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/filelock.h>
56#include <linux/fs.h>
57#include <linux/init.h>
58#include <linux/security.h>
59#include <linux/slab.h>
60#include <linux/syscalls.h>
61#include <linux/time.h>
62#include <linux/rcupdate.h>
63#include <linux/pid_namespace.h>
64#include <linux/hashtable.h>
65#include <linux/percpu.h>
66#include <linux/sysctl.h>
67
68#define CREATE_TRACE_POINTS
69#include <trace/events/filelock.h>
70
71#include <linux/uaccess.h>
72
73#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
74#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
75#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
76#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
77#define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
78
79static bool lease_breaking(struct file_lock *fl)
80{
81 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
82}
83
84static int target_leasetype(struct file_lock *fl)
85{
86 if (fl->fl_flags & FL_UNLOCK_PENDING)
87 return F_UNLCK;
88 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
89 return F_RDLCK;
90 return fl->fl_type;
91}
92
93static int leases_enable = 1;
94static int lease_break_time = 45;
95
96#ifdef CONFIG_SYSCTL
97static struct ctl_table locks_sysctls[] = {
98 {
99 .procname = "leases-enable",
100 .data = &leases_enable,
101 .maxlen = sizeof(int),
102 .mode = 0644,
103 .proc_handler = proc_dointvec,
104 },
105#ifdef CONFIG_MMU
106 {
107 .procname = "lease-break-time",
108 .data = &lease_break_time,
109 .maxlen = sizeof(int),
110 .mode = 0644,
111 .proc_handler = proc_dointvec,
112 },
113#endif /* CONFIG_MMU */
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 __ro_after_init;
170static struct kmem_cache *filelock_cache __ro_after_init;
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 = file_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, int 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, int 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, int 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. Used on xx_GETLK
871 * path so checks for additional GETLK-specific things like F_UNLCK.
872 */
873static bool posix_test_locks_conflict(struct file_lock *caller_fl,
874 struct file_lock *sys_fl)
875{
876 /* F_UNLCK checks any locks on the same fd. */
877 if (caller_fl->fl_type == F_UNLCK) {
878 if (!posix_same_owner(caller_fl, sys_fl))
879 return false;
880 return locks_overlap(caller_fl, sys_fl);
881 }
882 return posix_locks_conflict(caller_fl, sys_fl);
883}
884
885/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
886 * checking before calling the locks_conflict().
887 */
888static bool flock_locks_conflict(struct file_lock *caller_fl,
889 struct file_lock *sys_fl)
890{
891 /* FLOCK locks referring to the same filp do not conflict with
892 * each other.
893 */
894 if (caller_fl->fl_file == sys_fl->fl_file)
895 return false;
896
897 return locks_conflict(caller_fl, sys_fl);
898}
899
900void
901posix_test_lock(struct file *filp, struct file_lock *fl)
902{
903 struct file_lock *cfl;
904 struct file_lock_context *ctx;
905 struct inode *inode = file_inode(filp);
906 void *owner;
907 void (*func)(void);
908
909 ctx = locks_inode_context(inode);
910 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
911 fl->fl_type = F_UNLCK;
912 return;
913 }
914
915retry:
916 spin_lock(&ctx->flc_lock);
917 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
918 if (!posix_test_locks_conflict(fl, cfl))
919 continue;
920 if (cfl->fl_lmops && cfl->fl_lmops->lm_lock_expirable
921 && (*cfl->fl_lmops->lm_lock_expirable)(cfl)) {
922 owner = cfl->fl_lmops->lm_mod_owner;
923 func = cfl->fl_lmops->lm_expire_lock;
924 __module_get(owner);
925 spin_unlock(&ctx->flc_lock);
926 (*func)();
927 module_put(owner);
928 goto retry;
929 }
930 locks_copy_conflock(fl, cfl);
931 goto out;
932 }
933 fl->fl_type = F_UNLCK;
934out:
935 spin_unlock(&ctx->flc_lock);
936 return;
937}
938EXPORT_SYMBOL(posix_test_lock);
939
940/*
941 * Deadlock detection:
942 *
943 * We attempt to detect deadlocks that are due purely to posix file
944 * locks.
945 *
946 * We assume that a task can be waiting for at most one lock at a time.
947 * So for any acquired lock, the process holding that lock may be
948 * waiting on at most one other lock. That lock in turns may be held by
949 * someone waiting for at most one other lock. Given a requested lock
950 * caller_fl which is about to wait for a conflicting lock block_fl, we
951 * follow this chain of waiters to ensure we are not about to create a
952 * cycle.
953 *
954 * Since we do this before we ever put a process to sleep on a lock, we
955 * are ensured that there is never a cycle; that is what guarantees that
956 * the while() loop in posix_locks_deadlock() eventually completes.
957 *
958 * Note: the above assumption may not be true when handling lock
959 * requests from a broken NFS client. It may also fail in the presence
960 * of tasks (such as posix threads) sharing the same open file table.
961 * To handle those cases, we just bail out after a few iterations.
962 *
963 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
964 * Because the owner is not even nominally tied to a thread of
965 * execution, the deadlock detection below can't reasonably work well. Just
966 * skip it for those.
967 *
968 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
969 * locks that just checks for the case where two tasks are attempting to
970 * upgrade from read to write locks on the same inode.
971 */
972
973#define MAX_DEADLK_ITERATIONS 10
974
975/* Find a lock that the owner of the given block_fl is blocking on. */
976static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
977{
978 struct file_lock *fl;
979
980 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
981 if (posix_same_owner(fl, block_fl)) {
982 while (fl->fl_blocker)
983 fl = fl->fl_blocker;
984 return fl;
985 }
986 }
987 return NULL;
988}
989
990/* Must be called with the blocked_lock_lock held! */
991static int posix_locks_deadlock(struct file_lock *caller_fl,
992 struct file_lock *block_fl)
993{
994 int i = 0;
995
996 lockdep_assert_held(&blocked_lock_lock);
997
998 /*
999 * This deadlock detector can't reasonably detect deadlocks with
1000 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1001 */
1002 if (IS_OFDLCK(caller_fl))
1003 return 0;
1004
1005 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
1006 if (i++ > MAX_DEADLK_ITERATIONS)
1007 return 0;
1008 if (posix_same_owner(caller_fl, block_fl))
1009 return 1;
1010 }
1011 return 0;
1012}
1013
1014/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1015 * after any leases, but before any posix locks.
1016 *
1017 * Note that if called with an FL_EXISTS argument, the caller may determine
1018 * whether or not a lock was successfully freed by testing the return
1019 * value for -ENOENT.
1020 */
1021static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1022{
1023 struct file_lock *new_fl = NULL;
1024 struct file_lock *fl;
1025 struct file_lock_context *ctx;
1026 int error = 0;
1027 bool found = false;
1028 LIST_HEAD(dispose);
1029
1030 ctx = locks_get_lock_context(inode, request->fl_type);
1031 if (!ctx) {
1032 if (request->fl_type != F_UNLCK)
1033 return -ENOMEM;
1034 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1035 }
1036
1037 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1038 new_fl = locks_alloc_lock();
1039 if (!new_fl)
1040 return -ENOMEM;
1041 }
1042
1043 percpu_down_read(&file_rwsem);
1044 spin_lock(&ctx->flc_lock);
1045 if (request->fl_flags & FL_ACCESS)
1046 goto find_conflict;
1047
1048 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1049 if (request->fl_file != fl->fl_file)
1050 continue;
1051 if (request->fl_type == fl->fl_type)
1052 goto out;
1053 found = true;
1054 locks_delete_lock_ctx(fl, &dispose);
1055 break;
1056 }
1057
1058 if (request->fl_type == F_UNLCK) {
1059 if ((request->fl_flags & FL_EXISTS) && !found)
1060 error = -ENOENT;
1061 goto out;
1062 }
1063
1064find_conflict:
1065 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1066 if (!flock_locks_conflict(request, fl))
1067 continue;
1068 error = -EAGAIN;
1069 if (!(request->fl_flags & FL_SLEEP))
1070 goto out;
1071 error = FILE_LOCK_DEFERRED;
1072 locks_insert_block(fl, request, flock_locks_conflict);
1073 goto out;
1074 }
1075 if (request->fl_flags & FL_ACCESS)
1076 goto out;
1077 locks_copy_lock(new_fl, request);
1078 locks_move_blocks(new_fl, request);
1079 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1080 new_fl = NULL;
1081 error = 0;
1082
1083out:
1084 spin_unlock(&ctx->flc_lock);
1085 percpu_up_read(&file_rwsem);
1086 if (new_fl)
1087 locks_free_lock(new_fl);
1088 locks_dispose_list(&dispose);
1089 trace_flock_lock_inode(inode, request, error);
1090 return error;
1091}
1092
1093static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1094 struct file_lock *conflock)
1095{
1096 struct file_lock *fl, *tmp;
1097 struct file_lock *new_fl = NULL;
1098 struct file_lock *new_fl2 = NULL;
1099 struct file_lock *left = NULL;
1100 struct file_lock *right = NULL;
1101 struct file_lock_context *ctx;
1102 int error;
1103 bool added = false;
1104 LIST_HEAD(dispose);
1105 void *owner;
1106 void (*func)(void);
1107
1108 ctx = locks_get_lock_context(inode, request->fl_type);
1109 if (!ctx)
1110 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1111
1112 /*
1113 * We may need two file_lock structures for this operation,
1114 * so we get them in advance to avoid races.
1115 *
1116 * In some cases we can be sure, that no new locks will be needed
1117 */
1118 if (!(request->fl_flags & FL_ACCESS) &&
1119 (request->fl_type != F_UNLCK ||
1120 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1121 new_fl = locks_alloc_lock();
1122 new_fl2 = locks_alloc_lock();
1123 }
1124
1125retry:
1126 percpu_down_read(&file_rwsem);
1127 spin_lock(&ctx->flc_lock);
1128 /*
1129 * New lock request. Walk all POSIX locks and look for conflicts. If
1130 * there are any, either return error or put the request on the
1131 * blocker's list of waiters and the global blocked_hash.
1132 */
1133 if (request->fl_type != F_UNLCK) {
1134 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1135 if (!posix_locks_conflict(request, fl))
1136 continue;
1137 if (fl->fl_lmops && fl->fl_lmops->lm_lock_expirable
1138 && (*fl->fl_lmops->lm_lock_expirable)(fl)) {
1139 owner = fl->fl_lmops->lm_mod_owner;
1140 func = fl->fl_lmops->lm_expire_lock;
1141 __module_get(owner);
1142 spin_unlock(&ctx->flc_lock);
1143 percpu_up_read(&file_rwsem);
1144 (*func)();
1145 module_put(owner);
1146 goto retry;
1147 }
1148 if (conflock)
1149 locks_copy_conflock(conflock, fl);
1150 error = -EAGAIN;
1151 if (!(request->fl_flags & FL_SLEEP))
1152 goto out;
1153 /*
1154 * Deadlock detection and insertion into the blocked
1155 * locks list must be done while holding the same lock!
1156 */
1157 error = -EDEADLK;
1158 spin_lock(&blocked_lock_lock);
1159 /*
1160 * Ensure that we don't find any locks blocked on this
1161 * request during deadlock detection.
1162 */
1163 __locks_wake_up_blocks(request);
1164 if (likely(!posix_locks_deadlock(request, fl))) {
1165 error = FILE_LOCK_DEFERRED;
1166 __locks_insert_block(fl, request,
1167 posix_locks_conflict);
1168 }
1169 spin_unlock(&blocked_lock_lock);
1170 goto out;
1171 }
1172 }
1173
1174 /* If we're just looking for a conflict, we're done. */
1175 error = 0;
1176 if (request->fl_flags & FL_ACCESS)
1177 goto out;
1178
1179 /* Find the first old lock with the same owner as the new lock */
1180 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1181 if (posix_same_owner(request, fl))
1182 break;
1183 }
1184
1185 /* Process locks with this owner. */
1186 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1187 if (!posix_same_owner(request, fl))
1188 break;
1189
1190 /* Detect adjacent or overlapping regions (if same lock type) */
1191 if (request->fl_type == fl->fl_type) {
1192 /* In all comparisons of start vs end, use
1193 * "start - 1" rather than "end + 1". If end
1194 * is OFFSET_MAX, end + 1 will become negative.
1195 */
1196 if (fl->fl_end < request->fl_start - 1)
1197 continue;
1198 /* If the next lock in the list has entirely bigger
1199 * addresses than the new one, insert the lock here.
1200 */
1201 if (fl->fl_start - 1 > request->fl_end)
1202 break;
1203
1204 /* If we come here, the new and old lock are of the
1205 * same type and adjacent or overlapping. Make one
1206 * lock yielding from the lower start address of both
1207 * locks to the higher end address.
1208 */
1209 if (fl->fl_start > request->fl_start)
1210 fl->fl_start = request->fl_start;
1211 else
1212 request->fl_start = fl->fl_start;
1213 if (fl->fl_end < request->fl_end)
1214 fl->fl_end = request->fl_end;
1215 else
1216 request->fl_end = fl->fl_end;
1217 if (added) {
1218 locks_delete_lock_ctx(fl, &dispose);
1219 continue;
1220 }
1221 request = fl;
1222 added = true;
1223 } else {
1224 /* Processing for different lock types is a bit
1225 * more complex.
