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 */
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>
133 #include <asm/uaccess.h>
135 #define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
136 #define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
137 #define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
139 static bool lease_breaking(struct file_lock *fl)
140 {
141 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
142 }
144 static int target_leasetype(struct file_lock *fl)
145 {
146 if (fl->fl_flags & FL_UNLOCK_PENDING)
147 return F_UNLCK;
148 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
149 return F_RDLCK;
150 return fl->fl_type;
151 }
153 int leases_enable = 1;
154 int lease_break_time = 45;
156 #define for_each_lock(inode, lockp) \
157 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
159 /*
160 * The global file_lock_list is only used for displaying /proc/locks, so we
161 * keep a list on each CPU, with each list protected by its own spinlock via
162 * the file_lock_lglock. Note that alterations to the list also require that
163 * the relevant i_lock is held.
164 */
165 DEFINE_STATIC_LGLOCK(file_lock_lglock);
166 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
168 /*
169 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
170 * It is protected by blocked_lock_lock.
171 *
172 * We hash locks by lockowner in order to optimize searching for the lock a
173 * particular lockowner is waiting on.
174 *
175 * FIXME: make this value scale via some heuristic? We generally will want more
176 * buckets when we have more lockowners holding locks, but that's a little
177 * difficult to determine without knowing what the workload will look like.
178 */
179 #define BLOCKED_HASH_BITS 7
180 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
182 /*
183 * This lock protects the blocked_hash. Generally, if you're accessing it, you
184 * want to be holding this lock.
185 *
186 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
187 * pointer for file_lock structures that are acting as lock requests (in
188 * contrast to those that are acting as records of acquired locks).
189 *
190 * Note that when we acquire this lock in order to change the above fields,
191 * we often hold the i_lock as well. In certain cases, when reading the fields
192 * protected by this lock, we can skip acquiring it iff we already hold the
193 * i_lock.
194 *
195 * In particular, adding an entry to the fl_block list requires that you hold
196 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
197 * an entry from the list however only requires the file_lock_lock.
198 */
199 static DEFINE_SPINLOCK(blocked_lock_lock);
201 static struct kmem_cache *filelock_cache __read_mostly;
203 static void locks_init_lock_heads(struct file_lock *fl)
204 {
205 INIT_HLIST_NODE(&fl->fl_link);
206 INIT_LIST_HEAD(&fl->fl_block);
207 init_waitqueue_head(&fl->fl_wait);
208 }
210 /* Allocate an empty lock structure. */
211 struct file_lock *locks_alloc_lock(void)
212 {
213 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
215 if (fl)
216 locks_init_lock_heads(fl);
218 return fl;
219 }
220 EXPORT_SYMBOL_GPL(locks_alloc_lock);
222 void locks_release_private(struct file_lock *fl)
223 {
224 if (fl->fl_ops) {
225 if (fl->fl_ops->fl_release_private)
226 fl->fl_ops->fl_release_private(fl);
227 fl->fl_ops = NULL;
228 }
229 fl->fl_lmops = NULL;
231 }
232 EXPORT_SYMBOL_GPL(locks_release_private);
234 /* Free a lock which is not in use. */
235 void locks_free_lock(struct file_lock *fl)
236 {
237 BUG_ON(waitqueue_active(&fl->fl_wait));
238 BUG_ON(!list_empty(&fl->fl_block));
239 BUG_ON(!hlist_unhashed(&fl->fl_link));
241 locks_release_private(fl);
242 kmem_cache_free(filelock_cache, fl);
243 }
244 EXPORT_SYMBOL(locks_free_lock);
246 void locks_init_lock(struct file_lock *fl)
247 {
248 memset(fl, 0, sizeof(struct file_lock));
249 locks_init_lock_heads(fl);
250 }
252 EXPORT_SYMBOL(locks_init_lock);
254 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
255 {
256 if (fl->fl_ops) {
257 if (fl->fl_ops->fl_copy_lock)
258 fl->fl_ops->fl_copy_lock(new, fl);
259 new->fl_ops = fl->fl_ops;
260 }
261 if (fl->fl_lmops)
262 new->fl_lmops = fl->fl_lmops;
263 }
265 /*
266 * Initialize a new lock from an existing file_lock structure.
267 */
268 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
269 {
270 new->fl_owner = fl->fl_owner;
271 new->fl_pid = fl->fl_pid;
272 new->fl_file = NULL;
273 new->fl_flags = fl->fl_flags;
274 new->fl_type = fl->fl_type;
275 new->fl_start = fl->fl_start;
276 new->fl_end = fl->fl_end;
277 new->fl_ops = NULL;
278 new->fl_lmops = NULL;
279 }
280 EXPORT_SYMBOL(__locks_copy_lock);
282 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
283 {
284 locks_release_private(new);
286 __locks_copy_lock(new, fl);
287 new->fl_file = fl->fl_file;
288 new->fl_ops = fl->fl_ops;
289 new->fl_lmops = fl->fl_lmops;
291 locks_copy_private(new, fl);
292 }
294 EXPORT_SYMBOL(locks_copy_lock);
296 static inline int flock_translate_cmd(int cmd) {
297 if (cmd & LOCK_MAND)
298 return cmd & (LOCK_MAND | LOCK_RW);
299 switch (cmd) {
300 case LOCK_SH:
301 return F_RDLCK;
302 case LOCK_EX:
303 return F_WRLCK;
304 case LOCK_UN:
305 return F_UNLCK;
306 }
307 return -EINVAL;
308 }
310 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
311 static int flock_make_lock(struct file *filp, struct file_lock **lock,
312 unsigned int cmd)
313 {
314 struct file_lock *fl;
315 int type = flock_translate_cmd(cmd);
316 if (type < 0)
317 return type;
319 fl = locks_alloc_lock();
320 if (fl == NULL)
321 return -ENOMEM;
323 fl->fl_file = filp;
324 fl->fl_pid = current->tgid;
325 fl->fl_flags = FL_FLOCK;
326 fl->fl_type = type;
327 fl->fl_end = OFFSET_MAX;
329 *lock = fl;
330 return 0;
331 }
333 static int assign_type(struct file_lock *fl, long type)
334 {
335 switch (type) {
336 case F_RDLCK:
337 case F_WRLCK:
338 case F_UNLCK:
339 fl->fl_type = type;
340 break;
341 default:
342 return -EINVAL;
343 }
344 return 0;
345 }
347 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
348 * style lock.
349 */
350 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
351 struct flock *l)
352 {
353 off_t start, end;
355 switch (l->l_whence) {
356 case SEEK_SET:
357 start = 0;
358 break;
359 case SEEK_CUR:
360 start = filp->f_pos;
361 break;
362 case SEEK_END:
363 start = i_size_read(file_inode(filp));
364 break;
365 default:
366 return -EINVAL;
367 }
369 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
370 POSIX-2001 defines it. */
371 start += l->l_start;
372 if (start < 0)
373 return -EINVAL;
374 fl->fl_end = OFFSET_MAX;
375 if (l->l_len > 0) {
376 end = start + l->l_len - 1;
377 fl->fl_end = end;
378 } else if (l->l_len < 0) {
379 end = start - 1;
380 fl->fl_end = end;
381 start += l->l_len;
382 if (start < 0)
383 return -EINVAL;
384 }
385 fl->fl_start = start; /* we record the absolute position */
386 if (fl->fl_end < fl->fl_start)
387 return -EOVERFLOW;
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;
396 return assign_type(fl, l->l_type);
397 }
399 #if BITS_PER_LONG == 32
400 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
401 struct flock64 *l)
402 {
403 loff_t start;
405 switch (l->l_whence) {
406 case SEEK_SET:
407 start = 0;
408 break;
409 case SEEK_CUR:
410 start = filp->f_pos;
411 break;
412 case SEEK_END:
413 start = i_size_read(file_inode(filp));
414 break;
415 default:
416 return -EINVAL;
417 }
419 start += l->l_start;
420 if (start < 0)
421 return -EINVAL;
422 fl->fl_end = OFFSET_MAX;
423 if (l->l_len > 0) {
424 fl->fl_end = start + l->l_len - 1;
425 } else if (l->l_len < 0) {
426 fl->fl_end = start - 1;
427 start += l->l_len;
428 if (start < 0)
429 return -EINVAL;
430 }
431 fl->fl_start = start; /* we record the absolute position */
432 if (fl->fl_end < fl->fl_start)
433 return -EOVERFLOW;
435 fl->fl_owner = current->files;
436 fl->fl_pid = current->tgid;
437 fl->fl_file = filp;
438 fl->fl_flags = FL_POSIX;
439 fl->fl_ops = NULL;
440 fl->fl_lmops = NULL;
442 return assign_type(fl, l->l_type);
443 }
444 #endif
446 /* default lease lock manager operations */
447 static void lease_break_callback(struct file_lock *fl)
448 {
449 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
450 }
452 static const struct lock_manager_operations lease_manager_ops = {
453 .lm_break = lease_break_callback,
454 .lm_change = lease_modify,
455 };
457 /*
458 * Initialize a lease, use the default lock manager operations
459 */
460 static int lease_init(struct file *filp, long type, struct file_lock *fl)
461 {
462 if (assign_type(fl, type) != 0)
463 return -EINVAL;
465 fl->fl_owner = current->files;
466 fl->fl_pid = current->tgid;
468 fl->fl_file = filp;
469 fl->fl_flags = FL_LEASE;
470 fl->fl_start = 0;
471 fl->fl_end = OFFSET_MAX;
472 fl->fl_ops = NULL;
473 fl->fl_lmops = &lease_manager_ops;
474 return 0;
475 }
477 /* Allocate a file_lock initialised to this type of lease */
478 static struct file_lock *lease_alloc(struct file *filp, long type)
479 {
480 struct file_lock *fl = locks_alloc_lock();
481 int error = -ENOMEM;
483 if (fl == NULL)
484 return ERR_PTR(error);
486 error = lease_init(filp, type, fl);
487 if (error) {
488 locks_free_lock(fl);
489 return ERR_PTR(error);
490 }
491 return fl;
492 }
494 /* Check if two locks overlap each other.
