode->i_writecount) != self_wcount ||
1696 atomic_read(&inode->i_readcount) != self_rcount)
1697 return -EAGAIN;
1698
1699 return 0;
1700}
1701
1702static int
1703generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1704{
1705 struct file_lock *fl, *my_fl = NULL, *lease;
1706 struct inode *inode = locks_inode(filp);
1707 struct file_lock_context *ctx;
1708 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1709 int error;
1710 LIST_HEAD(dispose);
1711
1712 lease = *flp;
1713 trace_generic_add_lease(inode, lease);
1714
1715 /* Note that arg is never F_UNLCK here */
1716 ctx = locks_get_lock_context(inode, arg);
1717 if (!ctx)
1718 return -ENOMEM;
1719
1720 /*
1721 * In the delegation case we need mutual exclusion with
1722 * a number of operations that take the i_mutex. We trylock
1723 * because delegations are an optional optimization, and if
1724 * there's some chance of a conflict--we'd rather not
1725 * bother, maybe that's a sign this just isn't a good file to
1726 * hand out a delegation on.
1727 */
1728 if (is_deleg && !inode_trylock(inode))
1729 return -EAGAIN;
1730
1731 if (is_deleg && arg == F_WRLCK) {
1732 /* Write delegations are not currently supported: */
1733 inode_unlock(inode);
1734 WARN_ON_ONCE(1);
1735 return -EINVAL;
1736 }
1737
1738 percpu_down_read(&file_rwsem);
1739 spin_lock(&ctx->flc_lock);
1740 time_out_leases(inode, &dispose);
1741 error = check_conflicting_open(filp, arg, lease->fl_flags);
1742 if (error)
1743 goto out;
1744
1745 /*
1746 * At this point, we know that if there is an exclusive
1747 * lease on this file, then we hold it on this filp
1748 * (otherwise our open of this file would have blocked).
1749 * And if we are trying to acquire an exclusive lease,
1750 * then the file is not open by anyone (including us)
1751 * except for this filp.
1752 */
1753 error = -EAGAIN;
1754 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1755 if (fl->fl_file == filp &&
1756 fl->fl_owner == lease->fl_owner) {
1757 my_fl = fl;
1758 continue;
1759 }
1760
1761 /*
1762 * No exclusive leases if someone else has a lease on
1763 * this file:
1764 */
1765 if (arg == F_WRLCK)
1766 goto out;
1767 /*
1768 * Modifying our existing lease is OK, but no getting a
1769 * new lease if someone else is opening for write:
1770 */
1771 if (fl->fl_flags & FL_UNLOCK_PENDING)
1772 goto out;
1773 }
1774
1775 if (my_fl != NULL) {
1776 lease = my_fl;
1777 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1778 if (error)
1779 goto out;
1780 goto out_setup;
1781 }
1782
1783 error = -EINVAL;
1784 if (!leases_enable)
1785 goto out;
1786
1787 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1788 /*
1789 * The check in break_lease() is lockless. It's possible for another
1790 * open to race in after we did the earlier check for a conflicting
1791 * open but before the lease was inserted. Check again for a
1792 * conflicting open and cancel the lease if there is one.
1793 *
1794 * We also add a barrier here to ensure that the insertion of the lock
1795 * precedes these checks.
1796 */
1797 smp_mb();
1798 error = check_conflicting_open(filp, arg, lease->fl_flags);
1799 if (error) {
1800 locks_unlink_lock_ctx(lease);
1801 goto out;
1802 }
1803
1804out_setup:
1805 if (lease->fl_lmops->lm_setup)
1806 lease->fl_lmops->lm_setup(lease, priv);
1807out:
1808 spin_unlock(&ctx->flc_lock);
1809 percpu_up_read(&file_rwsem);
1810 locks_dispose_list(&dispose);
1811 if (is_deleg)
1812 inode_unlock(inode);
1813 if (!error && !my_fl)
1814 *flp = NULL;
1815 return error;
1816}
1817
1818static int generic_delete_lease(struct file *filp, void *owner)
1819{
1820 int error = -EAGAIN;
1821 struct file_lock *fl, *victim = NULL;
1822 struct inode *inode = locks_inode(filp);
1823 struct file_lock_context *ctx;
1824 LIST_HEAD(dispose);
1825
1826 ctx = locks_inode_context(inode);
1827 if (!ctx) {
1828 trace_generic_delete_lease(inode, NULL);
1829 return error;
1830 }
1831
1832 percpu_down_read(&file_rwsem);
1833 spin_lock(&ctx->flc_lock);
1834 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1835 if (fl->fl_file == filp &&
1836 fl->fl_owner == owner) {
1837 victim = fl;
1838 break;
1839 }
1840 }
1841 trace_generic_delete_lease(inode, victim);
1842 if (victim)
1843 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1844 spin_unlock(&ctx->flc_lock);
1845 percpu_up_read(&file_rwsem);
1846 locks_dispose_list(&dispose);
1847 return error;
1848}
1849
1850/**
1851 * generic_setlease - sets a lease on an open file
1852 * @filp: file pointer
1853 * @arg: type of lease to obtain
1854 * @flp: input - file_lock to use, output - file_lock inserted
1855 * @priv: private data for lm_setup (may be NULL if lm_setup
1856 * doesn't require it)
1857 *
1858 * The (input) flp->fl_lmops->lm_break function is required
1859 * by break_lease().
1860 */
1861int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1862 void **priv)
1863{
1864 struct inode *inode = locks_inode(filp);
1865 int error;
1866
1867 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1868 return -EACCES;
1869 if (!S_ISREG(inode->i_mode))
1870 return -EINVAL;
1871 error = security_file_lock(filp, arg);
1872 if (error)
1873 return error;
1874
1875 switch (arg) {
1876 case F_UNLCK:
1877 return generic_delete_lease(filp, *priv);
1878 case F_RDLCK:
1879 case F_WRLCK:
1880 if (!(*flp)->fl_lmops->lm_break) {
1881 WARN_ON_ONCE(1);
1882 return -ENOLCK;
1883 }
1884
1885 return generic_add_lease(filp, arg, flp, priv);
1886 default:
1887 return -EINVAL;
1888 }
1889}
1890EXPORT_SYMBOL(generic_setlease);
1891
1892#if IS_ENABLED(CONFIG_SRCU)
1893/*
1894 * Kernel subsystems can register to be notified on any attempt to set
1895 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1896 * to close files that it may have cached when there is an attempt to set a
1897 * conflicting lease.
1898 */
1899static struct srcu_notifier_head lease_notifier_chain;
1900
1901static inline void
1902lease_notifier_chain_init(void)
1903{
1904 srcu_init_notifier_head(&lease_notifier_chain);
1905}
1906
1907static inline void
1908setlease_notifier(long arg, struct file_lock *lease)
1909{
1910 if (arg != F_UNLCK)
1911 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1912}
1913
1914int lease_register_notifier(struct notifier_block *nb)
1915{
1916 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1917}
1918EXPORT_SYMBOL_GPL(lease_register_notifier);
1919
1920void lease_unregister_notifier(struct notifier_block *nb)
1921{
1922 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1923}
1924EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1925
1926#else /* !IS_ENABLED(CONFIG_SRCU) */
1927static inline void
1928lease_notifier_chain_init(void)
1929{
1930}
1931
1932static inline void
1933setlease_notifier(long arg, struct file_lock *lease)
1934{
1935}
1936
1937int lease_register_notifier(struct notifier_block *nb)
1938{
1939 return 0;
1940}
1941EXPORT_SYMBOL_GPL(lease_register_notifier);
1942
1943void lease_unregister_notifier(struct notifier_block *nb)
1944{
1945}
1946EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1947
1948#endif /* IS_ENABLED(CONFIG_SRCU) */
1949
1950/**
1951 * vfs_setlease - sets a lease on an open file
1952 * @filp: file pointer
1953 * @arg: type of lease to obtain
1954 * @lease: file_lock to use when adding a lease
1955 * @priv: private info for lm_setup when adding a lease (may be
1956 * NULL if lm_setup doesn't require it)
1957 *
1958 * Call this to establish a lease on the file. The "lease" argument is not
1959 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1960 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1961 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1962 * stack trace).
1963 *
1964 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1965 * may be NULL if the lm_setup operation doesn't require it.
1966 */
1967int
1968vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1969{
1970 if (lease)
1971 setlease_notifier(arg, *lease);
1972 if (filp->f_op->setlease)
1973 return filp->f_op->setlease(filp, arg, lease, priv);
1974 else
1975 return generic_setlease(filp, arg, lease, priv);
1976}
1977EXPORT_SYMBOL_GPL(vfs_setlease);
1978
1979static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1980{
1981 struct file_lock *fl;
1982 struct fasync_struct *new;
1983 int error;
1984
1985 fl = lease_alloc(filp, arg);
1986 if (IS_ERR(fl))
1987 return PTR_ERR(fl);
1988
1989 new = fasync_alloc();
1990 if (!new) {
1991 locks_free_lock(fl);
1992 return -ENOMEM;
1993 }
1994 new->fa_fd = fd;
1995
1996 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1997 if (fl)
1998 locks_free_lock(fl);
1999 if (new)
2000 fasync_free(new);
2001 return error;
2002}
2003
2004/**
2005 * fcntl_setlease - sets a lease on an open file
2006 * @fd: open file descriptor
2007 * @filp: file pointer
2008 * @arg: type of lease to obtain
2009 *
2010 * Call this fcntl to establish a lease on the file.
2011 * Note that you also need to call %F_SETSIG to
2012 * receive a signal when the lease is broken.
2013 */
2014int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2015{
2016 if (arg == F_UNLCK)
2017 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2018 return do_fcntl_add_lease(fd, filp, arg);
2019}
2020
2021/**
2022 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2023 * @inode: inode of the file to apply to
2024 * @fl: The lock to be applied
2025 *
2026 * Apply a FLOCK style lock request to an inode.
2027 */
2028static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2029{
2030 int error;
2031 might_sleep();
2032 for (;;) {
2033 error = flock_lock_inode(inode, fl);
2034 if (error != FILE_LOCK_DEFERRED)
2035 break;
2036 error = wait_event_interruptible(fl->fl_wait,
2037 list_empty(&fl->fl_blocked_member));
2038 if (error)
2039 break;
2040 }
2041 locks_delete_block(fl);
2042 return error;
2043}
2044
2045/**
2046 * locks_lock_inode_wait - Apply a lock to an inode
2047 * @inode: inode of the file to apply to
2048 * @fl: The lock to be applied
2049 *
2050 * Apply a POSIX or FLOCK style lock request to an inode.
2051 */
2052int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2053{
2054 int res = 0;
2055 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2056 case FL_POSIX:
2057 res = posix_lock_inode_wait(inode, fl);
2058 break;
2059 case FL_FLOCK:
2060 res = flock_lock_inode_wait(inode, fl);
2061 break;
2062 default:
2063 BUG();
2064 }
2065 return res;
2066}
2067EXPORT_SYMBOL(locks_lock_inode_wait);
2068
2069/**
2070 * sys_flock: - flock() system call.
2071 * @fd: the file descriptor to lock.
2072 * @cmd: the type of lock to apply.
2073 *
2074 * Apply a %FL_FLOCK style lock to an open file descriptor.
2075 * The @cmd can be one of:
2076 *
2077 * - %LOCK_SH -- a shared lock.
2078 * - %LOCK_EX -- an exclusive lock.
2079 * - %LOCK_UN -- remove an existing lock.
2080 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2081 *
2082 * %LOCK_MAND support has been removed from the kernel.
2083 */
2084SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2085{
2086 int can_sleep, error, type;
2087 struct file_lock fl;
2088 struct fd f;
2089
2090 /*
2091 * LOCK_MAND locks were broken for a long time in that they never
2092 * conflicted with one another and didn't prevent any sort of open,
2093 * read or write activity.
2094 *
2095 * Just ignore these requests now, to preserve legacy behavior, but
2096 * throw a warning to let people know that they don't actually work.
2097 */
2098 if (cmd & LOCK_MAND) {
2099 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);
2100 return 0;
2101 }
2102
2103 type = flock_translate_cmd(cmd & ~LOCK_NB);
2104 if (type < 0)
2105 return type;
2106
2107 error = -EBADF;
2108 f = fdget(fd);
2109 if (!f.file)
2110 return error;
2111
2112 if (type != F_UNLCK && !(f.file->f_mode & (FMODE_READ | FMODE_WRITE)))
2113 goto out_putf;
2114
2115 flock_make_lock(f.file, &fl, type);
2116
2117 error = security_file_lock(f.file, fl.fl_type);
2118 if (error)
2119 goto out_putf;
2120
2121 can_sleep = !(cmd & LOCK_NB);
2122 if (can_sleep)
2123 fl.fl_flags |= FL_SLEEP;
2124
2125 if (f.file->f_op->flock)
2126 error = f.file->f_op->flock(f.file,
2127 (can_sleep) ? F_SETLKW : F_SETLK,
2128 &fl);
2129 else
2130 error = locks_lock_file_wait(f.file, &fl);
2131
2132 locks_release_private(&fl);
2133 out_putf:
2134 fdput(f);
2135
2136 return error;
2137}
2138
2139/**
2140 * vfs_test_lock - test file byte range lock
2141 * @filp: The file to test lock for
2142 * @fl: The lock to test; also used to hold result
2143 *
2144 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2145 * setting conf->fl_type to something other than F_UNLCK.