1226 */
1227 if (fl->fl_end < request->fl_start)
1228 continue;
1229 if (fl->fl_start > request->fl_end)
1230 break;
1231 if (request->fl_type == F_UNLCK)
1232 added = true;
1233 if (fl->fl_start < request->fl_start)
1234 left = fl;
1235 /* If the next lock in the list has a higher end
1236 * address than the new one, insert the new one here.
1237 */
1238 if (fl->fl_end > request->fl_end) {
1239 right = fl;
1240 break;
1241 }
1242 if (fl->fl_start >= request->fl_start) {
1243 /* The new lock completely replaces an old
1244 * one (This may happen several times).
1245 */
1246 if (added) {
1247 locks_delete_lock_ctx(fl, &dispose);
1248 continue;
1249 }
1250 /*
1251 * Replace the old lock with new_fl, and
1252 * remove the old one. It's safe to do the
1253 * insert here since we know that we won't be
1254 * using new_fl later, and that the lock is
1255 * just replacing an existing lock.
1256 */
1257 error = -ENOLCK;
1258 if (!new_fl)
1259 goto out;
1260 locks_copy_lock(new_fl, request);
1261 locks_move_blocks(new_fl, request);
1262 request = new_fl;
1263 new_fl = NULL;
1264 locks_insert_lock_ctx(request, &fl->fl_list);
1265 locks_delete_lock_ctx(fl, &dispose);
1266 added = true;
1267 }
1268 }
1269 }
1270
1271 /*
1272 * The above code only modifies existing locks in case of merging or
1273 * replacing. If new lock(s) need to be inserted all modifications are
1274 * done below this, so it's safe yet to bail out.
1275 */
1276 error = -ENOLCK; /* "no luck" */
1277 if (right && left == right && !new_fl2)
1278 goto out;
1279
1280 error = 0;
1281 if (!added) {
1282 if (request->fl_type == F_UNLCK) {
1283 if (request->fl_flags & FL_EXISTS)
1284 error = -ENOENT;
1285 goto out;
1286 }
1287
1288 if (!new_fl) {
1289 error = -ENOLCK;
1290 goto out;
1291 }
1292 locks_copy_lock(new_fl, request);
1293 locks_move_blocks(new_fl, request);
1294 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1295 fl = new_fl;
1296 new_fl = NULL;
1297 }
1298 if (right) {
1299 if (left == right) {
1300 /* The new lock breaks the old one in two pieces,
1301 * so we have to use the second new lock.
1302 */
1303 left = new_fl2;
1304 new_fl2 = NULL;
1305 locks_copy_lock(left, right);
1306 locks_insert_lock_ctx(left, &fl->fl_list);
1307 }
1308 right->fl_start = request->fl_end + 1;
1309 locks_wake_up_blocks(right);
1310 }
1311 if (left) {
1312 left->fl_end = request->fl_start - 1;
1313 locks_wake_up_blocks(left);
1314 }
1315 out:
1316 spin_unlock(&ctx->flc_lock);
1317 percpu_up_read(&file_rwsem);
1318 trace_posix_lock_inode(inode, request, error);
1319 /*
1320 * Free any unused locks.
1321 */
1322 if (new_fl)
1323 locks_free_lock(new_fl);
1324 if (new_fl2)
1325 locks_free_lock(new_fl2);
1326 locks_dispose_list(&dispose);
1327
1328 return error;
1329}
1330
1331/**
1332 * posix_lock_file - Apply a POSIX-style lock to a file
1333 * @filp: The file to apply the lock to
1334 * @fl: The lock to be applied
1335 * @conflock: Place to return a copy of the conflicting lock, if found.
1336 *
1337 * Add a POSIX style lock to a file.
1338 * We merge adjacent & overlapping locks whenever possible.
1339 * POSIX locks are sorted by owner task, then by starting address
1340 *
1341 * Note that if called with an FL_EXISTS argument, the caller may determine
1342 * whether or not a lock was successfully freed by testing the return
1343 * value for -ENOENT.
1344 */
1345int posix_lock_file(struct file *filp, struct file_lock *fl,
1346 struct file_lock *conflock)
1347{
1348 return posix_lock_inode(file_inode(filp), fl, conflock);
1349}
1350EXPORT_SYMBOL(posix_lock_file);
1351
1352/**
1353 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1354 * @inode: inode of file to which lock request should be applied
1355 * @fl: The lock to be applied
1356 *
1357 * Apply a POSIX style lock request to an inode.
1358 */
1359static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1360{
1361 int error;
1362 might_sleep ();
1363 for (;;) {
1364 error = posix_lock_inode(inode, fl, NULL);
1365 if (error != FILE_LOCK_DEFERRED)
1366 break;
1367 error = wait_event_interruptible(fl->fl_wait,
1368 list_empty(&fl->fl_blocked_member));
1369 if (error)
1370 break;
1371 }
1372 locks_delete_block(fl);
1373 return error;
1374}
1375
1376static void lease_clear_pending(struct file_lock *fl, int arg)
1377{
1378 switch (arg) {
1379 case F_UNLCK:
1380 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1381 fallthrough;
1382 case F_RDLCK:
1383 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1384 }
1385}
1386
1387/* We already had a lease on this file; just change its type */
1388int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1389{
1390 int error = assign_type(fl, arg);
1391
1392 if (error)
1393 return error;
1394 lease_clear_pending(fl, arg);
1395 locks_wake_up_blocks(fl);
1396 if (arg == F_UNLCK) {
1397 struct file *filp = fl->fl_file;
1398
1399 f_delown(filp);
1400 filp->f_owner.signum = 0;
1401 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1402 if (fl->fl_fasync != NULL) {
1403 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1404 fl->fl_fasync = NULL;
1405 }
1406 locks_delete_lock_ctx(fl, dispose);
1407 }
1408 return 0;
1409}
1410EXPORT_SYMBOL(lease_modify);
1411
1412static bool past_time(unsigned long then)
1413{
1414 if (!then)
1415 /* 0 is a special value meaning "this never expires": */
1416 return false;
1417 return time_after(jiffies, then);
1418}
1419
1420static void time_out_leases(struct inode *inode, struct list_head *dispose)
1421{
1422 struct file_lock_context *ctx = inode->i_flctx;
1423 struct file_lock *fl, *tmp;
1424
1425 lockdep_assert_held(&ctx->flc_lock);
1426
1427 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1428 trace_time_out_leases(inode, fl);
1429 if (past_time(fl->fl_downgrade_time))
1430 lease_modify(fl, F_RDLCK, dispose);
1431 if (past_time(fl->fl_break_time))
1432 lease_modify(fl, F_UNLCK, dispose);
1433 }
1434}
1435
1436static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1437{
1438 bool rc;
1439
1440 if (lease->fl_lmops->lm_breaker_owns_lease
1441 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1442 return false;
1443 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1444 rc = false;
1445 goto trace;
1446 }
1447 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1448 rc = false;
1449 goto trace;
1450 }
1451
1452 rc = locks_conflict(breaker, lease);
1453trace:
1454 trace_leases_conflict(rc, lease, breaker);
1455 return rc;
1456}
1457
1458static bool
1459any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1460{
1461 struct file_lock_context *ctx = inode->i_flctx;
1462 struct file_lock *fl;
1463
1464 lockdep_assert_held(&ctx->flc_lock);
1465
1466 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1467 if (leases_conflict(fl, breaker))
1468 return true;
1469 }
1470 return false;
1471}
1472
1473/**
1474 * __break_lease - revoke all outstanding leases on file
1475 * @inode: the inode of the file to return
1476 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1477 * break all leases
1478 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1479 * only delegations
1480 *
1481 * break_lease (inlined for speed) has checked there already is at least
1482 * some kind of lock (maybe a lease) on this file. Leases are broken on
1483 * a call to open() or truncate(). This function can sleep unless you
1484 * specified %O_NONBLOCK to your open().
1485 */
1486int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1487{
1488 int error = 0;
1489 struct file_lock_context *ctx;
1490 struct file_lock *new_fl, *fl, *tmp;
1491 unsigned long break_time;
1492 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1493 LIST_HEAD(dispose);
1494
1495 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1496 if (IS_ERR(new_fl))
1497 return PTR_ERR(new_fl);
1498 new_fl->fl_flags = type;
1499
1500 /* typically we will check that ctx is non-NULL before calling */
1501 ctx = locks_inode_context(inode);
1502 if (!ctx) {
1503 WARN_ON_ONCE(1);
1504 goto free_lock;
1505 }
1506
1507 percpu_down_read(&file_rwsem);
1508 spin_lock(&ctx->flc_lock);
1509
1510 time_out_leases(inode, &dispose);
1511
1512 if (!any_leases_conflict(inode, new_fl))
1513 goto out;
1514
1515 break_time = 0;
1516 if (lease_break_time > 0) {
1517 break_time = jiffies + lease_break_time * HZ;
1518 if (break_time == 0)
1519 break_time++; /* so that 0 means no break time */
1520 }
1521
1522 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1523 if (!leases_conflict(fl, new_fl))
1524 continue;
1525 if (want_write) {
1526 if (fl->fl_flags & FL_UNLOCK_PENDING)
1527 continue;
1528 fl->fl_flags |= FL_UNLOCK_PENDING;
1529 fl->fl_break_time = break_time;
1530 } else {
1531 if (lease_breaking(fl))
1532 continue;
1533 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1534 fl->fl_downgrade_time = break_time;
1535 }
1536 if (fl->fl_lmops->lm_break(fl))
1537 locks_delete_lock_ctx(fl, &dispose);
1538 }
1539
1540 if (list_empty(&ctx->flc_lease))
1541 goto out;
1542
1543 if (mode & O_NONBLOCK) {
1544 trace_break_lease_noblock(inode, new_fl);
1545 error = -EWOULDBLOCK;
1546 goto out;
1547 }
1548
1549restart:
1550 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1551 break_time = fl->fl_break_time;
1552 if (break_time != 0)
1553 break_time -= jiffies;
1554 if (break_time == 0)
1555 break_time++;
1556 locks_insert_block(fl, new_fl, leases_conflict);
1557 trace_break_lease_block(inode, new_fl);
1558 spin_unlock(&ctx->flc_lock);
1559 percpu_up_read(&file_rwsem);
1560
1561 locks_dispose_list(&dispose);
1562 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1563 list_empty(&new_fl->fl_blocked_member),
1564 break_time);
1565
1566 percpu_down_read(&file_rwsem);
1567 spin_lock(&ctx->flc_lock);
1568 trace_break_lease_unblock(inode, new_fl);
1569 locks_delete_block(new_fl);
1570 if (error >= 0) {
1571 /*
1572 * Wait for the next conflicting lease that has not been
1573 * broken yet
1574 */
1575 if (error == 0)
1576 time_out_leases(inode, &dispose);
1577 if (any_leases_conflict(inode, new_fl))
1578 goto restart;
1579 error = 0;
1580 }
1581out:
1582 spin_unlock(&ctx->flc_lock);
1583 percpu_up_read(&file_rwsem);
1584 locks_dispose_list(&dispose);
1585free_lock:
1586 locks_free_lock(new_fl);
1587 return error;
1588}
1589EXPORT_SYMBOL(__break_lease);
1590
1591/**
1592 * lease_get_mtime - update modified time of an inode with exclusive lease
1593 * @inode: the inode
1594 * @time: pointer to a timespec which contains the last modified time
1595 *
1596 * This is to force NFS clients to flush their caches for files with
1597 * exclusive leases. The justification is that if someone has an
1598 * exclusive lease, then they could be modifying it.
1599 */
1600void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1601{
1602 bool has_lease = false;
1603 struct file_lock_context *ctx;
1604 struct file_lock *fl;
1605
1606 ctx = locks_inode_context(inode);
1607 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1608 spin_lock(&ctx->flc_lock);
1609 fl = list_first_entry_or_null(&ctx->flc_lease,
1610 struct file_lock, fl_list);
1611 if (fl && (fl->fl_type == F_WRLCK))
1612 has_lease = true;
1613 spin_unlock(&ctx->flc_lock);
1614 }
1615
1616 if (has_lease)
1617 *time = current_time(inode);
1618}
1619EXPORT_SYMBOL(lease_get_mtime);
1620
1621/**
1622 * fcntl_getlease - Enquire what lease is currently active
1623 * @filp: the file
1624 *
1625 * The value returned by this function will be one of
1626 * (if no lease break is pending):
1627 *
1628 * %F_RDLCK to indicate a shared lease is held.
1629 *
1630 * %F_WRLCK to indicate an exclusive lease is held.
1631 *
1632 * %F_UNLCK to indicate no lease is held.
1633 *
1634 * (if a lease break is pending):
1635 *
1636 * %F_RDLCK to indicate an exclusive lease needs to be
1637 * changed to a shared lease (or removed).
1638 *
1639 * %F_UNLCK to indicate the lease needs to be removed.
1640 *
1641 * XXX: sfr & willy disagree over whether F_INPROGRESS
1642 * should be returned to userspace.
