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// 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#include <linux/capability.h>
52#include <linux/file.h>
53#include <linux/fdtable.h>
54#include <linux/filelock.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
72static struct file_lock *file_lock(struct file_lock_core *flc)
73{
74 return container_of(flc, struct file_lock, c);
75}
76
77static struct file_lease *file_lease(struct file_lock_core *flc)
78{
79 return container_of(flc, struct file_lease, c);
80}
81
82static bool lease_breaking(struct file_lease *fl)
83{
84 return fl->c.flc_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
85}
86
87static int target_leasetype(struct file_lease *fl)
88{
89 if (fl->c.flc_flags & FL_UNLOCK_PENDING)
90 return F_UNLCK;
91 if (fl->c.flc_flags & FL_DOWNGRADE_PENDING)
92 return F_RDLCK;
93 return fl->c.flc_type;
94}
95
96static int leases_enable = 1;
97static int lease_break_time = 45;
98
99#ifdef CONFIG_SYSCTL
100static struct ctl_table locks_sysctls[] = {
101 {
102 .procname = "leases-enable",
103 .data = &leases_enable,
104 .maxlen = sizeof(int),
105 .mode = 0644,
106 .proc_handler = proc_dointvec,
107 },
108#ifdef CONFIG_MMU
109 {
110 .procname = "lease-break-time",
111 .data = &lease_break_time,
112 .maxlen = sizeof(int),
113 .mode = 0644,
114 .proc_handler = proc_dointvec,
115 },
116#endif /* CONFIG_MMU */
117};
118
119static int __init init_fs_locks_sysctls(void)
120{
121 register_sysctl_init("fs", locks_sysctls);
122 return 0;
123}
124early_initcall(init_fs_locks_sysctls);
125#endif /* CONFIG_SYSCTL */
126
127/*
128 * The global file_lock_list is only used for displaying /proc/locks, so we
129 * keep a list on each CPU, with each list protected by its own spinlock.
130 * Global serialization is done using file_rwsem.
131 *
132 * Note that alterations to the list also require that the relevant flc_lock is
133 * held.
134 */
135struct file_lock_list_struct {
136 spinlock_t lock;
137 struct hlist_head hlist;
138};
139static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
140DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
141
142
143/*
144 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
145 * It is protected by blocked_lock_lock.
146 *
147 * We hash locks by lockowner in order to optimize searching for the lock a
148 * particular lockowner is waiting on.
149 *
150 * FIXME: make this value scale via some heuristic? We generally will want more
151 * buckets when we have more lockowners holding locks, but that's a little
152 * difficult to determine without knowing what the workload will look like.
153 */
154#define BLOCKED_HASH_BITS 7
155static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
156
157/*
158 * This lock protects the blocked_hash. Generally, if you're accessing it, you
159 * want to be holding this lock.
160 *
161 * In addition, it also protects the fl->fl_blocked_requests list, and the
162 * fl->fl_blocker pointer for file_lock structures that are acting as lock
163 * requests (in contrast to those that are acting as records of acquired locks).
164 *
165 * Note that when we acquire this lock in order to change the above fields,
166 * we often hold the flc_lock as well. In certain cases, when reading the fields
167 * protected by this lock, we can skip acquiring it iff we already hold the
168 * flc_lock.
169 */
170static DEFINE_SPINLOCK(blocked_lock_lock);
171
172static struct kmem_cache *flctx_cache __ro_after_init;
173static struct kmem_cache *filelock_cache __ro_after_init;
174static struct kmem_cache *filelease_cache __ro_after_init;
175
176static struct file_lock_context *
177locks_get_lock_context(struct inode *inode, int type)
178{
179 struct file_lock_context *ctx;
180
181 /* paired with cmpxchg() below */
182 ctx = locks_inode_context(inode);
183 if (likely(ctx) || type == F_UNLCK)
184 goto out;
185
186 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
187 if (!ctx)
188 goto out;
189
190 spin_lock_init(&ctx->flc_lock);
191 INIT_LIST_HEAD(&ctx->flc_flock);
192 INIT_LIST_HEAD(&ctx->flc_posix);
193 INIT_LIST_HEAD(&ctx->flc_lease);
194
195 /*
196 * Assign the pointer if it's not already assigned. If it is, then
197 * free the context we just allocated.
198 */
199 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
200 kmem_cache_free(flctx_cache, ctx);
201 ctx = locks_inode_context(inode);
202 }
203out:
204 trace_locks_get_lock_context(inode, type, ctx);
205 return ctx;
206}
207
208static void
209locks_dump_ctx_list(struct list_head *list, char *list_type)
210{
211 struct file_lock_core *flc;
212
213 list_for_each_entry(flc, list, flc_list)
214 pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
215 list_type, flc->flc_owner, flc->flc_flags,
216 flc->flc_type, flc->flc_pid);
217}
218
219static void
220locks_check_ctx_lists(struct inode *inode)
221{
222 struct file_lock_context *ctx = inode->i_flctx;
223
224 if (unlikely(!list_empty(&ctx->flc_flock) ||
225 !list_empty(&ctx->flc_posix) ||
226 !list_empty(&ctx->flc_lease))) {
227 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
228 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
229 inode->i_ino);
230 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
231 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
232 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
233 }
234}
235
236static void
237locks_check_ctx_file_list(struct file *filp, struct list_head *list, char *list_type)
238{
239 struct file_lock_core *flc;
240 struct inode *inode = file_inode(filp);
241
242 list_for_each_entry(flc, list, flc_list)
243 if (flc->flc_file == filp)
244 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
245 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
246 list_type, MAJOR(inode->i_sb->s_dev),
247 MINOR(inode->i_sb->s_dev), inode->i_ino,
248 flc->flc_owner, flc->flc_flags,
249 flc->flc_type, flc->flc_pid);
250}
251
252void
253locks_free_lock_context(struct inode *inode)
254{
255 struct file_lock_context *ctx = locks_inode_context(inode);
256
257 if (unlikely(ctx)) {
258 locks_check_ctx_lists(inode);
259 kmem_cache_free(flctx_cache, ctx);
260 }
261}
262
263static void locks_init_lock_heads(struct file_lock_core *flc)
264{
265 INIT_HLIST_NODE(&flc->flc_link);
266 INIT_LIST_HEAD(&flc->flc_list);
267 INIT_LIST_HEAD(&flc->flc_blocked_requests);
268 INIT_LIST_HEAD(&flc->flc_blocked_member);
269 init_waitqueue_head(&flc->flc_wait);
270}
271
272/* Allocate an empty lock structure. */
273struct file_lock *locks_alloc_lock(void)
274{
275 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
276
277 if (fl)
278 locks_init_lock_heads(&fl->c);
279
280 return fl;
281}
282EXPORT_SYMBOL_GPL(locks_alloc_lock);
283
284/* Allocate an empty lock structure. */
285struct file_lease *locks_alloc_lease(void)
286{
287 struct file_lease *fl = kmem_cache_zalloc(filelease_cache, GFP_KERNEL);
288
289 if (fl)
290 locks_init_lock_heads(&fl->c);
291
292 return fl;
293}
294EXPORT_SYMBOL_GPL(locks_alloc_lease);
295
296void locks_release_private(struct file_lock *fl)
297{
298 struct file_lock_core *flc = &fl->c;
299
300 BUG_ON(waitqueue_active(&flc->flc_wait));
301 BUG_ON(!list_empty(&flc->flc_list));
302 BUG_ON(!list_empty(&flc->flc_blocked_requests));
303 BUG_ON(!list_empty(&flc->flc_blocked_member));
304 BUG_ON(!hlist_unhashed(&flc->flc_link));
305
306 if (fl->fl_ops) {
307 if (fl->fl_ops->fl_release_private)
308 fl->fl_ops->fl_release_private(fl);
309 fl->fl_ops = NULL;
310 }
311
312 if (fl->fl_lmops) {
313 if (fl->fl_lmops->lm_put_owner) {
314 fl->fl_lmops->lm_put_owner(flc->flc_owner);
315 flc->flc_owner = NULL;
316 }
317 fl->fl_lmops = NULL;
318 }
319}
320EXPORT_SYMBOL_GPL(locks_release_private);
321
322/**
323 * locks_owner_has_blockers - Check for blocking lock requests
324 * @flctx: file lock context
325 * @owner: lock owner
326 *
327 * Return values:
328 * %true: @owner has at least one blocker
329 * %false: @owner has no blockers
330 */
331bool locks_owner_has_blockers(struct file_lock_context *flctx, fl_owner_t owner)
332{
333 struct file_lock_core *flc;
334
335 spin_lock(&flctx->flc_lock);
336 list_for_each_entry(flc, &flctx->flc_posix, flc_list) {
337 if (flc->flc_owner != owner)
338 continue;
339 if (!list_empty(&flc->flc_blocked_requests)) {
340 spin_unlock(&flctx->flc_lock);
341 return true;
342 }
343 }
344 spin_unlock(&flctx->flc_lock);
345 return false;
346}
347EXPORT_SYMBOL_GPL(locks_owner_has_blockers);
348
349/* Free a lock which is not in use. */
350void locks_free_lock(struct file_lock *fl)
351{
352 locks_release_private(fl);
353 kmem_cache_free(filelock_cache, fl);
354}
355EXPORT_SYMBOL(locks_free_lock);
356
357/* Free a lease which is not in use. */
358void locks_free_lease(struct file_lease *fl)
359{
360 kmem_cache_free(filelease_cache, fl);
361}
362EXPORT_SYMBOL(locks_free_lease);
363
364static void
365locks_dispose_list(struct list_head *dispose)
366{
367 struct file_lock_core *flc;
368
369 while (!list_empty(dispose)) {
370 flc = list_first_entry(dispose, struct file_lock_core, flc_list);
371 list_del_init(&flc->flc_list);
372 if (flc->flc_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
373 locks_free_lease(file_lease(flc));
374 else
375 locks_free_lock(file_lock(flc));
376 }
377}
378
379void locks_init_lock(struct file_lock *fl)
380{
381 memset(fl, 0, sizeof(struct file_lock));
382 locks_init_lock_heads(&fl->c);
383}
384EXPORT_SYMBOL(locks_init_lock);
385
386void locks_init_lease(struct file_lease *fl)
387{
388 memset(fl, 0, sizeof(*fl));
389 locks_init_lock_heads(&fl->c);
390}
391EXPORT_SYMBOL(locks_init_lease);
392
393/*
394 * Initialize a new lock from an existing file_lock structure.
395 */
396void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
397{
398 new->c.flc_owner = fl->c.flc_owner;
399 new->c.flc_pid = fl->c.flc_pid;
400 new->c.flc_file = NULL;
401 new->c.flc_flags = fl->c.flc_flags;
402 new->c.flc_type = fl->c.flc_type;
403 new->fl_start = fl->fl_start;
404 new->fl_end = fl->fl_end;
405 new->fl_lmops = fl->fl_lmops;
406 new->fl_ops = NULL;
407
408 if (fl->fl_lmops) {
409 if (fl->fl_lmops->lm_get_owner)
410 fl->fl_lmops->lm_get_owner(fl->c.flc_owner);
411 }
412}
413EXPORT_SYMBOL(locks_copy_conflock);
414
415void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
416{
417 /* "new" must be a freshly-initialized lock */
418 WARN_ON_ONCE(new->fl_ops);
419
420 locks_copy_conflock(new, fl);
421
422 new->c.flc_file = fl->c.flc_file;
423 new->fl_ops = fl->fl_ops;
424
425 if (fl->fl_ops) {
426 if (fl->fl_ops->fl_copy_lock)
427 fl->fl_ops->fl_copy_lock(new, fl);
428 }
429}
430EXPORT_SYMBOL(locks_copy_lock);
431
432static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
433{
434 struct file_lock *f;
435
436 /*
437 * As ctx->flc_lock is held, new requests cannot be added to
438 * ->flc_blocked_requests, so we don't need a lock to check if it
439 * is empty.