2146 */
2147int vfs_test_lock(struct file *filp, struct file_lock *fl)
2148{
2149 WARN_ON_ONCE(filp != fl->fl_file);
2150 if (filp->f_op->lock)
2151 return filp->f_op->lock(filp, F_GETLK, fl);
2152 posix_test_lock(filp, fl);
2153 return 0;
2154}
2155EXPORT_SYMBOL_GPL(vfs_test_lock);
2156
2157/**
2158 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2159 * @fl: The file_lock who's fl_pid should be translated
2160 * @ns: The namespace into which the pid should be translated
2161 *
2162 * Used to tranlate a fl_pid into a namespace virtual pid number
2163 */
2164static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2165{
2166 pid_t vnr;
2167 struct pid *pid;
2168
2169 if (IS_OFDLCK(fl))
2170 return -1;
2171 if (IS_REMOTELCK(fl))
2172 return fl->fl_pid;
2173 /*
2174 * If the flock owner process is dead and its pid has been already
2175 * freed, the translation below won't work, but we still want to show
2176 * flock owner pid number in init pidns.
2177 */
2178 if (ns == &init_pid_ns)
2179 return (pid_t)fl->fl_pid;
2180
2181 rcu_read_lock();
2182 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2183 vnr = pid_nr_ns(pid, ns);
2184 rcu_read_unlock();
2185 return vnr;
2186}
2187
2188static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2189{
2190 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2191#if BITS_PER_LONG == 32
2192 /*
2193 * Make sure we can represent the posix lock via
2194 * legacy 32bit flock.
2195 */
2196 if (fl->fl_start > OFFT_OFFSET_MAX)
2197 return -EOVERFLOW;
2198 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2199 return -EOVERFLOW;
2200#endif
2201 flock->l_start = fl->fl_start;
2202 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2203 fl->fl_end - fl->fl_start + 1;
2204 flock->l_whence = 0;
2205 flock->l_type = fl->fl_type;
2206 return 0;
2207}
2208
2209#if BITS_PER_LONG == 32
2210static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2211{
2212 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2213 flock->l_start = fl->fl_start;
2214 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2215 fl->fl_end - fl->fl_start + 1;
2216 flock->l_whence = 0;
2217 flock->l_type = fl->fl_type;
2218}
2219#endif
2220
2221/* Report the first existing lock that would conflict with l.
2222 * This implements the F_GETLK command of fcntl().
2223 */
2224int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2225{
2226 struct file_lock *fl;
2227 int error;
2228
2229 fl = locks_alloc_lock();
2230 if (fl == NULL)
2231 return -ENOMEM;
2232 error = -EINVAL;
2233 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2234 goto out;
2235
2236 error = flock_to_posix_lock(filp, fl, flock);
2237 if (error)
2238 goto out;
2239
2240 if (cmd == F_OFD_GETLK) {
2241 error = -EINVAL;
2242 if (flock->l_pid != 0)
2243 goto out;
2244
2245 fl->fl_flags |= FL_OFDLCK;
2246 fl->fl_owner = filp;
2247 }
2248
2249 error = vfs_test_lock(filp, fl);
2250 if (error)
2251 goto out;
2252
2253 flock->l_type = fl->fl_type;
2254 if (fl->fl_type != F_UNLCK) {
2255 error = posix_lock_to_flock(flock, fl);
2256 if (error)
2257 goto out;
2258 }
2259out:
2260 locks_free_lock(fl);
2261 return error;
2262}
2263
2264/**
2265 * vfs_lock_file - file byte range lock
2266 * @filp: The file to apply the lock to
2267 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2268 * @fl: The lock to be applied
2269 * @conf: Place to return a copy of the conflicting lock, if found.
2270 *
2271 * A caller that doesn't care about the conflicting lock may pass NULL
2272 * as the final argument.
2273 *
2274 * If the filesystem defines a private ->lock() method, then @conf will
2275 * be left unchanged; so a caller that cares should initialize it to
2276 * some acceptable default.
2277 *
2278 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2279 * locks, the ->lock() interface may return asynchronously, before the lock has
2280 * been granted or denied by the underlying filesystem, if (and only if)
2281 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2282 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2283 * the request is for a blocking lock. When ->lock() does return asynchronously,
2284 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2285 * request completes.
2286 * If the request is for non-blocking lock the file system should return
2287 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2288 * with the result. If the request timed out the callback routine will return a
2289 * nonzero return code and the file system should release the lock. The file
2290 * system is also responsible to keep a corresponding posix lock when it
2291 * grants a lock so the VFS can find out which locks are locally held and do
2292 * the correct lock cleanup when required.
2293 * The underlying filesystem must not drop the kernel lock or call
2294 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2295 * return code.
2296 */
2297int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2298{
2299 WARN_ON_ONCE(filp != fl->fl_file);
2300 if (filp->f_op->lock)
2301 return filp->f_op->lock(filp, cmd, fl);
2302 else
2303 return posix_lock_file(filp, fl, conf);
2304}
2305EXPORT_SYMBOL_GPL(vfs_lock_file);
2306
2307static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2308 struct file_lock *fl)
2309{
2310 int error;
2311
2312 error = security_file_lock(filp, fl->fl_type);
2313 if (error)
2314 return error;
2315
2316 for (;;) {
2317 error = vfs_lock_file(filp, cmd, fl, NULL);
2318 if (error != FILE_LOCK_DEFERRED)
2319 break;
2320 error = wait_event_interruptible(fl->fl_wait,
2321 list_empty(&fl->fl_blocked_member));
2322 if (error)
2323 break;
2324 }
2325 locks_delete_block(fl);
2326
2327 return error;
2328}
2329
2330/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2331static int
2332check_fmode_for_setlk(struct file_lock *fl)
2333{
2334 switch (fl->fl_type) {
2335 case F_RDLCK:
2336 if (!(fl->fl_file->f_mode & FMODE_READ))
2337 return -EBADF;
2338 break;
2339 case F_WRLCK:
2340 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2341 return -EBADF;
2342 }
2343 return 0;
2344}
2345
2346/* Apply the lock described by l to an open file descriptor.
2347 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2348 */
2349int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2350 struct flock *flock)
2351{
2352 struct file_lock *file_lock = locks_alloc_lock();
2353 struct inode *inode = locks_inode(filp);
2354 struct file *f;
2355 int error;
2356
2357 if (file_lock == NULL)
2358 return -ENOLCK;
2359
2360 error = flock_to_posix_lock(filp, file_lock, flock);
2361 if (error)
2362 goto out;
2363
2364 error = check_fmode_for_setlk(file_lock);
2365 if (error)
2366 goto out;
2367
2368 /*
2369 * If the cmd is requesting file-private locks, then set the
2370 * FL_OFDLCK flag and override the owner.
2371 */
2372 switch (cmd) {
2373 case F_OFD_SETLK:
2374 error = -EINVAL;
2375 if (flock->l_pid != 0)
2376 goto out;
2377
2378 cmd = F_SETLK;
2379 file_lock->fl_flags |= FL_OFDLCK;
2380 file_lock->fl_owner = filp;
2381 break;
2382 case F_OFD_SETLKW:
2383 error = -EINVAL;
2384 if (flock->l_pid != 0)
2385 goto out;
2386
2387 cmd = F_SETLKW;
2388 file_lock->fl_flags |= FL_OFDLCK;
2389 file_lock->fl_owner = filp;
2390 fallthrough;
2391 case F_SETLKW:
2392 file_lock->fl_flags |= FL_SLEEP;
2393 }
2394
2395 error = do_lock_file_wait(filp, cmd, file_lock);
2396
2397 /*
2398 * Attempt to detect a close/fcntl race and recover by releasing the
2399 * lock that was just acquired. There is no need to do that when we're
2400 * unlocking though, or for OFD locks.
2401 */
2402 if (!error && file_lock->fl_type != F_UNLCK &&
2403 !(file_lock->fl_flags & FL_OFDLCK)) {
2404 struct files_struct *files = current->files;
2405 /*
2406 * We need that spin_lock here - it prevents reordering between
2407 * update of i_flctx->flc_posix and check for it done in
2408 * close(). rcu_read_lock() wouldn't do.
2409 */
2410 spin_lock(&files->file_lock);
2411 f = files_lookup_fd_locked(files, fd);
2412 spin_unlock(&files->file_lock);
2413 if (f != filp) {
2414 file_lock->fl_type = F_UNLCK;
2415 error = do_lock_file_wait(filp, cmd, file_lock);
2416 WARN_ON_ONCE(error);
2417 error = -EBADF;
2418 }
2419 }
2420out:
2421 trace_fcntl_setlk(inode, file_lock, error);
2422 locks_free_lock(file_lock);
2423 return error;
2424}
2425
2426#if BITS_PER_LONG == 32
2427/* Report the first existing lock that would conflict with l.
2428 * This implements the F_GETLK command of fcntl().
2429 */
2430int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2431{
2432 struct file_lock *fl;
2433 int error;
2434
2435 fl = locks_alloc_lock();
2436 if (fl == NULL)
2437 return -ENOMEM;
2438
2439 error = -EINVAL;
2440 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2441 goto out;
2442
2443 error = flock64_to_posix_lock(filp, fl, flock);
2444 if (error)
2445 goto out;
2446
2447 if (cmd == F_OFD_GETLK) {
2448 error = -EINVAL;
2449 if (flock->l_pid != 0)
2450 goto out;
2451
2452 cmd = F_GETLK64;
2453 fl->fl_flags |= FL_OFDLCK;
2454 fl->fl_owner = filp;
2455 }
2456
2457 error = vfs_test_lock(filp, fl);
2458 if (error)
2459 goto out;
2460
2461 flock->l_type = fl->fl_type;
2462 if (fl->fl_type != F_UNLCK)
2463 posix_lock_to_flock64(flock, fl);
2464
2465out:
2466 locks_free_lock(fl);
2467 return error;
2468}
2469
2470/* Apply the lock described by l to an open file descriptor.
2471 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2472 */
2473int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2474 struct flock64 *flock)
2475{
2476 struct file_lock *file_lock = locks_alloc_lock();
2477 struct file *f;
2478 int error;
2479
2480 if (file_lock == NULL)
2481 return -ENOLCK;
2482
2483 error = flock64_to_posix_lock(filp, file_lock, flock);
2484 if (error)
2485 goto out;
2486
2487 error = check_fmode_for_setlk(file_lock);
2488 if (error)
2489 goto out;
2490
2491 /*
2492 * If the cmd is requesting file-private locks, then set the
2493 * FL_OFDLCK flag and override the owner.
2494 */
2495 switch (cmd) {
2496 case F_OFD_SETLK:
2497 error = -EINVAL;
2498 if (flock->l_pid != 0)
2499 goto out;
2500
2501 cmd = F_SETLK64;
2502 file_lock->fl_flags |= FL_OFDLCK;
2503 file_lock->fl_owner = filp;
2504 break;
2505 case F_OFD_SETLKW:
2506 error = -EINVAL;
2507 if (flock->l_pid != 0)
2508 goto out;
2509
2510 cmd = F_SETLKW64;
2511 file_lock->fl_flags |= FL_OFDLCK;
2512 file_lock->fl_owner = filp;
2513 fallthrough;
2514 case F_SETLKW64:
2515 file_lock->fl_flags |= FL_SLEEP;
2516 }
2517
2518 error = do_lock_file_wait(filp, cmd, file_lock);
2519
2520 /*
2521 * Attempt to detect a close/fcntl race and recover by releasing the
2522 * lock that was just acquired. There is no need to do that when we're
2523 * unlocking though, or for OFD locks.
2524 */
2525 if (!error && file_lock->fl_type != F_UNLCK &&
2526 !(file_lock->fl_flags & FL_OFDLCK)) {
2527 struct files_struct *files = current->files;
2528 /*
2529 * We need that spin_lock here - it prevents reordering between
2530 * update of i_flctx->flc_posix and check for it done in
2531 * close(). rcu_read_lock() wouldn't do.
2532 */
2533 spin_lock(&files->file_lock);
2534 f = files_lookup_fd_locked(files, fd);
2535 spin_unlock(&files->file_lock);
2536 if (f != filp) {
2537 file_lock->fl_type = F_UNLCK;
2538 error = do_lock_file_wait(filp, cmd, file_lock);
2539 WARN_ON_ONCE(error);
2540 error = -EBADF;
2541 }
2542 }
2543out:
2544 locks_free_lock(file_lock);
2545 return error;
2546}
2547#endif /* BITS_PER_LONG == 32 */
2548
2549/*
2550 * This function is called when the file is being removed
2551 * from the task's fd array. POSIX locks belonging to this task
2552 * are deleted at this time.