1643 */
1644int fcntl_getlease(struct file *filp)
1645{
1646 struct file_lock *fl;
1647 struct inode *inode = file_inode(filp);
1648 struct file_lock_context *ctx;
1649 int type = F_UNLCK;
1650 LIST_HEAD(dispose);
1651
1652 ctx = locks_inode_context(inode);
1653 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1654 percpu_down_read(&file_rwsem);
1655 spin_lock(&ctx->flc_lock);
1656 time_out_leases(inode, &dispose);
1657 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1658 if (fl->fl_file != filp)
1659 continue;
1660 type = target_leasetype(fl);
1661 break;
1662 }
1663 spin_unlock(&ctx->flc_lock);
1664 percpu_up_read(&file_rwsem);
1665
1666 locks_dispose_list(&dispose);
1667 }
1668 return type;
1669}
1670
1671/**
1672 * check_conflicting_open - see if the given file points to an inode that has
1673 * an existing open that would conflict with the
1674 * desired lease.
1675 * @filp: file to check
1676 * @arg: type of lease that we're trying to acquire
1677 * @flags: current lock flags
1678 *
1679 * Check to see if there's an existing open fd on this file that would
1680 * conflict with the lease we're trying to set.
1681 */
1682static int
1683check_conflicting_open(struct file *filp, const int arg, int flags)
1684{
1685 struct inode *inode = file_inode(filp);
1686 int self_wcount = 0, self_rcount = 0;
1687
1688 if (flags & FL_LAYOUT)
1689 return 0;
1690 if (flags & FL_DELEG)
1691 /* We leave these checks to the caller */
1692 return 0;
1693
1694 if (arg == F_RDLCK)
1695 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1696 else if (arg != F_WRLCK)
1697 return 0;
1698
1699 /*
1700 * Make sure that only read/write count is from lease requestor.
1701 * Note that this will result in denying write leases when i_writecount
1702 * is negative, which is what we want. (We shouldn't grant write leases
1703 * on files open for execution.)
1704 */
1705 if (filp->f_mode & FMODE_WRITE)
1706 self_wcount = 1;
1707 else if (filp->f_mode & FMODE_READ)
1708 self_rcount = 1;
1709
1710 if (atomic_read(&inode->i_writecount) != self_wcount ||
1711 atomic_read(&inode->i_readcount) != self_rcount)
1712 return -EAGAIN;
1713
1714 return 0;
1715}
1716
1717static int
1718generic_add_lease(struct file *filp, int arg, struct file_lock **flp, void **priv)
1719{
1720 struct file_lock *fl, *my_fl = NULL, *lease;
1721 struct inode *inode = file_inode(filp);
1722 struct file_lock_context *ctx;
1723 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1724 int error;
1725 LIST_HEAD(dispose);
1726
1727 lease = *flp;
1728 trace_generic_add_lease(inode, lease);
1729
1730 /* Note that arg is never F_UNLCK here */
1731 ctx = locks_get_lock_context(inode, arg);
1732 if (!ctx)
1733 return -ENOMEM;
1734
1735 /*
1736 * In the delegation case we need mutual exclusion with
1737 * a number of operations that take the i_mutex. We trylock
1738 * because delegations are an optional optimization, and if
1739 * there's some chance of a conflict--we'd rather not
1740 * bother, maybe that's a sign this just isn't a good file to
1741 * hand out a delegation on.
1742 */
1743 if (is_deleg && !inode_trylock(inode))
1744 return -EAGAIN;
1745
1746 percpu_down_read(&file_rwsem);
1747 spin_lock(&ctx->flc_lock);
1748 time_out_leases(inode, &dispose);
1749 error = check_conflicting_open(filp, arg, lease->fl_flags);
1750 if (error)
1751 goto out;
1752
1753 /*
1754 * At this point, we know that if there is an exclusive
1755 * lease on this file, then we hold it on this filp
1756 * (otherwise our open of this file would have blocked).
1757 * And if we are trying to acquire an exclusive lease,
1758 * then the file is not open by anyone (including us)
1759 * except for this filp.
1760 */
1761 error = -EAGAIN;
1762 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1763 if (fl->fl_file == filp &&
1764 fl->fl_owner == lease->fl_owner) {
1765 my_fl = fl;
1766 continue;
1767 }
1768
1769 /*
1770 * No exclusive leases if someone else has a lease on
1771 * this file:
1772 */
1773 if (arg == F_WRLCK)
1774 goto out;
1775 /*
1776 * Modifying our existing lease is OK, but no getting a
1777 * new lease if someone else is opening for write:
1778 */
1779 if (fl->fl_flags & FL_UNLOCK_PENDING)
1780 goto out;
1781 }
1782
1783 if (my_fl != NULL) {
1784 lease = my_fl;
1785 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1786 if (error)
1787 goto out;
1788 goto out_setup;
1789 }
1790
1791 error = -EINVAL;
1792 if (!leases_enable)
1793 goto out;
1794
1795 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1796 /*
1797 * The check in break_lease() is lockless. It's possible for another
1798 * open to race in after we did the earlier check for a conflicting
1799 * open but before the lease was inserted. Check again for a
1800 * conflicting open and cancel the lease if there is one.
1801 *
1802 * We also add a barrier here to ensure that the insertion of the lock
1803 * precedes these checks.
1804 */
1805 smp_mb();
1806 error = check_conflicting_open(filp, arg, lease->fl_flags);
1807 if (error) {
1808 locks_unlink_lock_ctx(lease);
1809 goto out;
1810 }
1811
1812out_setup:
1813 if (lease->fl_lmops->lm_setup)
1814 lease->fl_lmops->lm_setup(lease, priv);
1815out:
1816 spin_unlock(&ctx->flc_lock);
1817 percpu_up_read(&file_rwsem);
1818 locks_dispose_list(&dispose);
1819 if (is_deleg)
1820 inode_unlock(inode);
1821 if (!error && !my_fl)
1822 *flp = NULL;
1823 return error;
1824}
1825
1826static int generic_delete_lease(struct file *filp, void *owner)
1827{
1828 int error = -EAGAIN;
1829 struct file_lock *fl, *victim = NULL;
1830 struct inode *inode = file_inode(filp);
1831 struct file_lock_context *ctx;
1832 LIST_HEAD(dispose);
1833
1834 ctx = locks_inode_context(inode);
1835 if (!ctx) {
1836 trace_generic_delete_lease(inode, NULL);
1837 return error;
1838 }
1839
1840 percpu_down_read(&file_rwsem);
1841 spin_lock(&ctx->flc_lock);
1842 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1843 if (fl->fl_file == filp &&
1844 fl->fl_owner == owner) {
1845 victim = fl;
1846 break;
1847 }
1848 }
1849 trace_generic_delete_lease(inode, victim);
1850 if (victim)
1851 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1852 spin_unlock(&ctx->flc_lock);
1853 percpu_up_read(&file_rwsem);
1854 locks_dispose_list(&dispose);
1855 return error;
1856}
1857
1858/**
1859 * generic_setlease - sets a lease on an open file
1860 * @filp: file pointer
1861 * @arg: type of lease to obtain
1862 * @flp: input - file_lock to use, output - file_lock inserted
1863 * @priv: private data for lm_setup (may be NULL if lm_setup
1864 * doesn't require it)
1865 *
1866 * The (input) flp->fl_lmops->lm_break function is required
1867 * by break_lease().
1868 */
1869int generic_setlease(struct file *filp, int arg, struct file_lock **flp,
1870 void **priv)
1871{
1872 struct inode *inode = file_inode(filp);
1873 vfsuid_t vfsuid = i_uid_into_vfsuid(file_mnt_idmap(filp), inode);
1874 int error;
1875
1876 if ((!vfsuid_eq_kuid(vfsuid, current_fsuid())) && !capable(CAP_LEASE))
1877 return -EACCES;
1878 if (!S_ISREG(inode->i_mode))
1879 return -EINVAL;
1880 error = security_file_lock(filp, arg);
1881 if (error)
1882 return error;
1883
1884 switch (arg) {
1885 case F_UNLCK:
1886 return generic_delete_lease(filp, *priv);
1887 case F_RDLCK:
1888 case F_WRLCK:
1889 if (!(*flp)->fl_lmops->lm_break) {
1890 WARN_ON_ONCE(1);
1891 return -ENOLCK;
1892 }
1893
1894 return generic_add_lease(filp, arg, flp, priv);
1895 default:
1896 return -EINVAL;
1897 }
1898}
1899EXPORT_SYMBOL(generic_setlease);
1900
1901/*
1902 * Kernel subsystems can register to be notified on any attempt to set
1903 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1904 * to close files that it may have cached when there is an attempt to set a
1905 * conflicting lease.
1906 */
1907static struct srcu_notifier_head lease_notifier_chain;
1908
1909static inline void
1910lease_notifier_chain_init(void)
1911{
1912 srcu_init_notifier_head(&lease_notifier_chain);
1913}
1914
1915static inline void
1916setlease_notifier(int arg, struct file_lock *lease)
1917{
1918 if (arg != F_UNLCK)
1919 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1920}
1921
1922int lease_register_notifier(struct notifier_block *nb)
1923{
1924 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1925}
1926EXPORT_SYMBOL_GPL(lease_register_notifier);
1927
1928void lease_unregister_notifier(struct notifier_block *nb)
1929{
1930 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1931}
1932EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1933
1934/**
1935 * vfs_setlease - sets a lease on an open file
1936 * @filp: file pointer
1937 * @arg: type of lease to obtain
1938 * @lease: file_lock to use when adding a lease
1939 * @priv: private info for lm_setup when adding a lease (may be
1940 * NULL if lm_setup doesn't require it)
1941 *
1942 * Call this to establish a lease on the file. The "lease" argument is not
1943 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1944 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1945 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1946 * stack trace).
1947 *
1948 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1949 * may be NULL if the lm_setup operation doesn't require it.
1950 */
1951int
1952vfs_setlease(struct file *filp, int arg, struct file_lock **lease, void **priv)
1953{
1954 if (lease)
1955 setlease_notifier(arg, *lease);
1956 if (filp->f_op->setlease)
1957 return filp->f_op->setlease(filp, arg, lease, priv);
1958 else
1959 return generic_setlease(filp, arg, lease, priv);
1960}
1961EXPORT_SYMBOL_GPL(vfs_setlease);
1962
1963static int do_fcntl_add_lease(unsigned int fd, struct file *filp, int arg)
1964{
1965 struct file_lock *fl;
1966 struct fasync_struct *new;
1967 int error;
1968
1969 fl = lease_alloc(filp, arg);
1970 if (IS_ERR(fl))
1971 return PTR_ERR(fl);
1972
1973 new = fasync_alloc();
1974 if (!new) {
1975 locks_free_lock(fl);
1976 return -ENOMEM;
1977 }
1978 new->fa_fd = fd;
1979
1980 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1981 if (fl)
1982 locks_free_lock(fl);
1983 if (new)
1984 fasync_free(new);
1985 return error;
1986}
1987
1988/**
1989 * fcntl_setlease - sets a lease on an open file
1990 * @fd: open file descriptor
1991 * @filp: file pointer
1992 * @arg: type of lease to obtain
1993 *
1994 * Call this fcntl to establish a lease on the file.
1995 * Note that you also need to call %F_SETSIG to
1996 * receive a signal when the lease is broken.
1997 */
1998int fcntl_setlease(unsigned int fd, struct file *filp, int arg)
1999{
2000 if (arg == F_UNLCK)
2001 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2002 return do_fcntl_add_lease(fd, filp, arg);
2003}
2004
2005/**
2006 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2007 * @inode: inode of the file to apply to
2008 * @fl: The lock to be applied
2009 *
2010 * Apply a FLOCK style lock request to an inode.
2011 */
2012static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2013{
2014 int error;
2015 might_sleep();
2016 for (;;) {
2017 error = flock_lock_inode(inode, fl);
2018 if (error != FILE_LOCK_DEFERRED)
2019 break;
2020 error = wait_event_interruptible(fl->fl_wait,
2021 list_empty(&fl->fl_blocked_member));
2022 if (error)
2023 break;
2024 }
2025 locks_delete_block(fl);
2026 return error;
2027}
2028
2029/**
2030 * locks_lock_inode_wait - Apply a lock to an inode
2031 * @inode: inode of the file to apply to
2032 * @fl: The lock to be applied
2033 *
2034 * Apply a POSIX or FLOCK style lock request to an inode.
2035 */
2036int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2037{
2038 int res = 0;
2039 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2040 case FL_POSIX:
2041 res = posix_lock_inode_wait(inode, fl);
2042 break;
2043 case FL_FLOCK:
2044 res = flock_lock_inode_wait(inode, fl);
2045 break;
2046 default:
2047 BUG();
2048 }
2049 return res;
2050}
2051EXPORT_SYMBOL(locks_lock_inode_wait);
2052
2053/**
2054 * sys_flock: - flock() system call.
2055 * @fd: the file descriptor to lock.
2056 * @cmd: the type of lock to apply.
2057 *
2058 * Apply a %FL_FLOCK style lock to an open file descriptor.
2059 * The @cmd can be one of:
2060 *
2061 * - %LOCK_SH -- a shared lock.