495 */
496 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
497 {
498 return ((fl1->fl_end >= fl2->fl_start) &&
499 (fl2->fl_end >= fl1->fl_start));
500 }
502 /*
503 * Check whether two locks have the same owner.
504 */
505 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
506 {
507 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
508 return fl2->fl_lmops == fl1->fl_lmops &&
509 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
510 return fl1->fl_owner == fl2->fl_owner;
511 }
513 /* Must be called with the i_lock held! */
514 static inline void
515 locks_insert_global_locks(struct file_lock *fl)
516 {
517 lg_local_lock(&file_lock_lglock);
518 fl->fl_link_cpu = smp_processor_id();
519 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
520 lg_local_unlock(&file_lock_lglock);
521 }
523 /* Must be called with the i_lock held! */
524 static inline void
525 locks_delete_global_locks(struct file_lock *fl)
526 {
527 /*
528 * Avoid taking lock if already unhashed. This is safe since this check
529 * is done while holding the i_lock, and new insertions into the list
530 * also require that it be held.
531 */
532 if (hlist_unhashed(&fl->fl_link))
533 return;
534 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
535 hlist_del_init(&fl->fl_link);
536 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
537 }
539 static unsigned long
540 posix_owner_key(struct file_lock *fl)
541 {
542 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
543 return fl->fl_lmops->lm_owner_key(fl);
544 return (unsigned long)fl->fl_owner;
545 }
547 static inline void
548 locks_insert_global_blocked(struct file_lock *waiter)
549 {
550 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
551 }
553 static inline void
554 locks_delete_global_blocked(struct file_lock *waiter)
555 {
556 hash_del(&waiter->fl_link);
557 }
559 /* Remove waiter from blocker's block list.
560 * When blocker ends up pointing to itself then the list is empty.
561 *
562 * Must be called with blocked_lock_lock held.
563 */
564 static void __locks_delete_block(struct file_lock *waiter)
565 {
566 locks_delete_global_blocked(waiter);
567 list_del_init(&waiter->fl_block);
568 waiter->fl_next = NULL;
569 }
571 static void locks_delete_block(struct file_lock *waiter)
572 {
573 spin_lock(&blocked_lock_lock);
574 __locks_delete_block(waiter);
575 spin_unlock(&blocked_lock_lock);
576 }
578 /* Insert waiter into blocker's block list.
579 * We use a circular list so that processes can be easily woken up in
580 * the order they blocked. The documentation doesn't require this but
581 * it seems like the reasonable thing to do.
582 *
583 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
584 * list itself is protected by the file_lock_list, but by ensuring that the
585 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
586 * in some cases when we see that the fl_block list is empty.
587 */
588 static void __locks_insert_block(struct file_lock *blocker,
589 struct file_lock *waiter)
590 {
591 BUG_ON(!list_empty(&waiter->fl_block));
592 waiter->fl_next = blocker;
593 list_add_tail(&waiter->fl_block, &blocker->fl_block);
594 if (IS_POSIX(blocker))
595 locks_insert_global_blocked(waiter);
596 }
598 /* Must be called with i_lock held. */
599 static void locks_insert_block(struct file_lock *blocker,
600 struct file_lock *waiter)
601 {
602 spin_lock(&blocked_lock_lock);
603 __locks_insert_block(blocker, waiter);
604 spin_unlock(&blocked_lock_lock);
605 }
607 /*
608 * Wake up processes blocked waiting for blocker.
609 *
610 * Must be called with the inode->i_lock held!
611 */
612 static void locks_wake_up_blocks(struct file_lock *blocker)
613 {
614 /*
615 * Avoid taking global lock if list is empty. This is safe since new
616 * blocked requests are only added to the list under the i_lock, and
617 * the i_lock is always held here. Note that removal from the fl_block
618 * list does not require the i_lock, so we must recheck list_empty()
619 * after acquiring the blocked_lock_lock.
620 */
621 if (list_empty(&blocker->fl_block))
622 return;
624 spin_lock(&blocked_lock_lock);
625 while (!list_empty(&blocker->fl_block)) {
626 struct file_lock *waiter;
628 waiter = list_first_entry(&blocker->fl_block,
629 struct file_lock, fl_block);
630 __locks_delete_block(waiter);
631 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
632 waiter->fl_lmops->lm_notify(waiter);
633 else
634 wake_up(&waiter->fl_wait);
635 }
636 spin_unlock(&blocked_lock_lock);
637 }
639 /* Insert file lock fl into an inode's lock list at the position indicated
640 * by pos. At the same time add the lock to the global file lock list.
641 *
642 * Must be called with the i_lock held!
643 */
644 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
645 {
646 fl->fl_nspid = get_pid(task_tgid(current));
648 /* insert into file's list */
649 fl->fl_next = *pos;
650 *pos = fl;
652 locks_insert_global_locks(fl);
653 }
655 /*
656 * Delete a lock and then free it.
657 * Wake up processes that are blocked waiting for this lock,
658 * notify the FS that the lock has been cleared and
659 * finally free the lock.
660 *
661 * Must be called with the i_lock held!
662 */
663 static void locks_delete_lock(struct file_lock **thisfl_p)
664 {
665 struct file_lock *fl = *thisfl_p;
667 locks_delete_global_locks(fl);
669 *thisfl_p = fl->fl_next;
670 fl->fl_next = NULL;
672 if (fl->fl_nspid) {
673 put_pid(fl->fl_nspid);
674 fl->fl_nspid = NULL;
675 }
677 locks_wake_up_blocks(fl);
678 locks_free_lock(fl);
679 }
681 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
682 * checks for shared/exclusive status of overlapping locks.
683 */
684 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
685 {
686 if (sys_fl->fl_type == F_WRLCK)
687 return 1;
688 if (caller_fl->fl_type == F_WRLCK)
689 return 1;
690 return 0;
691 }
693 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
694 * checking before calling the locks_conflict().
695 */
696 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
697 {
698 /* POSIX locks owned by the same process do not conflict with
699 * each other.
700 */
701 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
702 return (0);
704 /* Check whether they overlap */
705 if (!locks_overlap(caller_fl, sys_fl))
706 return 0;
708 return (locks_conflict(caller_fl, sys_fl));
709 }
711 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
712 * checking before calling the locks_conflict().
713 */
714 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
715 {
716 /* FLOCK locks referring to the same filp do not conflict with
717 * each other.
718 */
719 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
720 return (0);
721 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
722 return 0;
724 return (locks_conflict(caller_fl, sys_fl));
725 }
727 void
728 posix_test_lock(struct file *filp, struct file_lock *fl)
729 {
730 struct file_lock *cfl;
731 struct inode *inode = file_inode(filp);
733 spin_lock(&inode->i_lock);
734 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
735 if (!IS_POSIX(cfl))
736 continue;
737 if (posix_locks_conflict(fl, cfl))
738 break;
739 }
740 if (cfl) {
741 __locks_copy_lock(fl, cfl);
742 if (cfl->fl_nspid)
743 fl->fl_pid = pid_vnr(cfl->fl_nspid);
744 } else
745 fl->fl_type = F_UNLCK;
746 spin_unlock(&inode->i_lock);
747 return;
748 }
749 EXPORT_SYMBOL(posix_test_lock);
751 /*
752 * Deadlock detection:
753 *
754 * We attempt to detect deadlocks that are due purely to posix file
755 * locks.
756 *
757 * We assume that a task can be waiting for at most one lock at a time.
758 * So for any acquired lock, the process holding that lock may be
759 * waiting on at most one other lock. That lock in turns may be held by
760 * someone waiting for at most one other lock. Given a requested lock
761 * caller_fl which is about to wait for a conflicting lock block_fl, we
762 * follow this chain of waiters to ensure we are not about to create a
763 * cycle.