440 */
441 if (list_empty(&fl->c.flc_blocked_requests))
442 return;
443 spin_lock(&blocked_lock_lock);
444 list_splice_init(&fl->c.flc_blocked_requests,
445 &new->c.flc_blocked_requests);
446 list_for_each_entry(f, &new->c.flc_blocked_requests,
447 c.flc_blocked_member)
448 f->c.flc_blocker = &new->c;
449 spin_unlock(&blocked_lock_lock);
450}
451
452static inline int flock_translate_cmd(int cmd) {
453 switch (cmd) {
454 case LOCK_SH:
455 return F_RDLCK;
456 case LOCK_EX:
457 return F_WRLCK;
458 case LOCK_UN:
459 return F_UNLCK;
460 }
461 return -EINVAL;
462}
463
464/* Fill in a file_lock structure with an appropriate FLOCK lock. */
465static void flock_make_lock(struct file *filp, struct file_lock *fl, int type)
466{
467 locks_init_lock(fl);
468
469 fl->c.flc_file = filp;
470 fl->c.flc_owner = filp;
471 fl->c.flc_pid = current->tgid;
472 fl->c.flc_flags = FL_FLOCK;
473 fl->c.flc_type = type;
474 fl->fl_end = OFFSET_MAX;
475}
476
477static int assign_type(struct file_lock_core *flc, int type)
478{
479 switch (type) {
480 case F_RDLCK:
481 case F_WRLCK:
482 case F_UNLCK:
483 flc->flc_type = type;
484 break;
485 default:
486 return -EINVAL;
487 }
488 return 0;
489}
490
491static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
492 struct flock64 *l)
493{
494 switch (l->l_whence) {
495 case SEEK_SET:
496 fl->fl_start = 0;
497 break;
498 case SEEK_CUR:
499 fl->fl_start = filp->f_pos;
500 break;
501 case SEEK_END:
502 fl->fl_start = i_size_read(file_inode(filp));
503 break;
504 default:
505 return -EINVAL;
506 }
507 if (l->l_start > OFFSET_MAX - fl->fl_start)
508 return -EOVERFLOW;
509 fl->fl_start += l->l_start;
510 if (fl->fl_start < 0)
511 return -EINVAL;
512
513 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
514 POSIX-2001 defines it. */
515 if (l->l_len > 0) {
516 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
517 return -EOVERFLOW;
518 fl->fl_end = fl->fl_start + (l->l_len - 1);
519
520 } else if (l->l_len < 0) {
521 if (fl->fl_start + l->l_len < 0)
522 return -EINVAL;
523 fl->fl_end = fl->fl_start - 1;
524 fl->fl_start += l->l_len;
525 } else
526 fl->fl_end = OFFSET_MAX;
527
528 fl->c.flc_owner = current->files;
529 fl->c.flc_pid = current->tgid;
530 fl->c.flc_file = filp;
531 fl->c.flc_flags = FL_POSIX;
532 fl->fl_ops = NULL;
533 fl->fl_lmops = NULL;
534
535 return assign_type(&fl->c, l->l_type);
536}
537
538/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
539 * style lock.
540 */
541static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
542 struct flock *l)
543{
544 struct flock64 ll = {
545 .l_type = l->l_type,
546 .l_whence = l->l_whence,
547 .l_start = l->l_start,
548 .l_len = l->l_len,
549 };
550
551 return flock64_to_posix_lock(filp, fl, &ll);
552}
553
554/* default lease lock manager operations */
555static bool
556lease_break_callback(struct file_lease *fl)
557{
558 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
559 return false;
560}
561
562static void
563lease_setup(struct file_lease *fl, void **priv)
564{
565 struct file *filp = fl->c.flc_file;
566 struct fasync_struct *fa = *priv;
567
568 /*
569 * fasync_insert_entry() returns the old entry if any. If there was no
570 * old entry, then it used "priv" and inserted it into the fasync list.
571 * Clear the pointer to indicate that it shouldn't be freed.
572 */
573 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
574 *priv = NULL;
575
576 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
577}
578
579static const struct lease_manager_operations lease_manager_ops = {
580 .lm_break = lease_break_callback,
581 .lm_change = lease_modify,
582 .lm_setup = lease_setup,
583};
584
585/*
586 * Initialize a lease, use the default lock manager operations
587 */
588static int lease_init(struct file *filp, int type, struct file_lease *fl)
589{
590 if (assign_type(&fl->c, type) != 0)
591 return -EINVAL;
592
593 fl->c.flc_owner = filp;
594 fl->c.flc_pid = current->tgid;
595
596 fl->c.flc_file = filp;
597 fl->c.flc_flags = FL_LEASE;
598 fl->fl_lmops = &lease_manager_ops;
599 return 0;
600}
601
602/* Allocate a file_lock initialised to this type of lease */
603static struct file_lease *lease_alloc(struct file *filp, int type)
604{
605 struct file_lease *fl = locks_alloc_lease();
606 int error = -ENOMEM;
607
608 if (fl == NULL)
609 return ERR_PTR(error);
610
611 error = lease_init(filp, type, fl);
612 if (error) {
613 locks_free_lease(fl);
614 return ERR_PTR(error);
615 }
616 return fl;
617}
618
619/* Check if two locks overlap each other.
620 */
621static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
622{
623 return ((fl1->fl_end >= fl2->fl_start) &&
624 (fl2->fl_end >= fl1->fl_start));
625}
626
627/*
628 * Check whether two locks have the same owner.
629 */
630static int posix_same_owner(struct file_lock_core *fl1, struct file_lock_core *fl2)
631{
632 return fl1->flc_owner == fl2->flc_owner;
633}
634
635/* Must be called with the flc_lock held! */
636static void locks_insert_global_locks(struct file_lock_core *flc)
637{
638 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
639
640 percpu_rwsem_assert_held(&file_rwsem);
641
642 spin_lock(&fll->lock);
643 flc->flc_link_cpu = smp_processor_id();
644 hlist_add_head(&flc->flc_link, &fll->hlist);
645 spin_unlock(&fll->lock);
646}
647
648/* Must be called with the flc_lock held! */
649static void locks_delete_global_locks(struct file_lock_core *flc)
650{
651 struct file_lock_list_struct *fll;
652
653 percpu_rwsem_assert_held(&file_rwsem);
654
655 /*
656 * Avoid taking lock if already unhashed. This is safe since this check
657 * is done while holding the flc_lock, and new insertions into the list
658 * also require that it be held.
659 */
660 if (hlist_unhashed(&flc->flc_link))
661 return;
662
663 fll = per_cpu_ptr(&file_lock_list, flc->flc_link_cpu);
664 spin_lock(&fll->lock);
665 hlist_del_init(&flc->flc_link);
666 spin_unlock(&fll->lock);
667}
668
669static unsigned long
670posix_owner_key(struct file_lock_core *flc)
671{
672 return (unsigned long) flc->flc_owner;
673}
674
675static void locks_insert_global_blocked(struct file_lock_core *waiter)
676{
677 lockdep_assert_held(&blocked_lock_lock);
678
679 hash_add(blocked_hash, &waiter->flc_link, posix_owner_key(waiter));
680}
681
682static void locks_delete_global_blocked(struct file_lock_core *waiter)
683{
684 lockdep_assert_held(&blocked_lock_lock);
685
686 hash_del(&waiter->flc_link);
687}
688
689/* Remove waiter from blocker's block list.
690 * When blocker ends up pointing to itself then the list is empty.
691 *
692 * Must be called with blocked_lock_lock held.
693 */
694static void __locks_unlink_block(struct file_lock_core *waiter)
695{
696 locks_delete_global_blocked(waiter);
697 list_del_init(&waiter->flc_blocked_member);
698}
699
700static void __locks_wake_up_blocks(struct file_lock_core *blocker)
701{
702 while (!list_empty(&blocker->flc_blocked_requests)) {
703 struct file_lock_core *waiter;
704 struct file_lock *fl;
705
706 waiter = list_first_entry(&blocker->flc_blocked_requests,
707 struct file_lock_core, flc_blocked_member);
708
709 fl = file_lock(waiter);
710 __locks_unlink_block(waiter);
711 if ((waiter->flc_flags & (FL_POSIX | FL_FLOCK)) &&
712 fl->fl_lmops && fl->fl_lmops->lm_notify)
713 fl->fl_lmops->lm_notify(fl);
714 else
715 locks_wake_up(fl);
716
717 /*
718 * The setting of flc_blocker to NULL marks the "done"
719 * point in deleting a block. Paired with acquire at the top
720 * of locks_delete_block().
721 */
722 smp_store_release(&waiter->flc_blocker, NULL);
723 }
724}
725
726static int __locks_delete_block(struct file_lock_core *waiter)
727{
728 int status = -ENOENT;
729
730 /*
731 * If fl_blocker is NULL, it won't be set again as this thread "owns"
732 * the lock and is the only one that might try to claim the lock.
733 *
734 * We use acquire/release to manage fl_blocker so that we can
735 * optimize away taking the blocked_lock_lock in many cases.
736 *
737 * The smp_load_acquire guarantees two things:
738 *
739 * 1/ that fl_blocked_requests can be tested locklessly. If something
740 * was recently added to that list it must have been in a locked region
741 * *before* the locked region when fl_blocker was set to NULL.
742 *
743 * 2/ that no other thread is accessing 'waiter', so it is safe to free
744 * it. __locks_wake_up_blocks is careful not to touch waiter after
745 * fl_blocker is released.
746 *
747 * If a lockless check of fl_blocker shows it to be NULL, we know that
748 * no new locks can be inserted into its fl_blocked_requests list, and
749 * can avoid doing anything further if the list is empty.
750 */
751 if (!smp_load_acquire(&waiter->flc_blocker) &&
752 list_empty(&waiter->flc_blocked_requests))
753 return status;
754
755 spin_lock(&blocked_lock_lock);
756 if (waiter->flc_blocker)
757 status = 0;
758 __locks_wake_up_blocks(waiter);
759 __locks_unlink_block(waiter);
760
761 /*
762 * The setting of fl_blocker to NULL marks the "done" point in deleting
763 * a block. Paired with acquire at the top of this function.
764 */
765 smp_store_release(&waiter->flc_blocker, NULL);
766 spin_unlock(&blocked_lock_lock);
767 return status;
768}
769
770/**
771 * locks_delete_block - stop waiting for a file lock
772 * @waiter: the lock which was waiting
773 *
774 * lockd/nfsd need to disconnect the lock while working on it.
775 */
776int locks_delete_block(struct file_lock *waiter)
777{
778 return __locks_delete_block(&waiter->c);
779}
780EXPORT_SYMBOL(locks_delete_block);
781
782/* Insert waiter into blocker's block list.
783 * We use a circular list so that processes can be easily woken up in
784 * the order they blocked. The documentation doesn't require this but
785 * it seems like the reasonable thing to do.
786 *
787 * Must be called with both the flc_lock and blocked_lock_lock held. The
788 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
789 * but by ensuring that the flc_lock is also held on insertions we can avoid
790 * taking the blocked_lock_lock in some cases when we see that the
791 * fl_blocked_requests list is empty.
792 *
793 * Rather than just adding to the list, we check for conflicts with any existing
794 * waiters, and add beneath any waiter that blocks the new waiter.
795 * Thus wakeups don't happen until needed.
796 */
797static void __locks_insert_block(struct file_lock_core *blocker,
798 struct file_lock_core *waiter,
799 bool conflict(struct file_lock_core *,
800 struct file_lock_core *))
801{
802 struct file_lock_core *flc;
803
804 BUG_ON(!list_empty(&waiter->flc_blocked_member));
805new_blocker:
806 list_for_each_entry(flc, &blocker->flc_blocked_requests, flc_blocked_member)
807 if (conflict(flc, waiter)) {
808 blocker = flc;
809 goto new_blocker;
810 }
811 waiter->flc_blocker = blocker;
812 list_add_tail(&waiter->flc_blocked_member,
813 &blocker->flc_blocked_requests);
814
815 if ((blocker->flc_flags & (FL_POSIX|FL_OFDLCK)) == FL_POSIX)
816 locks_insert_global_blocked(waiter);
817
818 /* The requests in waiter->flc_blocked are known to conflict with
819 * waiter, but might not conflict with blocker, or the requests
820 * and lock which block it. So they all need to be woken.