2062 * - %LOCK_EX -- an exclusive lock.
2063 * - %LOCK_UN -- remove an existing lock.
2064 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2065 *
2066 * %LOCK_MAND support has been removed from the kernel.
2067 */
2068SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2069{
2070 int can_sleep, error, type;
2071 struct file_lock fl;
2072 struct fd f;
2073
2074 /*
2075 * LOCK_MAND locks were broken for a long time in that they never
2076 * conflicted with one another and didn't prevent any sort of open,
2077 * read or write activity.
2078 *
2079 * Just ignore these requests now, to preserve legacy behavior, but
2080 * throw a warning to let people know that they don't actually work.
2081 */
2082 if (cmd & LOCK_MAND) {
2083 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);
2084 return 0;
2085 }
2086
2087 type = flock_translate_cmd(cmd & ~LOCK_NB);
2088 if (type < 0)
2089 return type;
2090
2091 error = -EBADF;
2092 f = fdget(fd);
2093 if (!f.file)
2094 return error;
2095
2096 if (type != F_UNLCK && !(f.file->f_mode & (FMODE_READ | FMODE_WRITE)))
2097 goto out_putf;
2098
2099 flock_make_lock(f.file, &fl, type);
2100
2101 error = security_file_lock(f.file, fl.fl_type);
2102 if (error)
2103 goto out_putf;
2104
2105 can_sleep = !(cmd & LOCK_NB);
2106 if (can_sleep)
2107 fl.fl_flags |= FL_SLEEP;
2108
2109 if (f.file->f_op->flock)
2110 error = f.file->f_op->flock(f.file,
2111 (can_sleep) ? F_SETLKW : F_SETLK,
2112 &fl);
2113 else
2114 error = locks_lock_file_wait(f.file, &fl);
2115
2116 locks_release_private(&fl);
2117 out_putf:
2118 fdput(f);
2119
2120 return error;
2121}
2122
2123/**
2124 * vfs_test_lock - test file byte range lock
2125 * @filp: The file to test lock for
2126 * @fl: The lock to test; also used to hold result
2127 *
2128 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2129 * setting conf->fl_type to something other than F_UNLCK.
2130 */
2131int vfs_test_lock(struct file *filp, struct file_lock *fl)
2132{
2133 WARN_ON_ONCE(filp != fl->fl_file);
2134 if (filp->f_op->lock)
2135 return filp->f_op->lock(filp, F_GETLK, fl);
2136 posix_test_lock(filp, fl);
2137 return 0;
2138}
2139EXPORT_SYMBOL_GPL(vfs_test_lock);
2140
2141/**
2142 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2143 * @fl: The file_lock who's fl_pid should be translated
2144 * @ns: The namespace into which the pid should be translated
2145 *
2146 * Used to translate a fl_pid into a namespace virtual pid number
2147 */
2148static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2149{
2150 pid_t vnr;
2151 struct pid *pid;
2152
2153 if (IS_OFDLCK(fl))
2154 return -1;
2155 if (IS_REMOTELCK(fl))
2156 return fl->fl_pid;
2157 /*
2158 * If the flock owner process is dead and its pid has been already
2159 * freed, the translation below won't work, but we still want to show
2160 * flock owner pid number in init pidns.
2161 */
2162 if (ns == &init_pid_ns)
2163 return (pid_t)fl->fl_pid;
2164
2165 rcu_read_lock();
2166 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2167 vnr = pid_nr_ns(pid, ns);
2168 rcu_read_unlock();
2169 return vnr;
2170}
2171
2172static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2173{
2174 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2175#if BITS_PER_LONG == 32
2176 /*
2177 * Make sure we can represent the posix lock via
2178 * legacy 32bit flock.
2179 */
2180 if (fl->fl_start > OFFT_OFFSET_MAX)
2181 return -EOVERFLOW;
2182 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2183 return -EOVERFLOW;
2184#endif
2185 flock->l_start = fl->fl_start;
2186 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2187 fl->fl_end - fl->fl_start + 1;
2188 flock->l_whence = 0;
2189 flock->l_type = fl->fl_type;
2190 return 0;
2191}
2192
2193#if BITS_PER_LONG == 32
2194static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2195{
2196 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2197 flock->l_start = fl->fl_start;
2198 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2199 fl->fl_end - fl->fl_start + 1;
2200 flock->l_whence = 0;
2201 flock->l_type = fl->fl_type;
2202}
2203#endif
2204
2205/* Report the first existing lock that would conflict with l.
2206 * This implements the F_GETLK command of fcntl().
2207 */
2208int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2209{
2210 struct file_lock *fl;
2211 int error;
2212
2213 fl = locks_alloc_lock();
2214 if (fl == NULL)
2215 return -ENOMEM;
2216 error = -EINVAL;
2217 if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK
2218 && flock->l_type != F_WRLCK)
2219 goto out;
2220
2221 error = flock_to_posix_lock(filp, fl, flock);
2222 if (error)
2223 goto out;
2224
2225 if (cmd == F_OFD_GETLK) {
2226 error = -EINVAL;
2227 if (flock->l_pid != 0)
2228 goto out;
2229
2230 fl->fl_flags |= FL_OFDLCK;
2231 fl->fl_owner = filp;
2232 }
2233
2234 error = vfs_test_lock(filp, fl);
2235 if (error)
2236 goto out;
2237
2238 flock->l_type = fl->fl_type;
2239 if (fl->fl_type != F_UNLCK) {
2240 error = posix_lock_to_flock(flock, fl);
2241 if (error)
2242 goto out;
2243 }
2244out:
2245 locks_free_lock(fl);
2246 return error;
2247}
2248
2249/**
2250 * vfs_lock_file - file byte range lock
2251 * @filp: The file to apply the lock to
2252 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2253 * @fl: The lock to be applied
2254 * @conf: Place to return a copy of the conflicting lock, if found.
2255 *
2256 * A caller that doesn't care about the conflicting lock may pass NULL
2257 * as the final argument.
2258 *
2259 * If the filesystem defines a private ->lock() method, then @conf will
2260 * be left unchanged; so a caller that cares should initialize it to
2261 * some acceptable default.
2262 *
2263 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2264 * locks, the ->lock() interface may return asynchronously, before the lock has
2265 * been granted or denied by the underlying filesystem, if (and only if)
2266 * lm_grant is set. Additionally EXPORT_OP_ASYNC_LOCK in export_operations
2267 * flags need to be set.
2268 *
2269 * Callers expecting ->lock() to return asynchronously will only use F_SETLK,
2270 * not F_SETLKW; they will set FL_SLEEP if (and only if) the request is for a
2271 * blocking lock. When ->lock() does return asynchronously, it must return
2272 * FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock request completes.
2273 * If the request is for non-blocking lock the file system should return
2274 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2275 * with the result. If the request timed out the callback routine will return a
2276 * nonzero return code and the file system should release the lock. The file
2277 * system is also responsible to keep a corresponding posix lock when it
2278 * grants a lock so the VFS can find out which locks are locally held and do
2279 * the correct lock cleanup when required.
2280 * The underlying filesystem must not drop the kernel lock or call
2281 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2282 * return code.
2283 */
2284int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2285{
2286 WARN_ON_ONCE(filp != fl->fl_file);
2287 if (filp->f_op->lock)
2288 return filp->f_op->lock(filp, cmd, fl);
2289 else
2290 return posix_lock_file(filp, fl, conf);
2291}
2292EXPORT_SYMBOL_GPL(vfs_lock_file);
2293
2294static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2295 struct file_lock *fl)
2296{
2297 int error;
2298
2299 error = security_file_lock(filp, fl->fl_type);
2300 if (error)
2301 return error;
2302
2303 for (;;) {
2304 error = vfs_lock_file(filp, cmd, fl, NULL);
2305 if (error != FILE_LOCK_DEFERRED)
2306 break;
2307 error = wait_event_interruptible(fl->fl_wait,
2308 list_empty(&fl->fl_blocked_member));
2309 if (error)
2310 break;
2311 }
2312 locks_delete_block(fl);
2313
2314 return error;
2315}
2316
2317/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2318static int
2319check_fmode_for_setlk(struct file_lock *fl)
2320{
2321 switch (fl->fl_type) {
2322 case F_RDLCK:
2323 if (!(fl->fl_file->f_mode & FMODE_READ))
2324 return -EBADF;
2325 break;
2326 case F_WRLCK:
2327 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2328 return -EBADF;
2329 }
2330 return 0;
2331}
2332
2333/* Apply the lock described by l to an open file descriptor.
2334 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2335 */
2336int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2337 struct flock *flock)
2338{
2339 struct file_lock *file_lock = locks_alloc_lock();
2340 struct inode *inode = file_inode(filp);
2341 struct file *f;
2342 int error;
2343
2344 if (file_lock == NULL)
2345 return -ENOLCK;
2346
2347 error = flock_to_posix_lock(filp, file_lock, flock);
2348 if (error)
2349 goto out;
2350
2351 error = check_fmode_for_setlk(file_lock);
2352 if (error)
2353 goto out;
2354
2355 /*
2356 * If the cmd is requesting file-private locks, then set the
2357 * FL_OFDLCK flag and override the owner.
2358 */
2359 switch (cmd) {
2360 case F_OFD_SETLK:
2361 error = -EINVAL;
2362 if (flock->l_pid != 0)
2363 goto out;
2364
2365 cmd = F_SETLK;
2366 file_lock->fl_flags |= FL_OFDLCK;
2367 file_lock->fl_owner = filp;
2368 break;
2369 case F_OFD_SETLKW:
2370 error = -EINVAL;
2371 if (flock->l_pid != 0)
2372 goto out;
2373
2374 cmd = F_SETLKW;
2375 file_lock->fl_flags |= FL_OFDLCK;
2376 file_lock->fl_owner = filp;
2377 fallthrough;
2378 case F_SETLKW:
2379 file_lock->fl_flags |= FL_SLEEP;
2380 }
2381
2382 error = do_lock_file_wait(filp, cmd, file_lock);
2383
2384 /*
2385 * Attempt to detect a close/fcntl race and recover by releasing the
2386 * lock that was just acquired. There is no need to do that when we're
2387 * unlocking though, or for OFD locks.
2388 */
2389 if (!error && file_lock->fl_type != F_UNLCK &&
2390 !(file_lock->fl_flags & FL_OFDLCK)) {
2391 struct files_struct *files = current->files;
2392 /*
2393 * We need that spin_lock here - it prevents reordering between
2394 * update of i_flctx->flc_posix and check for it done in
2395 * close(). rcu_read_lock() wouldn't do.
2396 */
2397 spin_lock(&files->file_lock);
2398 f = files_lookup_fd_locked(files, fd);
2399 spin_unlock(&files->file_lock);
2400 if (f != filp) {
2401 file_lock->fl_type = F_UNLCK;
2402 error = do_lock_file_wait(filp, cmd, file_lock);
2403 WARN_ON_ONCE(error);
2404 error = -EBADF;
2405 }
2406 }
2407out:
2408 trace_fcntl_setlk(inode, file_lock, error);
2409 locks_free_lock(file_lock);
2410 return error;
2411}
2412
2413#if BITS_PER_LONG == 32
2414/* Report the first existing lock that would conflict with l.
2415 * This implements the F_GETLK command of fcntl().
2416 */
2417int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2418{
2419 struct file_lock *fl;
2420 int error;
2421
2422 fl = locks_alloc_lock();
2423 if (fl == NULL)
2424 return -ENOMEM;
2425
2426 error = -EINVAL;
2427 if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK
2428 && flock->l_type != F_WRLCK)
2429 goto out;
2430
2431 error = flock64_to_posix_lock(filp, fl, flock);
2432 if (error)
2433 goto out;
2434
2435 if (cmd == F_OFD_GETLK) {
2436 error = -EINVAL;
2437 if (flock->l_pid != 0)
2438 goto out;
2439
2440 fl->fl_flags |= FL_OFDLCK;
2441 fl->fl_owner = filp;
2442 }
2443
2444 error = vfs_test_lock(filp, fl);
2445 if (error)
2446 goto out;
2447
2448 flock->l_type = fl->fl_type;
2449 if (fl->fl_type != F_UNLCK)
2450 posix_lock_to_flock64(flock, fl);
2451
2452out:
2453 locks_free_lock(fl);
2454 return error;
2455}
2456
2457/* Apply the lock described by l to an open file descriptor.
2458 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2459 */
2460int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2461 struct flock64 *flock)
2462{
2463 struct file_lock *file_lock = locks_alloc_lock();
2464 struct file *f;
2465 int error;
2466
2467 if (file_lock == NULL)
2468 return -ENOLCK;
2469
2470 error = flock64_to_posix_lock(filp, file_lock, flock);
2471 if (error)
2472 goto out;
2473
2474 error = check_fmode_for_setlk(file_lock);
2475 if (error)
2476 goto out;
2477
2478 /*
2479 * If the cmd is requesting file-private locks, then set the
2480 * FL_OFDLCK flag and override the owner.
