xref: /openbmc/linux/fs/locks.c (revision ee8a99bd)
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 #include <asm/uaccess.h>
134 
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)
138 
139 static bool lease_breaking(struct file_lock *fl)
140 {
141 	return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
142 }
143 
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 }
152 
153 int leases_enable = 1;
154 int lease_break_time = 45;
155 
156 #define for_each_lock(inode, lockp) \
157 	for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
158 
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);
167 
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);
181 
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);
200 
201 static struct kmem_cache *filelock_cache __read_mostly;
202 
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 }
209 
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);
214 
215 	if (fl)
216 		locks_init_lock_heads(fl);
217 
218 	return fl;
219 }
220 EXPORT_SYMBOL_GPL(locks_alloc_lock);
221 
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;
230 
231 }
232 EXPORT_SYMBOL_GPL(locks_release_private);
233 
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));
240 
241 	locks_release_private(fl);
242 	kmem_cache_free(filelock_cache, fl);
243 }
244 EXPORT_SYMBOL(locks_free_lock);
245 
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 }
251 
252 EXPORT_SYMBOL(locks_init_lock);
253 
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 }
264 
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);
281 
282 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
283 {
284 	locks_release_private(new);
285 
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;
290 
291 	locks_copy_private(new, fl);
292 }
293 
294 EXPORT_SYMBOL(locks_copy_lock);
295 
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 }
309 
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;
318 
319 	fl = locks_alloc_lock();
320 	if (fl == NULL)
321 		return -ENOMEM;
322 
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;
328 
329 	*lock = fl;
330 	return 0;
331 }
332 
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 }
346 
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;
354 
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 	}
368 
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;
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 #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;
404 
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 	}
418 
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;
434 
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;
441 
442 	return assign_type(fl, l->l_type);
443 }
444 #endif
445 
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 }
451 
452 static const struct lock_manager_operations lease_manager_ops = {
453 	.lm_break = lease_break_callback,
454 	.lm_change = lease_modify,
455 };
456 
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;
464 
465 	fl->fl_owner = current->files;
466 	fl->fl_pid = current->tgid;
467 
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 }
476 
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;
482 
483 	if (fl == NULL)
484 		return ERR_PTR(error);
485 
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 }
493 
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 }
501 
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 }
512 
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 }
522 
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 }
538 
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 }
546 
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 }
552 
553 static inline void
554 locks_delete_global_blocked(struct file_lock *waiter)
555 {
556 	hash_del(&waiter->fl_link);
557 }
558 
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 }
570 
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 }
577 
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 }
597 
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 }
606 
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;
623 
624 	spin_lock(&blocked_lock_lock);
625 	while (!list_empty(&blocker->fl_block)) {
626 		struct file_lock *waiter;
627 
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 }
638 
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));
647 
648 	/* insert into file's list */
649 	fl->fl_next = *pos;
650 	*pos = fl;
651 
652 	locks_insert_global_locks(fl);
653 }
654 
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;
666 
667 	locks_delete_global_locks(fl);
668 
669 	*thisfl_p = fl->fl_next;
670 	fl->fl_next = NULL;
671 
672 	if (fl->fl_nspid) {
673 		put_pid(fl->fl_nspid);
674 		fl->fl_nspid = NULL;
675 	}
676 
677 	locks_wake_up_blocks(fl);
678 	locks_free_lock(fl);
679 }
680 
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 }
692 
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);
703 
704 	/* Check whether they overlap */
705 	if (!locks_overlap(caller_fl, sys_fl))
706 		return 0;
707 
708 	return (locks_conflict(caller_fl, sys_fl));
709 }
710 
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;
723 
724 	return (locks_conflict(caller_fl, sys_fl));
725 }
726 
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);
732 
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);
750 
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  */
775 
776 #define MAX_DEADLK_ITERATIONS 10
777 
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;
782 
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 }
789 
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;
795 
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 }
804 
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;
819 
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 	}
825 
826 	spin_lock(&inode->i_lock);
827 	if (request->fl_flags & FL_ACCESS)
828 		goto find_conflict;
829 
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 	}
844 
845 	if (request->fl_type == F_UNLCK) {
846 		if ((request->fl_flags & FL_EXISTS) && !found)
847 			error = -ENOENT;
848 		goto out;
849 	}
850 
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 	}
860 
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;
883 
884 out:
885 	spin_unlock(&inode->i_lock);
886 	if (new_fl)
887 		locks_free_lock(new_fl);
888 	return error;
889 }
890 
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;
901 
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 	}
914 
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   	}
947 
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;
952 
953 	/*
954 	 * Find the first old lock with the same owner as the new lock.