2553 */
2554void locks_remove_posix(struct file *filp, fl_owner_t owner)
2555{
2556 int error;
2557 struct inode *inode = locks_inode(filp);
2558 struct file_lock lock;
2559 struct file_lock_context *ctx;
2560
2561 /*
2562 * If there are no locks held on this file, we don't need to call
2563 * posix_lock_file(). Another process could be setting a lock on this
2564 * file at the same time, but we wouldn't remove that lock anyway.
2565 */
2566 ctx = locks_inode_context(inode);
2567 if (!ctx || list_empty(&ctx->flc_posix))
2568 return;
2569
2570 locks_init_lock(&lock);
2571 lock.fl_type = F_UNLCK;
2572 lock.fl_flags = FL_POSIX | FL_CLOSE;
2573 lock.fl_start = 0;
2574 lock.fl_end = OFFSET_MAX;
2575 lock.fl_owner = owner;
2576 lock.fl_pid = current->tgid;
2577 lock.fl_file = filp;
2578 lock.fl_ops = NULL;
2579 lock.fl_lmops = NULL;
2580
2581 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2582
2583 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2584 lock.fl_ops->fl_release_private(&lock);
2585 trace_locks_remove_posix(inode, &lock, error);
2586}
2587EXPORT_SYMBOL(locks_remove_posix);
2588
2589/* The i_flctx must be valid when calling into here */
2590static void
2591locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2592{
2593 struct file_lock fl;
2594 struct inode *inode = locks_inode(filp);
2595
2596 if (list_empty(&flctx->flc_flock))
2597 return;
2598
2599 flock_make_lock(filp, &fl, F_UNLCK);
2600 fl.fl_flags |= FL_CLOSE;
2601
2602 if (filp->f_op->flock)
2603 filp->f_op->flock(filp, F_SETLKW, &fl);
2604 else
2605 flock_lock_inode(inode, &fl);
2606
2607 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2608 fl.fl_ops->fl_release_private(&fl);
2609}
2610
2611/* The i_flctx must be valid when calling into here */
2612static void
2613locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2614{
2615 struct file_lock *fl, *tmp;
2616 LIST_HEAD(dispose);
2617
2618 if (list_empty(&ctx->flc_lease))
2619 return;
2620
2621 percpu_down_read(&file_rwsem);
2622 spin_lock(&ctx->flc_lock);
2623 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2624 if (filp == fl->fl_file)
2625 lease_modify(fl, F_UNLCK, &dispose);
2626 spin_unlock(&ctx->flc_lock);
2627 percpu_up_read(&file_rwsem);
2628
2629 locks_dispose_list(&dispose);
2630}
2631
2632/*
2633 * This function is called on the last close of an open file.
2634 */
2635void locks_remove_file(struct file *filp)
2636{
2637 struct file_lock_context *ctx;
2638
2639 ctx = locks_inode_context(locks_inode(filp));
2640 if (!ctx)
2641 return;
2642
2643 /* remove any OFD locks */
2644 locks_remove_posix(filp, filp);
2645
2646 /* remove flock locks */
2647 locks_remove_flock(filp, ctx);
2648
2649 /* remove any leases */
2650 locks_remove_lease(filp, ctx);
2651
2652 spin_lock(&ctx->flc_lock);
2653 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2654 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2655 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2656 spin_unlock(&ctx->flc_lock);
2657}
2658
2659/**
2660 * vfs_cancel_lock - file byte range unblock lock
2661 * @filp: The file to apply the unblock to
2662 * @fl: The lock to be unblocked
2663 *
2664 * Used by lock managers to cancel blocked requests
2665 */
2666int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2667{
2668 WARN_ON_ONCE(filp != fl->fl_file);
2669 if (filp->f_op->lock)
2670 return filp->f_op->lock(filp, F_CANCELLK, fl);
2671 return 0;
2672}
2673EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2674
2675/**
2676 * vfs_inode_has_locks - are any file locks held on @inode?
2677 * @inode: inode to check for locks
2678 *
2679 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2680 * set on @inode.
2681 */
2682bool vfs_inode_has_locks(struct inode *inode)
2683{
2684 struct file_lock_context *ctx;
2685 bool ret;
2686
2687 ctx = locks_inode_context(inode);
2688 if (!ctx)
2689 return false;
2690
2691 spin_lock(&ctx->flc_lock);
2692 ret = !list_empty(&ctx->flc_posix) || !list_empty(&ctx->flc_flock);
2693 spin_unlock(&ctx->flc_lock);
2694 return ret;
2695}
2696EXPORT_SYMBOL_GPL(vfs_inode_has_locks);
2697
2698#ifdef CONFIG_PROC_FS
2699#include <linux/proc_fs.h>
2700#include <linux/seq_file.h>
2701
2702struct locks_iterator {
2703 int li_cpu;
2704 loff_t li_pos;
2705};
2706
2707static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2708 loff_t id, char *pfx, int repeat)
2709{
2710 struct inode *inode = NULL;
2711 unsigned int fl_pid;
2712 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2713 int type;
2714
2715 fl_pid = locks_translate_pid(fl, proc_pidns);
2716 /*
2717 * If lock owner is dead (and pid is freed) or not visible in current
2718 * pidns, zero is shown as a pid value. Check lock info from
2719 * init_pid_ns to get saved lock pid value.
2720 */
2721
2722 if (fl->fl_file != NULL)
2723 inode = locks_inode(fl->fl_file);
2724
2725 seq_printf(f, "%lld: ", id);
2726
2727 if (repeat)
2728 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2729
2730 if (IS_POSIX(fl)) {
2731 if (fl->fl_flags & FL_ACCESS)
2732 seq_puts(f, "ACCESS");
2733 else if (IS_OFDLCK(fl))
2734 seq_puts(f, "OFDLCK");
2735 else
2736 seq_puts(f, "POSIX ");
2737
2738 seq_printf(f, " %s ",
2739 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2740 } else if (IS_FLOCK(fl)) {
2741 seq_puts(f, "FLOCK ADVISORY ");
2742 } else if (IS_LEASE(fl)) {
2743 if (fl->fl_flags & FL_DELEG)
2744 seq_puts(f, "DELEG ");
2745 else
2746 seq_puts(f, "LEASE ");
2747
2748 if (lease_breaking(fl))
2749 seq_puts(f, "BREAKING ");
2750 else if (fl->fl_file)
2751 seq_puts(f, "ACTIVE ");
2752 else
2753 seq_puts(f, "BREAKER ");
2754 } else {
2755 seq_puts(f, "UNKNOWN UNKNOWN ");
2756 }
2757 type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2758
2759 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2760 (type == F_RDLCK) ? "READ" : "UNLCK");
2761 if (inode) {
2762 /* userspace relies on this representation of dev_t */
2763 seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2764 MAJOR(inode->i_sb->s_dev),
2765 MINOR(inode->i_sb->s_dev), inode->i_ino);
2766 } else {
2767 seq_printf(f, "%d <none>:0 ", fl_pid);
2768 }
2769 if (IS_POSIX(fl)) {
2770 if (fl->fl_end == OFFSET_MAX)
2771 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2772 else
2773 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2774 } else {
2775 seq_puts(f, "0 EOF\n");
2776 }
2777}
2778
2779static struct file_lock *get_next_blocked_member(struct file_lock *node)
2780{
2781 struct file_lock *tmp;
2782
2783 /* NULL node or root node */
2784 if (node == NULL || node->fl_blocker == NULL)
2785 return NULL;
2786
2787 /* Next member in the linked list could be itself */
2788 tmp = list_next_entry(node, fl_blocked_member);
2789 if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2790 || tmp == node) {
2791 return NULL;
2792 }
2793
2794 return tmp;
2795}
2796
2797static int locks_show(struct seq_file *f, void *v)
2798{
2799 struct locks_iterator *iter = f->private;
2800 struct file_lock *cur, *tmp;
2801 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2802 int level = 0;
2803
2804 cur = hlist_entry(v, struct file_lock, fl_link);
2805
2806 if (locks_translate_pid(cur, proc_pidns) == 0)
2807 return 0;
2808
2809 /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2810 * is the left child of current node, the next silibing in fl_blocked_member is the
2811 * right child, we can alse get the parent of current node from fl_blocker, so this
2812 * question becomes traversal of a binary tree
2813 */
2814 while (cur != NULL) {
2815 if (level)
2816 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2817 else
2818 lock_get_status(f, cur, iter->li_pos, "", level);
2819
2820 if (!list_empty(&cur->fl_blocked_requests)) {
2821 /* Turn left */
2822 cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2823 struct file_lock, fl_blocked_member);
2824 level++;
2825 } else {
2826 /* Turn right */
2827 tmp = get_next_blocked_member(cur);
2828 /* Fall back to parent node */
2829 while (tmp == NULL && cur->fl_blocker != NULL) {
2830 cur = cur->fl_blocker;
2831 level--;
2832 tmp = get_next_blocked_member(cur);
2833 }
2834 cur = tmp;
2835 }
2836 }
2837
2838 return 0;
2839}
2840
2841static void __show_fd_locks(struct seq_file *f,
2842 struct list_head *head, int *id,
2843 struct file *filp, struct files_struct *files)
2844{
2845 struct file_lock *fl;
2846
2847 list_for_each_entry(fl, head, fl_list) {
2848
2849 if (filp != fl->fl_file)
2850 continue;
2851 if (fl->fl_owner != files &&
2852 fl->fl_owner != filp)
2853 continue;
2854
2855 (*id)++;
2856 seq_puts(f, "lock:\t");
2857 lock_get_status(f, fl, *id, "", 0);
2858 }
2859}
2860
2861void show_fd_locks(struct seq_file *f,
2862 struct file *filp, struct files_struct *files)
2863{
2864 struct inode *inode = locks_inode(filp);
2865 struct file_lock_context *ctx;
2866 int id = 0;
2867
2868 ctx = locks_inode_context(inode);
2869 if (!ctx)
2870 return;
2871
2872 spin_lock(&ctx->flc_lock);
2873 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2874 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2875 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2876 spin_unlock(&ctx->flc_lock);
2877}
2878
2879static void *locks_start(struct seq_file *f, loff_t *pos)
2880 __acquires(&blocked_lock_lock)
2881{
2882 struct locks_iterator *iter = f->private;
2883
2884 iter->li_pos = *pos + 1;
2885 percpu_down_write(&file_rwsem);
2886 spin_lock(&blocked_lock_lock);
2887 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2888}
2889
2890static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2891{
2892 struct locks_iterator *iter = f->private;
2893
2894 ++iter->li_pos;
2895 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2896}
2897
2898static void locks_stop(struct seq_file *f, void *v)
2899 __releases(&blocked_lock_lock)
2900{
2901 spin_unlock(&blocked_lock_lock);
2902 percpu_up_write(&file_rwsem);
2903}
2904
2905static const struct seq_operations locks_seq_operations = {
2906 .start = locks_start,
2907 .next = locks_next,
2908 .stop = locks_stop,
2909 .show = locks_show,
2910};
2911
2912static int __init proc_locks_init(void)
2913{
2914 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2915 sizeof(struct locks_iterator), NULL);
2916 return 0;
2917}
2918fs_initcall(proc_locks_init);
2919#endif
2920
2921static int __init filelock_init(void)
2922{
2923 int i;
2924
2925 flctx_cache = kmem_cache_create("file_lock_ctx",
2926 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2927
2928 filelock_cache = kmem_cache_create("file_lock_cache",
2929 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2930
2931 for_each_possible_cpu(i) {
2932 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2933
2934 spin_lock_init(&fll->lock);
2935 INIT_HLIST_HEAD(&fll->hlist);
2936 }
2937
2938 lease_notifier_chain_init();
2939 return 0;
2940}
2941core_initcall(filelock_init);
1/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
119#include <linux/fdtable.h>
120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
125#include <linux/syscalls.h>
126#include <linux/time.h>
127#include <linux/rcupdate.h>
128#include <linux/pid_namespace.h>
129
130#include <asm/uaccess.h>
131
132#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
135
136static bool lease_breaking(struct file_lock *fl)
137{
138 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
139}
140
141static int target_leasetype(struct file_lock *fl)
142{
143 if (fl->fl_flags & FL_UNLOCK_PENDING)
144 return F_UNLCK;
145 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
146 return F_RDLCK;
147 return fl->fl_type;
148}
149
150int leases_enable = 1;
151int lease_break_time = 45;
152
153#define for_each_lock(inode, lockp) \
154 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
155
156static LIST_HEAD(file_lock_list);
157static LIST_HEAD(blocked_list);
158static DEFINE_SPINLOCK(file_lock_lock);
159
160/*
161 * Protects the two list heads above, plus the inode->i_flock list
162 */
163void lock_flocks(void)
164{
165 spin_lock(&file_lock_lock);
166}
167EXPORT_SYMBOL_GPL(lock_flocks);
168
169void unlock_flocks(void)
170{
171 spin_unlock(&file_lock_lock);
172}
173EXPORT_SYMBOL_GPL(unlock_flocks);
174
175static struct kmem_cache *filelock_cache __read_mostly;
176
177static void locks_init_lock_heads(struct file_lock *fl)
178{
179 INIT_LIST_HEAD(&fl->fl_link);
180 INIT_LIST_HEAD(&fl->fl_block);
181 init_waitqueue_head(&fl->fl_wait);
182}
183
184/* Allocate an empty lock structure. */
185struct file_lock *locks_alloc_lock(void)
186{
187 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
188
189 if (fl)
190 locks_init_lock_heads(fl);
191
192 return fl;
193}
194EXPORT_SYMBOL_GPL(locks_alloc_lock);
195
196void locks_release_private(struct file_lock *fl)
197{
198 if (fl->fl_ops) {
199 if (fl->fl_ops->fl_release_private)
200 fl->fl_ops->fl_release_private(fl);
201 fl->fl_ops = NULL;
202 }
203 if (fl->fl_lmops) {
204 if (fl->fl_lmops->lm_release_private)
205 fl->fl_lmops->lm_release_private(fl);
206 fl->fl_lmops = NULL;
207 }
208
209}
210EXPORT_SYMBOL_GPL(locks_release_private);
211
212/* Free a lock which is not in use. */
213void locks_free_lock(struct file_lock *fl)
214{
215 BUG_ON(waitqueue_active(&fl->fl_wait));
216 BUG_ON(!list_empty(&fl->fl_block));
217 BUG_ON(!list_empty(&fl->fl_link));
218
219 locks_release_private(fl);
220 kmem_cache_free(filelock_cache, fl);
221}
222EXPORT_SYMBOL(locks_free_lock);
223
224void locks_init_lock(struct file_lock *fl)
225{
226 memset(fl, 0, sizeof(struct file_lock));
227 locks_init_lock_heads(fl);
228}
229
230EXPORT_SYMBOL(locks_init_lock);
231
232static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
233{
234 if (fl->fl_ops) {
235 if (fl->fl_ops->fl_copy_lock)
236 fl->fl_ops->fl_copy_lock(new, fl);
237 new->fl_ops = fl->fl_ops;
238 }
239 if (fl->fl_lmops)
240 new->fl_lmops = fl->fl_lmops;
241}
242
243/*
244 * Initialize a new lock from an existing file_lock structure.