2481 */
2482 switch (cmd) {
2483 case F_OFD_SETLK:
2484 error = -EINVAL;
2485 if (flock->l_pid != 0)
2486 goto out;
2487
2488 cmd = F_SETLK64;
2489 file_lock->fl_flags |= FL_OFDLCK;
2490 file_lock->fl_owner = filp;
2491 break;
2492 case F_OFD_SETLKW:
2493 error = -EINVAL;
2494 if (flock->l_pid != 0)
2495 goto out;
2496
2497 cmd = F_SETLKW64;
2498 file_lock->fl_flags |= FL_OFDLCK;
2499 file_lock->fl_owner = filp;
2500 fallthrough;
2501 case F_SETLKW64:
2502 file_lock->fl_flags |= FL_SLEEP;
2503 }
2504
2505 error = do_lock_file_wait(filp, cmd, file_lock);
2506
2507 /*
2508 * Attempt to detect a close/fcntl race and recover by releasing the
2509 * lock that was just acquired. There is no need to do that when we're
2510 * unlocking though, or for OFD locks.
2511 */
2512 if (!error && file_lock->fl_type != F_UNLCK &&
2513 !(file_lock->fl_flags & FL_OFDLCK)) {
2514 struct files_struct *files = current->files;
2515 /*
2516 * We need that spin_lock here - it prevents reordering between
2517 * update of i_flctx->flc_posix and check for it done in
2518 * close(). rcu_read_lock() wouldn't do.
2519 */
2520 spin_lock(&files->file_lock);
2521 f = files_lookup_fd_locked(files, fd);
2522 spin_unlock(&files->file_lock);
2523 if (f != filp) {
2524 file_lock->fl_type = F_UNLCK;
2525 error = do_lock_file_wait(filp, cmd, file_lock);
2526 WARN_ON_ONCE(error);
2527 error = -EBADF;
2528 }
2529 }
2530out:
2531 locks_free_lock(file_lock);
2532 return error;
2533}
2534#endif /* BITS_PER_LONG == 32 */
2535
2536/*
2537 * This function is called when the file is being removed
2538 * from the task's fd array. POSIX locks belonging to this task
2539 * are deleted at this time.
2540 */
2541void locks_remove_posix(struct file *filp, fl_owner_t owner)
2542{
2543 int error;
2544 struct inode *inode = file_inode(filp);
2545 struct file_lock lock;
2546 struct file_lock_context *ctx;
2547
2548 /*
2549 * If there are no locks held on this file, we don't need to call
2550 * posix_lock_file(). Another process could be setting a lock on this
2551 * file at the same time, but we wouldn't remove that lock anyway.
2552 */
2553 ctx = locks_inode_context(inode);
2554 if (!ctx || list_empty(&ctx->flc_posix))
2555 return;
2556
2557 locks_init_lock(&lock);
2558 lock.fl_type = F_UNLCK;
2559 lock.fl_flags = FL_POSIX | FL_CLOSE;
2560 lock.fl_start = 0;
2561 lock.fl_end = OFFSET_MAX;
2562 lock.fl_owner = owner;
2563 lock.fl_pid = current->tgid;
2564 lock.fl_file = filp;
2565 lock.fl_ops = NULL;
2566 lock.fl_lmops = NULL;
2567
2568 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2569
2570 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2571 lock.fl_ops->fl_release_private(&lock);
2572 trace_locks_remove_posix(inode, &lock, error);
2573}
2574EXPORT_SYMBOL(locks_remove_posix);
2575
2576/* The i_flctx must be valid when calling into here */
2577static void
2578locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2579{
2580 struct file_lock fl;
2581 struct inode *inode = file_inode(filp);
2582
2583 if (list_empty(&flctx->flc_flock))
2584 return;
2585
2586 flock_make_lock(filp, &fl, F_UNLCK);
2587 fl.fl_flags |= FL_CLOSE;
2588
2589 if (filp->f_op->flock)
2590 filp->f_op->flock(filp, F_SETLKW, &fl);
2591 else
2592 flock_lock_inode(inode, &fl);
2593
2594 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2595 fl.fl_ops->fl_release_private(&fl);
2596}
2597
2598/* The i_flctx must be valid when calling into here */
2599static void
2600locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2601{
2602 struct file_lock *fl, *tmp;
2603 LIST_HEAD(dispose);
2604
2605 if (list_empty(&ctx->flc_lease))
2606 return;
2607
2608 percpu_down_read(&file_rwsem);
2609 spin_lock(&ctx->flc_lock);
2610 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2611 if (filp == fl->fl_file)
2612 lease_modify(fl, F_UNLCK, &dispose);
2613 spin_unlock(&ctx->flc_lock);
2614 percpu_up_read(&file_rwsem);
2615
2616 locks_dispose_list(&dispose);
2617}
2618
2619/*
2620 * This function is called on the last close of an open file.
2621 */
2622void locks_remove_file(struct file *filp)
2623{
2624 struct file_lock_context *ctx;
2625
2626 ctx = locks_inode_context(file_inode(filp));
2627 if (!ctx)
2628 return;
2629
2630 /* remove any OFD locks */
2631 locks_remove_posix(filp, filp);
2632
2633 /* remove flock locks */
2634 locks_remove_flock(filp, ctx);
2635
2636 /* remove any leases */
2637 locks_remove_lease(filp, ctx);
2638
2639 spin_lock(&ctx->flc_lock);
2640 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2641 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2642 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2643 spin_unlock(&ctx->flc_lock);
2644}
2645
2646/**
2647 * vfs_cancel_lock - file byte range unblock lock
2648 * @filp: The file to apply the unblock to
2649 * @fl: The lock to be unblocked
2650 *
2651 * Used by lock managers to cancel blocked requests
2652 */
2653int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2654{
2655 WARN_ON_ONCE(filp != fl->fl_file);
2656 if (filp->f_op->lock)
2657 return filp->f_op->lock(filp, F_CANCELLK, fl);
2658 return 0;
2659}
2660EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2661
2662/**
2663 * vfs_inode_has_locks - are any file locks held on @inode?
2664 * @inode: inode to check for locks
2665 *
2666 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2667 * set on @inode.
2668 */
2669bool vfs_inode_has_locks(struct inode *inode)
2670{
2671 struct file_lock_context *ctx;
2672 bool ret;
2673
2674 ctx = locks_inode_context(inode);
2675 if (!ctx)
2676 return false;
2677
2678 spin_lock(&ctx->flc_lock);
2679 ret = !list_empty(&ctx->flc_posix) || !list_empty(&ctx->flc_flock);
2680 spin_unlock(&ctx->flc_lock);
2681 return ret;
2682}
2683EXPORT_SYMBOL_GPL(vfs_inode_has_locks);
2684
2685#ifdef CONFIG_PROC_FS
2686#include <linux/proc_fs.h>
2687#include <linux/seq_file.h>
2688
2689struct locks_iterator {
2690 int li_cpu;
2691 loff_t li_pos;
2692};
2693
2694static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2695 loff_t id, char *pfx, int repeat)
2696{
2697 struct inode *inode = NULL;
2698 unsigned int fl_pid;
2699 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2700 int type;
2701
2702 fl_pid = locks_translate_pid(fl, proc_pidns);
2703 /*
2704 * If lock owner is dead (and pid is freed) or not visible in current
2705 * pidns, zero is shown as a pid value. Check lock info from
2706 * init_pid_ns to get saved lock pid value.
2707 */
2708
2709 if (fl->fl_file != NULL)
2710 inode = file_inode(fl->fl_file);
2711
2712 seq_printf(f, "%lld: ", id);
2713
2714 if (repeat)
2715 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2716
2717 if (IS_POSIX(fl)) {
2718 if (fl->fl_flags & FL_ACCESS)
2719 seq_puts(f, "ACCESS");
2720 else if (IS_OFDLCK(fl))
2721 seq_puts(f, "OFDLCK");
2722 else
2723 seq_puts(f, "POSIX ");
2724
2725 seq_printf(f, " %s ",
2726 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2727 } else if (IS_FLOCK(fl)) {
2728 seq_puts(f, "FLOCK ADVISORY ");
2729 } else if (IS_LEASE(fl)) {
2730 if (fl->fl_flags & FL_DELEG)
2731 seq_puts(f, "DELEG ");
2732 else
2733 seq_puts(f, "LEASE ");
2734
2735 if (lease_breaking(fl))
2736 seq_puts(f, "BREAKING ");
2737 else if (fl->fl_file)
2738 seq_puts(f, "ACTIVE ");
2739 else
2740 seq_puts(f, "BREAKER ");
2741 } else {
2742 seq_puts(f, "UNKNOWN UNKNOWN ");
2743 }
2744 type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2745
2746 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2747 (type == F_RDLCK) ? "READ" : "UNLCK");
2748 if (inode) {
2749 /* userspace relies on this representation of dev_t */
2750 seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2751 MAJOR(inode->i_sb->s_dev),
2752 MINOR(inode->i_sb->s_dev), inode->i_ino);
2753 } else {
2754 seq_printf(f, "%d <none>:0 ", fl_pid);
2755 }
2756 if (IS_POSIX(fl)) {
2757 if (fl->fl_end == OFFSET_MAX)
2758 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2759 else
2760 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2761 } else {
2762 seq_puts(f, "0 EOF\n");
2763 }
2764}
2765
2766static struct file_lock *get_next_blocked_member(struct file_lock *node)
2767{
2768 struct file_lock *tmp;
2769
2770 /* NULL node or root node */
2771 if (node == NULL || node->fl_blocker == NULL)
2772 return NULL;
2773
2774 /* Next member in the linked list could be itself */
2775 tmp = list_next_entry(node, fl_blocked_member);
2776 if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2777 || tmp == node) {
2778 return NULL;
2779 }
2780
2781 return tmp;
2782}
2783
2784static int locks_show(struct seq_file *f, void *v)
2785{
2786 struct locks_iterator *iter = f->private;
2787 struct file_lock *cur, *tmp;
2788 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2789 int level = 0;
2790
2791 cur = hlist_entry(v, struct file_lock, fl_link);
2792
2793 if (locks_translate_pid(cur, proc_pidns) == 0)
2794 return 0;
2795
2796 /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2797 * is the left child of current node, the next silibing in fl_blocked_member is the
2798 * right child, we can alse get the parent of current node from fl_blocker, so this
2799 * question becomes traversal of a binary tree
2800 */
2801 while (cur != NULL) {
2802 if (level)
2803 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2804 else
2805 lock_get_status(f, cur, iter->li_pos, "", level);
2806
2807 if (!list_empty(&cur->fl_blocked_requests)) {
2808 /* Turn left */
2809 cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2810 struct file_lock, fl_blocked_member);
2811 level++;
2812 } else {
2813 /* Turn right */
2814 tmp = get_next_blocked_member(cur);
2815 /* Fall back to parent node */
2816 while (tmp == NULL && cur->fl_blocker != NULL) {
2817 cur = cur->fl_blocker;
2818 level--;
2819 tmp = get_next_blocked_member(cur);
2820 }
2821 cur = tmp;
2822 }
2823 }
2824
2825 return 0;
2826}
2827
2828static void __show_fd_locks(struct seq_file *f,
2829 struct list_head *head, int *id,
2830 struct file *filp, struct files_struct *files)
2831{
2832 struct file_lock *fl;
2833
2834 list_for_each_entry(fl, head, fl_list) {
2835
2836 if (filp != fl->fl_file)
2837 continue;
2838 if (fl->fl_owner != files &&
2839 fl->fl_owner != filp)
2840 continue;
2841
2842 (*id)++;
2843 seq_puts(f, "lock:\t");
2844 lock_get_status(f, fl, *id, "", 0);
2845 }
2846}
2847
2848void show_fd_locks(struct seq_file *f,
2849 struct file *filp, struct files_struct *files)
2850{
2851 struct inode *inode = file_inode(filp);
2852 struct file_lock_context *ctx;
2853 int id = 0;
2854
2855 ctx = locks_inode_context(inode);
2856 if (!ctx)
2857 return;
2858
2859 spin_lock(&ctx->flc_lock);
2860 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2861 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2862 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2863 spin_unlock(&ctx->flc_lock);
2864}
2865
2866static void *locks_start(struct seq_file *f, loff_t *pos)
2867 __acquires(&blocked_lock_lock)
2868{
2869 struct locks_iterator *iter = f->private;
2870
2871 iter->li_pos = *pos + 1;
2872 percpu_down_write(&file_rwsem);
2873 spin_lock(&blocked_lock_lock);
2874 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2875}
2876
2877static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2878{
2879 struct locks_iterator *iter = f->private;
2880
2881 ++iter->li_pos;
2882 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2883}
2884
2885static void locks_stop(struct seq_file *f, void *v)
2886 __releases(&blocked_lock_lock)
2887{
2888 spin_unlock(&blocked_lock_lock);
2889 percpu_up_write(&file_rwsem);
2890}
2891
2892static const struct seq_operations locks_seq_operations = {
2893 .start = locks_start,
2894 .next = locks_next,
2895 .stop = locks_stop,
2896 .show = locks_show,
2897};
2898
2899static int __init proc_locks_init(void)
2900{
2901 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2902 sizeof(struct locks_iterator), NULL);
2903 return 0;
2904}
2905fs_initcall(proc_locks_init);
2906#endif
2907
2908static int __init filelock_init(void)
2909{
2910 int i;
2911
2912 flctx_cache = kmem_cache_create("file_lock_ctx",
2913 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2914
2915 filelock_cache = kmem_cache_create("file_lock_cache",
2916 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2917
2918 for_each_possible_cpu(i) {
2919 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2920
2921 spin_lock_init(&fll->lock);
2922 INIT_HLIST_HEAD(&fll->hlist);
2923 }
2924
2925 lease_notifier_chain_init();
2926 return 0;
2927}
2928core_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 trace_posix_lock_inode(inode, request, error);
1371 spin_unlock(&ctx->flc_lock);
1372 percpu_up_read(&file_rwsem);
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 file_f_owner(filp)->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 error = file_f_owner_allocate(filp);
1787 if (error)
1788 return error;
1789
1790 /* Note that arg is never F_UNLCK here */
1791 ctx = locks_get_lock_context(inode, arg);
1792 if (!ctx)
1793 return -ENOMEM;
1794
1795 /*
1796 * In the delegation case we need mutual exclusion with
1797 * a number of operations that take the i_mutex. We trylock
1798 * because delegations are an optional optimization, and if
1799 * there's some chance of a conflict--we'd rather not
1800 * bother, maybe that's a sign this just isn't a good file to
1801 * hand out a delegation on.