821 */
822 __locks_wake_up_blocks(waiter);
823}
824
825/* Must be called with flc_lock held. */
826static void locks_insert_block(struct file_lock_core *blocker,
827 struct file_lock_core *waiter,
828 bool conflict(struct file_lock_core *,
829 struct file_lock_core *))
830{
831 spin_lock(&blocked_lock_lock);
832 __locks_insert_block(blocker, waiter, conflict);
833 spin_unlock(&blocked_lock_lock);
834}
835
836/*
837 * Wake up processes blocked waiting for blocker.
838 *
839 * Must be called with the inode->flc_lock held!
840 */
841static void locks_wake_up_blocks(struct file_lock_core *blocker)
842{
843 /*
844 * Avoid taking global lock if list is empty. This is safe since new
845 * blocked requests are only added to the list under the flc_lock, and
846 * the flc_lock is always held here. Note that removal from the
847 * fl_blocked_requests list does not require the flc_lock, so we must
848 * recheck list_empty() after acquiring the blocked_lock_lock.
849 */
850 if (list_empty(&blocker->flc_blocked_requests))
851 return;
852
853 spin_lock(&blocked_lock_lock);
854 __locks_wake_up_blocks(blocker);
855 spin_unlock(&blocked_lock_lock);
856}
857
858static void
859locks_insert_lock_ctx(struct file_lock_core *fl, struct list_head *before)
860{
861 list_add_tail(&fl->flc_list, before);
862 locks_insert_global_locks(fl);
863}
864
865static void
866locks_unlink_lock_ctx(struct file_lock_core *fl)
867{
868 locks_delete_global_locks(fl);
869 list_del_init(&fl->flc_list);
870 locks_wake_up_blocks(fl);
871}
872
873static void
874locks_delete_lock_ctx(struct file_lock_core *fl, struct list_head *dispose)
875{
876 locks_unlink_lock_ctx(fl);
877 if (dispose)
878 list_add(&fl->flc_list, dispose);
879 else
880 locks_free_lock(file_lock(fl));
881}
882
883/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
884 * checks for shared/exclusive status of overlapping locks.
885 */
886static bool locks_conflict(struct file_lock_core *caller_flc,
887 struct file_lock_core *sys_flc)
888{
889 if (sys_flc->flc_type == F_WRLCK)
890 return true;
891 if (caller_flc->flc_type == F_WRLCK)
892 return true;
893 return false;
894}
895
896/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
897 * checking before calling the locks_conflict().
898 */
899static bool posix_locks_conflict(struct file_lock_core *caller_flc,
900 struct file_lock_core *sys_flc)
901{
902 struct file_lock *caller_fl = file_lock(caller_flc);
903 struct file_lock *sys_fl = file_lock(sys_flc);
904
905 /* POSIX locks owned by the same process do not conflict with
906 * each other.
907 */
908 if (posix_same_owner(caller_flc, sys_flc))
909 return false;
910
911 /* Check whether they overlap */
912 if (!locks_overlap(caller_fl, sys_fl))
913 return false;
914
915 return locks_conflict(caller_flc, sys_flc);
916}
917
918/* Determine if lock sys_fl blocks lock caller_fl. Used on xx_GETLK
919 * path so checks for additional GETLK-specific things like F_UNLCK.
920 */
921static bool posix_test_locks_conflict(struct file_lock *caller_fl,
922 struct file_lock *sys_fl)
923{
924 struct file_lock_core *caller = &caller_fl->c;
925 struct file_lock_core *sys = &sys_fl->c;
926
927 /* F_UNLCK checks any locks on the same fd. */
928 if (lock_is_unlock(caller_fl)) {
929 if (!posix_same_owner(caller, sys))
930 return false;
931 return locks_overlap(caller_fl, sys_fl);
932 }
933 return posix_locks_conflict(caller, sys);
934}
935
936/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
937 * checking before calling the locks_conflict().
938 */
939static bool flock_locks_conflict(struct file_lock_core *caller_flc,
940 struct file_lock_core *sys_flc)
941{
942 /* FLOCK locks referring to the same filp do not conflict with
943 * each other.
944 */
945 if (caller_flc->flc_file == sys_flc->flc_file)
946 return false;
947
948 return locks_conflict(caller_flc, sys_flc);
949}
950
951void
952posix_test_lock(struct file *filp, struct file_lock *fl)
953{
954 struct file_lock *cfl;
955 struct file_lock_context *ctx;
956 struct inode *inode = file_inode(filp);
957 void *owner;
958 void (*func)(void);
959
960 ctx = locks_inode_context(inode);
961 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
962 fl->c.flc_type = F_UNLCK;
963 return;
964 }
965
966retry:
967 spin_lock(&ctx->flc_lock);
968 list_for_each_entry(cfl, &ctx->flc_posix, c.flc_list) {
969 if (!posix_test_locks_conflict(fl, cfl))
970 continue;
971 if (cfl->fl_lmops && cfl->fl_lmops->lm_lock_expirable
972 && (*cfl->fl_lmops->lm_lock_expirable)(cfl)) {
973 owner = cfl->fl_lmops->lm_mod_owner;
974 func = cfl->fl_lmops->lm_expire_lock;
975 __module_get(owner);
976 spin_unlock(&ctx->flc_lock);
977 (*func)();
978 module_put(owner);
979 goto retry;
980 }
981 locks_copy_conflock(fl, cfl);
982 goto out;
983 }
984 fl->c.flc_type = F_UNLCK;
985out:
986 spin_unlock(&ctx->flc_lock);
987 return;
988}
989EXPORT_SYMBOL(posix_test_lock);
990
991/*
992 * Deadlock detection:
993 *
994 * We attempt to detect deadlocks that are due purely to posix file
995 * locks.
996 *
997 * We assume that a task can be waiting for at most one lock at a time.
998 * So for any acquired lock, the process holding that lock may be
999 * waiting on at most one other lock. That lock in turns may be held by
1000 * someone waiting for at most one other lock. Given a requested lock
1001 * caller_fl which is about to wait for a conflicting lock block_fl, we
1002 * follow this chain of waiters to ensure we are not about to create a
1003 * cycle.
1004 *
1005 * Since we do this before we ever put a process to sleep on a lock, we
1006 * are ensured that there is never a cycle; that is what guarantees that
1007 * the while() loop in posix_locks_deadlock() eventually completes.
1008 *
1009 * Note: the above assumption may not be true when handling lock
1010 * requests from a broken NFS client. It may also fail in the presence
1011 * of tasks (such as posix threads) sharing the same open file table.
1012 * To handle those cases, we just bail out after a few iterations.
1013 *
1014 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
1015 * Because the owner is not even nominally tied to a thread of
1016 * execution, the deadlock detection below can't reasonably work well. Just
1017 * skip it for those.
1018 *
1019 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
1020 * locks that just checks for the case where two tasks are attempting to
1021 * upgrade from read to write locks on the same inode.
1022 */
1023
1024#define MAX_DEADLK_ITERATIONS 10
1025
1026/* Find a lock that the owner of the given @blocker is blocking on. */
1027static struct file_lock_core *what_owner_is_waiting_for(struct file_lock_core *blocker)
1028{
1029 struct file_lock_core *flc;
1030
1031 hash_for_each_possible(blocked_hash, flc, flc_link, posix_owner_key(blocker)) {
1032 if (posix_same_owner(flc, blocker)) {
1033 while (flc->flc_blocker)
1034 flc = flc->flc_blocker;
1035 return flc;
1036 }
1037 }
1038 return NULL;
1039}
1040
1041/* Must be called with the blocked_lock_lock held! */
1042static bool posix_locks_deadlock(struct file_lock *caller_fl,
1043 struct file_lock *block_fl)
1044{
1045 struct file_lock_core *caller = &caller_fl->c;
1046 struct file_lock_core *blocker = &block_fl->c;
1047 int i = 0;
1048
1049 lockdep_assert_held(&blocked_lock_lock);
1050
1051 /*
1052 * This deadlock detector can't reasonably detect deadlocks with
1053 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1054 */
1055 if (caller->flc_flags & FL_OFDLCK)
1056 return false;
1057
1058 while ((blocker = what_owner_is_waiting_for(blocker))) {
1059 if (i++ > MAX_DEADLK_ITERATIONS)
1060 return false;
1061 if (posix_same_owner(caller, blocker))
1062 return true;
1063 }
1064 return false;
1065}
1066
1067/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1068 * after any leases, but before any posix locks.
1069 *
1070 * Note that if called with an FL_EXISTS argument, the caller may determine
1071 * whether or not a lock was successfully freed by testing the return
1072 * value for -ENOENT.
1073 */
1074static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1075{
1076 struct file_lock *new_fl = NULL;
1077 struct file_lock *fl;
1078 struct file_lock_context *ctx;
1079 int error = 0;
1080 bool found = false;
1081 LIST_HEAD(dispose);
1082
1083 ctx = locks_get_lock_context(inode, request->c.flc_type);
1084 if (!ctx) {
1085 if (request->c.flc_type != F_UNLCK)
1086 return -ENOMEM;
1087 return (request->c.flc_flags & FL_EXISTS) ? -ENOENT : 0;
1088 }
1089
1090 if (!(request->c.flc_flags & FL_ACCESS) && (request->c.flc_type != F_UNLCK)) {
1091 new_fl = locks_alloc_lock();
1092 if (!new_fl)
1093 return -ENOMEM;
1094 }
1095
1096 percpu_down_read(&file_rwsem);
1097 spin_lock(&ctx->flc_lock);
1098 if (request->c.flc_flags & FL_ACCESS)
1099 goto find_conflict;
1100
1101 list_for_each_entry(fl, &ctx->flc_flock, c.flc_list) {
1102 if (request->c.flc_file != fl->c.flc_file)
1103 continue;
1104 if (request->c.flc_type == fl->c.flc_type)
1105 goto out;
1106 found = true;
1107 locks_delete_lock_ctx(&fl->c, &dispose);
1108 break;
1109 }
1110
1111 if (lock_is_unlock(request)) {
1112 if ((request->c.flc_flags & FL_EXISTS) && !found)
1113 error = -ENOENT;
1114 goto out;
1115 }
1116
1117find_conflict:
1118 list_for_each_entry(fl, &ctx->flc_flock, c.flc_list) {
1119 if (!flock_locks_conflict(&request->c, &fl->c))
1120 continue;
1121 error = -EAGAIN;
1122 if (!(request->c.flc_flags & FL_SLEEP))
1123 goto out;
1124 error = FILE_LOCK_DEFERRED;
1125 locks_insert_block(&fl->c, &request->c, flock_locks_conflict);
1126 goto out;
1127 }
1128 if (request->c.flc_flags & FL_ACCESS)
1129 goto out;
1130 locks_copy_lock(new_fl, request);
1131 locks_move_blocks(new_fl, request);
1132 locks_insert_lock_ctx(&new_fl->c, &ctx->flc_flock);
1133 new_fl = NULL;
1134 error = 0;
1135
1136out:
1137 spin_unlock(&ctx->flc_lock);
1138 percpu_up_read(&file_rwsem);
1139 if (new_fl)
1140 locks_free_lock(new_fl);
1141 locks_dispose_list(&dispose);
1142 trace_flock_lock_inode(inode, request, error);
1143 return error;
1144}
1145
1146static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1147 struct file_lock *conflock)
1148{
1149 struct file_lock *fl, *tmp;
1150 struct file_lock *new_fl = NULL;
1151 struct file_lock *new_fl2 = NULL;
1152 struct file_lock *left = NULL;
1153 struct file_lock *right = NULL;
1154 struct file_lock_context *ctx;
1155 int error;
1156 bool added = false;
1157 LIST_HEAD(dispose);
1158 void *owner;
1159 void (*func)(void);
1160
1161 ctx = locks_get_lock_context(inode, request->c.flc_type);
1162 if (!ctx)
1163 return lock_is_unlock(request) ? 0 : -ENOMEM;
1164
1165 /*
1166 * We may need two file_lock structures for this operation,
1167 * so we get them in advance to avoid races.