245 */
246void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
247{
248 new->fl_owner = fl->fl_owner;
249 new->fl_pid = fl->fl_pid;
250 new->fl_file = NULL;
251 new->fl_flags = fl->fl_flags;
252 new->fl_type = fl->fl_type;
253 new->fl_start = fl->fl_start;
254 new->fl_end = fl->fl_end;
255 new->fl_ops = NULL;
256 new->fl_lmops = NULL;
257}
258EXPORT_SYMBOL(__locks_copy_lock);
259
260void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
261{
262 locks_release_private(new);
263
264 __locks_copy_lock(new, fl);
265 new->fl_file = fl->fl_file;
266 new->fl_ops = fl->fl_ops;
267 new->fl_lmops = fl->fl_lmops;
268
269 locks_copy_private(new, fl);
270}
271
272EXPORT_SYMBOL(locks_copy_lock);
273
274static inline int flock_translate_cmd(int cmd) {
275 if (cmd & LOCK_MAND)
276 return cmd & (LOCK_MAND | LOCK_RW);
277 switch (cmd) {
278 case LOCK_SH:
279 return F_RDLCK;
280 case LOCK_EX:
281 return F_WRLCK;
282 case LOCK_UN:
283 return F_UNLCK;
284 }
285 return -EINVAL;
286}
287
288/* Fill in a file_lock structure with an appropriate FLOCK lock. */
289static int flock_make_lock(struct file *filp, struct file_lock **lock,
290 unsigned int cmd)
291{
292 struct file_lock *fl;
293 int type = flock_translate_cmd(cmd);
294 if (type < 0)
295 return type;
296
297 fl = locks_alloc_lock();
298 if (fl == NULL)
299 return -ENOMEM;
300
301 fl->fl_file = filp;
302 fl->fl_pid = current->tgid;
303 fl->fl_flags = FL_FLOCK;
304 fl->fl_type = type;
305 fl->fl_end = OFFSET_MAX;
306
307 *lock = fl;
308 return 0;
309}
310
311static int assign_type(struct file_lock *fl, long type)
312{
313 switch (type) {
314 case F_RDLCK:
315 case F_WRLCK:
316 case F_UNLCK:
317 fl->fl_type = type;
318 break;
319 default:
320 return -EINVAL;
321 }
322 return 0;
323}
324
325/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
326 * style lock.
327 */
328static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
329 struct flock *l)
330{
331 off_t start, end;
332
333 switch (l->l_whence) {
334 case SEEK_SET:
335 start = 0;
336 break;
337 case SEEK_CUR:
338 start = filp->f_pos;
339 break;
340 case SEEK_END:
341 start = i_size_read(filp->f_path.dentry->d_inode);
342 break;
343 default:
344 return -EINVAL;
345 }
346
347 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
348 POSIX-2001 defines it. */
349 start += l->l_start;
350 if (start < 0)
351 return -EINVAL;
352 fl->fl_end = OFFSET_MAX;
353 if (l->l_len > 0) {
354 end = start + l->l_len - 1;
355 fl->fl_end = end;
356 } else if (l->l_len < 0) {
357 end = start - 1;
358 fl->fl_end = end;
359 start += l->l_len;
360 if (start < 0)
361 return -EINVAL;
362 }
363 fl->fl_start = start; /* we record the absolute position */
364 if (fl->fl_end < fl->fl_start)
365 return -EOVERFLOW;
366
367 fl->fl_owner = current->files;
368 fl->fl_pid = current->tgid;
369 fl->fl_file = filp;
370 fl->fl_flags = FL_POSIX;
371 fl->fl_ops = NULL;
372 fl->fl_lmops = NULL;
373
374 return assign_type(fl, l->l_type);
375}
376
377#if BITS_PER_LONG == 32
378static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
379 struct flock64 *l)
380{
381 loff_t start;
382
383 switch (l->l_whence) {
384 case SEEK_SET:
385 start = 0;
386 break;
387 case SEEK_CUR:
388 start = filp->f_pos;
389 break;
390 case SEEK_END:
391 start = i_size_read(filp->f_path.dentry->d_inode);
392 break;
393 default:
394 return -EINVAL;
395 }
396
397 start += l->l_start;
398 if (start < 0)
399 return -EINVAL;
400 fl->fl_end = OFFSET_MAX;
401 if (l->l_len > 0) {
402 fl->fl_end = start + l->l_len - 1;
403 } else if (l->l_len < 0) {
404 fl->fl_end = start - 1;
405 start += l->l_len;
406 if (start < 0)
407 return -EINVAL;
408 }
409 fl->fl_start = start; /* we record the absolute position */
410 if (fl->fl_end < fl->fl_start)
411 return -EOVERFLOW;
412
413 fl->fl_owner = current->files;
414 fl->fl_pid = current->tgid;
415 fl->fl_file = filp;
416 fl->fl_flags = FL_POSIX;
417 fl->fl_ops = NULL;
418 fl->fl_lmops = NULL;
419
420 return assign_type(fl, l->l_type);
421}
422#endif
423
424/* default lease lock manager operations */
425static void lease_break_callback(struct file_lock *fl)
426{
427 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
428}
429
430static void lease_release_private_callback(struct file_lock *fl)
431{
432 if (!fl->fl_file)
433 return;
434
435 f_delown(fl->fl_file);
436 fl->fl_file->f_owner.signum = 0;
437}
438
439static const struct lock_manager_operations lease_manager_ops = {
440 .lm_break = lease_break_callback,
441 .lm_release_private = lease_release_private_callback,
442 .lm_change = lease_modify,
443};
444
445/*
446 * Initialize a lease, use the default lock manager operations
447 */
448static int lease_init(struct file *filp, long type, struct file_lock *fl)
449 {
450 if (assign_type(fl, type) != 0)
451 return -EINVAL;
452
453 fl->fl_owner = current->files;
454 fl->fl_pid = current->tgid;
455
456 fl->fl_file = filp;
457 fl->fl_flags = FL_LEASE;
458 fl->fl_start = 0;
459 fl->fl_end = OFFSET_MAX;
460 fl->fl_ops = NULL;
461 fl->fl_lmops = &lease_manager_ops;
462 return 0;
463}
464
465/* Allocate a file_lock initialised to this type of lease */
466static struct file_lock *lease_alloc(struct file *filp, long type)
467{
468 struct file_lock *fl = locks_alloc_lock();
469 int error = -ENOMEM;
470
471 if (fl == NULL)
472 return ERR_PTR(error);
473
474 error = lease_init(filp, type, fl);
475 if (error) {
476 locks_free_lock(fl);
477 return ERR_PTR(error);
478 }
479 return fl;
480}
481
482/* Check if two locks overlap each other.
483 */
484static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
485{
486 return ((fl1->fl_end >= fl2->fl_start) &&
487 (fl2->fl_end >= fl1->fl_start));
488}
489
490/*
491 * Check whether two locks have the same owner.
492 */
493static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
494{
495 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
496 return fl2->fl_lmops == fl1->fl_lmops &&
497 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
498 return fl1->fl_owner == fl2->fl_owner;
499}
500
501/* Remove waiter from blocker's block list.
502 * When blocker ends up pointing to itself then the list is empty.
503 */
504static void __locks_delete_block(struct file_lock *waiter)
505{
506 list_del_init(&waiter->fl_block);
507 list_del_init(&waiter->fl_link);
508 waiter->fl_next = NULL;
509}
510
511/*
512 */
513void locks_delete_block(struct file_lock *waiter)
514{
515 lock_flocks();
516 __locks_delete_block(waiter);
517 unlock_flocks();
518}
519EXPORT_SYMBOL(locks_delete_block);
520
521/* Insert waiter into blocker's block list.
522 * We use a circular list so that processes can be easily woken up in
523 * the order they blocked. The documentation doesn't require this but
524 * it seems like the reasonable thing to do.
525 */
526static void locks_insert_block(struct file_lock *blocker,
527 struct file_lock *waiter)
528{
529 BUG_ON(!list_empty(&waiter->fl_block));
530 list_add_tail(&waiter->fl_block, &blocker->fl_block);
531 waiter->fl_next = blocker;
532 if (IS_POSIX(blocker))
533 list_add(&waiter->fl_link, &blocked_list);
534}
535
536/* Wake up processes blocked waiting for blocker.
537 * If told to wait then schedule the processes until the block list
538 * is empty, otherwise empty the block list ourselves.
539 */
540static void locks_wake_up_blocks(struct file_lock *blocker)
541{
542 while (!list_empty(&blocker->fl_block)) {
543 struct file_lock *waiter;
544
545 waiter = list_first_entry(&blocker->fl_block,
546 struct file_lock, fl_block);
547 __locks_delete_block(waiter);
548 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
549 waiter->fl_lmops->lm_notify(waiter);
550 else
551 wake_up(&waiter->fl_wait);
552 }
553}
554
555/* Insert file lock fl into an inode's lock list at the position indicated
556 * by pos. At the same time add the lock to the global file lock list.
557 */
558static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
559{
560 list_add(&fl->fl_link, &file_lock_list);
561
562 fl->fl_nspid = get_pid(task_tgid(current));
563
564 /* insert into file's list */
565 fl->fl_next = *pos;
566 *pos = fl;
567}
568
569/*
570 * Delete a lock and then free it.
571 * Wake up processes that are blocked waiting for this lock,
572 * notify the FS that the lock has been cleared and
573 * finally free the lock.
574 */
575static void locks_delete_lock(struct file_lock **thisfl_p)
576{
577 struct file_lock *fl = *thisfl_p;
578
579 *thisfl_p = fl->fl_next;
580 fl->fl_next = NULL;
581 list_del_init(&fl->fl_link);
582
583 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
584 if (fl->fl_fasync != NULL) {
585 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
586 fl->fl_fasync = NULL;
587 }
588
589 if (fl->fl_nspid) {
590 put_pid(fl->fl_nspid);
591 fl->fl_nspid = NULL;
592 }
593
594 locks_wake_up_blocks(fl);
595 locks_free_lock(fl);
596}
597
598/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
599 * checks for shared/exclusive status of overlapping locks.
600 */
601static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
602{
603 if (sys_fl->fl_type == F_WRLCK)
604 return 1;
605 if (caller_fl->fl_type == F_WRLCK)
606 return 1;
607 return 0;
608}
609
610/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
611 * checking before calling the locks_conflict().
612 */
613static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
614{
615 /* POSIX locks owned by the same process do not conflict with
616 * each other.