955 	 */
956 
957 	before = &inode->i_flock;
958 
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 	}
964 
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;
981 
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 	}
1049 
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;
1058 
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 		}
1066 
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 }
1103 
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);
1124 
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;
1145 
1146 		locks_delete_block(fl);
1147 		break;
1148 	}
1149 	return error;
1150 }
1151 EXPORT_SYMBOL(posix_lock_file_wait);
1152 
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;
1164 
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 }
1178 
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;
1198 
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;
1209 
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 		}
1223 
1224 		locks_delete_block(&fl);
1225 		break;
1226 	}
1227 
1228 	return error;
1229 }
1230 
1231 EXPORT_SYMBOL(locks_mandatory_area);
1232 
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 }
1243 
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);
1249 
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;
1256 
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 }
1268 
1269 EXPORT_SYMBOL(lease_modify);
1270 
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 }
1278 
1279 static void time_out_leases(struct inode *inode)
1280 {
1281 	struct file_lock **before;
1282 	struct file_lock *fl;
1283 
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 }
1294 
1295 /**
1296  *	__break_lease	-	revoke all outstanding leases on file
1297  *	@inode: the inode of the file to return
1298  *	@mode: the open mode (read or write)
1299  *
1300  *	break_lease (inlined for speed) has checked there already is at least
1301  *	some kind of lock (maybe a lease) on this file.  Leases are broken on
1302  *	a call to open() or truncate().  This function can sleep unless you
1303  *	specified %O_NONBLOCK to your open().
1304  */
1305 int __break_lease(struct inode *inode, unsigned int mode)
1306 {
1307 	int error = 0;
1308 	struct file_lock *new_fl, *flock;
1309 	struct file_lock *fl;
1310 	unsigned long break_time;
1311 	int i_have_this_lease = 0;
1312 	int want_write = (mode & O_ACCMODE) != O_RDONLY;
1313 
1314 	new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1315 	if (IS_ERR(new_fl))
1316 		return PTR_ERR(new_fl);
1317 
1318 	spin_lock(&inode->i_lock);
1319 
1320 	time_out_leases(inode);
1321 
1322 	flock = inode->i_flock;
1323 	if ((flock == NULL) || !IS_LEASE(flock))
1324 		goto out;
1325 
1326 	if (!locks_conflict(flock, new_fl))
1327 		goto out;
1328 
1329 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1330 		if (fl->fl_owner == current->files)
1331 			i_have_this_lease = 1;
1332 
1333 	break_time = 0;
1334 	if (lease_break_time > 0) {
1335 		break_time = jiffies + lease_break_time * HZ;
1336 		if (break_time == 0)
1337 			break_time++;	/* so that 0 means no break time */
1338 	}
1339 
1340 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1341 		if (want_write) {
1342 			if (fl->fl_flags & FL_UNLOCK_PENDING)
1343 				continue;
1344 			fl->fl_flags |= FL_UNLOCK_PENDING;
1345 			fl->fl_break_time = break_time;
1346 		} else {
1347 			if (lease_breaking(flock))
1348 				continue;
1349 			fl->fl_flags |= FL_DOWNGRADE_PENDING;
1350 			fl->fl_downgrade_time = break_time;
1351 		}
1352 		fl->fl_lmops->lm_break(fl);
1353 	}
1354 
1355 	if (i_have_this_lease || (mode & O_NONBLOCK)) {
1356 		error = -EWOULDBLOCK;
