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