617 */
618 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
619 return (0);
620
621 /* Check whether they overlap */
622 if (!locks_overlap(caller_fl, sys_fl))
623 return 0;
624
625 return (locks_conflict(caller_fl, sys_fl));
626}
627
628/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
629 * checking before calling the locks_conflict().
630 */
631static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
632{
633 /* FLOCK locks referring to the same filp do not conflict with
634 * each other.
635 */
636 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
637 return (0);
638 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
639 return 0;
640
641 return (locks_conflict(caller_fl, sys_fl));
642}
643
644void
645posix_test_lock(struct file *filp, struct file_lock *fl)
646{
647 struct file_lock *cfl;
648
649 lock_flocks();
650 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
651 if (!IS_POSIX(cfl))
652 continue;
653 if (posix_locks_conflict(fl, cfl))
654 break;
655 }
656 if (cfl) {
657 __locks_copy_lock(fl, cfl);
658 if (cfl->fl_nspid)
659 fl->fl_pid = pid_vnr(cfl->fl_nspid);
660 } else
661 fl->fl_type = F_UNLCK;
662 unlock_flocks();
663 return;
664}
665EXPORT_SYMBOL(posix_test_lock);
666
667/*
668 * Deadlock detection:
669 *
670 * We attempt to detect deadlocks that are due purely to posix file
671 * locks.
672 *
673 * We assume that a task can be waiting for at most one lock at a time.
674 * So for any acquired lock, the process holding that lock may be
675 * waiting on at most one other lock. That lock in turns may be held by
676 * someone waiting for at most one other lock. Given a requested lock
677 * caller_fl which is about to wait for a conflicting lock block_fl, we
678 * follow this chain of waiters to ensure we are not about to create a
679 * cycle.
680 *
681 * Since we do this before we ever put a process to sleep on a lock, we
682 * are ensured that there is never a cycle; that is what guarantees that
683 * the while() loop in posix_locks_deadlock() eventually completes.
684 *
685 * Note: the above assumption may not be true when handling lock
686 * requests from a broken NFS client. It may also fail in the presence
687 * of tasks (such as posix threads) sharing the same open file table.
688 *
689 * To handle those cases, we just bail out after a few iterations.
690 */
691
692#define MAX_DEADLK_ITERATIONS 10
693
694/* Find a lock that the owner of the given block_fl is blocking on. */
695static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
696{
697 struct file_lock *fl;
698
699 list_for_each_entry(fl, &blocked_list, fl_link) {
700 if (posix_same_owner(fl, block_fl))
701 return fl->fl_next;
702 }
703 return NULL;
704}
705
706static int posix_locks_deadlock(struct file_lock *caller_fl,
707 struct file_lock *block_fl)
708{
709 int i = 0;
710
711 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
712 if (i++ > MAX_DEADLK_ITERATIONS)
713 return 0;
714 if (posix_same_owner(caller_fl, block_fl))
715 return 1;
716 }
717 return 0;
718}
719
720/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
721 * after any leases, but before any posix locks.
722 *
723 * Note that if called with an FL_EXISTS argument, the caller may determine
724 * whether or not a lock was successfully freed by testing the return
725 * value for -ENOENT.
726 */
727static int flock_lock_file(struct file *filp, struct file_lock *request)
728{
729 struct file_lock *new_fl = NULL;
730 struct file_lock **before;
731 struct inode * inode = filp->f_path.dentry->d_inode;
732 int error = 0;
733 int found = 0;
734
735 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
736 new_fl = locks_alloc_lock();
737 if (!new_fl)
738 return -ENOMEM;
739 }
740
741 lock_flocks();
742 if (request->fl_flags & FL_ACCESS)
743 goto find_conflict;
744
745 for_each_lock(inode, before) {
746 struct file_lock *fl = *before;
747 if (IS_POSIX(fl))
748 break;
749 if (IS_LEASE(fl))
750 continue;
751 if (filp != fl->fl_file)
752 continue;
753 if (request->fl_type == fl->fl_type)
754 goto out;
755 found = 1;
756 locks_delete_lock(before);
757 break;
758 }
759
760 if (request->fl_type == F_UNLCK) {
761 if ((request->fl_flags & FL_EXISTS) && !found)
762 error = -ENOENT;
763 goto out;
764 }
765
766 /*
767 * If a higher-priority process was blocked on the old file lock,
768 * give it the opportunity to lock the file.
769 */
770 if (found) {
771 unlock_flocks();
772 cond_resched();
773 lock_flocks();
774 }
775
776find_conflict:
777 for_each_lock(inode, before) {
778 struct file_lock *fl = *before;
779 if (IS_POSIX(fl))
780 break;
781 if (IS_LEASE(fl))
782 continue;
783 if (!flock_locks_conflict(request, fl))
784 continue;
785 error = -EAGAIN;
786 if (!(request->fl_flags & FL_SLEEP))
787 goto out;
788 error = FILE_LOCK_DEFERRED;
789 locks_insert_block(fl, request);
790 goto out;
791 }
792 if (request->fl_flags & FL_ACCESS)
793 goto out;
794 locks_copy_lock(new_fl, request);
795 locks_insert_lock(before, new_fl);
796 new_fl = NULL;
797 error = 0;
798
799out:
800 unlock_flocks();
801 if (new_fl)
802 locks_free_lock(new_fl);
803 return error;
804}
805
806static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
807{
808 struct file_lock *fl;
809 struct file_lock *new_fl = NULL;
810 struct file_lock *new_fl2 = NULL;
811 struct file_lock *left = NULL;
812 struct file_lock *right = NULL;
813 struct file_lock **before;
814 int error, added = 0;
815
816 /*
817 * We may need two file_lock structures for this operation,
818 * so we get them in advance to avoid races.
819 *
820 * In some cases we can be sure, that no new locks will be needed
821 */
822 if (!(request->fl_flags & FL_ACCESS) &&
823 (request->fl_type != F_UNLCK ||
824 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
825 new_fl = locks_alloc_lock();
826 new_fl2 = locks_alloc_lock();
827 }
828
829 lock_flocks();
830 if (request->fl_type != F_UNLCK) {
831 for_each_lock(inode, before) {
832 fl = *before;
833 if (!IS_POSIX(fl))
834 continue;
835 if (!posix_locks_conflict(request, fl))
836 continue;
837 if (conflock)
838 __locks_copy_lock(conflock, fl);
839 error = -EAGAIN;
840 if (!(request->fl_flags & FL_SLEEP))
841 goto out;
842 error = -EDEADLK;
843 if (posix_locks_deadlock(request, fl))
844 goto out;
845 error = FILE_LOCK_DEFERRED;
846 locks_insert_block(fl, request);
847 goto out;
848 }
849 }
850
851 /* If we're just looking for a conflict, we're done. */
852 error = 0;
853 if (request->fl_flags & FL_ACCESS)
854 goto out;
855
856 /*
857 * Find the first old lock with the same owner as the new lock.
858 */
859
860 before = &inode->i_flock;
861
862 /* First skip locks owned by other processes. */
863 while ((fl = *before) && (!IS_POSIX(fl) ||
864 !posix_same_owner(request, fl))) {
865 before = &fl->fl_next;
866 }
867
868 /* Process locks with this owner. */
869 while ((fl = *before) && posix_same_owner(request, fl)) {
870 /* Detect adjacent or overlapping regions (if same lock type)
871 */
872 if (request->fl_type == fl->fl_type) {
873 /* In all comparisons of start vs end, use
874 * "start - 1" rather than "end + 1". If end
875 * is OFFSET_MAX, end + 1 will become negative.
876 */
877 if (fl->fl_end < request->fl_start - 1)
878 goto next_lock;
879 /* If the next lock in the list has entirely bigger
880 * addresses than the new one, insert the lock here.
881 */
882 if (fl->fl_start - 1 > request->fl_end)
883 break;
884
885 /* If we come here, the new and old lock are of the
886 * same type and adjacent or overlapping. Make one
887 * lock yielding from the lower start address of both
888 * locks to the higher end address.
889 */
890 if (fl->fl_start > request->fl_start)
891 fl->fl_start = request->fl_start;
892 else
893 request->fl_start = fl->fl_start;
894 if (fl->fl_end < request->fl_end)
895 fl->fl_end = request->fl_end;
896 else
897 request->fl_end = fl->fl_end;
898 if (added) {
899 locks_delete_lock(before);
900 continue;
901 }
902 request = fl;
903 added = 1;
904 }
905 else {
906 /* Processing for different lock types is a bit
907 * more complex.
908 */
909 if (fl->fl_end < request->fl_start)
910 goto next_lock;
911 if (fl->fl_start > request->fl_end)
912 break;
913 if (request->fl_type == F_UNLCK)
914 added = 1;
915 if (fl->fl_start < request->fl_start)
916 left = fl;
917 /* If the next lock in the list has a higher end
918 * address than the new one, insert the new one here.
919 */
920 if (fl->fl_end > request->fl_end) {
921 right = fl;
922 break;
923 }
924 if (fl->fl_start >= request->fl_start) {
925 /* The new lock completely replaces an old
926 * one (This may happen several times).
927 */
928 if (added) {
929 locks_delete_lock(before);
930 continue;
931 }
932 /* Replace the old lock with the new one.
933 * Wake up anybody waiting for the old one,
934 * as the change in lock type might satisfy
935 * their needs.
936 */
937 locks_wake_up_blocks(fl);
938 fl->fl_start = request->fl_start;
939 fl->fl_end = request->fl_end;
940 fl->fl_type = request->fl_type;
941 locks_release_private(fl);
942 locks_copy_private(fl, request);
943 request = fl;
944 added = 1;
945 }
946 }
947 /* Go on to next lock.
948 */
949 next_lock:
950 before = &fl->fl_next;
951 }
952
953 /*
954 * The above code only modifies existing locks in case of
955 * merging or replacing. If new lock(s) need to be inserted
956 * all modifications are done bellow this, so it's safe yet to
957 * bail out.
958 */
959 error = -ENOLCK; /* "no luck" */
960 if (right && left == right && !new_fl2)
961 goto out;
962
963 error = 0;
964 if (!added) {
965 if (request->fl_type == F_UNLCK) {
966 if (request->fl_flags & FL_EXISTS)
967 error = -ENOENT;
968 goto out;
969 }
970
971 if (!new_fl) {
972 error = -ENOLCK;
973 goto out;
974 }
975 locks_copy_lock(new_fl, request);
976 locks_insert_lock(before, new_fl);
977 new_fl = NULL;
978 }
979 if (right) {
980 if (left == right) {
981 /* The new lock breaks the old one in two pieces,
982 * so we have to use the second new lock.
983 */
984 left = new_fl2;
985 new_fl2 = NULL;
986 locks_copy_lock(left, right);
987 locks_insert_lock(before, left);
988 }
989 right->fl_start = request->fl_end + 1;
990 locks_wake_up_blocks(right);
991 }
992 if (left) {
993 left->fl_end = request->fl_start - 1;
994 locks_wake_up_blocks(left);
995 }
996 out:
997 unlock_flocks();
998 /*
999 * Free any unused locks.
1000 */
1001 if (new_fl)
1002 locks_free_lock(new_fl);
1003 if (new_fl2)
1004 locks_free_lock(new_fl2);
1005 return error;
1006}
1007
1008/**
1009 * posix_lock_file - Apply a POSIX-style lock to a file
1010 * @filp: The file to apply the lock to
1011 * @fl: The lock to be applied
1012 * @conflock: Place to return a copy of the conflicting lock, if found.
1013 *
1014 * Add a POSIX style lock to a file.
1015 * We merge adjacent & overlapping locks whenever possible.
1016 * POSIX locks are sorted by owner task, then by starting address
1017 *
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.
1021 */
1022int posix_lock_file(struct file *filp, struct file_lock *fl,
1023 struct file_lock *conflock)
1024{
1025 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1026}
1027EXPORT_SYMBOL(posix_lock_file);
1028
1029/**
1030 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1031 * @filp: The file to apply the lock to
1032 * @fl: The lock to be applied
1033 *
1034 * Add a POSIX style lock to a file.
1035 * We merge adjacent & overlapping locks whenever possible.
1036 * POSIX locks are sorted by owner task, then by starting address
1037 */
1038int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1039{
1040 int error;
1041 might_sleep ();
1042 for (;;) {
1043 error = posix_lock_file(filp, fl, NULL);
1044 if (error != FILE_LOCK_DEFERRED)
1045 break;
1046 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1047 if (!error)
1048 continue;
1049
1050 locks_delete_block(fl);
1051 break;
1052 }
1053 return error;
1054}
1055EXPORT_SYMBOL(posix_lock_file_wait);
1056
1057/**
1058 * locks_mandatory_locked - Check for an active lock
1059 * @inode: the file to check
1060 *
1061 * Searches the inode's list of locks to find any POSIX locks which conflict.
1062 * This function is called from locks_verify_locked() only.