1802 */
1803 if (is_deleg && !inode_trylock(inode))
1804 return -EAGAIN;
1805
1806 percpu_down_read(&file_rwsem);
1807 spin_lock(&ctx->flc_lock);
1808 time_out_leases(inode, &dispose);
1809 error = check_conflicting_open(filp, arg, lease->c.flc_flags);
1810 if (error)
1811 goto out;
1812
1813 /*
1814 * At this point, we know that if there is an exclusive
1815 * lease on this file, then we hold it on this filp
1816 * (otherwise our open of this file would have blocked).
1817 * And if we are trying to acquire an exclusive lease,
1818 * then the file is not open by anyone (including us)
1819 * except for this filp.
1820 */
1821 error = -EAGAIN;
1822 list_for_each_entry(fl, &ctx->flc_lease, c.flc_list) {
1823 if (fl->c.flc_file == filp &&
1824 fl->c.flc_owner == lease->c.flc_owner) {
1825 my_fl = fl;
1826 continue;
1827 }
1828
1829 /*
1830 * No exclusive leases if someone else has a lease on
1831 * this file:
1832 */
1833 if (arg == F_WRLCK)
1834 goto out;
1835 /*
1836 * Modifying our existing lease is OK, but no getting a
1837 * new lease if someone else is opening for write:
1838 */
1839 if (fl->c.flc_flags & FL_UNLOCK_PENDING)
1840 goto out;
1841 }
1842
1843 if (my_fl != NULL) {
1844 lease = my_fl;
1845 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1846 if (error)
1847 goto out;
1848 goto out_setup;
1849 }
1850
1851 error = -EINVAL;
1852 if (!leases_enable)
1853 goto out;
1854
1855 locks_insert_lock_ctx(&lease->c, &ctx->flc_lease);
1856 /*
1857 * The check in break_lease() is lockless. It's possible for another
1858 * open to race in after we did the earlier check for a conflicting
1859 * open but before the lease was inserted. Check again for a
1860 * conflicting open and cancel the lease if there is one.
1861 *
1862 * We also add a barrier here to ensure that the insertion of the lock
1863 * precedes these checks.
1864 */
1865 smp_mb();
1866 error = check_conflicting_open(filp, arg, lease->c.flc_flags);
1867 if (error) {
1868 locks_unlink_lock_ctx(&lease->c);
1869 goto out;
1870 }
1871
1872out_setup:
1873 if (lease->fl_lmops->lm_setup)
1874 lease->fl_lmops->lm_setup(lease, priv);
1875out:
1876 spin_unlock(&ctx->flc_lock);
1877 percpu_up_read(&file_rwsem);
1878 locks_dispose_list(&dispose);
1879 if (is_deleg)
1880 inode_unlock(inode);
1881 if (!error && !my_fl)
1882 *flp = NULL;
1883 return error;
1884}
1885
1886static int generic_delete_lease(struct file *filp, void *owner)
1887{
1888 int error = -EAGAIN;
1889 struct file_lease *fl, *victim = NULL;
1890 struct inode *inode = file_inode(filp);
1891 struct file_lock_context *ctx;
1892 LIST_HEAD(dispose);
1893
1894 ctx = locks_inode_context(inode);
1895 if (!ctx) {
1896 trace_generic_delete_lease(inode, NULL);
1897 return error;
1898 }
1899
1900 percpu_down_read(&file_rwsem);
1901 spin_lock(&ctx->flc_lock);
1902 list_for_each_entry(fl, &ctx->flc_lease, c.flc_list) {
1903 if (fl->c.flc_file == filp &&
1904 fl->c.flc_owner == owner) {
1905 victim = fl;
1906 break;
1907 }
1908 }
1909 trace_generic_delete_lease(inode, victim);
1910 if (victim)
1911 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1912 spin_unlock(&ctx->flc_lock);
1913 percpu_up_read(&file_rwsem);
1914 locks_dispose_list(&dispose);
1915 return error;
1916}
1917
1918/**
1919 * generic_setlease - sets a lease on an open file
1920 * @filp: file pointer
1921 * @arg: type of lease to obtain
1922 * @flp: input - file_lock to use, output - file_lock inserted
1923 * @priv: private data for lm_setup (may be NULL if lm_setup
1924 * doesn't require it)
1925 *
1926 * The (input) flp->fl_lmops->lm_break function is required
1927 * by break_lease().
1928 */
1929int generic_setlease(struct file *filp, int arg, struct file_lease **flp,
1930 void **priv)
1931{
1932 switch (arg) {
1933 case F_UNLCK:
1934 return generic_delete_lease(filp, *priv);
1935 case F_RDLCK:
1936 case F_WRLCK:
1937 if (!(*flp)->fl_lmops->lm_break) {
1938 WARN_ON_ONCE(1);
1939 return -ENOLCK;
1940 }
1941
1942 return generic_add_lease(filp, arg, flp, priv);
1943 default:
1944 return -EINVAL;
1945 }
1946}
1947EXPORT_SYMBOL(generic_setlease);
1948
1949/*
1950 * Kernel subsystems can register to be notified on any attempt to set
1951 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1952 * to close files that it may have cached when there is an attempt to set a
1953 * conflicting lease.
1954 */
1955static struct srcu_notifier_head lease_notifier_chain;
1956
1957static inline void
1958lease_notifier_chain_init(void)
1959{
1960 srcu_init_notifier_head(&lease_notifier_chain);
1961}
1962
1963static inline void
1964setlease_notifier(int arg, struct file_lease *lease)
1965{
1966 if (arg != F_UNLCK)
1967 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1968}
1969
1970int lease_register_notifier(struct notifier_block *nb)
1971{
1972 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1973}
1974EXPORT_SYMBOL_GPL(lease_register_notifier);
1975
1976void lease_unregister_notifier(struct notifier_block *nb)
1977{
1978 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1979}
1980EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1981
1982
1983int
1984kernel_setlease(struct file *filp, int arg, struct file_lease **lease, void **priv)
1985{
1986 if (lease)
1987 setlease_notifier(arg, *lease);
1988 if (filp->f_op->setlease)
1989 return filp->f_op->setlease(filp, arg, lease, priv);
1990 else
1991 return generic_setlease(filp, arg, lease, priv);
1992}
1993EXPORT_SYMBOL_GPL(kernel_setlease);
1994
1995/**
1996 * vfs_setlease - sets a lease on an open file
1997 * @filp: file pointer
1998 * @arg: type of lease to obtain
1999 * @lease: file_lock to use when adding a lease
2000 * @priv: private info for lm_setup when adding a lease (may be
2001 * NULL if lm_setup doesn't require it)
2002 *
2003 * Call this to establish a lease on the file. The "lease" argument is not
2004 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2005 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2006 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2007 * stack trace).
2008 *
2009 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2010 * may be NULL if the lm_setup operation doesn't require it.
2011 */
2012int
2013vfs_setlease(struct file *filp, int arg, struct file_lease **lease, void **priv)
2014{
2015 struct inode *inode = file_inode(filp);
2016 vfsuid_t vfsuid = i_uid_into_vfsuid(file_mnt_idmap(filp), inode);
2017 int error;
2018
2019 if ((!vfsuid_eq_kuid(vfsuid, current_fsuid())) && !capable(CAP_LEASE))
2020 return -EACCES;
2021 if (!S_ISREG(inode->i_mode))
2022 return -EINVAL;
2023 error = security_file_lock(filp, arg);
2024 if (error)
2025 return error;
2026 return kernel_setlease(filp, arg, lease, priv);
2027}
2028EXPORT_SYMBOL_GPL(vfs_setlease);
2029
2030static int do_fcntl_add_lease(unsigned int fd, struct file *filp, int arg)
2031{
2032 struct file_lease *fl;
2033 struct fasync_struct *new;
2034 int error;
2035
2036 fl = lease_alloc(filp, arg);
2037 if (IS_ERR(fl))
2038 return PTR_ERR(fl);
2039
2040 new = fasync_alloc();
2041 if (!new) {
2042 locks_free_lease(fl);
2043 return -ENOMEM;
2044 }
2045 new->fa_fd = fd;
2046
2047 error = vfs_setlease(filp, arg, &fl, (void **)&new);
2048 if (fl)
2049 locks_free_lease(fl);
2050 if (new)
2051 fasync_free(new);
2052 return error;
2053}
2054
2055/**
2056 * fcntl_setlease - sets a lease on an open file
2057 * @fd: open file descriptor
2058 * @filp: file pointer
2059 * @arg: type of lease to obtain
2060 *
2061 * Call this fcntl to establish a lease on the file.
2062 * Note that you also need to call %F_SETSIG to
2063 * receive a signal when the lease is broken.
2064 */
2065int fcntl_setlease(unsigned int fd, struct file *filp, int arg)
2066{
2067 if (arg == F_UNLCK)
2068 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2069 return do_fcntl_add_lease(fd, filp, arg);
2070}
2071
2072/**
2073 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2074 * @inode: inode of the file to apply to
2075 * @fl: The lock to be applied
2076 *
2077 * Apply a FLOCK style lock request to an inode.
2078 */
2079static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2080{
2081 int error;
2082 might_sleep();
2083 for (;;) {
2084 error = flock_lock_inode(inode, fl);
2085 if (error != FILE_LOCK_DEFERRED)
2086 break;
2087 error = wait_event_interruptible(fl->c.flc_wait,
2088 list_empty(&fl->c.flc_blocked_member));
2089 if (error)
2090 break;
2091 }
2092 locks_delete_block(fl);
2093 return error;
2094}
2095
2096/**
2097 * locks_lock_inode_wait - Apply a lock to an inode
2098 * @inode: inode of the file to apply to
2099 * @fl: The lock to be applied
2100 *
2101 * Apply a POSIX or FLOCK style lock request to an inode.
2102 */
2103int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2104{
2105 int res = 0;
2106 switch (fl->c.flc_flags & (FL_POSIX|FL_FLOCK)) {
2107 case FL_POSIX:
2108 res = posix_lock_inode_wait(inode, fl);
2109 break;
2110 case FL_FLOCK:
2111 res = flock_lock_inode_wait(inode, fl);
2112 break;
2113 default:
2114 BUG();
2115 }
2116 return res;
2117}
2118EXPORT_SYMBOL(locks_lock_inode_wait);
2119
2120/**
2121 * sys_flock: - flock() system call.
2122 * @fd: the file descriptor to lock.
2123 * @cmd: the type of lock to apply.
2124 *
2125 * Apply a %FL_FLOCK style lock to an open file descriptor.
2126 * The @cmd can be one of:
2127 *
2128 * - %LOCK_SH -- a shared lock.
2129 * - %LOCK_EX -- an exclusive lock.
2130 * - %LOCK_UN -- remove an existing lock.
2131 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2132 *
2133 * %LOCK_MAND support has been removed from the kernel.
2134 */
2135SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2136{
2137 int can_sleep, error, type;
2138 struct file_lock fl;
2139
2140 /*
2141 * LOCK_MAND locks were broken for a long time in that they never
2142 * conflicted with one another and didn't prevent any sort of open,
2143 * read or write activity.
2144 *
2145 * Just ignore these requests now, to preserve legacy behavior, but
2146 * throw a warning to let people know that they don't actually work.