764 *
765 * Since we do this before we ever put a process to sleep on a lock, we
766 * are ensured that there is never a cycle; that is what guarantees that
767 * the while() loop in posix_locks_deadlock() eventually completes.
768 *
769 * Note: the above assumption may not be true when handling lock
770 * requests from a broken NFS client. It may also fail in the presence
771 * of tasks (such as posix threads) sharing the same open file table.
772 *
773 * To handle those cases, we just bail out after a few iterations.
774 */
776 #define MAX_DEADLK_ITERATIONS 10
778 /* Find a lock that the owner of the given block_fl is blocking on. */
779 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
780 {
781 struct file_lock *fl;
783 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
784 if (posix_same_owner(fl, block_fl))
785 return fl->fl_next;
786 }
787 return NULL;
788 }
790 /* Must be called with the blocked_lock_lock held! */
791 static int posix_locks_deadlock(struct file_lock *caller_fl,
792 struct file_lock *block_fl)
793 {
794 int i = 0;
796 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
797 if (i++ > MAX_DEADLK_ITERATIONS)
798 return 0;
799 if (posix_same_owner(caller_fl, block_fl))
800 return 1;
801 }
802 return 0;
803 }
805 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
806 * after any leases, but before any posix locks.
807 *
808 * Note that if called with an FL_EXISTS argument, the caller may determine
809 * whether or not a lock was successfully freed by testing the return
810 * value for -ENOENT.
811 */
812 static int flock_lock_file(struct file *filp, struct file_lock *request)
813 {
814 struct file_lock *new_fl = NULL;
815 struct file_lock **before;
816 struct inode * inode = file_inode(filp);
817 int error = 0;
818 int found = 0;
820 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
821 new_fl = locks_alloc_lock();
822 if (!new_fl)
823 return -ENOMEM;
824 }
826 spin_lock(&inode->i_lock);
827 if (request->fl_flags & FL_ACCESS)
828 goto find_conflict;
830 for_each_lock(inode, before) {
831 struct file_lock *fl = *before;
832 if (IS_POSIX(fl))
833 break;
834 if (IS_LEASE(fl))
835 continue;
836 if (filp != fl->fl_file)
837 continue;
838 if (request->fl_type == fl->fl_type)
839 goto out;
840 found = 1;
841 locks_delete_lock(before);
842 break;
843 }
845 if (request->fl_type == F_UNLCK) {
846 if ((request->fl_flags & FL_EXISTS) && !found)
847 error = -ENOENT;
848 goto out;
849 }
851 /*
852 * If a higher-priority process was blocked on the old file lock,
853 * give it the opportunity to lock the file.
854 */
855 if (found) {
856 spin_unlock(&inode->i_lock);
857 cond_resched();
858 spin_lock(&inode->i_lock);
859 }
861 find_conflict:
862 for_each_lock(inode, before) {
863 struct file_lock *fl = *before;
864 if (IS_POSIX(fl))
865 break;
866 if (IS_LEASE(fl))
867 continue;
868 if (!flock_locks_conflict(request, fl))
869 continue;
870 error = -EAGAIN;
871 if (!(request->fl_flags & FL_SLEEP))
872 goto out;
873 error = FILE_LOCK_DEFERRED;
874 locks_insert_block(fl, request);
875 goto out;
876 }
877 if (request->fl_flags & FL_ACCESS)
878 goto out;
879 locks_copy_lock(new_fl, request);
880 locks_insert_lock(before, new_fl);
881 new_fl = NULL;
882 error = 0;
884 out:
885 spin_unlock(&inode->i_lock);
886 if (new_fl)
887 locks_free_lock(new_fl);
888 return error;
889 }
891 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
892 {
893 struct file_lock *fl;
894 struct file_lock *new_fl = NULL;
895 struct file_lock *new_fl2 = NULL;
896 struct file_lock *left = NULL;
897 struct file_lock *right = NULL;
898 struct file_lock **before;
899 int error;
900 bool added = false;
902 /*
903 * We may need two file_lock structures for this operation,
904 * so we get them in advance to avoid races.
905 *
906 * In some cases we can be sure, that no new locks will be needed
907 */
908 if (!(request->fl_flags & FL_ACCESS) &&
909 (request->fl_type != F_UNLCK ||
910 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
911 new_fl = locks_alloc_lock();
912 new_fl2 = locks_alloc_lock();
913 }
915 spin_lock(&inode->i_lock);
916 /*
917 * New lock request. Walk all POSIX locks and look for conflicts. If
918 * there are any, either return error or put the request on the
919 * blocker's list of waiters and the global blocked_hash.
920 */
921 if (request->fl_type != F_UNLCK) {
922 for_each_lock(inode, before) {
923 fl = *before;
924 if (!IS_POSIX(fl))
925 continue;
926 if (!posix_locks_conflict(request, fl))
927 continue;
928 if (conflock)
929 __locks_copy_lock(conflock, fl);
930 error = -EAGAIN;
931 if (!(request->fl_flags & FL_SLEEP))
932 goto out;
933 /*
934 * Deadlock detection and insertion into the blocked
935 * locks list must be done while holding the same lock!
936 */
937 error = -EDEADLK;
938 spin_lock(&blocked_lock_lock);
939 if (likely(!posix_locks_deadlock(request, fl))) {
940 error = FILE_LOCK_DEFERRED;
941 __locks_insert_block(fl, request);
942 }
943 spin_unlock(&blocked_lock_lock);
944 goto out;
945 }
946 }
948 /* If we're just looking for a conflict, we're done. */
949 error = 0;
950 if (request->fl_flags & FL_ACCESS)
951 goto out;
953 /*
954 * Find the first old lock with the same owner as the new lock.
955 */
957 before = &inode->i_flock;
959 /* First skip locks owned by other processes. */
960 while ((fl = *before) && (!IS_POSIX(fl) ||
961 !posix_same_owner(request, fl))) {
962 before = &fl->fl_next;
963 }
965 /* Process locks with this owner. */
966 while ((fl = *before) && posix_same_owner(request, fl)) {
967 /* Detect adjacent or overlapping regions (if same lock type)
968 */
969 if (request->fl_type == fl->fl_type) {
970 /* In all comparisons of start vs end, use
971 * "start - 1" rather than "end + 1". If end
972 * is OFFSET_MAX, end + 1 will become negative.
973 */
974 if (fl->fl_end < request->fl_start - 1)
975 goto next_lock;
976 /* If the next lock in the list has entirely bigger
977 * addresses than the new one, insert the lock here.
978 */
979 if (fl->fl_start - 1 > request->fl_end)
980 break;
982 /* If we come here, the new and old lock are of the
983 * same type and adjacent or overlapping. Make one
984 * lock yielding from the lower start address of both
985 * locks to the higher end address.
986 */
987 if (fl->fl_start > request->fl_start)
988 fl->fl_start = request->fl_start;
989 else
990 request->fl_start = fl->fl_start;
991 if (fl->fl_end < request->fl_end)
992 fl->fl_end = request->fl_end;
993 else
994 request->fl_end = fl->fl_end;
995 if (added) {
996 locks_delete_lock(before);
997 continue;
998 }
999 request = fl;
1000 added = true;
1001 }
1002 else {
1003 /* Processing for different lock types is a bit
1004 * more complex.
1005 */
1006 if (fl->fl_end < request->fl_start)
1007 goto next_lock;
1008 if (fl->fl_start > request->fl_end)
1009 break;
1010 if (request->fl_type == F_UNLCK)
1011 added = true;
1012 if (fl->fl_start < request->fl_start)
1013 left = fl;
1014 /* If the next lock in the list has a higher end
1015 * address than the new one, insert the new one here.
1016 */
1017 if (fl->fl_end > request->fl_end) {
1018 right = fl;
1019 break;
1020 }
1021 if (fl->fl_start >= request->fl_start) {
1022 /* The new lock completely replaces an old
1023 * one (This may happen several times).
1024 */
1025 if (added) {
1026 locks_delete_lock(before);
1027 continue;
1028 }
1029 /* Replace the old lock with the new one.
1030 * Wake up anybody waiting for the old one,
1031 * as the change in lock type might satisfy
1032 * their needs.
1033 */
1034 locks_wake_up_blocks(fl);
1035 fl->fl_start = request->fl_start;
1036 fl->fl_end = request->fl_end;
1037 fl->fl_type = request->fl_type;
1038 locks_release_private(fl);
1039 locks_copy_private(fl, request);
1040 request = fl;
1041 added = true;
1042 }
1043 }
1044 /* Go on to next lock.
1045 */
1046 next_lock:
1047 before = &fl->fl_next;
1048 }
1050 /*
1051 * The above code only modifies existing locks in case of merging or
1052 * replacing. If new lock(s) need to be inserted all modifications are
1053 * done below this, so it's safe yet to bail out.