1063 */
1064int locks_mandatory_locked(struct inode *inode)
1065{
1066 fl_owner_t owner = current->files;
1067 struct file_lock *fl;
1068
1069 /*
1070 * Search the lock list for this inode for any POSIX locks.
1071 */
1072 lock_flocks();
1073 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1074 if (!IS_POSIX(fl))
1075 continue;
1076 if (fl->fl_owner != owner)
1077 break;
1078 }
1079 unlock_flocks();
1080 return fl ? -EAGAIN : 0;
1081}
1082
1083/**
1084 * locks_mandatory_area - Check for a conflicting lock
1085 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1086 * for shared
1087 * @inode: the file to check
1088 * @filp: how the file was opened (if it was)
1089 * @offset: start of area to check
1090 * @count: length of area to check
1091 *
1092 * Searches the inode's list of locks to find any POSIX locks which conflict.
1093 * This function is called from rw_verify_area() and
1094 * locks_verify_truncate().
1095 */
1096int locks_mandatory_area(int read_write, struct inode *inode,
1097 struct file *filp, loff_t offset,
1098 size_t count)
1099{
1100 struct file_lock fl;
1101 int error;
1102
1103 locks_init_lock(&fl);
1104 fl.fl_owner = current->files;
1105 fl.fl_pid = current->tgid;
1106 fl.fl_file = filp;
1107 fl.fl_flags = FL_POSIX | FL_ACCESS;
1108 if (filp && !(filp->f_flags & O_NONBLOCK))
1109 fl.fl_flags |= FL_SLEEP;
1110 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1111 fl.fl_start = offset;
1112 fl.fl_end = offset + count - 1;
1113
1114 for (;;) {
1115 error = __posix_lock_file(inode, &fl, NULL);
1116 if (error != FILE_LOCK_DEFERRED)
1117 break;
1118 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1119 if (!error) {
1120 /*
1121 * If we've been sleeping someone might have
1122 * changed the permissions behind our back.
1123 */
1124 if (__mandatory_lock(inode))
1125 continue;
1126 }
1127
1128 locks_delete_block(&fl);
1129 break;
1130 }
1131
1132 return error;
1133}
1134
1135EXPORT_SYMBOL(locks_mandatory_area);
1136
1137static void lease_clear_pending(struct file_lock *fl, int arg)
1138{
1139 switch (arg) {
1140 case F_UNLCK:
1141 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1142 /* fall through: */
1143 case F_RDLCK:
1144 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1145 }
1146}
1147
1148/* We already had a lease on this file; just change its type */
1149int lease_modify(struct file_lock **before, int arg)
1150{
1151 struct file_lock *fl = *before;
1152 int error = assign_type(fl, arg);
1153
1154 if (error)
1155 return error;
1156 lease_clear_pending(fl, arg);
1157 locks_wake_up_blocks(fl);
1158 if (arg == F_UNLCK)
1159 locks_delete_lock(before);
1160 return 0;
1161}
1162
1163EXPORT_SYMBOL(lease_modify);
1164
1165static bool past_time(unsigned long then)
1166{
1167 if (!then)
1168 /* 0 is a special value meaning "this never expires": */
1169 return false;
1170 return time_after(jiffies, then);
1171}
1172
1173static void time_out_leases(struct inode *inode)
1174{
1175 struct file_lock **before;
1176 struct file_lock *fl;
1177
1178 before = &inode->i_flock;
1179 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1180 if (past_time(fl->fl_downgrade_time))
1181 lease_modify(before, F_RDLCK);
1182 if (past_time(fl->fl_break_time))
1183 lease_modify(before, F_UNLCK);
1184 if (fl == *before) /* lease_modify may have freed fl */
1185 before = &fl->fl_next;
1186 }
1187}
1188
1189/**
1190 * __break_lease - revoke all outstanding leases on file
1191 * @inode: the inode of the file to return
1192 * @mode: the open mode (read or write)
1193 *
1194 * break_lease (inlined for speed) has checked there already is at least
1195 * some kind of lock (maybe a lease) on this file. Leases are broken on
1196 * a call to open() or truncate(). This function can sleep unless you
1197 * specified %O_NONBLOCK to your open().
1198 */
1199int __break_lease(struct inode *inode, unsigned int mode)
1200{
1201 int error = 0;
1202 struct file_lock *new_fl, *flock;
1203 struct file_lock *fl;
1204 unsigned long break_time;
1205 int i_have_this_lease = 0;
1206 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1207
1208 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1209 if (IS_ERR(new_fl))
1210 return PTR_ERR(new_fl);
1211
1212 lock_flocks();
1213
1214 time_out_leases(inode);
1215
1216 flock = inode->i_flock;
1217 if ((flock == NULL) || !IS_LEASE(flock))
1218 goto out;
1219
1220 if (!locks_conflict(flock, new_fl))
1221 goto out;
1222
1223 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1224 if (fl->fl_owner == current->files)
1225 i_have_this_lease = 1;
1226
1227 break_time = 0;
1228 if (lease_break_time > 0) {
1229 break_time = jiffies + lease_break_time * HZ;
1230 if (break_time == 0)
1231 break_time++; /* so that 0 means no break time */
1232 }
1233
1234 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1235 if (want_write) {
1236 if (fl->fl_flags & FL_UNLOCK_PENDING)
1237 continue;
1238 fl->fl_flags |= FL_UNLOCK_PENDING;
1239 fl->fl_break_time = break_time;
1240 } else {
1241 if (lease_breaking(flock))
1242 continue;
1243 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1244 fl->fl_downgrade_time = break_time;
1245 }
1246 fl->fl_lmops->lm_break(fl);
1247 }
1248
1249 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1250 error = -EWOULDBLOCK;
1251 goto out;
1252 }
1253
1254restart:
1255 break_time = flock->fl_break_time;
1256 if (break_time != 0) {
1257 break_time -= jiffies;
1258 if (break_time == 0)
1259 break_time++;
1260 }
1261 locks_insert_block(flock, new_fl);
1262 unlock_flocks();
1263 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1264 !new_fl->fl_next, break_time);
1265 lock_flocks();
1266 __locks_delete_block(new_fl);
1267 if (error >= 0) {
1268 if (error == 0)
1269 time_out_leases(inode);
1270 /*
1271 * Wait for the next conflicting lease that has not been
1272 * broken yet
1273 */
1274 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1275 flock = flock->fl_next) {
1276 if (locks_conflict(new_fl, flock))
1277 goto restart;
1278 }
1279 error = 0;
1280 }
1281
1282out:
1283 unlock_flocks();
1284 locks_free_lock(new_fl);
1285 return error;
1286}
1287
1288EXPORT_SYMBOL(__break_lease);
1289
1290/**
1291 * lease_get_mtime - get the last modified time of an inode
1292 * @inode: the inode
1293 * @time: pointer to a timespec which will contain the last modified time
1294 *
1295 * This is to force NFS clients to flush their caches for files with
1296 * exclusive leases. The justification is that if someone has an
1297 * exclusive lease, then they could be modifying it.
1298 */
1299void lease_get_mtime(struct inode *inode, struct timespec *time)
1300{
1301 struct file_lock *flock = inode->i_flock;
1302 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1303 *time = current_fs_time(inode->i_sb);
1304 else
1305 *time = inode->i_mtime;
1306}
1307
1308EXPORT_SYMBOL(lease_get_mtime);
1309
1310/**
1311 * fcntl_getlease - Enquire what lease is currently active
1312 * @filp: the file
1313 *
1314 * The value returned by this function will be one of
1315 * (if no lease break is pending):
1316 *
1317 * %F_RDLCK to indicate a shared lease is held.
1318 *
1319 * %F_WRLCK to indicate an exclusive lease is held.
1320 *
1321 * %F_UNLCK to indicate no lease is held.
1322 *
1323 * (if a lease break is pending):
1324 *
1325 * %F_RDLCK to indicate an exclusive lease needs to be
1326 * changed to a shared lease (or removed).
1327 *
1328 * %F_UNLCK to indicate the lease needs to be removed.
1329 *
1330 * XXX: sfr & willy disagree over whether F_INPROGRESS
1331 * should be returned to userspace.
1332 */
1333int fcntl_getlease(struct file *filp)
1334{
1335 struct file_lock *fl;
1336 int type = F_UNLCK;
1337
1338 lock_flocks();
1339 time_out_leases(filp->f_path.dentry->d_inode);
1340 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1341 fl = fl->fl_next) {
1342 if (fl->fl_file == filp) {
1343 type = target_leasetype(fl);
1344 break;
1345 }
1346 }
1347 unlock_flocks();
1348 return type;
1349}
1350
1351int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1352{
1353 struct file_lock *fl, **before, **my_before = NULL, *lease;
1354 struct dentry *dentry = filp->f_path.dentry;
1355 struct inode *inode = dentry->d_inode;
1356 int error;
1357
1358 lease = *flp;
1359
1360 error = -EAGAIN;
1361 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1362 goto out;
1363 if ((arg == F_WRLCK)
1364 && ((dentry->d_count > 1)
1365 || (atomic_read(&inode->i_count) > 1)))
1366 goto out;
1367
1368 /*
1369 * At this point, we know that if there is an exclusive
1370 * lease on this file, then we hold it on this filp
1371 * (otherwise our open of this file would have blocked).
1372 * And if we are trying to acquire an exclusive lease,
1373 * then the file is not open by anyone (including us)
1374 * except for this filp.
1375 */
1376 error = -EAGAIN;
1377 for (before = &inode->i_flock;
1378 ((fl = *before) != NULL) && IS_LEASE(fl);
1379 before = &fl->fl_next) {
1380 if (fl->fl_file == filp) {
1381 my_before = before;
1382 continue;
1383 }
1384 /*
1385 * No exclusive leases if someone else has a lease on
1386 * this file:
1387 */
1388 if (arg == F_WRLCK)
1389 goto out;
1390 /*
1391 * Modifying our existing lease is OK, but no getting a
1392 * new lease if someone else is opening for write:
1393 */
1394 if (fl->fl_flags & FL_UNLOCK_PENDING)
1395 goto out;
1396 }
1397
1398 if (my_before != NULL) {
1399 error = lease->fl_lmops->lm_change(my_before, arg);
1400 if (!error)
1401 *flp = *my_before;
1402 goto out;
1403 }
1404
1405 error = -EINVAL;
1406 if (!leases_enable)
1407 goto out;
1408
1409 locks_insert_lock(before, lease);
1410 return 0;
1411
1412out:
1413 return error;
1414}
1415
1416int generic_delete_lease(struct file *filp, struct file_lock **flp)
1417{
1418 struct file_lock *fl, **before;
1419 struct dentry *dentry = filp->f_path.dentry;
1420 struct inode *inode = dentry->d_inode;
1421
1422 for (before = &inode->i_flock;
1423 ((fl = *before) != NULL) && IS_LEASE(fl);
1424 before = &fl->fl_next) {
1425 if (fl->fl_file != filp)
1426 continue;
1427 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1428 }
1429 return -EAGAIN;
1430}
1431
1432/**
1433 * generic_setlease - sets a lease on an open file
1434 * @filp: file pointer
1435 * @arg: type of lease to obtain
1436 * @flp: input - file_lock to use, output - file_lock inserted
1437 *
1438 * The (input) flp->fl_lmops->lm_break function is required
1439 * by break_lease().
1440 *
1441 * Called with file_lock_lock held.
1442 */
1443int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1444{
1445 struct dentry *dentry = filp->f_path.dentry;
1446 struct inode *inode = dentry->d_inode;
1447 int error;
1448
1449 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1450 return -EACCES;
1451 if (!S_ISREG(inode->i_mode))
1452 return -EINVAL;
1453 error = security_file_lock(filp, arg);
1454 if (error)
1455 return error;
1456
1457 time_out_leases(inode);
1458
1459 BUG_ON(!(*flp)->fl_lmops->lm_break);
1460
1461 switch (arg) {
1462 case F_UNLCK:
1463 return generic_delete_lease(filp, flp);
1464 case F_RDLCK:
1465 case F_WRLCK:
1466 return generic_add_lease(filp, arg, flp);
1467 default:
1468 return -EINVAL;
1469 }
1470}
1471EXPORT_SYMBOL(generic_setlease);
1472
1473static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1474{
1475 if (filp->f_op && filp->f_op->setlease)
1476 return filp->f_op->setlease(filp, arg, lease);
1477 else
1478 return generic_setlease(filp, arg, lease);
1479}
1480
1481/**
1482 * vfs_setlease - sets a lease on an open file
1483 * @filp: file pointer
1484 * @arg: type of lease to obtain
1485 * @lease: file_lock to use
1486 *
1487 * Call this to establish a lease on the file.
1488 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1489 * break_lease will oops!
1490 *
1491 * This will call the filesystem's setlease file method, if
1492 * defined. Note that there is no getlease method; instead, the
1493 * filesystem setlease method should call back to setlease() to
1494 * add a lease to the inode's lease list, where fcntl_getlease() can
1495 * find it. Since fcntl_getlease() only reports whether the current
1496 * task holds a lease, a cluster filesystem need only do this for
1497 * leases held by processes on this node.
