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