1054 */
1055 error = -ENOLCK; /* "no luck" */
1056 if (right && left == right && !new_fl2)
1057 goto out;
1059 error = 0;
1060 if (!added) {
1061 if (request->fl_type == F_UNLCK) {
1062 if (request->fl_flags & FL_EXISTS)
1063 error = -ENOENT;
1064 goto out;
1065 }
1067 if (!new_fl) {
1068 error = -ENOLCK;
1069 goto out;
1070 }
1071 locks_copy_lock(new_fl, request);
1072 locks_insert_lock(before, new_fl);
1073 new_fl = NULL;
1074 }
1075 if (right) {
1076 if (left == right) {
1077 /* The new lock breaks the old one in two pieces,
1078 * so we have to use the second new lock.
1079 */
1080 left = new_fl2;
1081 new_fl2 = NULL;
1082 locks_copy_lock(left, right);
1083 locks_insert_lock(before, left);
1084 }
1085 right->fl_start = request->fl_end + 1;
1086 locks_wake_up_blocks(right);
1087 }
1088 if (left) {
1089 left->fl_end = request->fl_start - 1;
1090 locks_wake_up_blocks(left);
1091 }
1092 out:
1093 spin_unlock(&inode->i_lock);
1094 /*
1095 * Free any unused locks.
1096 */
1097 if (new_fl)
1098 locks_free_lock(new_fl);
1099 if (new_fl2)
1100 locks_free_lock(new_fl2);
1101 return error;
1102 }
1104 /**
1105 * posix_lock_file - Apply a POSIX-style lock to a file
1106 * @filp: The file to apply the lock to
1107 * @fl: The lock to be applied
1108 * @conflock: Place to return a copy of the conflicting lock, if found.
1109 *
1110 * Add a POSIX style lock to a file.
1111 * We merge adjacent & overlapping locks whenever possible.
1112 * POSIX locks are sorted by owner task, then by starting address
1113 *
1114 * Note that if called with an FL_EXISTS argument, the caller may determine
1115 * whether or not a lock was successfully freed by testing the return
1116 * value for -ENOENT.
1117 */
1118 int posix_lock_file(struct file *filp, struct file_lock *fl,
1119 struct file_lock *conflock)
1120 {
1121 return __posix_lock_file(file_inode(filp), fl, conflock);
1122 }
1123 EXPORT_SYMBOL(posix_lock_file);
1125 /**
1126 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1127 * @filp: The file to apply the lock to
1128 * @fl: The lock to be applied
1129 *
1130 * Add a POSIX style lock to a file.
1131 * We merge adjacent & overlapping locks whenever possible.
1132 * POSIX locks are sorted by owner task, then by starting address
1133 */
1134 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1135 {
1136 int error;
1137 might_sleep ();
1138 for (;;) {
1139 error = posix_lock_file(filp, fl, NULL);
1140 if (error != FILE_LOCK_DEFERRED)
1141 break;
1142 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1143 if (!error)
1144 continue;
1146 locks_delete_block(fl);
1147 break;
1148 }
1149 return error;
1150 }
1151 EXPORT_SYMBOL(posix_lock_file_wait);
1153 /**
1154 * locks_mandatory_locked - Check for an active lock
1155 * @inode: the file to check
1156 *
1157 * Searches the inode's list of locks to find any POSIX locks which conflict.
1158 * This function is called from locks_verify_locked() only.
1159 */
1160 int locks_mandatory_locked(struct inode *inode)
1161 {
1162 fl_owner_t owner = current->files;
1163 struct file_lock *fl;
1165 /*
1166 * Search the lock list for this inode for any POSIX locks.
1167 */
1168 spin_lock(&inode->i_lock);
1169 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1170 if (!IS_POSIX(fl))
1171 continue;
1172 if (fl->fl_owner != owner)
1173 break;
1174 }
1175 spin_unlock(&inode->i_lock);
1176 return fl ? -EAGAIN : 0;
1177 }
1179 /**
1180 * locks_mandatory_area - Check for a conflicting lock
1181 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1182 * for shared
1183 * @inode: the file to check
1184 * @filp: how the file was opened (if it was)
1185 * @offset: start of area to check
1186 * @count: length of area to check
1187 *
1188 * Searches the inode's list of locks to find any POSIX locks which conflict.
1189 * This function is called from rw_verify_area() and
1190 * locks_verify_truncate().
1191 */
1192 int locks_mandatory_area(int read_write, struct inode *inode,
1193 struct file *filp, loff_t offset,
1194 size_t count)
1195 {
1196 struct file_lock fl;
1197 int error;
1199 locks_init_lock(&fl);
1200 fl.fl_owner = current->files;
1201 fl.fl_pid = current->tgid;
1202 fl.fl_file = filp;
1203 fl.fl_flags = FL_POSIX | FL_ACCESS;
1204 if (filp && !(filp->f_flags & O_NONBLOCK))
1205 fl.fl_flags |= FL_SLEEP;
1206 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1207 fl.fl_start = offset;
1208 fl.fl_end = offset + count - 1;
1210 for (;;) {
1211 error = __posix_lock_file(inode, &fl, NULL);
1212 if (error != FILE_LOCK_DEFERRED)
1213 break;
1214 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1215 if (!error) {
1216 /*
1217 * If we've been sleeping someone might have
1218 * changed the permissions behind our back.
1219 */
1220 if (__mandatory_lock(inode))
1221 continue;
1222 }
1224 locks_delete_block(&fl);
1225 break;
1226 }
1228 return error;
1229 }
1231 EXPORT_SYMBOL(locks_mandatory_area);
1233 static void lease_clear_pending(struct file_lock *fl, int arg)
1234 {
1235 switch (arg) {
1236 case F_UNLCK:
1237 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1238 /* fall through: */
1239 case F_RDLCK:
1240 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1241 }
1242 }
1244 /* We already had a lease on this file; just change its type */
1245 int lease_modify(struct file_lock **before, int arg)
1246 {
1247 struct file_lock *fl = *before;
1248 int error = assign_type(fl, arg);
1250 if (error)
1251 return error;
1252 lease_clear_pending(fl, arg);
1253 locks_wake_up_blocks(fl);
1254 if (arg == F_UNLCK) {
1255 struct file *filp = fl->fl_file;
1257 f_delown(filp);
1258 filp->f_owner.signum = 0;
1259 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1260 if (fl->fl_fasync != NULL) {
1261 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1262 fl->fl_fasync = NULL;
1263 }
1264 locks_delete_lock(before);
1265 }
1266 return 0;
1267 }
1269 EXPORT_SYMBOL(lease_modify);
1271 static bool past_time(unsigned long then)
1272 {
1273 if (!then)
1274 /* 0 is a special value meaning "this never expires": */
1275 return false;
1276 return time_after(jiffies, then);
1277 }
1279 static void time_out_leases(struct inode *inode)
1280 {
1281 struct file_lock **before;
1282 struct file_lock *fl;
1284 before = &inode->i_flock;
1285 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1286 if (past_time(fl->fl_downgrade_time))
1287 lease_modify(before, F_RDLCK);
1288 if (past_time(fl->fl_break_time))
1289 lease_modify(before, F_UNLCK);
1290 if (fl == *before) /* lease_modify may have freed fl */
1291 before = &fl->fl_next;
1292 }
1293 }
1295 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1296 {
1297 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1298 return false;
1299 return locks_conflict(breaker, lease);
1300 }
1302 /**
1303 * __break_lease - revoke all outstanding leases on file
1304 * @inode: the inode of the file to return
1305 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1306 * break all leases
1307 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1308 * only delegations
1309 *
1310 * break_lease (inlined for speed) has checked there already is at least
1311 * some kind of lock (maybe a lease) on this file. Leases are broken on
1312 * a call to open() or truncate(). This function can sleep unless you
1313 * specified %O_NONBLOCK to your open().