1498 *
1499 * There is also no break_lease method; filesystems that
1500 * handle their own leases should break leases themselves from the
1501 * filesystem's open, create, and (on truncate) setattr methods.
1502 *
1503 * Warning: the only current setlease methods exist only to disable
1504 * leases in certain cases. More vfs changes may be required to
1505 * allow a full filesystem lease implementation.
1506 */
1507
1508int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1509{
1510 int error;
1511
1512 lock_flocks();
1513 error = __vfs_setlease(filp, arg, lease);
1514 unlock_flocks();
1515
1516 return error;
1517}
1518EXPORT_SYMBOL_GPL(vfs_setlease);
1519
1520static int do_fcntl_delete_lease(struct file *filp)
1521{
1522 struct file_lock fl, *flp = &fl;
1523
1524 lease_init(filp, F_UNLCK, flp);
1525
1526 return vfs_setlease(filp, F_UNLCK, &flp);
1527}
1528
1529static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1530{
1531 struct file_lock *fl, *ret;
1532 struct fasync_struct *new;
1533 int error;
1534
1535 fl = lease_alloc(filp, arg);
1536 if (IS_ERR(fl))
1537 return PTR_ERR(fl);
1538
1539 new = fasync_alloc();
1540 if (!new) {
1541 locks_free_lock(fl);
1542 return -ENOMEM;
1543 }
1544 ret = fl;
1545 lock_flocks();
1546 error = __vfs_setlease(filp, arg, &ret);
1547 if (error) {
1548 unlock_flocks();
1549 locks_free_lock(fl);
1550 goto out_free_fasync;
1551 }
1552 if (ret != fl)
1553 locks_free_lock(fl);
1554
1555 /*
1556 * fasync_insert_entry() returns the old entry if any.
1557 * If there was no old entry, then it used 'new' and
1558 * inserted it into the fasync list. Clear new so that
1559 * we don't release it here.
1560 */
1561 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1562 new = NULL;
1563
1564 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1565 unlock_flocks();
1566
1567out_free_fasync:
1568 if (new)
1569 fasync_free(new);
1570 return error;
1571}
1572
1573/**
1574 * fcntl_setlease - sets a lease on an open file
1575 * @fd: open file descriptor
1576 * @filp: file pointer
1577 * @arg: type of lease to obtain
1578 *
1579 * Call this fcntl to establish a lease on the file.
1580 * Note that you also need to call %F_SETSIG to
1581 * receive a signal when the lease is broken.
1582 */
1583int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1584{
1585 if (arg == F_UNLCK)
1586 return do_fcntl_delete_lease(filp);
1587 return do_fcntl_add_lease(fd, filp, arg);
1588}
1589
1590/**
1591 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1592 * @filp: The file to apply the lock to
1593 * @fl: The lock to be applied
1594 *
1595 * Add a FLOCK style lock to a file.
1596 */
1597int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1598{
1599 int error;
1600 might_sleep();
1601 for (;;) {
1602 error = flock_lock_file(filp, fl);
1603 if (error != FILE_LOCK_DEFERRED)
1604 break;
1605 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1606 if (!error)
1607 continue;
1608
1609 locks_delete_block(fl);
1610 break;
1611 }
1612 return error;
1613}
1614
1615EXPORT_SYMBOL(flock_lock_file_wait);
1616
1617/**
1618 * sys_flock: - flock() system call.
1619 * @fd: the file descriptor to lock.
1620 * @cmd: the type of lock to apply.
1621 *
1622 * Apply a %FL_FLOCK style lock to an open file descriptor.
1623 * The @cmd can be one of
1624 *
1625 * %LOCK_SH -- a shared lock.
1626 *
1627 * %LOCK_EX -- an exclusive lock.
1628 *
1629 * %LOCK_UN -- remove an existing lock.
1630 *
1631 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1632 *
1633 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1634 * processes read and write access respectively.
1635 */
1636SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1637{
1638 struct file *filp;
1639 int fput_needed;
1640 struct file_lock *lock;
1641 int can_sleep, unlock;
1642 int error;
1643
1644 error = -EBADF;
1645 filp = fget_light(fd, &fput_needed);
1646 if (!filp)
1647 goto out;
1648
1649 can_sleep = !(cmd & LOCK_NB);
1650 cmd &= ~LOCK_NB;
1651 unlock = (cmd == LOCK_UN);
1652
1653 if (!unlock && !(cmd & LOCK_MAND) &&
1654 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
1655 goto out_putf;
1656
1657 error = flock_make_lock(filp, &lock, cmd);
1658 if (error)
1659 goto out_putf;
1660 if (can_sleep)
1661 lock->fl_flags |= FL_SLEEP;
1662
1663 error = security_file_lock(filp, lock->fl_type);
1664 if (error)
1665 goto out_free;
1666
1667 if (filp->f_op && filp->f_op->flock)
1668 error = filp->f_op->flock(filp,
1669 (can_sleep) ? F_SETLKW : F_SETLK,
1670 lock);
1671 else
1672 error = flock_lock_file_wait(filp, lock);
1673
1674 out_free:
1675 locks_free_lock(lock);
1676
1677 out_putf:
1678 fput_light(filp, fput_needed);
1679 out:
1680 return error;
1681}
1682
1683/**
1684 * vfs_test_lock - test file byte range lock
1685 * @filp: The file to test lock for
1686 * @fl: The lock to test; also used to hold result
1687 *
1688 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1689 * setting conf->fl_type to something other than F_UNLCK.
1690 */
1691int vfs_test_lock(struct file *filp, struct file_lock *fl)
1692{
1693 if (filp->f_op && filp->f_op->lock)
1694 return filp->f_op->lock(filp, F_GETLK, fl);
1695 posix_test_lock(filp, fl);
1696 return 0;
1697}
1698EXPORT_SYMBOL_GPL(vfs_test_lock);
1699
1700static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1701{
1702 flock->l_pid = fl->fl_pid;
1703#if BITS_PER_LONG == 32
1704 /*
1705 * Make sure we can represent the posix lock via
1706 * legacy 32bit flock.
1707 */
1708 if (fl->fl_start > OFFT_OFFSET_MAX)
1709 return -EOVERFLOW;
1710 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1711 return -EOVERFLOW;
1712#endif
1713 flock->l_start = fl->fl_start;
1714 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1715 fl->fl_end - fl->fl_start + 1;
1716 flock->l_whence = 0;
1717 flock->l_type = fl->fl_type;
1718 return 0;
1719}
1720
1721#if BITS_PER_LONG == 32
1722static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1723{
1724 flock->l_pid = fl->fl_pid;
1725 flock->l_start = fl->fl_start;
1726 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1727 fl->fl_end - fl->fl_start + 1;
1728 flock->l_whence = 0;
1729 flock->l_type = fl->fl_type;
1730}
1731#endif
1732
1733/* Report the first existing lock that would conflict with l.
1734 * This implements the F_GETLK command of fcntl().
1735 */
1736int fcntl_getlk(struct file *filp, struct flock __user *l)
1737{
1738 struct file_lock file_lock;
1739 struct flock flock;
1740 int error;
1741
1742 error = -EFAULT;
1743 if (copy_from_user(&flock, l, sizeof(flock)))
1744 goto out;
1745 error = -EINVAL;
1746 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1747 goto out;
1748
1749 error = flock_to_posix_lock(filp, &file_lock, &flock);
1750 if (error)
1751 goto out;
1752
1753 error = vfs_test_lock(filp, &file_lock);
1754 if (error)
1755 goto out;
1756
1757 flock.l_type = file_lock.fl_type;
1758 if (file_lock.fl_type != F_UNLCK) {
1759 error = posix_lock_to_flock(&flock, &file_lock);
1760 if (error)
1761 goto out;
1762 }
1763 error = -EFAULT;
1764 if (!copy_to_user(l, &flock, sizeof(flock)))
1765 error = 0;
1766out:
1767 return error;
1768}
1769
1770/**
1771 * vfs_lock_file - file byte range lock
1772 * @filp: The file to apply the lock to
1773 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1774 * @fl: The lock to be applied
1775 * @conf: Place to return a copy of the conflicting lock, if found.
1776 *
1777 * A caller that doesn't care about the conflicting lock may pass NULL
1778 * as the final argument.
1779 *
1780 * If the filesystem defines a private ->lock() method, then @conf will
1781 * be left unchanged; so a caller that cares should initialize it to
1782 * some acceptable default.
1783 *
1784 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1785 * locks, the ->lock() interface may return asynchronously, before the lock has
1786 * been granted or denied by the underlying filesystem, if (and only if)
1787 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1788 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1789 * the request is for a blocking lock. When ->lock() does return asynchronously,
1790 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1791 * request completes.
1792 * If the request is for non-blocking lock the file system should return
1793 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1794 * with the result. If the request timed out the callback routine will return a
1795 * nonzero return code and the file system should release the lock. The file
1796 * system is also responsible to keep a corresponding posix lock when it
1797 * grants a lock so the VFS can find out which locks are locally held and do
1798 * the correct lock cleanup when required.
1799 * The underlying filesystem must not drop the kernel lock or call
1800 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1801 * return code.
1802 */
1803int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1804{
1805 if (filp->f_op && filp->f_op->lock)
1806 return filp->f_op->lock(filp, cmd, fl);
1807 else
1808 return posix_lock_file(filp, fl, conf);
1809}
1810EXPORT_SYMBOL_GPL(vfs_lock_file);
1811
1812static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1813 struct file_lock *fl)
1814{
1815 int error;
1816
1817 error = security_file_lock(filp, fl->fl_type);
1818 if (error)
1819 return error;
1820
1821 for (;;) {
1822 error = vfs_lock_file(filp, cmd, fl, NULL);
1823 if (error != FILE_LOCK_DEFERRED)
1824 break;
1825 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1826 if (!error)
1827 continue;
1828
1829 locks_delete_block(fl);
1830 break;
1831 }
1832
1833 return error;
1834}
1835
1836/* Apply the lock described by l to an open file descriptor.
1837 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1838 */
1839int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1840 struct flock __user *l)
1841{
1842 struct file_lock *file_lock = locks_alloc_lock();
1843 struct flock flock;
1844 struct inode *inode;
1845 struct file *f;
1846 int error;
1847
1848 if (file_lock == NULL)
1849 return -ENOLCK;
1850
1851 /*
1852 * This might block, so we do it before checking the inode.
1853 */
1854 error = -EFAULT;
1855 if (copy_from_user(&flock, l, sizeof(flock)))
1856 goto out;
1857
1858 inode = filp->f_path.dentry->d_inode;
1859
1860 /* Don't allow mandatory locks on files that may be memory mapped
1861 * and shared.
1862 */
1863 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1864 error = -EAGAIN;
1865 goto out;
1866 }
1867
1868again:
1869 error = flock_to_posix_lock(filp, file_lock, &flock);
1870 if (error)
1871 goto out;
1872 if (cmd == F_SETLKW) {
1873 file_lock->fl_flags |= FL_SLEEP;
1874 }
1875
1876 error = -EBADF;
1877 switch (flock.l_type) {
1878 case F_RDLCK:
1879 if (!(filp->f_mode & FMODE_READ))
1880 goto out;
1881 break;
1882 case F_WRLCK:
1883 if (!(filp->f_mode & FMODE_WRITE))
1884 goto out;
1885 break;
1886 case F_UNLCK:
1887 break;
1888 default:
1889 error = -EINVAL;
1890 goto out;
1891 }
1892
1893 error = do_lock_file_wait(filp, cmd, file_lock);
1894
1895 /*
1896 * Attempt to detect a close/fcntl race and recover by
1897 * releasing the lock that was just acquired.
1898 */
1899 /*
1900 * we need that spin_lock here - it prevents reordering between
1901 * update of inode->i_flock and check for it done in close().
1902 * rcu_read_lock() wouldn't do.
1903 */
1904 spin_lock(¤t->files->file_lock);
1905 f = fcheck(fd);
1906 spin_unlock(¤t->files->file_lock);
1907 if (!error && f != filp && flock.l_type != F_UNLCK) {
1908 flock.l_type = F_UNLCK;
1909 goto again;
1910 }
1911
1912out:
1913 locks_free_lock(file_lock);
1914 return error;
1915}
1916
1917#if BITS_PER_LONG == 32
1918/* Report the first existing lock that would conflict with l.
1919 * This implements the F_GETLK command of fcntl().
