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