1314 */
1315 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1316 {
1317 int error = 0;
1318 struct file_lock *new_fl, *flock;
1319 struct file_lock *fl;
1320 unsigned long break_time;
1321 int i_have_this_lease = 0;
1322 bool lease_conflict = false;
1323 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1325 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1326 if (IS_ERR(new_fl))
1327 return PTR_ERR(new_fl);
1328 new_fl->fl_flags = type;
1330 spin_lock(&inode->i_lock);
1332 time_out_leases(inode);
1334 flock = inode->i_flock;
1335 if ((flock == NULL) || !IS_LEASE(flock))
1336 goto out;
1338 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1339 if (leases_conflict(fl, new_fl)) {
1340 lease_conflict = true;
1341 if (fl->fl_owner == current->files)
1342 i_have_this_lease = 1;
1343 }
1344 }
1345 if (!lease_conflict)
1346 goto out;
1348 break_time = 0;
1349 if (lease_break_time > 0) {
1350 break_time = jiffies + lease_break_time * HZ;
1351 if (break_time == 0)
1352 break_time++; /* so that 0 means no break time */
1353 }
1355 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1356 if (!leases_conflict(fl, new_fl))
1357 continue;
1358 if (want_write) {
1359 if (fl->fl_flags & FL_UNLOCK_PENDING)
1360 continue;
1361 fl->fl_flags |= FL_UNLOCK_PENDING;
1362 fl->fl_break_time = break_time;
1363 } else {
1364 if (lease_breaking(flock))
1365 continue;
1366 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1367 fl->fl_downgrade_time = break_time;
1368 }
1369 fl->fl_lmops->lm_break(fl);
1370 }
1372 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1373 error = -EWOULDBLOCK;
1374 goto out;
1375 }
1377 restart:
1378 break_time = flock->fl_break_time;
1379 if (break_time != 0)
1380 break_time -= jiffies;
1381 if (break_time == 0)
1382 break_time++;
1383 locks_insert_block(flock, new_fl);
1384 spin_unlock(&inode->i_lock);
1385 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1386 !new_fl->fl_next, break_time);
1387 spin_lock(&inode->i_lock);
1388 locks_delete_block(new_fl);
1389 if (error >= 0) {
1390 if (error == 0)
1391 time_out_leases(inode);
1392 /*
1393 * Wait for the next conflicting lease that has not been
1394 * broken yet
1395 */
1396 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1397 flock = flock->fl_next) {
1398 if (leases_conflict(new_fl, flock))
1399 goto restart;
1400 }
1401 error = 0;
1402 }
1404 out:
1405 spin_unlock(&inode->i_lock);
1406 locks_free_lock(new_fl);
1407 return error;
1408 }
1410 EXPORT_SYMBOL(__break_lease);
1412 /**
1413 * lease_get_mtime - get the last modified time of an inode
1414 * @inode: the inode
1415 * @time: pointer to a timespec which will contain the last modified time
1416 *
1417 * This is to force NFS clients to flush their caches for files with
1418 * exclusive leases. The justification is that if someone has an
1419 * exclusive lease, then they could be modifying it.
1420 */
1421 void lease_get_mtime(struct inode *inode, struct timespec *time)
1422 {
1423 struct file_lock *flock = inode->i_flock;
1424 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1425 *time = current_fs_time(inode->i_sb);
1426 else
1427 *time = inode->i_mtime;
1428 }
1430 EXPORT_SYMBOL(lease_get_mtime);
1432 /**
1433 * fcntl_getlease - Enquire what lease is currently active
1434 * @filp: the file
1435 *
1436 * The value returned by this function will be one of
1437 * (if no lease break is pending):
1438 *
1439 * %F_RDLCK to indicate a shared lease is held.
1440 *
1441 * %F_WRLCK to indicate an exclusive lease is held.
1442 *
1443 * %F_UNLCK to indicate no lease is held.
1444 *
1445 * (if a lease break is pending):
1446 *
1447 * %F_RDLCK to indicate an exclusive lease needs to be
1448 * changed to a shared lease (or removed).
1449 *
1450 * %F_UNLCK to indicate the lease needs to be removed.
1451 *
1452 * XXX: sfr & willy disagree over whether F_INPROGRESS
1453 * should be returned to userspace.
1454 */
1455 int fcntl_getlease(struct file *filp)
1456 {
1457 struct file_lock *fl;
1458 struct inode *inode = file_inode(filp);
1459 int type = F_UNLCK;
1461 spin_lock(&inode->i_lock);
1462 time_out_leases(file_inode(filp));
1463 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1464 fl = fl->fl_next) {
1465 if (fl->fl_file == filp) {
1466 type = target_leasetype(fl);
1467 break;
1468 }
1469 }
1470 spin_unlock(&inode->i_lock);
1471 return type;
1472 }
1474 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1475 {
1476 struct file_lock *fl, **before, **my_before = NULL, *lease;
1477 struct dentry *dentry = filp->f_path.dentry;
1478 struct inode *inode = dentry->d_inode;
1479 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1480 int error;
1482 lease = *flp;
1483 /*
1484 * In the delegation case we need mutual exclusion with
1485 * a number of operations that take the i_mutex. We trylock
1486 * because delegations are an optional optimization, and if
1487 * there's some chance of a conflict--we'd rather not
1488 * bother, maybe that's a sign this just isn't a good file to
1489 * hand out a delegation on.
1490 */
1491 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1492 return -EAGAIN;
1494 if (is_deleg && arg == F_WRLCK) {
1495 /* Write delegations are not currently supported: */
1496 mutex_unlock(&inode->i_mutex);
1497 WARN_ON_ONCE(1);
1498 return -EINVAL;
1499 }
1501 error = -EAGAIN;
1502 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1503 goto out;
1504 if ((arg == F_WRLCK)
1505 && ((d_count(dentry) > 1)
1506 || (atomic_read(&inode->i_count) > 1)))
1507 goto out;
1509 /*
1510 * At this point, we know that if there is an exclusive
1511 * lease on this file, then we hold it on this filp
1512 * (otherwise our open of this file would have blocked).
1513 * And if we are trying to acquire an exclusive lease,
1514 * then the file is not open by anyone (including us)
1515 * except for this filp.
1516 */
1517 error = -EAGAIN;
1518 for (before = &inode->i_flock;
1519 ((fl = *before) != NULL) && IS_LEASE(fl);
1520 before = &fl->fl_next) {
1521 if (fl->fl_file == filp) {
1522 my_before = before;
1523 continue;
1524 }
1525 /*
1526 * No exclusive leases if someone else has a lease on
1527 * this file:
1528 */
1529 if (arg == F_WRLCK)
1530 goto out;
1531 /*
1532 * Modifying our existing lease is OK, but no getting a
1533 * new lease if someone else is opening for write:
1534 */
1535 if (fl->fl_flags & FL_UNLOCK_PENDING)
1536 goto out;
1537 }
1539 if (my_before != NULL) {
1540 error = lease->fl_lmops->lm_change(my_before, arg);
1541 if (!error)
1542 *flp = *my_before;
1543 goto out;
1544 }
1546 error = -EINVAL;
1547 if (!leases_enable)
1548 goto out;
1550 locks_insert_lock(before, lease);
1551 error = 0;
1552 out:
1553 if (is_deleg)
1554 mutex_unlock(&inode->i_mutex);
1555 return error;
1556 }
1558 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1559 {
1560 struct file_lock *fl, **before;
1561 struct dentry *dentry = filp->f_path.dentry;
1562 struct inode *inode = dentry->d_inode;
1564 for (before = &inode->i_flock;
1565 ((fl = *before) != NULL) && IS_LEASE(fl);
1566 before = &fl->fl_next) {
1567 if (fl->fl_file != filp)
1568 continue;
1569 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1570 }
1571 return -EAGAIN;
1572 }
1574 /**
1575 * generic_setlease - sets a lease on an open file
1576 * @filp: file pointer
1577 * @arg: type of lease to obtain
1578 * @flp: input - file_lock to use, output - file_lock inserted
1579 *
1580 * The (input) flp->fl_lmops->lm_break function is required
1581 * by break_lease().
1582 *
1583 * Called with inode->i_lock held.
1584 */
1585 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1586 {
1587 struct dentry *dentry = filp->f_path.dentry;
1588 struct inode *inode = dentry->d_inode;
1589 int error;
1591 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1592 return -EACCES;
1593 if (!S_ISREG(inode->i_mode))
1594 return -EINVAL;
1595 error = security_file_lock(filp, arg);
1596 if (error)
1597 return error;
1599 time_out_leases(inode);
1601 BUG_ON(!(*flp)->fl_lmops->lm_break);
1603 switch (arg) {
1604 case F_UNLCK:
1605 return generic_delete_lease(filp, flp);
1606 case F_RDLCK:
1607 case F_WRLCK:
1608 return generic_add_lease(filp, arg, flp);
1609 default:
1610 return -EINVAL;
1611 }
1612 }
1613 EXPORT_SYMBOL(generic_setlease);
1615 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1616 {
1617 if (filp->f_op->setlease)
1618 return filp->f_op->setlease(filp, arg, lease);
1619 else
1620 return generic_setlease(filp, arg, lease);
1621 }
1623 /**
1624 * vfs_setlease - sets a lease on an open file
1625 * @filp: file pointer
1626 * @arg: type of lease to obtain
1627 * @lease: file_lock to use
1628 *
1629 * Call this to establish a lease on the file.
1630 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
1631 * break_lease will oops!
1632 *
1633 * This will call the filesystem's setlease file method, if
1634 * defined. Note that there is no getlease method; instead, the
1635 * filesystem setlease method should call back to setlease() to
1636 * add a lease to the inode's lease list, where fcntl_getlease() can
1637 * find it. Since fcntl_getlease() only reports whether the current
1638 * task holds a lease, a cluster filesystem need only do this for
1639 * leases held by processes on this node.