1168 *
1169 * In some cases we can be sure, that no new locks will be needed
1170 */
1171 if (!(request->c.flc_flags & FL_ACCESS) &&
1172 (request->c.flc_type != F_UNLCK ||
1173 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1174 new_fl = locks_alloc_lock();
1175 new_fl2 = locks_alloc_lock();
1176 }
1177
1178retry:
1179 percpu_down_read(&file_rwsem);
1180 spin_lock(&ctx->flc_lock);
1181 /*
1182 * New lock request. Walk all POSIX locks and look for conflicts. If
1183 * there are any, either return error or put the request on the
1184 * blocker's list of waiters and the global blocked_hash.
1185 */
1186 if (request->c.flc_type != F_UNLCK) {
1187 list_for_each_entry(fl, &ctx->flc_posix, c.flc_list) {
1188 if (!posix_locks_conflict(&request->c, &fl->c))
1189 continue;
1190 if (fl->fl_lmops && fl->fl_lmops->lm_lock_expirable
1191 && (*fl->fl_lmops->lm_lock_expirable)(fl)) {
1192 owner = fl->fl_lmops->lm_mod_owner;
1193 func = fl->fl_lmops->lm_expire_lock;
1194 __module_get(owner);
1195 spin_unlock(&ctx->flc_lock);
1196 percpu_up_read(&file_rwsem);
1197 (*func)();
1198 module_put(owner);
1199 goto retry;
1200 }
1201 if (conflock)
1202 locks_copy_conflock(conflock, fl);
1203 error = -EAGAIN;
1204 if (!(request->c.flc_flags & FL_SLEEP))
1205 goto out;
1206 /*
1207 * Deadlock detection and insertion into the blocked
1208 * locks list must be done while holding the same lock!
1209 */
1210 error = -EDEADLK;
1211 spin_lock(&blocked_lock_lock);
1212 /*
1213 * Ensure that we don't find any locks blocked on this
1214 * request during deadlock detection.
1215 */
1216 __locks_wake_up_blocks(&request->c);
1217 if (likely(!posix_locks_deadlock(request, fl))) {
1218 error = FILE_LOCK_DEFERRED;
1219 __locks_insert_block(&fl->c, &request->c,
1220 posix_locks_conflict);
1221 }
1222 spin_unlock(&blocked_lock_lock);
1223 goto out;
1224 }
1225 }
1226
1227 /* If we're just looking for a conflict, we're done. */
1228 error = 0;
1229 if (request->c.flc_flags & FL_ACCESS)
1230 goto out;
1231
1232 /* Find the first old lock with the same owner as the new lock */
1233 list_for_each_entry(fl, &ctx->flc_posix, c.flc_list) {
1234 if (posix_same_owner(&request->c, &fl->c))
1235 break;
1236 }
1237
1238 /* Process locks with this owner. */
1239 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, c.flc_list) {
1240 if (!posix_same_owner(&request->c, &fl->c))
1241 break;
1242
1243 /* Detect adjacent or overlapping regions (if same lock type) */
1244 if (request->c.flc_type == fl->c.flc_type) {
1245 /* In all comparisons of start vs end, use
1246 * "start - 1" rather than "end + 1". If end
1247 * is OFFSET_MAX, end + 1 will become negative.
1248 */
1249 if (fl->fl_end < request->fl_start - 1)
1250 continue;
1251 /* If the next lock in the list has entirely bigger
1252 * addresses than the new one, insert the lock here.
1253 */
1254 if (fl->fl_start - 1 > request->fl_end)
1255 break;
1256
1257 /* If we come here, the new and old lock are of the
1258 * same type and adjacent or overlapping. Make one
1259 * lock yielding from the lower start address of both
1260 * locks to the higher end address.
1261 */
1262 if (fl->fl_start > request->fl_start)
1263 fl->fl_start = request->fl_start;
1264 else
1265 request->fl_start = fl->fl_start;
1266 if (fl->fl_end < request->fl_end)
1267 fl->fl_end = request->fl_end;
1268 else
1269 request->fl_end = fl->fl_end;
1270 if (added) {
1271 locks_delete_lock_ctx(&fl->c, &dispose);
1272 continue;
1273 }
1274 request = fl;
1275 added = true;
1276 } else {
1277 /* Processing for different lock types is a bit
1278 * more complex.
1279 */
1280 if (fl->fl_end < request->fl_start)
1281 continue;
1282 if (fl->fl_start > request->fl_end)
1283 break;
1284 if (lock_is_unlock(request))
1285 added = true;
1286 if (fl->fl_start < request->fl_start)
1287 left = fl;
1288 /* If the next lock in the list has a higher end
1289 * address than the new one, insert the new one here.
1290 */
1291 if (fl->fl_end > request->fl_end) {
1292 right = fl;
1293 break;
1294 }
1295 if (fl->fl_start >= request->fl_start) {
1296 /* The new lock completely replaces an old
1297 * one (This may happen several times).
1298 */
1299 if (added) {
1300 locks_delete_lock_ctx(&fl->c, &dispose);
1301 continue;
1302 }
1303 /*
1304 * Replace the old lock with new_fl, and
1305 * remove the old one. It's safe to do the
1306 * insert here since we know that we won't be
1307 * using new_fl later, and that the lock is
1308 * just replacing an existing lock.
1309 */
1310 error = -ENOLCK;
1311 if (!new_fl)
1312 goto out;
1313 locks_copy_lock(new_fl, request);
1314 locks_move_blocks(new_fl, request);
1315 request = new_fl;
1316 new_fl = NULL;
1317 locks_insert_lock_ctx(&request->c,
1318 &fl->c.flc_list);
1319 locks_delete_lock_ctx(&fl->c, &dispose);
1320 added = true;
1321 }
1322 }
1323 }
1324
1325 /*
1326 * The above code only modifies existing locks in case of merging or
1327 * replacing. If new lock(s) need to be inserted all modifications are
1328 * done below this, so it's safe yet to bail out.
1329 */
1330 error = -ENOLCK; /* "no luck" */
1331 if (right && left == right && !new_fl2)
1332 goto out;
1333
1334 error = 0;
1335 if (!added) {
1336 if (lock_is_unlock(request)) {
1337 if (request->c.flc_flags & FL_EXISTS)
1338 error = -ENOENT;
1339 goto out;
1340 }
1341
1342 if (!new_fl) {
1343 error = -ENOLCK;
1344 goto out;
1345 }
1346 locks_copy_lock(new_fl, request);
1347 locks_move_blocks(new_fl, request);
1348 locks_insert_lock_ctx(&new_fl->c, &fl->c.flc_list);
1349 fl = new_fl;
1350 new_fl = NULL;
1351 }
1352 if (right) {
1353 if (left == right) {
1354 /* The new lock breaks the old one in two pieces,
1355 * so we have to use the second new lock.
1356 */
1357 left = new_fl2;
1358 new_fl2 = NULL;
1359 locks_copy_lock(left, right);
1360 locks_insert_lock_ctx(&left->c, &fl->c.flc_list);
1361 }
1362 right->fl_start = request->fl_end + 1;
1363 locks_wake_up_blocks(&right->c);
1364 }
1365 if (left) {
1366 left->fl_end = request->fl_start - 1;
1367 locks_wake_up_blocks(&left->c);
1368 }
1369 out:
1370 spin_unlock(&ctx->flc_lock);
1371 percpu_up_read(&file_rwsem);
1372 trace_posix_lock_inode(inode, request, error);
1373 /*
1374 * Free any unused locks.
1375 */
1376 if (new_fl)
1377 locks_free_lock(new_fl);
1378 if (new_fl2)
1379 locks_free_lock(new_fl2);
1380 locks_dispose_list(&dispose);
1381
1382 return error;
1383}
1384
1385/**
1386 * posix_lock_file - Apply a POSIX-style lock to a file
1387 * @filp: The file to apply the lock to
1388 * @fl: The lock to be applied
1389 * @conflock: Place to return a copy of the conflicting lock, if found.
1390 *
1391 * Add a POSIX style lock to a file.
1392 * We merge adjacent & overlapping locks whenever possible.
1393 * POSIX locks are sorted by owner task, then by starting address
1394 *
1395 * Note that if called with an FL_EXISTS argument, the caller may determine
1396 * whether or not a lock was successfully freed by testing the return
1397 * value for -ENOENT.
1398 */
1399int posix_lock_file(struct file *filp, struct file_lock *fl,
1400 struct file_lock *conflock)
1401{
1402 return posix_lock_inode(file_inode(filp), fl, conflock);
1403}
1404EXPORT_SYMBOL(posix_lock_file);
1405
1406/**
1407 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1408 * @inode: inode of file to which lock request should be applied
1409 * @fl: The lock to be applied
1410 *
1411 * Apply a POSIX style lock request to an inode.
1412 */
1413static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1414{
1415 int error;
1416 might_sleep ();
1417 for (;;) {
1418 error = posix_lock_inode(inode, fl, NULL);
1419 if (error != FILE_LOCK_DEFERRED)
1420 break;
1421 error = wait_event_interruptible(fl->c.flc_wait,
1422 list_empty(&fl->c.flc_blocked_member));
1423 if (error)
1424 break;
1425 }
1426 locks_delete_block(fl);
1427 return error;
1428}
1429
1430static void lease_clear_pending(struct file_lease *fl, int arg)
1431{
1432 switch (arg) {
1433 case F_UNLCK:
1434 fl->c.flc_flags &= ~FL_UNLOCK_PENDING;
1435 fallthrough;
1436 case F_RDLCK:
1437 fl->c.flc_flags &= ~FL_DOWNGRADE_PENDING;
1438 }
1439}
1440
1441/* We already had a lease on this file; just change its type */
1442int lease_modify(struct file_lease *fl, int arg, struct list_head *dispose)
1443{
1444 int error = assign_type(&fl->c, arg);
1445
1446 if (error)
1447 return error;
1448 lease_clear_pending(fl, arg);
1449 locks_wake_up_blocks(&fl->c);
1450 if (arg == F_UNLCK) {
1451 struct file *filp = fl->c.flc_file;
1452
1453 f_delown(filp);
1454 filp->f_owner.signum = 0;
1455 fasync_helper(0, fl->c.flc_file, 0, &fl->fl_fasync);
1456 if (fl->fl_fasync != NULL) {
1457 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1458 fl->fl_fasync = NULL;
1459 }
1460 locks_delete_lock_ctx(&fl->c, dispose);
1461 }
1462 return 0;
1463}
1464EXPORT_SYMBOL(lease_modify);
1465
1466static bool past_time(unsigned long then)
1467{
1468 if (!then)
1469 /* 0 is a special value meaning "this never expires": */
1470 return false;
1471 return time_after(jiffies, then);
1472}
1473
1474static void time_out_leases(struct inode *inode, struct list_head *dispose)
1475{
1476 struct file_lock_context *ctx = inode->i_flctx;
1477 struct file_lease *fl, *tmp;
1478
1479 lockdep_assert_held(&ctx->flc_lock);
1480
1481 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, c.flc_list) {
1482 trace_time_out_leases(inode, fl);
1483 if (past_time(fl->fl_downgrade_time))
1484 lease_modify(fl, F_RDLCK, dispose);
1485 if (past_time(fl->fl_break_time))
1486 lease_modify(fl, F_UNLCK, dispose);
1487 }
1488}
1489
1490static bool leases_conflict(struct file_lock_core *lc, struct file_lock_core *bc)
1491{
1492 bool rc;
1493 struct file_lease *lease = file_lease(lc);
1494 struct file_lease *breaker = file_lease(bc);
1495
1496 if (lease->fl_lmops->lm_breaker_owns_lease
1497 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1498 return false;
1499 if ((bc->flc_flags & FL_LAYOUT) != (lc->flc_flags & FL_LAYOUT)) {
1500 rc = false;
1501 goto trace;
1502 }
1503 if ((bc->flc_flags & FL_DELEG) && (lc->flc_flags & FL_LEASE)) {
1504 rc = false;
1505 goto trace;
1506 }
1507
1508 rc = locks_conflict(bc, lc);
1509trace:
1510 trace_leases_conflict(rc, lease, breaker);
1511 return rc;
1512}
1513
1514static bool
1515any_leases_conflict(struct inode *inode, struct file_lease *breaker)
1516{
1517 struct file_lock_context *ctx = inode->i_flctx;
1518 struct file_lock_core *flc;
1519
1520 lockdep_assert_held(&ctx->flc_lock);
1521
1522 list_for_each_entry(flc, &ctx->flc_lease, flc_list) {
1523 if (leases_conflict(flc, &breaker->c))
1524 return true;
1525 }
1526 return false;
1527}
1528
1529/**
1530 * __break_lease - revoke all outstanding leases on file
1531 * @inode: the inode of the file to return
1532 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1533 * break all leases
1534 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1535 * only delegations
1536 *
1537 * break_lease (inlined for speed) has checked there already is at least
1538 * some kind of lock (maybe a lease) on this file. Leases are broken on
1539 * a call to open() or truncate(). This function can sleep unless you
1540 * specified %O_NONBLOCK to your open().