2147 */
2148 if (cmd & LOCK_MAND) {
2149 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);
2150 return 0;
2151 }
2152
2153 type = flock_translate_cmd(cmd & ~LOCK_NB);
2154 if (type < 0)
2155 return type;
2156
2157 CLASS(fd, f)(fd);
2158 if (fd_empty(f))
2159 return -EBADF;
2160
2161 if (type != F_UNLCK && !(fd_file(f)->f_mode & (FMODE_READ | FMODE_WRITE)))
2162 return -EBADF;
2163
2164 flock_make_lock(fd_file(f), &fl, type);
2165
2166 error = security_file_lock(fd_file(f), fl.c.flc_type);
2167 if (error)
2168 return error;
2169
2170 can_sleep = !(cmd & LOCK_NB);
2171 if (can_sleep)
2172 fl.c.flc_flags |= FL_SLEEP;
2173
2174 if (fd_file(f)->f_op->flock)
2175 error = fd_file(f)->f_op->flock(fd_file(f),
2176 (can_sleep) ? F_SETLKW : F_SETLK,
2177 &fl);
2178 else
2179 error = locks_lock_file_wait(fd_file(f), &fl);
2180
2181 locks_release_private(&fl);
2182 return error;
2183}
2184
2185/**
2186 * vfs_test_lock - test file byte range lock
2187 * @filp: The file to test lock for
2188 * @fl: The lock to test; also used to hold result
2189 *
2190 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2191 * setting conf->fl_type to something other than F_UNLCK.
2192 */
2193int vfs_test_lock(struct file *filp, struct file_lock *fl)
2194{
2195 WARN_ON_ONCE(filp != fl->c.flc_file);
2196 if (filp->f_op->lock)
2197 return filp->f_op->lock(filp, F_GETLK, fl);
2198 posix_test_lock(filp, fl);
2199 return 0;
2200}
2201EXPORT_SYMBOL_GPL(vfs_test_lock);
2202
2203/**
2204 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2205 * @fl: The file_lock who's fl_pid should be translated
2206 * @ns: The namespace into which the pid should be translated
2207 *
2208 * Used to translate a fl_pid into a namespace virtual pid number
2209 */
2210static pid_t locks_translate_pid(struct file_lock_core *fl, struct pid_namespace *ns)
2211{
2212 pid_t vnr;
2213 struct pid *pid;
2214
2215 if (fl->flc_flags & FL_OFDLCK)
2216 return -1;
2217
2218 /* Remote locks report a negative pid value */
2219 if (fl->flc_pid <= 0)
2220 return fl->flc_pid;
2221
2222 /*
2223 * If the flock owner process is dead and its pid has been already
2224 * freed, the translation below won't work, but we still want to show
2225 * flock owner pid number in init pidns.
2226 */
2227 if (ns == &init_pid_ns)
2228 return (pid_t) fl->flc_pid;
2229
2230 rcu_read_lock();
2231 pid = find_pid_ns(fl->flc_pid, &init_pid_ns);
2232 vnr = pid_nr_ns(pid, ns);
2233 rcu_read_unlock();
2234 return vnr;
2235}
2236
2237static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2238{
2239 flock->l_pid = locks_translate_pid(&fl->c, task_active_pid_ns(current));
2240#if BITS_PER_LONG == 32
2241 /*
2242 * Make sure we can represent the posix lock via
2243 * legacy 32bit flock.
2244 */
2245 if (fl->fl_start > OFFT_OFFSET_MAX)
2246 return -EOVERFLOW;
2247 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2248 return -EOVERFLOW;
2249#endif
2250 flock->l_start = fl->fl_start;
2251 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2252 fl->fl_end - fl->fl_start + 1;
2253 flock->l_whence = 0;
2254 flock->l_type = fl->c.flc_type;
2255 return 0;
2256}
2257
2258#if BITS_PER_LONG == 32
2259static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2260{
2261 flock->l_pid = locks_translate_pid(&fl->c, task_active_pid_ns(current));
2262 flock->l_start = fl->fl_start;
2263 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2264 fl->fl_end - fl->fl_start + 1;
2265 flock->l_whence = 0;
2266 flock->l_type = fl->c.flc_type;
2267}
2268#endif
2269
2270/* Report the first existing lock that would conflict with l.
2271 * This implements the F_GETLK command of fcntl().
2272 */
2273int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2274{
2275 struct file_lock *fl;
2276 int error;
2277
2278 fl = locks_alloc_lock();
2279 if (fl == NULL)
2280 return -ENOMEM;
2281 error = -EINVAL;
2282 if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK
2283 && flock->l_type != F_WRLCK)
2284 goto out;
2285
2286 error = flock_to_posix_lock(filp, fl, flock);
2287 if (error)
2288 goto out;
2289
2290 if (cmd == F_OFD_GETLK) {
2291 error = -EINVAL;
2292 if (flock->l_pid != 0)
2293 goto out;
2294
2295 fl->c.flc_flags |= FL_OFDLCK;
2296 fl->c.flc_owner = filp;
2297 }
2298
2299 error = vfs_test_lock(filp, fl);
2300 if (error)
2301 goto out;
2302
2303 flock->l_type = fl->c.flc_type;
2304 if (fl->c.flc_type != F_UNLCK) {
2305 error = posix_lock_to_flock(flock, fl);
2306 if (error)
2307 goto out;
2308 }
2309out:
2310 locks_free_lock(fl);
2311 return error;
2312}
2313
2314/**
2315 * vfs_lock_file - file byte range lock
2316 * @filp: The file to apply the lock to
2317 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2318 * @fl: The lock to be applied
2319 * @conf: Place to return a copy of the conflicting lock, if found.
2320 *
2321 * A caller that doesn't care about the conflicting lock may pass NULL
2322 * as the final argument.
2323 *
2324 * If the filesystem defines a private ->lock() method, then @conf will
2325 * be left unchanged; so a caller that cares should initialize it to
2326 * some acceptable default.
2327 *
2328 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2329 * locks, the ->lock() interface may return asynchronously, before the lock has
2330 * been granted or denied by the underlying filesystem, if (and only if)
2331 * lm_grant is set. Additionally EXPORT_OP_ASYNC_LOCK in export_operations
2332 * flags need to be set.
2333 *
2334 * Callers expecting ->lock() to return asynchronously will only use F_SETLK,
2335 * not F_SETLKW; they will set FL_SLEEP if (and only if) the request is for a
2336 * blocking lock. When ->lock() does return asynchronously, it must return
2337 * FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock request completes.
2338 * If the request is for non-blocking lock the file system should return
2339 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2340 * with the result. If the request timed out the callback routine will return a
2341 * nonzero return code and the file system should release the lock. The file
2342 * system is also responsible to keep a corresponding posix lock when it
2343 * grants a lock so the VFS can find out which locks are locally held and do
2344 * the correct lock cleanup when required.
2345 * The underlying filesystem must not drop the kernel lock or call
2346 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2347 * return code.
2348 */
2349int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2350{
2351 WARN_ON_ONCE(filp != fl->c.flc_file);
2352 if (filp->f_op->lock)
2353 return filp->f_op->lock(filp, cmd, fl);
2354 else
2355 return posix_lock_file(filp, fl, conf);
2356}
2357EXPORT_SYMBOL_GPL(vfs_lock_file);
2358
2359static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2360 struct file_lock *fl)
2361{
2362 int error;
2363
2364 error = security_file_lock(filp, fl->c.flc_type);
2365 if (error)
2366 return error;
2367
2368 for (;;) {
2369 error = vfs_lock_file(filp, cmd, fl, NULL);
2370 if (error != FILE_LOCK_DEFERRED)
2371 break;
2372 error = wait_event_interruptible(fl->c.flc_wait,
2373 list_empty(&fl->c.flc_blocked_member));
2374 if (error)
2375 break;
2376 }
2377 locks_delete_block(fl);
2378
2379 return error;
2380}
2381
2382/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2383static int
2384check_fmode_for_setlk(struct file_lock *fl)
2385{
2386 switch (fl->c.flc_type) {
2387 case F_RDLCK:
2388 if (!(fl->c.flc_file->f_mode & FMODE_READ))
2389 return -EBADF;
2390 break;
2391 case F_WRLCK:
2392 if (!(fl->c.flc_file->f_mode & FMODE_WRITE))
2393 return -EBADF;
2394 }
2395 return 0;
2396}
2397
2398/* Apply the lock described by l to an open file descriptor.
2399 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2400 */
2401int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2402 struct flock *flock)
2403{
2404 struct file_lock *file_lock = locks_alloc_lock();
2405 struct inode *inode = file_inode(filp);
2406 struct file *f;
2407 int error;
2408
2409 if (file_lock == NULL)
2410 return -ENOLCK;
2411
2412 error = flock_to_posix_lock(filp, file_lock, flock);
2413 if (error)
2414 goto out;
2415
2416 error = check_fmode_for_setlk(file_lock);
2417 if (error)
2418 goto out;
2419
2420 /*
2421 * If the cmd is requesting file-private locks, then set the
2422 * FL_OFDLCK flag and override the owner.
2423 */
2424 switch (cmd) {
2425 case F_OFD_SETLK:
2426 error = -EINVAL;
2427 if (flock->l_pid != 0)
2428 goto out;
2429
2430 cmd = F_SETLK;
2431 file_lock->c.flc_flags |= FL_OFDLCK;
2432 file_lock->c.flc_owner = filp;
2433 break;
2434 case F_OFD_SETLKW:
2435 error = -EINVAL;
2436 if (flock->l_pid != 0)
2437 goto out;
2438
2439 cmd = F_SETLKW;
2440 file_lock->c.flc_flags |= FL_OFDLCK;
2441 file_lock->c.flc_owner = filp;
2442 fallthrough;
2443 case F_SETLKW:
2444 file_lock->c.flc_flags |= FL_SLEEP;
2445 }
2446
2447 error = do_lock_file_wait(filp, cmd, file_lock);
2448
2449 /*
2450 * Detect close/fcntl races and recover by zapping all POSIX locks
2451 * associated with this file and our files_struct, just like on
2452 * filp_flush(). 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 locks_remove_posix(filp, files);
2468 error = -EBADF;
2469 }
2470 }
2471out:
2472 trace_fcntl_setlk(inode, file_lock, error);
2473 locks_free_lock(file_lock);
2474 return error;
2475}
2476
2477#if BITS_PER_LONG == 32
2478/* Report the first existing lock that would conflict with l.
2479 * This implements the F_GETLK command of fcntl().
2480 */
2481int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2482{
2483 struct file_lock *fl;
2484 int error;
2485
2486 fl = locks_alloc_lock();
2487 if (fl == NULL)
2488 return -ENOMEM;
2489
2490 error = -EINVAL;
2491 if (cmd != F_OFD_GETLK && flock->l_type != F_RDLCK
2492 && flock->l_type != F_WRLCK)
2493 goto out;
2494
2495 error = flock64_to_posix_lock(filp, fl, flock);
2496 if (error)
2497 goto out;
2498
2499 if (cmd == F_OFD_GETLK) {
2500 error = -EINVAL;
2501 if (flock->l_pid != 0)
2502 goto out;
2503
2504 fl->c.flc_flags |= FL_OFDLCK;
2505 fl->c.flc_owner = filp;
2506 }
2507
2508 error = vfs_test_lock(filp, fl);
2509 if (error)
2510 goto out;
2511
2512 flock->l_type = fl->c.flc_type;
2513 if (fl->c.flc_type != F_UNLCK)
2514 posix_lock_to_flock64(flock, fl);
2515
2516out:
2517 locks_free_lock(fl);
2518 return error;
2519}
2520
2521/* Apply the lock described by l to an open file descriptor.
2522 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2523 */
2524int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2525 struct flock64 *flock)
2526{
2527 struct file_lock *file_lock = locks_alloc_lock();
2528 struct file *f;
2529 int error;
2530
2531 if (file_lock == NULL)
2532 return -ENOLCK;
2533
2534 error = flock64_to_posix_lock(filp, file_lock, flock);
2535 if (error)
2536 goto out;
2537
2538 error = check_fmode_for_setlk(file_lock);
2539 if (error)
2540 goto out;
2541
2542 /*
2543 * If the cmd is requesting file-private locks, then set the
2544 * FL_OFDLCK flag and override the owner.
2545 */
2546 switch (cmd) {
2547 case F_OFD_SETLK:
2548 error = -EINVAL;
2549 if (flock->l_pid != 0)
2550 goto out;
2551
2552 cmd = F_SETLK64;
2553 file_lock->c.flc_flags |= FL_OFDLCK;
2554 file_lock->c.flc_owner = filp;
2555 break;
2556 case F_OFD_SETLKW:
2557 error = -EINVAL;
2558 if (flock->l_pid != 0)
2559 goto out;
2560
2561 cmd = F_SETLKW64;
2562 file_lock->c.flc_flags |= FL_OFDLCK;
2563 file_lock->c.flc_owner = filp;
2564 fallthrough;
2565 case F_SETLKW64:
2566 file_lock->c.flc_flags |= FL_SLEEP;
2567 }
2568
2569 error = do_lock_file_wait(filp, cmd, file_lock);
2570
2571 /*
2572 * Detect close/fcntl races and recover by zapping all POSIX locks
2573 * associated with this file and our files_struct, just like on
2574 * filp_flush(). There is no need to do that when we're
2575 * unlocking though, or for OFD locks.
2576 */
2577 if (!error && file_lock->c.flc_type != F_UNLCK &&
2578 !(file_lock->c.flc_flags & FL_OFDLCK)) {
2579 struct files_struct *files = current->files;
2580 /*
2581 * We need that spin_lock here - it prevents reordering between
2582 * update of i_flctx->flc_posix and check for it done in
2583 * close(). rcu_read_lock() wouldn't do.