1640 *
1641 * There is also no break_lease method; filesystems that
1642 * handle their own leases should break leases themselves from the
1643 * filesystem's open, create, and (on truncate) setattr methods.
1644 *
1645 * Warning: the only current setlease methods exist only to disable
1646 * leases in certain cases. More vfs changes may be required to
1647 * allow a full filesystem lease implementation.
1648 */
1650 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1651 {
1652 struct inode *inode = file_inode(filp);
1653 int error;
1655 spin_lock(&inode->i_lock);
1656 error = __vfs_setlease(filp, arg, lease);
1657 spin_unlock(&inode->i_lock);
1659 return error;
1660 }
1661 EXPORT_SYMBOL_GPL(vfs_setlease);
1663 static int do_fcntl_delete_lease(struct file *filp)
1664 {
1665 struct file_lock fl, *flp = &fl;
1667 lease_init(filp, F_UNLCK, flp);
1669 return vfs_setlease(filp, F_UNLCK, &flp);
1670 }
1672 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1673 {
1674 struct file_lock *fl, *ret;
1675 struct inode *inode = file_inode(filp);
1676 struct fasync_struct *new;
1677 int error;
1679 fl = lease_alloc(filp, arg);
1680 if (IS_ERR(fl))
1681 return PTR_ERR(fl);
1683 new = fasync_alloc();
1684 if (!new) {
1685 locks_free_lock(fl);
1686 return -ENOMEM;
1687 }
1688 ret = fl;
1689 spin_lock(&inode->i_lock);
1690 error = __vfs_setlease(filp, arg, &ret);
1691 if (error) {
1692 spin_unlock(&inode->i_lock);
1693 locks_free_lock(fl);
1694 goto out_free_fasync;
1695 }
1696 if (ret != fl)
1697 locks_free_lock(fl);
1699 /*
1700 * fasync_insert_entry() returns the old entry if any.
1701 * If there was no old entry, then it used 'new' and
1702 * inserted it into the fasync list. Clear new so that
1703 * we don't release it here.
1704 */
1705 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1706 new = NULL;
1708 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1709 spin_unlock(&inode->i_lock);
1711 out_free_fasync:
1712 if (new)
1713 fasync_free(new);
1714 return error;
1715 }
1717 /**
1718 * fcntl_setlease - sets a lease on an open file
1719 * @fd: open file descriptor
1720 * @filp: file pointer
1721 * @arg: type of lease to obtain
1722 *
1723 * Call this fcntl to establish a lease on the file.
1724 * Note that you also need to call %F_SETSIG to
1725 * receive a signal when the lease is broken.
1726 */
1727 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1728 {
1729 if (arg == F_UNLCK)
1730 return do_fcntl_delete_lease(filp);
1731 return do_fcntl_add_lease(fd, filp, arg);
1732 }
1734 /**
1735 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1736 * @filp: The file to apply the lock to
1737 * @fl: The lock to be applied
1738 *
1739 * Add a FLOCK style lock to a file.
1740 */
1741 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1742 {
1743 int error;
1744 might_sleep();
1745 for (;;) {
1746 error = flock_lock_file(filp, fl);
1747 if (error != FILE_LOCK_DEFERRED)
1748 break;
1749 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1750 if (!error)
1751 continue;
1753 locks_delete_block(fl);
1754 break;
1755 }
1756 return error;
1757 }
1759 EXPORT_SYMBOL(flock_lock_file_wait);
1761 /**
1762 * sys_flock: - flock() system call.
1763 * @fd: the file descriptor to lock.
1764 * @cmd: the type of lock to apply.
1765 *
1766 * Apply a %FL_FLOCK style lock to an open file descriptor.
1767 * The @cmd can be one of
1768 *
1769 * %LOCK_SH -- a shared lock.
1770 *
1771 * %LOCK_EX -- an exclusive lock.
1772 *
1773 * %LOCK_UN -- remove an existing lock.
1774 *
1775 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1776 *
1777 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1778 * processes read and write access respectively.
1779 */
1780 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1781 {
1782 struct fd f = fdget(fd);
1783 struct file_lock *lock;
1784 int can_sleep, unlock;
1785 int error;
1787 error = -EBADF;
1788 if (!f.file)
1789 goto out;
1791 can_sleep = !(cmd & LOCK_NB);
1792 cmd &= ~LOCK_NB;
1793 unlock = (cmd == LOCK_UN);
1795 if (!unlock && !(cmd & LOCK_MAND) &&
1796 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1797 goto out_putf;
1799 error = flock_make_lock(f.file, &lock, cmd);
1800 if (error)
1801 goto out_putf;
1802 if (can_sleep)
1803 lock->fl_flags |= FL_SLEEP;
1805 error = security_file_lock(f.file, lock->fl_type);
1806 if (error)
1807 goto out_free;
1809 if (f.file->f_op->flock)
1810 error = f.file->f_op->flock(f.file,
1811 (can_sleep) ? F_SETLKW : F_SETLK,
1812 lock);
1813 else
1814 error = flock_lock_file_wait(f.file, lock);
1816 out_free:
1817 locks_free_lock(lock);
1819 out_putf:
1820 fdput(f);
1821 out:
1822 return error;
1823 }
1825 /**
1826 * vfs_test_lock - test file byte range lock
1827 * @filp: The file to test lock for
1828 * @fl: The lock to test; also used to hold result
1829 *
1830 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1831 * setting conf->fl_type to something other than F_UNLCK.
1832 */
1833 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1834 {
1835 if (filp->f_op->lock)
1836 return filp->f_op->lock(filp, F_GETLK, fl);
1837 posix_test_lock(filp, fl);
1838 return 0;
1839 }
1840 EXPORT_SYMBOL_GPL(vfs_test_lock);
1842 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1843 {
1844 flock->l_pid = fl->fl_pid;
1845 #if BITS_PER_LONG == 32
1846 /*
1847 * Make sure we can represent the posix lock via
1848 * legacy 32bit flock.
1849 */
1850 if (fl->fl_start > OFFT_OFFSET_MAX)
1851 return -EOVERFLOW;
1852 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1853 return -EOVERFLOW;
1854 #endif
1855 flock->l_start = fl->fl_start;
1856 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1857 fl->fl_end - fl->fl_start + 1;
1858 flock->l_whence = 0;
1859 flock->l_type = fl->fl_type;
1860 return 0;
1861 }
1863 #if BITS_PER_LONG == 32
1864 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1865 {
1866 flock->l_pid = fl->fl_pid;
1867 flock->l_start = fl->fl_start;
1868 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1869 fl->fl_end - fl->fl_start + 1;
1870 flock->l_whence = 0;
1871 flock->l_type = fl->fl_type;
1872 }
1873 #endif
1875 /* Report the first existing lock that would conflict with l.
1876 * This implements the F_GETLK command of fcntl().
1877 */
1878 int fcntl_getlk(struct file *filp, struct flock __user *l)
1879 {
1880 struct file_lock file_lock;
1881 struct flock flock;
1882 int error;
1884 error = -EFAULT;
1885 if (copy_from_user(&flock, l, sizeof(flock)))
1886 goto out;
1887 error = -EINVAL;
1888 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1889 goto out;
1891 error = flock_to_posix_lock(filp, &file_lock, &flock);
1892 if (error)
1893 goto out;
1895 error = vfs_test_lock(filp, &file_lock);
1896 if (error)
1897 goto out;
1899 flock.l_type = file_lock.fl_type;
1900 if (file_lock.fl_type != F_UNLCK) {
1901 error = posix_lock_to_flock(&flock, &file_lock);
1902 if (error)
1903 goto out;
1904 }
1905 error = -EFAULT;
1906 if (!copy_to_user(l, &flock, sizeof(flock)))
1907 error = 0;
1908 out:
1909 return error;
1910 }
1912 /**
1913 * vfs_lock_file - file byte range lock
1914 * @filp: The file to apply the lock to
1915 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1916 * @fl: The lock to be applied
1917 * @conf: Place to return a copy of the conflicting lock, if found.
1918 *
1919 * A caller that doesn't care about the conflicting lock may pass NULL
1920 * as the final argument.
1921 *
1922 * If the filesystem defines a private ->lock() method, then @conf will
1923 * be left unchanged; so a caller that cares should initialize it to
1924 * some acceptable default.
1925 *
1926 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1927 * locks, the ->lock() interface may return asynchronously, before the lock has
1928 * been granted or denied by the underlying filesystem, if (and only if)
1929 * lm_grant is set. Callers expecting ->lock() to return asynchronously
1930 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1931 * the request is for a blocking lock. When ->lock() does return asynchronously,
1932 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1933 * request completes.