1357 		goto out;
1358 	}
1359 
1360 restart:
1361 	break_time = flock->fl_break_time;
1362 	if (break_time != 0) {
1363 		break_time -= jiffies;
1364 		if (break_time == 0)
1365 			break_time++;
1366 	}
1367 	locks_insert_block(flock, new_fl);
1368 	spin_unlock(&inode->i_lock);
1369 	error = wait_event_interruptible_timeout(new_fl->fl_wait,
1370 						!new_fl->fl_next, break_time);
1371 	spin_lock(&inode->i_lock);
1372 	locks_delete_block(new_fl);
1373 	if (error >= 0) {
1374 		if (error == 0)
1375 			time_out_leases(inode);
1376 		/*
1377 		 * Wait for the next conflicting lease that has not been
1378 		 * broken yet
1379 		 */
1380 		for (flock = inode->i_flock; flock && IS_LEASE(flock);
1381 				flock = flock->fl_next) {
1382 			if (locks_conflict(new_fl, flock))
1383 				goto restart;
1384 		}
1385 		error = 0;
1386 	}
1387 
1388 out:
1389 	spin_unlock(&inode->i_lock);
1390 	locks_free_lock(new_fl);
1391 	return error;
1392 }
1393 
1394 EXPORT_SYMBOL(__break_lease);
1395 
1396 /**
1397  *	lease_get_mtime - get the last modified time of an inode
1398  *	@inode: the inode
1399  *      @time:  pointer to a timespec which will contain the last modified time
1400  *
1401  * This is to force NFS clients to flush their caches for files with
1402  * exclusive leases.  The justification is that if someone has an
1403  * exclusive lease, then they could be modifying it.
1404  */
1405 void lease_get_mtime(struct inode *inode, struct timespec *time)
1406 {
1407 	struct file_lock *flock = inode->i_flock;
1408 	if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1409 		*time = current_fs_time(inode->i_sb);
1410 	else
1411 		*time = inode->i_mtime;
1412 }
1413 
1414 EXPORT_SYMBOL(lease_get_mtime);
1415 
1416 /**
1417  *	fcntl_getlease - Enquire what lease is currently active
1418  *	@filp: the file
1419  *
1420  *	The value returned by this function will be one of
1421  *	(if no lease break is pending):
1422  *
1423  *	%F_RDLCK to indicate a shared lease is held.
1424  *
1425  *	%F_WRLCK to indicate an exclusive lease is held.
1426  *
1427  *	%F_UNLCK to indicate no lease is held.
1428  *
1429  *	(if a lease break is pending):
1430  *
1431  *	%F_RDLCK to indicate an exclusive lease needs to be
1432  *		changed to a shared lease (or removed).
1433  *
1434  *	%F_UNLCK to indicate the lease needs to be removed.
1435  *
1436  *	XXX: sfr & willy disagree over whether F_INPROGRESS
1437  *	should be returned to userspace.
1438  */
1439 int fcntl_getlease(struct file *filp)
1440 {
1441 	struct file_lock *fl;
1442 	struct inode *inode = file_inode(filp);
1443 	int type = F_UNLCK;
1444 
1445 	spin_lock(&inode->i_lock);
1446 	time_out_leases(file_inode(filp));
1447 	for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1448 			fl = fl->fl_next) {
1449 		if (fl->fl_file == filp) {
1450 			type = target_leasetype(fl);
1451 			break;
1452 		}
1453 	}
1454 	spin_unlock(&inode->i_lock);
1455 	return type;
1456 }
1457 
1458 static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1459 {
1460 	struct file_lock *fl, **before, **my_before = NULL, *lease;
1461 	struct dentry *dentry = filp->f_path.dentry;
1462 	struct inode *inode = dentry->d_inode;
1463 	int error;
1464 
1465 	lease = *flp;
1466 
1467 	error = -EAGAIN;
1468 	if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1469 		goto out;
1470 	if ((arg == F_WRLCK)
1471 	    && ((d_count(dentry) > 1)
1472 		|| (atomic_read(&inode->i_count) > 1)))
1473 		goto out;
1474 
1475 	/*
1476 	 * At this point, we know that if there is an exclusive
1477 	 * lease on this file, then we hold it on this filp
1478 	 * (otherwise our open of this file would have blocked).