1541 */
1542int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1543{
1544 int error = 0;
1545 struct file_lock_context *ctx;
1546 struct file_lease *new_fl, *fl, *tmp;
1547 unsigned long break_time;
1548 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1549 LIST_HEAD(dispose);
1550
1551 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1552 if (IS_ERR(new_fl))
1553 return PTR_ERR(new_fl);
1554 new_fl->c.flc_flags = type;
1555
1556 /* typically we will check that ctx is non-NULL before calling */
1557 ctx = locks_inode_context(inode);
1558 if (!ctx) {
1559 WARN_ON_ONCE(1);
1560 goto free_lock;
1561 }
1562
1563 percpu_down_read(&file_rwsem);
1564 spin_lock(&ctx->flc_lock);
1565
1566 time_out_leases(inode, &dispose);
1567
1568 if (!any_leases_conflict(inode, new_fl))
1569 goto out;
1570
1571 break_time = 0;
1572 if (lease_break_time > 0) {
1573 break_time = jiffies + lease_break_time * HZ;
1574 if (break_time == 0)
1575 break_time++; /* so that 0 means no break time */
1576 }
1577
1578 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, c.flc_list) {
1579 if (!leases_conflict(&fl->c, &new_fl->c))
1580 continue;
1581 if (want_write) {
1582 if (fl->c.flc_flags & FL_UNLOCK_PENDING)
1583 continue;
1584 fl->c.flc_flags |= FL_UNLOCK_PENDING;
1585 fl->fl_break_time = break_time;
1586 } else {
1587 if (lease_breaking(fl))
1588 continue;
1589 fl->c.flc_flags |= FL_DOWNGRADE_PENDING;
1590 fl->fl_downgrade_time = break_time;
1591 }
1592 if (fl->fl_lmops->lm_break(fl))
1593 locks_delete_lock_ctx(&fl->c, &dispose);
1594 }
1595
1596 if (list_empty(&ctx->flc_lease))
1597 goto out;
1598
1599 if (mode & O_NONBLOCK) {
1600 trace_break_lease_noblock(inode, new_fl);
1601 error = -EWOULDBLOCK;
1602 goto out;
1603 }
1604
1605restart:
1606 fl = list_first_entry(&ctx->flc_lease, struct file_lease, c.flc_list);
1607 break_time = fl->fl_break_time;
1608 if (break_time != 0)
1609 break_time -= jiffies;
1610 if (break_time == 0)
1611 break_time++;
1612 locks_insert_block(&fl->c, &new_fl->c, leases_conflict);
1613 trace_break_lease_block(inode, new_fl);
1614 spin_unlock(&ctx->flc_lock);
1615 percpu_up_read(&file_rwsem);
1616
1617 locks_dispose_list(&dispose);
1618 error = wait_event_interruptible_timeout(new_fl->c.flc_wait,
1619 list_empty(&new_fl->c.flc_blocked_member),
1620 break_time);
1621
1622 percpu_down_read(&file_rwsem);
1623 spin_lock(&ctx->flc_lock);
1624 trace_break_lease_unblock(inode, new_fl);
1625 __locks_delete_block(&new_fl->c);
1626 if (error >= 0) {
1627 /*
1628 * Wait for the next conflicting lease that has not been
1629 * broken yet
1630 */
1631 if (error == 0)
1632 time_out_leases(inode, &dispose);
1633 if (any_leases_conflict(inode, new_fl))
1634 goto restart;
1635 error = 0;
1636 }
1637out:
1638 spin_unlock(&ctx->flc_lock);
1639 percpu_up_read(&file_rwsem);
1640 locks_dispose_list(&dispose);
1641free_lock:
1642 locks_free_lease(new_fl);
1643 return error;
1644}
1645EXPORT_SYMBOL(__break_lease);
1646
1647/**
1648 * lease_get_mtime - update modified time of an inode with exclusive lease
1649 * @inode: the inode
1650 * @time: pointer to a timespec which contains the last modified time
1651 *
1652 * This is to force NFS clients to flush their caches for files with
1653 * exclusive leases. The justification is that if someone has an
1654 * exclusive lease, then they could be modifying it.
1655 */
1656void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1657{
1658 bool has_lease = false;
1659 struct file_lock_context *ctx;
1660 struct file_lock_core *flc;
1661
1662 ctx = locks_inode_context(inode);
1663 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1664 spin_lock(&ctx->flc_lock);
1665 flc = list_first_entry_or_null(&ctx->flc_lease,
1666 struct file_lock_core, flc_list);
1667 if (flc && flc->flc_type == F_WRLCK)
1668 has_lease = true;
1669 spin_unlock(&ctx->flc_lock);
1670 }
1671
1672 if (has_lease)
1673 *time = current_time(inode);
1674}
1675EXPORT_SYMBOL(lease_get_mtime);
1676
1677/**
1678 * fcntl_getlease - Enquire what lease is currently active
1679 * @filp: the file
1680 *
1681 * The value returned by this function will be one of
1682 * (if no lease break is pending):
1683 *
1684 * %F_RDLCK to indicate a shared lease is held.
1685 *
1686 * %F_WRLCK to indicate an exclusive lease is held.
1687 *
1688 * %F_UNLCK to indicate no lease is held.
1689 *
1690 * (if a lease break is pending):
1691 *
1692 * %F_RDLCK to indicate an exclusive lease needs to be
1693 * changed to a shared lease (or removed).
1694 *
1695 * %F_UNLCK to indicate the lease needs to be removed.
1696 *
1697 * XXX: sfr & willy disagree over whether F_INPROGRESS
1698 * should be returned to userspace.
1699 */
1700int fcntl_getlease(struct file *filp)
1701{
1702 struct file_lease *fl;
1703 struct inode *inode = file_inode(filp);
1704 struct file_lock_context *ctx;
1705 int type = F_UNLCK;
1706 LIST_HEAD(dispose);
1707
1708 ctx = locks_inode_context(inode);
1709 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1710 percpu_down_read(&file_rwsem);
1711 spin_lock(&ctx->flc_lock);
1712 time_out_leases(inode, &dispose);
1713 list_for_each_entry(fl, &ctx->flc_lease, c.flc_list) {
1714 if (fl->c.flc_file != filp)
1715 continue;
1716 type = target_leasetype(fl);
1717 break;
1718 }
1719 spin_unlock(&ctx->flc_lock);
1720 percpu_up_read(&file_rwsem);
1721
1722 locks_dispose_list(&dispose);
1723 }
1724 return type;
1725}
1726
1727/**
1728 * check_conflicting_open - see if the given file points to an inode that has
1729 * an existing open that would conflict with the
1730 * desired lease.
1731 * @filp: file to check
1732 * @arg: type of lease that we're trying to acquire
1733 * @flags: current lock flags
1734 *
1735 * Check to see if there's an existing open fd on this file that would
1736 * conflict with the lease we're trying to set.
1737 */
1738static int
1739check_conflicting_open(struct file *filp, const int arg, int flags)
1740{
1741 struct inode *inode = file_inode(filp);
1742 int self_wcount = 0, self_rcount = 0;
1743
1744 if (flags & FL_LAYOUT)
1745 return 0;
1746 if (flags & FL_DELEG)
1747 /* We leave these checks to the caller */
1748 return 0;
1749
1750 if (arg == F_RDLCK)
1751 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1752 else if (arg != F_WRLCK)
1753 return 0;
1754
1755 /*
1756 * Make sure that only read/write count is from lease requestor.
1757 * Note that this will result in denying write leases when i_writecount
1758 * is negative, which is what we want. (We shouldn't grant write leases
1759 * on files open for execution.)
1760 */
1761 if (filp->f_mode & FMODE_WRITE)
1762 self_wcount = 1;
1763 else if (filp->f_mode & FMODE_READ)
1764 self_rcount = 1;
1765
1766 if (atomic_read(&inode->i_writecount) != self_wcount ||
1767 atomic_read(&inode->i_readcount) != self_rcount)
1768 return -EAGAIN;
1769
1770 return 0;
1771}
1772
1773static int
1774generic_add_lease(struct file *filp, int arg, struct file_lease **flp, void **priv)
1775{
1776 struct file_lease *fl, *my_fl = NULL, *lease;
1777 struct inode *inode = file_inode(filp);
1778 struct file_lock_context *ctx;
1779 bool is_deleg = (*flp)->c.flc_flags & FL_DELEG;
1780 int error;
1781 LIST_HEAD(dispose);
1782
1783 lease = *flp;
1784 trace_generic_add_lease(inode, lease);
1785
1786 /* Note that arg is never F_UNLCK here */
1787 ctx = locks_get_lock_context(inode, arg);
1788 if (!ctx)
1789 return -ENOMEM;
1790
1791 /*
1792 * In the delegation case we need mutual exclusion with
1793 * a number of operations that take the i_mutex. We trylock
1794 * because delegations are an optional optimization, and if
1795 * there's some chance of a conflict--we'd rather not
1796 * bother, maybe that's a sign this just isn't a good file to
1797 * hand out a delegation on.
1798 */
1799 if (is_deleg && !inode_trylock(inode))
1800 return -EAGAIN;
1801
1802 percpu_down_read(&file_rwsem);
1803 spin_lock(&ctx->flc_lock);
1804 time_out_leases(inode, &dispose);
1805 error = check_conflicting_open(filp, arg, lease->c.flc_flags);
1806 if (error)
1807 goto out;
1808
1809 /*
1810 * At this point, we know that if there is an exclusive
1811 * lease on this file, then we hold it on this filp
1812 * (otherwise our open of this file would have blocked).
1813 * And if we are trying to acquire an exclusive lease,
1814 * then the file is not open by anyone (including us)
1815 * except for this filp.
1816 */
1817 error = -EAGAIN;
1818 list_for_each_entry(fl, &ctx->flc_lease, c.flc_list) {
1819 if (fl->c.flc_file == filp &&
1820 fl->c.flc_owner == lease->c.flc_owner) {
1821 my_fl = fl;
1822 continue;
1823 }
1824
1825 /*
1826 * No exclusive leases if someone else has a lease on
1827 * this file:
1828 */
1829 if (arg == F_WRLCK)
1830 goto out;
1831 /*
1832 * Modifying our existing lease is OK, but no getting a
1833 * new lease if someone else is opening for write:
1834 */
1835 if (fl->c.flc_flags & FL_UNLOCK_PENDING)
1836 goto out;
1837 }
1838
1839 if (my_fl != NULL) {
1840 lease = my_fl;
1841 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1842 if (error)
1843 goto out;
1844 goto out_setup;
1845 }
1846
1847 error = -EINVAL;
1848 if (!leases_enable)
1849 goto out;
1850
1851 locks_insert_lock_ctx(&lease->c, &ctx->flc_lease);
1852 /*
1853 * The check in break_lease() is lockless. It's possible for another
1854 * open to race in after we did the earlier check for a conflicting
1855 * open but before the lease was inserted. Check again for a
1856 * conflicting open and cancel the lease if there is one.
1857 *
1858 * We also add a barrier here to ensure that the insertion of the lock
1859 * precedes these checks.