1920 */
1921int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1922{
1923 struct file_lock file_lock;
1924 struct flock64 flock;
1925 int error;
1926
1927 error = -EFAULT;
1928 if (copy_from_user(&flock, l, sizeof(flock)))
1929 goto out;
1930 error = -EINVAL;
1931 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1932 goto out;
1933
1934 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1935 if (error)
1936 goto out;
1937
1938 error = vfs_test_lock(filp, &file_lock);
1939 if (error)
1940 goto out;
1941
1942 flock.l_type = file_lock.fl_type;
1943 if (file_lock.fl_type != F_UNLCK)
1944 posix_lock_to_flock64(&flock, &file_lock);
1945
1946 error = -EFAULT;
1947 if (!copy_to_user(l, &flock, sizeof(flock)))
1948 error = 0;
1949
1950out:
1951 return error;
1952}
1953
1954/* Apply the lock described by l to an open file descriptor.
1955 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1956 */
1957int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1958 struct flock64 __user *l)
1959{
1960 struct file_lock *file_lock = locks_alloc_lock();
1961 struct flock64 flock;
1962 struct inode *inode;
1963 struct file *f;
1964 int error;
1965
1966 if (file_lock == NULL)
1967 return -ENOLCK;
1968
1969 /*
1970 * This might block, so we do it before checking the inode.
1971 */
1972 error = -EFAULT;
1973 if (copy_from_user(&flock, l, sizeof(flock)))
1974 goto out;
1975
1976 inode = filp->f_path.dentry->d_inode;
1977
1978 /* Don't allow mandatory locks on files that may be memory mapped
1979 * and shared.
1980 */
1981 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1982 error = -EAGAIN;
1983 goto out;
1984 }
1985
1986again:
1987 error = flock64_to_posix_lock(filp, file_lock, &flock);
1988 if (error)
1989 goto out;
1990 if (cmd == F_SETLKW64) {
1991 file_lock->fl_flags |= FL_SLEEP;
1992 }
1993
1994 error = -EBADF;
1995 switch (flock.l_type) {
1996 case F_RDLCK:
1997 if (!(filp->f_mode & FMODE_READ))
1998 goto out;
1999 break;
2000 case F_WRLCK:
2001 if (!(filp->f_mode & FMODE_WRITE))
2002 goto out;
2003 break;
2004 case F_UNLCK:
2005 break;
2006 default:
2007 error = -EINVAL;
2008 goto out;
2009 }
2010
2011 error = do_lock_file_wait(filp, cmd, file_lock);
2012
2013 /*
2014 * Attempt to detect a close/fcntl race and recover by
2015 * releasing the lock that was just acquired.
2016 */
2017 spin_lock(¤t->files->file_lock);
2018 f = fcheck(fd);
2019 spin_unlock(¤t->files->file_lock);
2020 if (!error && f != filp && flock.l_type != F_UNLCK) {
2021 flock.l_type = F_UNLCK;
2022 goto again;
2023 }
2024
2025out:
2026 locks_free_lock(file_lock);
2027 return error;
2028}
2029#endif /* BITS_PER_LONG == 32 */
2030
2031/*
2032 * This function is called when the file is being removed
2033 * from the task's fd array. POSIX locks belonging to this task
2034 * are deleted at this time.
2035 */
2036void locks_remove_posix(struct file *filp, fl_owner_t owner)
2037{
2038 struct file_lock lock;
2039
2040 /*
2041 * If there are no locks held on this file, we don't need to call
2042 * posix_lock_file(). Another process could be setting a lock on this
2043 * file at the same time, but we wouldn't remove that lock anyway.
2044 */
2045 if (!filp->f_path.dentry->d_inode->i_flock)
2046 return;
2047
2048 lock.fl_type = F_UNLCK;
2049 lock.fl_flags = FL_POSIX | FL_CLOSE;
2050 lock.fl_start = 0;
2051 lock.fl_end = OFFSET_MAX;
2052 lock.fl_owner = owner;
2053 lock.fl_pid = current->tgid;
2054 lock.fl_file = filp;
2055 lock.fl_ops = NULL;
2056 lock.fl_lmops = NULL;
2057
2058 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2059
2060 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2061 lock.fl_ops->fl_release_private(&lock);
2062}
2063
2064EXPORT_SYMBOL(locks_remove_posix);
2065
2066/*
2067 * This function is called on the last close of an open file.
2068 */
2069void locks_remove_flock(struct file *filp)
2070{
2071 struct inode * inode = filp->f_path.dentry->d_inode;
2072 struct file_lock *fl;
2073 struct file_lock **before;
2074
2075 if (!inode->i_flock)
2076 return;
2077
2078 if (filp->f_op && filp->f_op->flock) {
2079 struct file_lock fl = {
2080 .fl_pid = current->tgid,
2081 .fl_file = filp,
2082 .fl_flags = FL_FLOCK,
2083 .fl_type = F_UNLCK,
2084 .fl_end = OFFSET_MAX,
2085 };
2086 filp->f_op->flock(filp, F_SETLKW, &fl);
2087 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2088 fl.fl_ops->fl_release_private(&fl);
2089 }
2090
2091 lock_flocks();
2092 before = &inode->i_flock;
2093
2094 while ((fl = *before) != NULL) {
2095 if (fl->fl_file == filp) {
2096 if (IS_FLOCK(fl)) {
2097 locks_delete_lock(before);
2098 continue;
2099 }
2100 if (IS_LEASE(fl)) {
2101 lease_modify(before, F_UNLCK);
2102 continue;
2103 }
2104 /* What? */
2105 BUG();
2106 }
2107 before = &fl->fl_next;
2108 }
2109 unlock_flocks();
2110}
2111
2112/**
2113 * posix_unblock_lock - stop waiting for a file lock
2114 * @filp: how the file was opened
2115 * @waiter: the lock which was waiting
2116 *
2117 * lockd needs to block waiting for locks.
2118 */
2119int
2120posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2121{
2122 int status = 0;
2123
2124 lock_flocks();
2125 if (waiter->fl_next)
2126 __locks_delete_block(waiter);
2127 else
2128 status = -ENOENT;
2129 unlock_flocks();
2130 return status;
2131}
2132
2133EXPORT_SYMBOL(posix_unblock_lock);
2134
2135/**
2136 * vfs_cancel_lock - file byte range unblock lock
2137 * @filp: The file to apply the unblock to
2138 * @fl: The lock to be unblocked
2139 *
2140 * Used by lock managers to cancel blocked requests
2141 */
2142int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2143{
2144 if (filp->f_op && filp->f_op->lock)
2145 return filp->f_op->lock(filp, F_CANCELLK, fl);
2146 return 0;
2147}
2148
2149EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2150
2151#ifdef CONFIG_PROC_FS
2152#include <linux/proc_fs.h>
2153#include <linux/seq_file.h>
2154
2155static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2156 loff_t id, char *pfx)
2157{
2158 struct inode *inode = NULL;
2159 unsigned int fl_pid;
2160
2161 if (fl->fl_nspid)
2162 fl_pid = pid_vnr(fl->fl_nspid);
2163 else
2164 fl_pid = fl->fl_pid;
2165
2166 if (fl->fl_file != NULL)
2167 inode = fl->fl_file->f_path.dentry->d_inode;
2168
2169 seq_printf(f, "%lld:%s ", id, pfx);
2170 if (IS_POSIX(fl)) {
2171 seq_printf(f, "%6s %s ",
2172 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2173 (inode == NULL) ? "*NOINODE*" :
2174 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2175 } else if (IS_FLOCK(fl)) {
2176 if (fl->fl_type & LOCK_MAND) {
2177 seq_printf(f, "FLOCK MSNFS ");
2178 } else {
2179 seq_printf(f, "FLOCK ADVISORY ");
2180 }
2181 } else if (IS_LEASE(fl)) {
2182 seq_printf(f, "LEASE ");
2183 if (lease_breaking(fl))
2184 seq_printf(f, "BREAKING ");
2185 else if (fl->fl_file)
2186 seq_printf(f, "ACTIVE ");
2187 else
2188 seq_printf(f, "BREAKER ");
2189 } else {
2190 seq_printf(f, "UNKNOWN UNKNOWN ");
2191 }
2192 if (fl->fl_type & LOCK_MAND) {
2193 seq_printf(f, "%s ",
2194 (fl->fl_type & LOCK_READ)
2195 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2196 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2197 } else {
2198 seq_printf(f, "%s ",
2199 (lease_breaking(fl))
2200 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2201 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2202 }
2203 if (inode) {
2204#ifdef WE_CAN_BREAK_LSLK_NOW
2205 seq_printf(f, "%d %s:%ld ", fl_pid,
2206 inode->i_sb->s_id, inode->i_ino);
2207#else
2208 /* userspace relies on this representation of dev_t ;-( */
2209 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2210 MAJOR(inode->i_sb->s_dev),
2211 MINOR(inode->i_sb->s_dev), inode->i_ino);
2212#endif
2213 } else {
2214 seq_printf(f, "%d <none>:0 ", fl_pid);
2215 }
2216 if (IS_POSIX(fl)) {
2217 if (fl->fl_end == OFFSET_MAX)
2218 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2219 else
2220 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2221 } else {
2222 seq_printf(f, "0 EOF\n");
2223 }
2224}
2225
2226static int locks_show(struct seq_file *f, void *v)
2227{
2228 struct file_lock *fl, *bfl;
2229
2230 fl = list_entry(v, struct file_lock, fl_link);
2231
2232 lock_get_status(f, fl, *((loff_t *)f->private), "");
2233
2234 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2235 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2236
2237 return 0;
2238}
2239
2240static void *locks_start(struct seq_file *f, loff_t *pos)
2241{
2242 loff_t *p = f->private;
2243
2244 lock_flocks();
2245 *p = (*pos + 1);
2246 return seq_list_start(&file_lock_list, *pos);
2247}
2248
2249static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2250{
2251 loff_t *p = f->private;
2252 ++*p;
2253 return seq_list_next(v, &file_lock_list, pos);
2254}
2255
2256static void locks_stop(struct seq_file *f, void *v)
2257{
2258 unlock_flocks();
2259}
2260
2261static const struct seq_operations locks_seq_operations = {
2262 .start = locks_start,
2263 .next = locks_next,
2264 .stop = locks_stop,
2265 .show = locks_show,
2266};
2267
2268static int locks_open(struct inode *inode, struct file *filp)
2269{
2270 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
2271}
2272
2273static const struct file_operations proc_locks_operations = {
2274 .open = locks_open,
2275 .read = seq_read,
2276 .llseek = seq_lseek,
2277 .release = seq_release_private,
2278};
2279
2280static int __init proc_locks_init(void)
2281{
2282 proc_create("locks", 0, NULL, &proc_locks_operations);
2283 return 0;
2284}
2285module_init(proc_locks_init);
2286#endif
2287
2288/**
2289 * lock_may_read - checks that the region is free of locks
2290 * @inode: the inode that is being read
2291 * @start: the first byte to read
2292 * @len: the number of bytes to read
2293 *
2294 * Emulates Windows locking requirements. Whole-file
2295 * mandatory locks (share modes) can prohibit a read and
2296 * byte-range POSIX locks can prohibit a read if they overlap.
2297 *
2298 * N.B. this function is only ever called
2299 * from knfsd and ownership of locks is never checked.
2300 */
2301int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2302{
2303 struct file_lock *fl;
2304 int result = 1;
2305 lock_flocks();
2306 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2307 if (IS_POSIX(fl)) {
2308 if (fl->fl_type == F_RDLCK)
2309 continue;
2310 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2311 continue;
2312 } else if (IS_FLOCK(fl)) {
2313 if (!(fl->fl_type & LOCK_MAND))
2314 continue;
2315 if (fl->fl_type & LOCK_READ)
2316 continue;
2317 } else
2318 continue;
2319 result = 0;
2320 break;
2321 }
2322 unlock_flocks();
2323 return result;
2324}
2325
2326EXPORT_SYMBOL(lock_may_read);
2327
2328/**
2329 * lock_may_write - checks that the region is free of locks
2330 * @inode: the inode that is being written
2331 * @start: the first byte to write
2332 * @len: the number of bytes to write
2333 *
2334 * Emulates Windows locking requirements. Whole-file
2335 * mandatory locks (share modes) can prohibit a write and
2336 * byte-range POSIX locks can prohibit a write if they overlap.
2337 *
2338 * N.B. this function is only ever called
2339 * from knfsd and ownership of locks is never checked.
2340 */
2341int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2342{
2343 struct file_lock *fl;
2344 int result = 1;
2345 lock_flocks();
2346 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2347 if (IS_POSIX(fl)) {
2348 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2349 continue;
2350 } else if (IS_FLOCK(fl)) {
2351 if (!(fl->fl_type & LOCK_MAND))
2352 continue;
2353 if (fl->fl_type & LOCK_WRITE)
2354 continue;
2355 } else
2356 continue;
2357 result = 0;
2358 break;
2359 }
2360 unlock_flocks();
2361 return result;
2362}
2363
2364EXPORT_SYMBOL(lock_may_write);
2365
2366static int __init filelock_init(void)
2367{
2368 filelock_cache = kmem_cache_create("file_lock_cache",
2369 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2370
2371 return 0;
2372}
2373
2374core_initcall(filelock_init);