1479 	 * And if we are trying to acquire an exclusive lease,
1480 	 * then the file is not open by anyone (including us)
1481 	 * except for this filp.
1482 	 */
1483 	error = -EAGAIN;
1484 	for (before = &inode->i_flock;
1485 			((fl = *before) != NULL) && IS_LEASE(fl);
1486 			before = &fl->fl_next) {
1487 		if (fl->fl_file == filp) {
1488 			my_before = before;
1489 			continue;
1490 		}
1491 		/*
1492 		 * No exclusive leases if someone else has a lease on
1493 		 * this file:
1494 		 */
1495 		if (arg == F_WRLCK)
1496 			goto out;
1497 		/*
1498 		 * Modifying our existing lease is OK, but no getting a
1499 		 * new lease if someone else is opening for write:
1500 		 */
1501 		if (fl->fl_flags & FL_UNLOCK_PENDING)
1502 			goto out;
1503 	}
1504 
1505 	if (my_before != NULL) {
1506 		error = lease->fl_lmops->lm_change(my_before, arg);
1507 		if (!error)
1508 			*flp = *my_before;
1509 		goto out;
1510 	}
1511 
1512 	error = -EINVAL;
1513 	if (!leases_enable)
1514 		goto out;
1515 
1516 	locks_insert_lock(before, lease);
1517 	return 0;
1518 
1519 out:
1520 	return error;
1521 }
1522 
1523 static int generic_delete_lease(struct file *filp, struct file_lock **flp)
1524 {
1525 	struct file_lock *fl, **before;
1526 	struct dentry *dentry = filp->f_path.dentry;
1527 	struct inode *inode = dentry->d_inode;
1528 
1529 	for (before = &inode->i_flock;
1530 			((fl = *before) != NULL) && IS_LEASE(fl);
1531 			before = &fl->fl_next) {
1532 		if (fl->fl_file != filp)
1533 			continue;
1534 		return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1535 	}
1536 	return -EAGAIN;
1537 }
1538 
1539 /**
1540  *	generic_setlease	-	sets a lease on an open file
1541  *	@filp: file pointer
1542  *	@arg: type of lease to obtain
1543  *	@flp: input - file_lock to use, output - file_lock inserted
1544  *
1545  *	The (input) flp->fl_lmops->lm_break function is required
1546  *	by break_lease().
1547  *
1548  *	Called with inode->i_lock held.
1549  */
1550 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1551 {
1552 	struct dentry *dentry = filp->f_path.dentry;
1553 	struct inode *inode = dentry->d_inode;
1554 	int error;
1555 
1556 	if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1557 		return -EACCES;
1558 	if (!S_ISREG(inode->i_mode))
1559 		return -EINVAL;
1560 	error = security_file_lock(filp, arg);
1561 	if (error)
1562 		return error;
1563 
1564 	time_out_leases(inode);
1565 
1566 	BUG_ON(!(*flp)->fl_lmops->lm_break);
1567 
1568 	switch (arg) {
1569 	case F_UNLCK:
1570 		return generic_delete_lease(filp, flp);
1571 	case F_RDLCK:
1572 	case F_WRLCK:
1573 		return generic_add_lease(filp, arg, flp);
1574 	default:
1575 		return -EINVAL;
1576 	}
1577 }
1578 EXPORT_SYMBOL(generic_setlease);
1579 
1580 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1581 {
1582 	if (filp->f_op && filp->f_op->setlease)
1583 		return filp->f_op->setlease(filp, arg, lease);
1584 	else
1585 		return generic_setlease(filp, arg, lease);
1586 }
1587 
1588 /**
1589  *	vfs_setlease        -       sets a lease on an open file
1590  *	@filp: file pointer
1591  *	@arg: type of lease to obtain
1592  *	@lease: file_lock to use
1593  *
1594  *	Call this to establish a lease on the file.