2584 */
2585 spin_lock(&files->file_lock);
2586 f = files_lookup_fd_locked(files, fd);
2587 spin_unlock(&files->file_lock);
2588 if (f != filp) {
2589 locks_remove_posix(filp, files);
2590 error = -EBADF;
2591 }
2592 }
2593out:
2594 locks_free_lock(file_lock);
2595 return error;
2596}
2597#endif /* BITS_PER_LONG == 32 */
2598
2599/*
2600 * This function is called when the file is being removed
2601 * from the task's fd array. POSIX locks belonging to this task
2602 * are deleted at this time.
2603 */
2604void locks_remove_posix(struct file *filp, fl_owner_t owner)
2605{
2606 int error;
2607 struct inode *inode = file_inode(filp);
2608 struct file_lock lock;
2609 struct file_lock_context *ctx;
2610
2611 /*
2612 * If there are no locks held on this file, we don't need to call
2613 * posix_lock_file(). Another process could be setting a lock on this
2614 * file at the same time, but we wouldn't remove that lock anyway.
2615 */
2616 ctx = locks_inode_context(inode);
2617 if (!ctx || list_empty(&ctx->flc_posix))
2618 return;
2619
2620 locks_init_lock(&lock);
2621 lock.c.flc_type = F_UNLCK;
2622 lock.c.flc_flags = FL_POSIX | FL_CLOSE;
2623 lock.fl_start = 0;
2624 lock.fl_end = OFFSET_MAX;
2625 lock.c.flc_owner = owner;
2626 lock.c.flc_pid = current->tgid;
2627 lock.c.flc_file = filp;
2628 lock.fl_ops = NULL;
2629 lock.fl_lmops = NULL;
2630
2631 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2632
2633 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2634 lock.fl_ops->fl_release_private(&lock);
2635 trace_locks_remove_posix(inode, &lock, error);
2636}
2637EXPORT_SYMBOL(locks_remove_posix);
2638
2639/* The i_flctx must be valid when calling into here */
2640static void
2641locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2642{
2643 struct file_lock fl;
2644 struct inode *inode = file_inode(filp);
2645
2646 if (list_empty(&flctx->flc_flock))
2647 return;
2648
2649 flock_make_lock(filp, &fl, F_UNLCK);
2650 fl.c.flc_flags |= FL_CLOSE;
2651
2652 if (filp->f_op->flock)
2653 filp->f_op->flock(filp, F_SETLKW, &fl);
2654 else
2655 flock_lock_inode(inode, &fl);
2656
2657 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2658 fl.fl_ops->fl_release_private(&fl);
2659}
2660
2661/* The i_flctx must be valid when calling into here */
2662static void
2663locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2664{
2665 struct file_lease *fl, *tmp;
2666 LIST_HEAD(dispose);
2667
2668 if (list_empty(&ctx->flc_lease))
2669 return;
2670
2671 percpu_down_read(&file_rwsem);
2672 spin_lock(&ctx->flc_lock);
2673 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, c.flc_list)
2674 if (filp == fl->c.flc_file)
2675 lease_modify(fl, F_UNLCK, &dispose);
2676 spin_unlock(&ctx->flc_lock);
2677 percpu_up_read(&file_rwsem);
2678
2679 locks_dispose_list(&dispose);
2680}
2681
2682/*
2683 * This function is called on the last close of an open file.
2684 */
2685void locks_remove_file(struct file *filp)
2686{
2687 struct file_lock_context *ctx;
2688
2689 ctx = locks_inode_context(file_inode(filp));
2690 if (!ctx)
2691 return;
2692
2693 /* remove any OFD locks */
2694 locks_remove_posix(filp, filp);
2695
2696 /* remove flock locks */
2697 locks_remove_flock(filp, ctx);
2698
2699 /* remove any leases */
2700 locks_remove_lease(filp, ctx);
2701
2702 spin_lock(&ctx->flc_lock);
2703 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2704 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2705 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2706 spin_unlock(&ctx->flc_lock);
2707}
2708
2709/**
2710 * vfs_cancel_lock - file byte range unblock lock
2711 * @filp: The file to apply the unblock to
2712 * @fl: The lock to be unblocked
2713 *
2714 * Used by lock managers to cancel blocked requests
2715 */
2716int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2717{
2718 WARN_ON_ONCE(filp != fl->c.flc_file);
2719 if (filp->f_op->lock)
2720 return filp->f_op->lock(filp, F_CANCELLK, fl);
2721 return 0;
2722}
2723EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2724
2725/**
2726 * vfs_inode_has_locks - are any file locks held on @inode?
2727 * @inode: inode to check for locks
2728 *
2729 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2730 * set on @inode.
2731 */
2732bool vfs_inode_has_locks(struct inode *inode)
2733{
2734 struct file_lock_context *ctx;
2735 bool ret;
2736
2737 ctx = locks_inode_context(inode);
2738 if (!ctx)
2739 return false;
2740
2741 spin_lock(&ctx->flc_lock);
2742 ret = !list_empty(&ctx->flc_posix) || !list_empty(&ctx->flc_flock);
2743 spin_unlock(&ctx->flc_lock);
2744 return ret;
2745}
2746EXPORT_SYMBOL_GPL(vfs_inode_has_locks);
2747
2748#ifdef CONFIG_PROC_FS
2749#include <linux/proc_fs.h>
2750#include <linux/seq_file.h>
2751
2752struct locks_iterator {
2753 int li_cpu;
2754 loff_t li_pos;
2755};
2756
2757static void lock_get_status(struct seq_file *f, struct file_lock_core *flc,
2758 loff_t id, char *pfx, int repeat)
2759{
2760 struct inode *inode = NULL;
2761 unsigned int pid;
2762 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2763 int type = flc->flc_type;
2764 struct file_lock *fl = file_lock(flc);
2765
2766 pid = locks_translate_pid(flc, proc_pidns);
2767
2768 /*
2769 * If lock owner is dead (and pid is freed) or not visible in current
2770 * pidns, zero is shown as a pid value. Check lock info from
2771 * init_pid_ns to get saved lock pid value.
2772 */
2773 if (flc->flc_file != NULL)
2774 inode = file_inode(flc->flc_file);
2775
2776 seq_printf(f, "%lld: ", id);
2777
2778 if (repeat)
2779 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2780
2781 if (flc->flc_flags & FL_POSIX) {
2782 if (flc->flc_flags & FL_ACCESS)
2783 seq_puts(f, "ACCESS");
2784 else if (flc->flc_flags & FL_OFDLCK)
2785 seq_puts(f, "OFDLCK");
2786 else
2787 seq_puts(f, "POSIX ");
2788
2789 seq_printf(f, " %s ",
2790 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2791 } else if (flc->flc_flags & FL_FLOCK) {
2792 seq_puts(f, "FLOCK ADVISORY ");
2793 } else if (flc->flc_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT)) {
2794 struct file_lease *lease = file_lease(flc);
2795
2796 type = target_leasetype(lease);
2797
2798 if (flc->flc_flags & FL_DELEG)
2799 seq_puts(f, "DELEG ");
2800 else
2801 seq_puts(f, "LEASE ");
2802
2803 if (lease_breaking(lease))
2804 seq_puts(f, "BREAKING ");
2805 else if (flc->flc_file)
2806 seq_puts(f, "ACTIVE ");
2807 else
2808 seq_puts(f, "BREAKER ");
2809 } else {
2810 seq_puts(f, "UNKNOWN UNKNOWN ");
2811 }
2812
2813 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2814 (type == F_RDLCK) ? "READ" : "UNLCK");
2815 if (inode) {
2816 /* userspace relies on this representation of dev_t */
2817 seq_printf(f, "%d %02x:%02x:%lu ", pid,
2818 MAJOR(inode->i_sb->s_dev),
2819 MINOR(inode->i_sb->s_dev), inode->i_ino);
2820 } else {
2821 seq_printf(f, "%d <none>:0 ", pid);
2822 }
2823 if (flc->flc_flags & FL_POSIX) {
2824 if (fl->fl_end == OFFSET_MAX)
2825 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2826 else
2827 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2828 } else {
2829 seq_puts(f, "0 EOF\n");
2830 }
2831}
2832
2833static struct file_lock_core *get_next_blocked_member(struct file_lock_core *node)
2834{
2835 struct file_lock_core *tmp;
2836
2837 /* NULL node or root node */
2838 if (node == NULL || node->flc_blocker == NULL)
2839 return NULL;
2840
2841 /* Next member in the linked list could be itself */
2842 tmp = list_next_entry(node, flc_blocked_member);
2843 if (list_entry_is_head(tmp, &node->flc_blocker->flc_blocked_requests,
2844 flc_blocked_member)
2845 || tmp == node) {
2846 return NULL;
2847 }
2848
2849 return tmp;
2850}
2851
2852static int locks_show(struct seq_file *f, void *v)
2853{
2854 struct locks_iterator *iter = f->private;
2855 struct file_lock_core *cur, *tmp;
2856 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2857 int level = 0;
2858
2859 cur = hlist_entry(v, struct file_lock_core, flc_link);
2860
2861 if (locks_translate_pid(cur, proc_pidns) == 0)
2862 return 0;
2863
2864 /* View this crossed linked list as a binary tree, the first member of flc_blocked_requests
2865 * is the left child of current node, the next silibing in flc_blocked_member is the
2866 * right child, we can alse get the parent of current node from flc_blocker, so this
2867 * question becomes traversal of a binary tree
2868 */
2869 while (cur != NULL) {
2870 if (level)
2871 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2872 else
2873 lock_get_status(f, cur, iter->li_pos, "", level);
2874
2875 if (!list_empty(&cur->flc_blocked_requests)) {
2876 /* Turn left */
2877 cur = list_first_entry_or_null(&cur->flc_blocked_requests,
2878 struct file_lock_core,
2879 flc_blocked_member);
2880 level++;
2881 } else {
2882 /* Turn right */
2883 tmp = get_next_blocked_member(cur);
2884 /* Fall back to parent node */
2885 while (tmp == NULL && cur->flc_blocker != NULL) {
2886 cur = cur->flc_blocker;
2887 level--;
2888 tmp = get_next_blocked_member(cur);
2889 }
2890 cur = tmp;
2891 }
2892 }
2893
2894 return 0;
2895}
2896
2897static void __show_fd_locks(struct seq_file *f,
2898 struct list_head *head, int *id,
2899 struct file *filp, struct files_struct *files)
2900{
2901 struct file_lock_core *fl;
2902
2903 list_for_each_entry(fl, head, flc_list) {
2904
2905 if (filp != fl->flc_file)
2906 continue;
2907 if (fl->flc_owner != files && fl->flc_owner != filp)
2908 continue;
2909
2910 (*id)++;
2911 seq_puts(f, "lock:\t");
2912 lock_get_status(f, fl, *id, "", 0);
2913 }
2914}
2915
2916void show_fd_locks(struct seq_file *f,
2917 struct file *filp, struct files_struct *files)
2918{
2919 struct inode *inode = file_inode(filp);
2920 struct file_lock_context *ctx;
2921 int id = 0;
2922
2923 ctx = locks_inode_context(inode);
2924 if (!ctx)
2925 return;
2926
2927 spin_lock(&ctx->flc_lock);
2928 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2929 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2930 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2931 spin_unlock(&ctx->flc_lock);
2932}
2933
2934static void *locks_start(struct seq_file *f, loff_t *pos)
2935 __acquires(&blocked_lock_lock)
2936{
2937 struct locks_iterator *iter = f->private;
2938
2939 iter->li_pos = *pos + 1;
2940 percpu_down_write(&file_rwsem);
2941 spin_lock(&blocked_lock_lock);
2942 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2943}
2944
2945static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2946{
2947 struct locks_iterator *iter = f->private;
2948
2949 ++iter->li_pos;
2950 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2951}
2952
2953static void locks_stop(struct seq_file *f, void *v)
2954 __releases(&blocked_lock_lock)
2955{
2956 spin_unlock(&blocked_lock_lock);
2957 percpu_up_write(&file_rwsem);
2958}
2959
2960static const struct seq_operations locks_seq_operations = {
2961 .start = locks_start,
2962 .next = locks_next,
2963 .stop = locks_stop,
2964 .show = locks_show,
2965};
2966
2967static int __init proc_locks_init(void)
2968{
2969 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2970 sizeof(struct locks_iterator), NULL);
2971 return 0;
2972}
2973fs_initcall(proc_locks_init);
2974#endif
2975
2976static int __init filelock_init(void)
2977{
2978 int i;
2979
2980 flctx_cache = kmem_cache_create("file_lock_ctx",
2981 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2982
2983 filelock_cache = kmem_cache_create("file_lock_cache",
2984 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2985
2986 filelease_cache = kmem_cache_create("file_lease_cache",
2987 sizeof(struct file_lease), 0, SLAB_PANIC, NULL);
2988
2989 for_each_possible_cpu(i) {
2990 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2991
2992 spin_lock_init(&fll->lock);
2993 INIT_HLIST_HEAD(&fll->hlist);
2994 }
2995
2996 lease_notifier_chain_init();
2997 return 0;
2998}
2999core_initcall(filelock_init);