1860 */
1861 smp_mb();
1862 error = check_conflicting_open(filp, arg, lease->c.flc_flags);
1863 if (error) {
1864 locks_unlink_lock_ctx(&lease->c);
1865 goto out;
1866 }
1867
1868out_setup:
1869 if (lease->fl_lmops->lm_setup)
1870 lease->fl_lmops->lm_setup(lease, priv);
1871out:
1872 spin_unlock(&ctx->flc_lock);
1873 percpu_up_read(&file_rwsem);
1874 locks_dispose_list(&dispose);
1875 if (is_deleg)
1876 inode_unlock(inode);
1877 if (!error && !my_fl)
1878 *flp = NULL;
1879 return error;
1880}
1881
1882static int generic_delete_lease(struct file *filp, void *owner)
1883{
1884 int error = -EAGAIN;
1885 struct file_lease *fl, *victim = NULL;
1886 struct inode *inode = file_inode(filp);
1887 struct file_lock_context *ctx;
1888 LIST_HEAD(dispose);
1889
1890 ctx = locks_inode_context(inode);
1891 if (!ctx) {
1892 trace_generic_delete_lease(inode, NULL);
1893 return error;
1894 }
1895
1896 percpu_down_read(&file_rwsem);
1897 spin_lock(&ctx->flc_lock);
1898 list_for_each_entry(fl, &ctx->flc_lease, c.flc_list) {
1899 if (fl->c.flc_file == filp &&
1900 fl->c.flc_owner == owner) {
1901 victim = fl;
1902 break;
1903 }
1904 }
1905 trace_generic_delete_lease(inode, victim);
1906 if (victim)
1907 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1908 spin_unlock(&ctx->flc_lock);
1909 percpu_up_read(&file_rwsem);
1910 locks_dispose_list(&dispose);
1911 return error;
1912}
1913
1914/**
1915 * generic_setlease - sets a lease on an open file
1916 * @filp: file pointer
1917 * @arg: type of lease to obtain
1918 * @flp: input - file_lock to use, output - file_lock inserted
1919 * @priv: private data for lm_setup (may be NULL if lm_setup
1920 * doesn't require it)
1921 *
1922 * The (input) flp->fl_lmops->lm_break function is required
1923 * by break_lease().
1924 */
1925int generic_setlease(struct file *filp, int arg, struct file_lease **flp,
1926 void **priv)
1927{
1928 switch (arg) {
1929 case F_UNLCK:
1930 return generic_delete_lease(filp, *priv);
1931 case F_RDLCK:
1932 case F_WRLCK:
1933 if (!(*flp)->fl_lmops->lm_break) {
1934 WARN_ON_ONCE(1);
1935 return -ENOLCK;
1936 }
1937
1938 return generic_add_lease(filp, arg, flp, priv);
1939 default:
1940 return -EINVAL;
1941 }
1942}
1943EXPORT_SYMBOL(generic_setlease);
1944
1945/*
1946 * Kernel subsystems can register to be notified on any attempt to set
1947 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1948 * to close files that it may have cached when there is an attempt to set a
1949 * conflicting lease.
1950 */
1951static struct srcu_notifier_head lease_notifier_chain;
1952
1953static inline void
1954lease_notifier_chain_init(void)
1955{
1956 srcu_init_notifier_head(&lease_notifier_chain);
1957}
1958
1959static inline void
1960setlease_notifier(int arg, struct file_lease *lease)
1961{
1962 if (arg != F_UNLCK)
1963 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1964}
1965
1966int lease_register_notifier(struct notifier_block *nb)
1967{
1968 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1969}
1970EXPORT_SYMBOL_GPL(lease_register_notifier);
1971
1972void lease_unregister_notifier(struct notifier_block *nb)
1973{
1974 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1975}
1976EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1977
1978
1979int
1980kernel_setlease(struct file *filp, int arg, struct file_lease **lease, void **priv)
1981{
1982 if (lease)
1983 setlease_notifier(arg, *lease);
1984 if (filp->f_op->setlease)
1985 return filp->f_op->setlease(filp, arg, lease, priv);
1986 else
1987 return generic_setlease(filp, arg, lease, priv);
1988}
1989EXPORT_SYMBOL_GPL(kernel_setlease);
1990
1991/**
1992 * vfs_setlease - sets a lease on an open file
1993 * @filp: file pointer
1994 * @arg: type of lease to obtain
1995 * @lease: file_lock to use when adding a lease
1996 * @priv: private info for lm_setup when adding a lease (may be
1997 * NULL if lm_setup doesn't require it)
1998 *
1999 * Call this to establish a lease on the file. The "lease" argument is not
2000 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2001 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2002 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2003 * stack trace).
2004 *
2005 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2006 * may be NULL if the lm_setup operation doesn't require it.
2007 */
2008int
2009vfs_setlease(struct file *filp, int arg, struct file_lease **lease, void **priv)
2010{
2011 struct inode *inode = file_inode(filp);
2012 vfsuid_t vfsuid = i_uid_into_vfsuid(file_mnt_idmap(filp), inode);
2013 int error;
2014
2015 if ((!vfsuid_eq_kuid(vfsuid, current_fsuid())) && !capable(CAP_LEASE))
2016 return -EACCES;
2017 if (!S_ISREG(inode->i_mode))
2018 return -EINVAL;
2019 error = security_file_lock(filp, arg);
2020 if (error)
2021 return error;
2022 return kernel_setlease(filp, arg, lease, priv);
2023}
2024EXPORT_SYMBOL_GPL(vfs_setlease);
2025
2026static int do_fcntl_add_lease(unsigned int fd, struct file *filp, int arg)
2027{
2028 struct file_lease *fl;
2029 struct fasync_struct *new;
2030 int error;
2031
2032 fl = lease_alloc(filp, arg);
2033 if (IS_ERR(fl))
2034 return PTR_ERR(fl);
2035
2036 new = fasync_alloc();
2037 if (!new) {
2038 locks_free_lease(fl);
2039 return -ENOMEM;
2040 }
2041 new->fa_fd = fd;
2042
2043 error = vfs_setlease(filp, arg, &fl, (void **)&new);
2044 if (fl)
2045 locks_free_lease(fl);
2046 if (new)
2047 fasync_free(new);
2048 return error;
2049}
2050
2051/**
2052 * fcntl_setlease - sets a lease on an open file
2053 * @fd: open file descriptor
2054 * @filp: file pointer
2055 * @arg: type of lease to obtain
2056 *
2057 * Call this fcntl to establish a lease on the file.
2058 * Note that you also need to call %F_SETSIG to
2059 * receive a signal when the lease is broken.
2060 */
2061int fcntl_setlease(unsigned int fd, struct file *filp, int arg)
2062{
2063 if (arg == F_UNLCK)
2064 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2065 return do_fcntl_add_lease(fd, filp, arg);
2066}
2067
2068/**
2069 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2070 * @inode: inode of the file to apply to
2071 * @fl: The lock to be applied
2072 *
2073 * Apply a FLOCK style lock request to an inode.
2074 */
2075static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2076{
2077 int error;
2078 might_sleep();
2079 for (;;) {
2080 error = flock_lock_inode(inode, fl);
2081 if (error != FILE_LOCK_DEFERRED)
2082 break;
2083 error = wait_event_interruptible(fl->c.flc_wait,
2084 list_empty(&fl->c.flc_blocked_member));
2085 if (error)
2086 break;
2087 }
2088 locks_delete_block(fl);
2089 return error;
2090}
2091
2092/**
2093 * locks_lock_inode_wait - Apply a lock to an inode
2094 * @inode: inode of the file to apply to
2095 * @fl: The lock to be applied
2096 *
2097 * Apply a POSIX or FLOCK style lock request to an inode.
2098 */
2099int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2100{
2101 int res = 0;
2102 switch (fl->c.flc_flags & (FL_POSIX|FL_FLOCK)) {
2103 case FL_POSIX:
2104 res = posix_lock_inode_wait(inode, fl);
2105 break;
2106 case FL_FLOCK:
2107 res = flock_lock_inode_wait(inode, fl);
2108 break;
2109 default:
2110 BUG();
2111 }
2112 return res;
2113}
2114EXPORT_SYMBOL(locks_lock_inode_wait);
2115
2116/**
2117 * sys_flock: - flock() system call.
2118 * @fd: the file descriptor to lock.
2119 * @cmd: the type of lock to apply.
2120 *
2121 * Apply a %FL_FLOCK style lock to an open file descriptor.
2122 * The @cmd can be one of:
2123 *
2124 * - %LOCK_SH -- a shared lock.
2125 * - %LOCK_EX -- an exclusive lock.
2126 * - %LOCK_UN -- remove an existing lock.
2127 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2128 *
2129 * %LOCK_MAND support has been removed from the kernel.
2130 */
2131SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2132{
2133 int can_sleep, error, type;
2134 struct file_lock fl;
2135 struct fd f;
2136
2137 /*
2138 * LOCK_MAND locks were broken for a long time in that they never
2139 * conflicted with one another and didn't prevent any sort of open,
2140 * read or write activity.
2141 *
2142 * Just ignore these requests now, to preserve legacy behavior, but
2143 * throw a warning to let people know that they don't actually work.
2144 */
2145 if (cmd & LOCK_MAND) {
2146 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);
2147 return 0;
2148 }
2149
2150 type = flock_translate_cmd(cmd & ~LOCK_NB);
2151 if (type < 0)
2152 return type;
2153
2154 error = -EBADF;
2155 f = fdget(fd);
2156 if (!f.file)
2157 return error;
2158
2159 if (type != F_UNLCK && !(f.file->f_mode & (FMODE_READ | FMODE_WRITE)))
2160 goto out_putf;
2161
2162 flock_make_lock(f.file, &fl, type);
2163
2164 error = security_file_lock(f.file, fl.c.flc_type);
2165 if (error)
2166 goto out_putf;
2167
2168 can_sleep = !(cmd & LOCK_NB);
2169 if (can_sleep)
2170 fl.c.flc_flags |= FL_SLEEP;
2171
2172 if (f.file->f_op->flock)
2173 error = f.file->f_op->flock(f.file,
2174 (can_sleep) ? F_SETLKW : F_SETLK,
2175 &fl);
2176 else
2177 error = locks_lock_file_wait(f.file, &fl);
2178
2179 locks_release_private(&fl);
2180 out_putf:
2181 fdput(f);
2182
2183 return error;
2184}
2185
2186/**
2187 * vfs_test_lock - test file byte range lock
2188 * @filp: The file to test lock for
2189 * @fl: The lock to test; also used to hold result
2190 *
2191 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2192 * setting conf->fl_type to something other than F_UNLCK.
2193 */
2194int vfs_test_lock(struct file *filp, struct file_lock *fl)
2195{
2196 WARN_ON_ONCE(filp != fl->c.flc_file);
2197 if (filp->f_op->lock)
2198 return filp->f_op->lock(filp, F_GETLK, fl);
2199 posix_test_lock(filp, fl);
2200 return 0;
2201}
2202EXPORT_SYMBOL_GPL(vfs_test_lock);
2203
2204/**
2205 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2206 * @fl: The file_lock who's fl_pid should be translated
2207 * @ns: The namespace into which the pid should be translated
2208 *
2209 * Used to translate a fl_pid into a namespace virtual pid number
2210 */
2211static pid_t locks_translate_pid(struct file_lock_core *fl, struct pid_namespace *ns)
2212{
2213 pid_t vnr;
2214 struct pid *pid;
2215
2216 if (fl->flc_flags & FL_OFDLCK)
2217 return -1;
2218
2219 /* Remote locks report a negative pid value */
2220 if (fl->flc_pid <= 0)
2221 return fl->flc_pid;
2222
2223 /*
2224 * If the flock owner process is dead and its pid has been already
2225 * freed, the translation below won't work, but we still want to show
2226 * flock owner pid number in init pidns.
2227 */
2228 if (ns == &init_pid_ns)
2229 return (pid_t) fl->flc_pid;
2230
2231 rcu_read_lock();
2232 pid = find_pid_ns(fl->flc_pid, &init_pid_ns);
2233 vnr = pid_nr_ns(pid, ns);
2234 rcu_read_unlock();
2235 return vnr;
2236}
2237
2238static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2239{
2240 flock->l_pid = locks_translate_pid(&fl->c, task_active_pid_ns(current));
2241#if BITS_PER_LONG == 32
2242 /*
2243 * Make sure we can represent the posix lock via
2244 * legacy 32bit flock.