1934 * If the request is for non-blocking lock the file system should return
1935 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1936 * with the result. If the request timed out the callback routine will return a
1937 * nonzero return code and the file system should release the lock. The file
1938 * system is also responsible to keep a corresponding posix lock when it
1939 * grants a lock so the VFS can find out which locks are locally held and do
1940 * the correct lock cleanup when required.
1941 * The underlying filesystem must not drop the kernel lock or call
1942 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1943 * return code.
1944 */
1945 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1946 {
1947 if (filp->f_op->lock)
1948 return filp->f_op->lock(filp, cmd, fl);
1949 else
1950 return posix_lock_file(filp, fl, conf);
1951 }
1952 EXPORT_SYMBOL_GPL(vfs_lock_file);
1954 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1955 struct file_lock *fl)
1956 {
1957 int error;
1959 error = security_file_lock(filp, fl->fl_type);
1960 if (error)
1961 return error;
1963 for (;;) {
1964 error = vfs_lock_file(filp, cmd, fl, NULL);
1965 if (error != FILE_LOCK_DEFERRED)
1966 break;
1967 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1968 if (!error)
1969 continue;
1971 locks_delete_block(fl);
1972 break;
1973 }
1975 return error;
1976 }
1978 /* Apply the lock described by l to an open file descriptor.
1979 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1980 */
1981 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1982 struct flock __user *l)
1983 {
1984 struct file_lock *file_lock = locks_alloc_lock();
1985 struct flock flock;
1986 struct inode *inode;
1987 struct file *f;
1988 int error;
1990 if (file_lock == NULL)
1991 return -ENOLCK;
1993 /*
1994 * This might block, so we do it before checking the inode.
1995 */
1996 error = -EFAULT;
1997 if (copy_from_user(&flock, l, sizeof(flock)))
1998 goto out;
2000 inode = file_inode(filp);
2002 /* Don't allow mandatory locks on files that may be memory mapped
2003 * and shared.
2004 */
2005 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2006 error = -EAGAIN;
2007 goto out;
2008 }
2010 again:
2011 error = flock_to_posix_lock(filp, file_lock, &flock);
2012 if (error)
2013 goto out;
2014 if (cmd == F_SETLKW) {
2015 file_lock->fl_flags |= FL_SLEEP;
2016 }
2018 error = -EBADF;
2019 switch (flock.l_type) {
2020 case F_RDLCK:
2021 if (!(filp->f_mode & FMODE_READ))
2022 goto out;
2023 break;
2024 case F_WRLCK:
2025 if (!(filp->f_mode & FMODE_WRITE))
2026 goto out;
2027 break;
2028 case F_UNLCK:
2029 break;
2030 default:
2031 error = -EINVAL;
2032 goto out;
2033 }
2035 error = do_lock_file_wait(filp, cmd, file_lock);
2037 /*
2038 * Attempt to detect a close/fcntl race and recover by
2039 * releasing the lock that was just acquired.
2040 */
2041 /*
2042 * we need that spin_lock here - it prevents reordering between
2043 * update of inode->i_flock and check for it done in close().
2044 * rcu_read_lock() wouldn't do.
2045 */
2046 spin_lock(¤t->files->file_lock);
2047 f = fcheck(fd);
2048 spin_unlock(¤t->files->file_lock);
2049 if (!error && f != filp && flock.l_type != F_UNLCK) {
2050 flock.l_type = F_UNLCK;
2051 goto again;
2052 }
2054 out:
2055 locks_free_lock(file_lock);
2056 return error;
2057 }
2059 #if BITS_PER_LONG == 32
2060 /* Report the first existing lock that would conflict with l.
2061 * This implements the F_GETLK command of fcntl().
2062 */
2063 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
2064 {
2065 struct file_lock file_lock;
2066 struct flock64 flock;
2067 int error;
2069 error = -EFAULT;
2070 if (copy_from_user(&flock, l, sizeof(flock)))
2071 goto out;
2072 error = -EINVAL;
2073 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2074 goto out;
2076 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2077 if (error)
2078 goto out;
2080 error = vfs_test_lock(filp, &file_lock);
2081 if (error)
2082 goto out;
2084 flock.l_type = file_lock.fl_type;
2085 if (file_lock.fl_type != F_UNLCK)
2086 posix_lock_to_flock64(&flock, &file_lock);
2088 error = -EFAULT;
2089 if (!copy_to_user(l, &flock, sizeof(flock)))
2090 error = 0;
2092 out:
2093 return error;
2094 }
2096 /* Apply the lock described by l to an open file descriptor.
2097 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2098 */
2099 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2100 struct flock64 __user *l)
2101 {
2102 struct file_lock *file_lock = locks_alloc_lock();
2103 struct flock64 flock;
2104 struct inode *inode;
2105 struct file *f;
2106 int error;
2108 if (file_lock == NULL)
2109 return -ENOLCK;
2111 /*
2112 * This might block, so we do it before checking the inode.
2113 */
2114 error = -EFAULT;
2115 if (copy_from_user(&flock, l, sizeof(flock)))
2116 goto out;
2118 inode = file_inode(filp);
2120 /* Don't allow mandatory locks on files that may be memory mapped
2121 * and shared.
2122 */
2123 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2124 error = -EAGAIN;
2125 goto out;
2126 }
2128 again:
2129 error = flock64_to_posix_lock(filp, file_lock, &flock);
2130 if (error)
2131 goto out;
2132 if (cmd == F_SETLKW64) {
2133 file_lock->fl_flags |= FL_SLEEP;
2134 }
2136 error = -EBADF;
2137 switch (flock.l_type) {
2138 case F_RDLCK:
2139 if (!(filp->f_mode & FMODE_READ))
2140 goto out;
2141 break;
2142 case F_WRLCK:
2143 if (!(filp->f_mode & FMODE_WRITE))
2144 goto out;
2145 break;
2146 case F_UNLCK:
2147 break;
2148 default:
2149 error = -EINVAL;
2150 goto out;
2151 }
2153 error = do_lock_file_wait(filp, cmd, file_lock);
2155 /*
2156 * Attempt to detect a close/fcntl race and recover by
2157 * releasing the lock that was just acquired.
2158 */
2159 spin_lock(¤t->files->file_lock);
2160 f = fcheck(fd);
2161 spin_unlock(¤t->files->file_lock);
2162 if (!error && f != filp && flock.l_type != F_UNLCK) {
2163 flock.l_type = F_UNLCK;
2164 goto again;
2165 }
2167 out:
2168 locks_free_lock(file_lock);
2169 return error;
2170 }
2171 #endif /* BITS_PER_LONG == 32 */
2173 /*
2174 * This function is called when the file is being removed
2175 * from the task's fd array. POSIX locks belonging to this task
2176 * are deleted at this time.
2177 */
2178 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2179 {
2180 struct file_lock lock;
2182 /*
2183 * If there are no locks held on this file, we don't need to call
2184 * posix_lock_file(). Another process could be setting a lock on this
2185 * file at the same time, but we wouldn't remove that lock anyway.
2186 */
2187 if (!file_inode(filp)->i_flock)
2188 return;
2190 lock.fl_type = F_UNLCK;
2191 lock.fl_flags = FL_POSIX | FL_CLOSE;
2192 lock.fl_start = 0;
2193 lock.fl_end = OFFSET_MAX;
2194 lock.fl_owner = owner;
2195 lock.fl_pid = current->tgid;
2196 lock.fl_file = filp;
2197 lock.fl_ops = NULL;
2198 lock.fl_lmops = NULL;
2200 vfs_lock_file(filp, F_SETLK, &lock, NULL);
2202 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2203 lock.fl_ops->fl_release_private(&lock);
2204 }
2206 EXPORT_SYMBOL(locks_remove_posix);
2208 /*
2209 * This function is called on the last close of an open file.
2210 */
2211 void locks_remove_flock(struct file *filp)
2212 {
2213 struct inode * inode = file_inode(filp);
2214 struct file_lock *fl;
2215 struct file_lock **before;
2217 if (!inode->i_flock)
2218 return;
2220 if (filp->f_op->flock) {
2221 struct file_lock fl = {
2222 .fl_pid = current->tgid,
2223 .fl_file = filp,
2224 .fl_flags = FL_FLOCK,
2225 .fl_type = F_UNLCK,
2226 .fl_end = OFFSET_MAX,
2227 };
2228 filp->f_op->flock(filp, F_SETLKW, &fl);
2229 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2230 fl.fl_ops->fl_release_private(&fl);
2231 }
2233 spin_lock(&inode->i_lock);
2234 before = &inode->i_flock;
2236 while ((fl = *before) != NULL) {
2237 if (fl->fl_file == filp) {
2238 if (IS_LEASE(fl)) {
2239 lease_modify(before, F_UNLCK);
2240 continue;
2241 }
2243 /*
2244 * There's a leftover lock on the list of a type that
2245 * we didn't expect to see. Most likely a classic
2246 * POSIX lock that ended up not getting released
2247 * properly, or that raced onto the list somehow. Log
2248 * some info about it and then just remove it from
2249 * the list.
2250 */
2251 WARN(!IS_FLOCK(fl),
2252 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2253 MAJOR(inode->i_sb->s_dev),
2254 MINOR(inode->i_sb->s_dev), inode->i_ino,
2255 fl->fl_type, fl->fl_flags,
2256 fl->fl_start, fl->fl_end);
2258 locks_delete_lock(before);
2259 continue;
2260 }
2261 before = &fl->fl_next;
2262 }
2263 spin_unlock(&inode->i_lock);
2264 }
2266 /**
2267 * posix_unblock_lock - stop waiting for a file lock
2268 * @waiter: the lock which was waiting
2269 *
2270 * lockd needs to block waiting for locks.
2271 */
2272 int
2273 posix_unblock_lock(struct file_lock *waiter)
2274 {
2275 int status = 0;
2277 spin_lock(&blocked_lock_lock);
2278 if (waiter->fl_next)
2279 __locks_delete_block(waiter);
2280 else
2281 status = -ENOENT;
2282 spin_unlock(&blocked_lock_lock);
2283 return status;
2284 }
2285 EXPORT_SYMBOL(posix_unblock_lock);
2287 /**
2288 * vfs_cancel_lock - file byte range unblock lock
2289 * @filp: The file to apply the unblock to
2290 * @fl: The lock to be unblocked
2291 *
2292 * Used by lock managers to cancel blocked requests
2293 */
2294 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2295 {
2296 if (filp->f_op->lock)
2297 return filp->f_op->lock(filp, F_CANCELLK, fl);
2298 return 0;
2299 }
2301 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2303 #ifdef CONFIG_PROC_FS
2304 #include <linux/proc_fs.h>
2305 #include <linux/seq_file.h>
2307 struct locks_iterator {
2308 int li_cpu;
2309 loff_t li_pos;
2310 };
2312 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2313 loff_t id, char *pfx)
2314 {
2315 struct inode *inode = NULL;
2316 unsigned int fl_pid;
2318 if (fl->fl_nspid)
2319 fl_pid = pid_vnr(fl->fl_nspid);
2320 else
2321 fl_pid = fl->fl_pid;
2323 if (fl->fl_file != NULL)
2324 inode = file_inode(fl->fl_file);
2326 seq_printf(f, "%lld:%s ", id, pfx);
2327 if (IS_POSIX(fl)) {
2328 seq_printf(f, "%6s %s ",
2329 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2330 (inode == NULL) ? "*NOINODE*" :
2331 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2332 } else if (IS_FLOCK(fl)) {
2333 if (fl->fl_type & LOCK_MAND) {
2334 seq_printf(f, "FLOCK MSNFS ");
2335 } else {
2336 seq_printf(f, "FLOCK ADVISORY ");
2337 }
2338 } else if (IS_LEASE(fl)) {
2339 seq_printf(f, "LEASE ");
2340 if (lease_breaking(fl))
2341 seq_printf(f, "BREAKING ");
2342 else if (fl->fl_file)
2343 seq_printf(f, "ACTIVE ");
2344 else
2345 seq_printf(f, "BREAKER ");
2346 } else {
2347 seq_printf(f, "UNKNOWN UNKNOWN ");
2348 }
2349 if (fl->fl_type & LOCK_MAND) {
2350 seq_printf(f, "%s ",
2351 (fl->fl_type & LOCK_READ)
2352 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2353 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2354 } else {
2355 seq_printf(f, "%s ",
2356 (lease_breaking(fl))
2357 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2358 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2359 }
2360 if (inode) {
2361 #ifdef WE_CAN_BREAK_LSLK_NOW
2362 seq_printf(f, "%d %s:%ld ", fl_pid,
2363 inode->i_sb->s_id, inode->i_ino);
2364 #else
2365 /* userspace relies on this representation of dev_t ;-( */
2366 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2367 MAJOR(inode->i_sb->s_dev),
2368 MINOR(inode->i_sb->s_dev), inode->i_ino);
2369 #endif
2370 } else {
2371 seq_printf(f, "%d <none>:0 ", fl_pid);
2372 }
2373 if (IS_POSIX(fl)) {
2374 if (fl->fl_end == OFFSET_MAX)
2375 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2376 else
2377 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2378 } else {
2379 seq_printf(f, "0 EOF\n");
2380 }
2381 }
2383 static int locks_show(struct seq_file *f, void *v)
2384 {
2385 struct locks_iterator *iter = f->private;
2386 struct file_lock *fl, *bfl;
2388 fl = hlist_entry(v, struct file_lock, fl_link);
2390 lock_get_status(f, fl, iter->li_pos, "");
2392 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2393 lock_get_status(f, bfl, iter->li_pos, " ->");
2395 return 0;
2396 }
2398 static void *locks_start(struct seq_file *f, loff_t *pos)
2399 {
2400 struct locks_iterator *iter = f->private;
2402 iter->li_pos = *pos + 1;
2403 lg_global_lock(&file_lock_lglock);
2404 spin_lock(&blocked_lock_lock);
2405 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2406 }
2408 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2409 {
2410 struct locks_iterator *iter = f->private;
2412 ++iter->li_pos;
2413 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2414 }
2416 static void locks_stop(struct seq_file *f, void *v)
2417 {
2418 spin_unlock(&blocked_lock_lock);
2419 lg_global_unlock(&file_lock_lglock);
2420 }
2422 static const struct seq_operations locks_seq_operations = {
2423 .start = locks_start,
2424 .next = locks_next,
2425 .stop = locks_stop,
2426 .show = locks_show,
2427 };
2429 static int locks_open(struct inode *inode, struct file *filp)
2430 {
2431 return seq_open_private(filp, &locks_seq_operations,
2432 sizeof(struct locks_iterator));
2433 }
2435 static const struct file_operations proc_locks_operations = {
2436 .open = locks_open,
2437 .read = seq_read,
2438 .llseek = seq_lseek,
2439 .release = seq_release_private,
2440 };
2442 static int __init proc_locks_init(void)
2443 {
2444 proc_create("locks", 0, NULL, &proc_locks_operations);
2445 return 0;
2446 }
2447 module_init(proc_locks_init);
2448 #endif
2450 /**
2451 * lock_may_read - checks that the region is free of locks
2452 * @inode: the inode that is being read
2453 * @start: the first byte to read
2454 * @len: the number of bytes to read
2455 *
2456 * Emulates Windows locking requirements. Whole-file
2457 * mandatory locks (share modes) can prohibit a read and
2458 * byte-range POSIX locks can prohibit a read if they overlap.
2459 *
2460 * N.B. this function is only ever called
2461 * from knfsd and ownership of locks is never checked.
2462 */
2463 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2464 {
2465 struct file_lock *fl;
2466 int result = 1;
2468 spin_lock(&inode->i_lock);
2469 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2470 if (IS_POSIX(fl)) {
2471 if (fl->fl_type == F_RDLCK)
2472 continue;
2473 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2474 continue;
2475 } else if (IS_FLOCK(fl)) {
2476 if (!(fl->fl_type & LOCK_MAND))
2477 continue;
2478 if (fl->fl_type & LOCK_READ)
2479 continue;
2480 } else
2481 continue;
2482 result = 0;
2483 break;
2484 }
2485 spin_unlock(&inode->i_lock);
2486 return result;
2487 }
2489 EXPORT_SYMBOL(lock_may_read);
2491 /**
2492 * lock_may_write - checks that the region is free of locks
2493 * @inode: the inode that is being written
2494 * @start: the first byte to write
2495 * @len: the number of bytes to write
2496 *
2497 * Emulates Windows locking requirements. Whole-file
2498 * mandatory locks (share modes) can prohibit a write and
2499 * byte-range POSIX locks can prohibit a write if they overlap.
2500 *
2501 * N.B. this function is only ever called
2502 * from knfsd and ownership of locks is never checked.
2503 */
2504 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2505 {
2506 struct file_lock *fl;
2507 int result = 1;
2509 spin_lock(&inode->i_lock);
2510 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2511 if (IS_POSIX(fl)) {
2512 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2513 continue;
2514 } else if (IS_FLOCK(fl)) {
2515 if (!(fl->fl_type & LOCK_MAND))
2516 continue;
2517 if (fl->fl_type & LOCK_WRITE)
2518 continue;
2519 } else
2520 continue;
2521 result = 0;
2522 break;
2523 }
2524 spin_unlock(&inode->i_lock);
2525 return result;
2526 }
2528 EXPORT_SYMBOL(lock_may_write);
2530 static int __init filelock_init(void)
2531 {
2532 int i;
2534 filelock_cache = kmem_cache_create("file_lock_cache",
2535 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2537 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2539 for_each_possible_cpu(i)
2540 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2542 return 0;
2543 }
2545 core_initcall(filelock_init);