2245 */
2246 if (fl->fl_start > OFFT_OFFSET_MAX)
2247 return -EOVERFLOW;
2248 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2249 return -EOVERFLOW;
2250#endif
2251 flock->l_start = fl->fl_start;
2252 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2253 fl->fl_end - fl->fl_start + 1;
2254 flock->l_whence = 0;
2255 flock->l_type = fl->c.flc_type;
2256 return 0;
2257}
2258
2259#if BITS_PER_LONG == 32
2260static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2261{
2262 flock->l_pid = locks_translate_pid(&fl->c, task_active_pid_ns(current));
2263 flock->l_start = fl->fl_start;
2264 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2265 fl->fl_end - fl->fl_start + 1;
2266 flock->l_whence = 0;
2267 flock->l_type = fl->c.flc_type;
2268}
2269#endif
2270
2271/* Report the first existing lock that would conflict with l.
2272 * This implements the F_GETLK command of fcntl().
2273 */
2274int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2275{
2276 struct file_lock *fl;
2277 int error;
2278
2279 fl = locks_alloc_lock();
2280 if (fl == NULL)
2281 return -ENOMEM;
2282 error = -EINVAL;
2283 if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK
2284 && flock->l_type != F_WRLCK)
2285 goto out;
2286
2287 error = flock_to_posix_lock(filp, fl, flock);
2288 if (error)
2289 goto out;
2290
2291 if (cmd == F_OFD_GETLK) {
2292 error = -EINVAL;
2293 if (flock->l_pid != 0)
2294 goto out;
2295
2296 fl->c.flc_flags |= FL_OFDLCK;
2297 fl->c.flc_owner = filp;
2298 }
2299
2300 error = vfs_test_lock(filp, fl);
2301 if (error)
2302 goto out;
2303
2304 flock->l_type = fl->c.flc_type;
2305 if (fl->c.flc_type != F_UNLCK) {
2306 error = posix_lock_to_flock(flock, fl);
2307 if (error)
2308 goto out;
2309 }
2310out:
2311 locks_free_lock(fl);
2312 return error;
2313}
2314
2315/**
2316 * vfs_lock_file - file byte range lock
2317 * @filp: The file to apply the lock to
2318 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2319 * @fl: The lock to be applied
2320 * @conf: Place to return a copy of the conflicting lock, if found.
2321 *
2322 * A caller that doesn't care about the conflicting lock may pass NULL
2323 * as the final argument.
2324 *
2325 * If the filesystem defines a private ->lock() method, then @conf will
2326 * be left unchanged; so a caller that cares should initialize it to
2327 * some acceptable default.
2328 *
2329 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2330 * locks, the ->lock() interface may return asynchronously, before the lock has
2331 * been granted or denied by the underlying filesystem, if (and only if)
2332 * lm_grant is set. Additionally EXPORT_OP_ASYNC_LOCK in export_operations
2333 * flags need to be set.
2334 *
2335 * Callers expecting ->lock() to return asynchronously will only use F_SETLK,
2336 * not F_SETLKW; they will set FL_SLEEP if (and only if) the request is for a
2337 * blocking lock. When ->lock() does return asynchronously, it must return
2338 * FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock request completes.
2339 * If the request is for non-blocking lock the file system should return
2340 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2341 * with the result. If the request timed out the callback routine will return a
2342 * nonzero return code and the file system should release the lock. The file
2343 * system is also responsible to keep a corresponding posix lock when it
2344 * grants a lock so the VFS can find out which locks are locally held and do
2345 * the correct lock cleanup when required.
2346 * The underlying filesystem must not drop the kernel lock or call
2347 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2348 * return code.
2349 */
2350int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2351{
2352 WARN_ON_ONCE(filp != fl->c.flc_file);
2353 if (filp->f_op->lock)
2354 return filp->f_op->lock(filp, cmd, fl);
2355 else
2356 return posix_lock_file(filp, fl, conf);
2357}
2358EXPORT_SYMBOL_GPL(vfs_lock_file);
2359
2360static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2361 struct file_lock *fl)
2362{
2363 int error;
2364
2365 error = security_file_lock(filp, fl->c.flc_type);
2366 if (error)
2367 return error;
2368
2369 for (;;) {
2370 error = vfs_lock_file(filp, cmd, fl, NULL);
2371 if (error != FILE_LOCK_DEFERRED)
2372 break;
2373 error = wait_event_interruptible(fl->c.flc_wait,
2374 list_empty(&fl->c.flc_blocked_member));
2375 if (error)
2376 break;
2377 }
2378 locks_delete_block(fl);
2379
2380 return error;
2381}
2382
2383/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2384static int
2385check_fmode_for_setlk(struct file_lock *fl)
2386{
2387 switch (fl->c.flc_type) {
2388 case F_RDLCK:
2389 if (!(fl->c.flc_file->f_mode & FMODE_READ))
2390 return -EBADF;
2391 break;
2392 case F_WRLCK:
2393 if (!(fl->c.flc_file->f_mode & FMODE_WRITE))
2394 return -EBADF;
2395 }
2396 return 0;
2397}
2398
2399/* Apply the lock described by l to an open file descriptor.
2400 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2401 */
2402int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2403 struct flock *flock)
2404{
2405 struct file_lock *file_lock = locks_alloc_lock();
2406 struct inode *inode = file_inode(filp);
2407 struct file *f;
2408 int error;
2409
2410 if (file_lock == NULL)
2411 return -ENOLCK;
2412
2413 error = flock_to_posix_lock(filp, file_lock, flock);
2414 if (error)
2415 goto out;
2416
2417 error = check_fmode_for_setlk(file_lock);
2418 if (error)
2419 goto out;
2420
2421 /*
2422 * If the cmd is requesting file-private locks, then set the
2423 * FL_OFDLCK flag and override the owner.
2424 */
2425 switch (cmd) {
2426 case F_OFD_SETLK:
2427 error = -EINVAL;
2428 if (flock->l_pid != 0)
2429 goto out;
2430
2431 cmd = F_SETLK;
2432 file_lock->c.flc_flags |= FL_OFDLCK;
2433 file_lock->c.flc_owner = filp;
2434 break;
2435 case F_OFD_SETLKW:
2436 error = -EINVAL;
2437 if (flock->l_pid != 0)
2438 goto out;
2439
2440 cmd = F_SETLKW;
2441 file_lock->c.flc_flags |= FL_OFDLCK;
2442 file_lock->c.flc_owner = filp;
2443 fallthrough;
2444 case F_SETLKW:
2445 file_lock->c.flc_flags |= FL_SLEEP;
2446 }
2447
2448 error = do_lock_file_wait(filp, cmd, file_lock);
2449
2450 /*
2451 * Attempt to detect a close/fcntl race and recover by releasing the
2452 * lock that was just acquired. There is no need to do that when we're
2453 * unlocking though, or for OFD locks.
2454 */
2455 if (!error && file_lock->c.flc_type != F_UNLCK &&
2456 !(file_lock->c.flc_flags & FL_OFDLCK)) {
2457 struct files_struct *files = current->files;
2458 /*
2459 * We need that spin_lock here - it prevents reordering between
2460 * update of i_flctx->flc_posix and check for it done in
2461 * close(). rcu_read_lock() wouldn't do.
2462 */
2463 spin_lock(&files->file_lock);
2464 f = files_lookup_fd_locked(files, fd);
2465 spin_unlock(&files->file_lock);
2466 if (f != filp) {
2467 file_lock->c.flc_type = F_UNLCK;
2468 error = do_lock_file_wait(filp, cmd, file_lock);
2469 WARN_ON_ONCE(error);
2470 error = -EBADF;
2471 }
2472 }
2473out:
2474 trace_fcntl_setlk(inode, file_lock, error);
2475 locks_free_lock(file_lock);
2476 return error;
2477}
2478
2479#if BITS_PER_LONG == 32
2480/* Report the first existing lock that would conflict with l.
2481 * This implements the F_GETLK command of fcntl().
2482 */
2483int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2484{
2485 struct file_lock *fl;
2486 int error;
2487
2488 fl = locks_alloc_lock();
2489 if (fl == NULL)
2490 return -ENOMEM;
2491
2492 error = -EINVAL;
2493 if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK
2494 && flock->l_type != F_WRLCK)
2495 goto out;
2496
2497 error = flock64_to_posix_lock(filp, fl, flock);
2498 if (error)
2499 goto out;
2500
2501 if (cmd == F_OFD_GETLK) {
2502 error = -EINVAL;
2503 if (flock->l_pid != 0)
2504 goto out;
2505
2506 fl->c.flc_flags |= FL_OFDLCK;
2507 fl->c.flc_owner = filp;
2508 }
2509
2510 error = vfs_test_lock(filp, fl);
2511 if (error)
2512 goto out;
2513
2514 flock->l_type = fl->c.flc_type;
2515 if (fl->c.flc_type != F_UNLCK)
2516 posix_lock_to_flock64(flock, fl);
2517
2518out:
2519 locks_free_lock(fl);
2520 return error;
2521}
2522
2523/* Apply the lock described by l to an open file descriptor.
2524 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2525 */
2526int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2527 struct flock64 *flock)
2528{
2529 struct file_lock *file_lock = locks_alloc_lock();
2530 struct file *f;
2531 int error;
2532
2533 if (file_lock == NULL)
2534 return -ENOLCK;
2535
2536 error = flock64_to_posix_lock(filp, file_lock, flock);
2537 if (error)
2538 goto out;
2539
2540 error = check_fmode_for_setlk(file_lock);
2541 if (error)
2542 goto out;
2543
2544 /*
2545 * If the cmd is requesting file-private locks, then set the
2546 * FL_OFDLCK flag and override the owner.
2547 */
2548 switch (cmd) {
2549 case F_OFD_SETLK:
2550 error = -EINVAL;
2551 if (flock->l_pid != 0)
2552 goto out;
2553
2554 cmd = F_SETLK64;
2555 file_lock->c.flc_flags |= FL_OFDLCK;
2556 file_lock->c.flc_owner = filp;
2557 break;
2558 case F_OFD_SETLKW:
2559 error = -EINVAL;
2560 if (flock->l_pid != 0)
2561 goto out;
2562
2563 cmd = F_SETLKW64;
2564 file_lock->c.flc_flags |= FL_OFDLCK;
2565 file_lock->c.flc_owner = filp;
2566 fallthrough;
2567 case F_SETLKW64:
2568 file_lock->c.flc_flags |= FL_SLEEP;
2569 }
2570
2571 error = do_lock_file_wait(filp, cmd, file_lock);
2572
2573 /*
2574 * Attempt to detect a close/fcntl race and recover by releasing the
2575 * lock that was just acquired. There is no need to do that when we're
2576 * unlocking though, or for OFD locks.
2577 */
2578 if (!error && file_lock->c.flc_type != F_UNLCK &&
2579 !(file_lock->c.flc_flags & FL_OFDLCK)) {
2580 struct files_struct *files = current->files;
2581 /*
2582 * We need that spin_lock here - it prevents reordering between
2583 * update of i_flctx->flc_posix and check for it done in
2584 * close(). rcu_read_lock() wouldn't do.
2585 */
2586 spin_lock(&files->file_lock);
2587 f = files_lookup_fd_locked(files, fd);
2588 spin_unlock(&files->file_lock);
2589 if (f != filp) {
2590 file_lock->c.flc_type = F_UNLCK;
2591 error = do_lock_file_wait(filp, cmd, file_lock);
2592 WARN_ON_ONCE(error);
2593 error = -EBADF;
2594 }
2595 }
2596out:
2597 locks_free_lock(file_lock);
2598 return error;
2599}
2600#endif /* BITS_PER_LONG == 32 */
2601
2602/*
2603 * This function is called when the file is being removed
2604 * from the task's fd array. POSIX locks belonging to this task
2605 * are deleted at this time.
2606 */
2607void locks_remove_posix(struct file *filp, fl_owner_t owner)
2608{
2609 int error;
2610 struct inode *inode = file_inode(filp);
2611 struct file_lock lock;
2612 struct file_lock_context *ctx;
2613
2614 /*
2615 * If there are no locks held on this file, we don't need to call
2616 * posix_lock_file(). Another process could be setting a lock on this
2617 * file at the same time, but we wouldn't remove that lock anyway.
2618 */
2619 ctx = locks_inode_context(inode);
2620 if (!ctx || list_empty(&ctx->flc_posix))
2621 return;
2622
2623 locks_init_lock(&lock);
2624 lock.c.flc_type = F_UNLCK;
2625 lock.c.flc_flags = FL_POSIX | FL_CLOSE;
2626 lock.fl_start = 0;
2627 lock.fl_end = OFFSET_MAX;
2628 lock.c.flc_owner = owner;
2629 lock.c.flc_pid = current->tgid;
2630 lock.c.flc_file = filp;
2631 lock.fl_ops = NULL;
2632 lock.fl_lmops = NULL;
2633
2634 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2635
2636 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2637 lock.fl_ops->fl_release_private(&lock);
2638 trace_locks_remove_posix(inode, &lock, error);
2639}
2640EXPORT_SYMBOL(locks_remove_posix);
2641
2642/* The i_flctx must be valid when calling into here */
2643static void
2644locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2645{
2646 struct file_lock fl;
2647 struct inode *inode = file_inode(filp);
2648
2649 if (list_empty(&flctx->flc_flock))
2650 return;
2651
2652 flock_make_lock(filp, &fl, F_UNLCK);
2653 fl.c.flc_flags |= FL_CLOSE;
2654
2655 if (filp->f_op->flock)
2656 filp->f_op->flock(filp, F_SETLKW, &fl);
2657 else
2658 flock_lock_inode(inode, &fl);
2659
2660 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2661 fl.fl_ops->fl_release_private(&fl);
2662}
2663
2664/* The i_flctx must be valid when calling into here */
2665static void
2666locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2667{
2668 struct file_lease *fl, *tmp;
2669 LIST_HEAD(dispose);
2670
2671 if (list_empty(&ctx->flc_lease))
2672 return;
2673
2674 percpu_down_read(&file_rwsem);
2675 spin_lock(&ctx->flc_lock);
2676 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, c.flc_list)
2677 if (filp == fl->c.flc_file)
2678 lease_modify(fl, F_UNLCK, &dispose);
2679 spin_unlock(&ctx->flc_lock);
2680 percpu_up_read(&file_rwsem);
2681
2682 locks_dispose_list(&dispose);
2683}
2684
2685/*
2686 * This function is called on the last close of an open file.
2687 */
2688void locks_remove_file(struct file *filp)
2689{
2690 struct file_lock_context *ctx;
2691
2692 ctx = locks_inode_context(file_inode(filp));
2693 if (!ctx)
2694 return;
2695
2696 /* remove any OFD locks */
2697 locks_remove_posix(filp, filp);
2698
2699 /* remove flock locks */
2700 locks_remove_flock(filp, ctx);
2701
2702 /* remove any leases */
2703 locks_remove_lease(filp, ctx);
2704
2705 spin_lock(&ctx->flc_lock);
2706 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2707 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2708 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2709 spin_unlock(&ctx->flc_lock);
2710}
2711
2712/**
2713 * vfs_cancel_lock - file byte range unblock lock
2714 * @filp: The file to apply the unblock to
2715 * @fl: The lock to be unblocked
2716 *
2717 * Used by lock managers to cancel blocked requests
2718 */
2719int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2720{
2721 WARN_ON_ONCE(filp != fl->c.flc_file);
2722 if (filp->f_op->lock)
2723 return filp->f_op->lock(filp, F_CANCELLK, fl);
2724 return 0;
2725}
2726EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2727
2728/**
2729 * vfs_inode_has_locks - are any file locks held on @inode?
2730 * @inode: inode to check for locks
2731 *
2732 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2733 * set on @inode.
2734 */
2735bool vfs_inode_has_locks(struct inode *inode)
2736{
2737 struct file_lock_context *ctx;
2738 bool ret;
2739
2740 ctx = locks_inode_context(inode);
2741 if (!ctx)
2742 return false;
2743
2744 spin_lock(&ctx->flc_lock);
2745 ret = !list_empty(&ctx->flc_posix) || !list_empty(&ctx->flc_flock);
2746 spin_unlock(&ctx->flc_lock);
2747 return ret;
2748}
2749EXPORT_SYMBOL_GPL(vfs_inode_has_locks);
2750
2751#ifdef CONFIG_PROC_FS
2752#include <linux/proc_fs.h>
2753#include <linux/seq_file.h>
2754
2755struct locks_iterator {
2756 int li_cpu;
2757 loff_t li_pos;
2758};
2759
2760static void lock_get_status(struct seq_file *f, struct file_lock_core *flc,
2761 loff_t id, char *pfx, int repeat)
2762{
2763 struct inode *inode = NULL;
2764 unsigned int pid;
2765 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2766 int type = flc->flc_type;
2767 struct file_lock *fl = file_lock(flc);
2768
2769 pid = locks_translate_pid(flc, proc_pidns);
2770
2771 /*
2772 * If lock owner is dead (and pid is freed) or not visible in current
2773 * pidns, zero is shown as a pid value. Check lock info from
2774 * init_pid_ns to get saved lock pid value.
2775 */
2776 if (flc->flc_file != NULL)
2777 inode = file_inode(flc->flc_file);
2778
2779 seq_printf(f, "%lld: ", id);
2780
2781 if (repeat)
2782 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2783
2784 if (flc->flc_flags & FL_POSIX) {
2785 if (flc->flc_flags & FL_ACCESS)
2786 seq_puts(f, "ACCESS");
2787 else if (flc->flc_flags & FL_OFDLCK)
2788 seq_puts(f, "OFDLCK");
2789 else
2790 seq_puts(f, "POSIX ");
2791
2792 seq_printf(f, " %s ",
2793 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2794 } else if (flc->flc_flags & FL_FLOCK) {
2795 seq_puts(f, "FLOCK ADVISORY ");
2796 } else if (flc->flc_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT)) {
2797 struct file_lease *lease = file_lease(flc);
2798
2799 type = target_leasetype(lease);
2800
2801 if (flc->flc_flags & FL_DELEG)
2802 seq_puts(f, "DELEG ");
2803 else
2804 seq_puts(f, "LEASE ");
2805
2806 if (lease_breaking(lease))
2807 seq_puts(f, "BREAKING ");
2808 else if (flc->flc_file)
2809 seq_puts(f, "ACTIVE ");
2810 else
2811 seq_puts(f, "BREAKER ");
2812 } else {
2813 seq_puts(f, "UNKNOWN UNKNOWN ");
2814 }
2815
2816 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2817 (type == F_RDLCK) ? "READ" : "UNLCK");
2818 if (inode) {
2819 /* userspace relies on this representation of dev_t */
2820 seq_printf(f, "%d %02x:%02x:%lu ", pid,
2821 MAJOR(inode->i_sb->s_dev),
2822 MINOR(inode->i_sb->s_dev), inode->i_ino);
2823 } else {
2824 seq_printf(f, "%d <none>:0 ", pid);
2825 }
2826 if (flc->flc_flags & FL_POSIX) {
2827 if (fl->fl_end == OFFSET_MAX)
2828 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2829 else
2830 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2831 } else {
2832 seq_puts(f, "0 EOF\n");
2833 }
2834}
2835
2836static struct file_lock_core *get_next_blocked_member(struct file_lock_core *node)
2837{
2838 struct file_lock_core *tmp;
2839
2840 /* NULL node or root node */
2841 if (node == NULL || node->flc_blocker == NULL)
2842 return NULL;
2843
2844 /* Next member in the linked list could be itself */
2845 tmp = list_next_entry(node, flc_blocked_member);
2846 if (list_entry_is_head(tmp, &node->flc_blocker->flc_blocked_requests,
2847 flc_blocked_member)
2848 || tmp == node) {
2849 return NULL;
2850 }
2851
2852 return tmp;
2853}
2854
2855static int locks_show(struct seq_file *f, void *v)
2856{
2857 struct locks_iterator *iter = f->private;
2858 struct file_lock_core *cur, *tmp;
2859 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2860 int level = 0;
2861
2862 cur = hlist_entry(v, struct file_lock_core, flc_link);
2863
2864 if (locks_translate_pid(cur, proc_pidns) == 0)
2865 return 0;
2866
2867 /* View this crossed linked list as a binary tree, the first member of flc_blocked_requests
2868 * is the left child of current node, the next silibing in flc_blocked_member is the
2869 * right child, we can alse get the parent of current node from flc_blocker, so this
2870 * question becomes traversal of a binary tree
2871 */
2872 while (cur != NULL) {
2873 if (level)
2874 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2875 else
2876 lock_get_status(f, cur, iter->li_pos, "", level);
2877
2878 if (!list_empty(&cur->flc_blocked_requests)) {
2879 /* Turn left */
2880 cur = list_first_entry_or_null(&cur->flc_blocked_requests,
2881 struct file_lock_core,
2882 flc_blocked_member);
2883 level++;
2884 } else {
2885 /* Turn right */
2886 tmp = get_next_blocked_member(cur);
2887 /* Fall back to parent node */
2888 while (tmp == NULL && cur->flc_blocker != NULL) {
2889 cur = cur->flc_blocker;
2890 level--;
2891 tmp = get_next_blocked_member(cur);
2892 }
2893 cur = tmp;
2894 }
2895 }
2896
2897 return 0;
2898}
2899
2900static void __show_fd_locks(struct seq_file *f,
2901 struct list_head *head, int *id,
2902 struct file *filp, struct files_struct *files)
2903{
2904 struct file_lock_core *fl;
2905
2906 list_for_each_entry(fl, head, flc_list) {
2907
2908 if (filp != fl->flc_file)
2909 continue;
2910 if (fl->flc_owner != files && fl->flc_owner != filp)
2911 continue;
2912
2913 (*id)++;
2914 seq_puts(f, "lock:\t");
2915 lock_get_status(f, fl, *id, "", 0);
2916 }
2917}
2918
2919void show_fd_locks(struct seq_file *f,
2920 struct file *filp, struct files_struct *files)
2921{
2922 struct inode *inode = file_inode(filp);
2923 struct file_lock_context *ctx;
2924 int id = 0;
2925
2926 ctx = locks_inode_context(inode);
2927 if (!ctx)
2928 return;
2929
2930 spin_lock(&ctx->flc_lock);
2931 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2932 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2933 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2934 spin_unlock(&ctx->flc_lock);
2935}
2936
2937static void *locks_start(struct seq_file *f, loff_t *pos)
2938 __acquires(&blocked_lock_lock)
2939{
2940 struct locks_iterator *iter = f->private;
2941
2942 iter->li_pos = *pos + 1;
2943 percpu_down_write(&file_rwsem);
2944 spin_lock(&blocked_lock_lock);
2945 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2946}
2947
2948static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2949{
2950 struct locks_iterator *iter = f->private;
2951
2952 ++iter->li_pos;
2953 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2954}
2955
2956static void locks_stop(struct seq_file *f, void *v)
2957 __releases(&blocked_lock_lock)
2958{
2959 spin_unlock(&blocked_lock_lock);
2960 percpu_up_write(&file_rwsem);
2961}
2962
2963static const struct seq_operations locks_seq_operations = {
2964 .start = locks_start,
2965 .next = locks_next,
2966 .stop = locks_stop,
2967 .show = locks_show,
2968};
2969
2970static int __init proc_locks_init(void)
2971{
2972 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2973 sizeof(struct locks_iterator), NULL);
2974 return 0;
2975}
2976fs_initcall(proc_locks_init);
2977#endif
2978
2979static int __init filelock_init(void)
2980{
2981 int i;
2982
2983 flctx_cache = kmem_cache_create("file_lock_ctx",
2984 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2985
2986 filelock_cache = kmem_cache_create("file_lock_cache",
2987 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2988
2989 filelease_cache = kmem_cache_create("file_lock_cache",
2990 sizeof(struct file_lease), 0, SLAB_PANIC, NULL);
2991
2992 for_each_possible_cpu(i) {
2993 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2994
2995 spin_lock_init(&fll->lock);
2996 INIT_HLIST_HEAD(&fll->hlist);
2997 }
2998
2999 lease_notifier_chain_init();
3000 return 0;
3001}
3002core_initcall(filelock_init);