1595  *	The (*lease)->fl_lmops->lm_break operation must be set; if not,
1596  *	break_lease will oops!
1597  *
1598  *	This will call the filesystem's setlease file method, if
1599  *	defined.  Note that there is no getlease method; instead, the
1600  *	filesystem setlease method should call back to setlease() to
1601  *	add a lease to the inode's lease list, where fcntl_getlease() can
1602  *	find it.  Since fcntl_getlease() only reports whether the current
1603  *	task holds a lease, a cluster filesystem need only do this for
1604  *	leases held by processes on this node.
1605  *
1606  *	There is also no break_lease method; filesystems that
1607  *	handle their own leases should break leases themselves from the
1608  *	filesystem's open, create, and (on truncate) setattr methods.
1609  *
1610  *	Warning: the only current setlease methods exist only to disable
1611  *	leases in certain cases.  More vfs changes may be required to
1612  *	allow a full filesystem lease implementation.
1613  */
1614 
1615 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1616 {
1617 	struct inode *inode = file_inode(filp);
1618 	int error;
1619 
1620 	spin_lock(&inode->i_lock);
1621 	error = __vfs_setlease(filp, arg, lease);
1622 	spin_unlock(&inode->i_lock);
1623 
1624 	return error;
1625 }
1626 EXPORT_SYMBOL_GPL(vfs_setlease);
1627 
1628 static int do_fcntl_delete_lease(struct file *filp)
1629 {
1630 	struct file_lock fl, *flp = &fl;
1631 
1632 	lease_init(filp, F_UNLCK, flp);
1633 
1634 	return vfs_setlease(filp, F_UNLCK, &flp);
1635 }
1636 
1637 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1638 {
1639 	struct file_lock *fl, *ret;
1640 	struct inode *inode = file_inode(filp);
1641 	struct fasync_struct *new;
1642 	int error;
1643 
1644 	fl = lease_alloc(filp, arg);
1645 	if (IS_ERR(fl))
1646 		return PTR_ERR(fl);
1647 
1648 	new = fasync_alloc();
1649 	if (!new) {
1650 		locks_free_lock(fl);
1651 		return -ENOMEM;
1652 	}
1653 	ret = fl;
1654 	spin_lock(&inode->i_lock);
1655 	error = __vfs_setlease(filp, arg, &ret);
1656 	if (error) {
1657 		spin_unlock(&inode->i_lock);
1658 		locks_free_lock(fl);
1659 		goto out_free_fasync;
1660 	}
1661 	if (ret != fl)
1662 		locks_free_lock(fl);
1663 
1664 	/*
1665 	 * fasync_insert_entry() returns the old entry if any.
1666 	 * If there was no old entry, then it used 'new' and
1667 	 * inserted it into the fasync list. Clear new so that
1668 	 * we don't release it here.
1669 	 */
1670 	if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1671 		new = NULL;
1672 
1673 	error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1674 	spin_unlock(&inode->i_lock);
1675 
1676 out_free_fasync:
1677 	if (new)
1678 		fasync_free(new);
1679 	return error;
1680 }
1681 
1682 /**
1683  *	fcntl_setlease	-	sets a lease on an open file
1684  *	@fd: open file descriptor
1685  *	@filp: file pointer
1686  *	@arg: type of lease to obtain
1687  *
1688  *	Call this fcntl to establish a lease on the file.
1689  *	Note that you also need to call %F_SETSIG to
1690  *	receive a signal when the lease is broken.
1691  */
1692 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1693 {
1694 	if (arg == F_UNLCK)
1695 		return do_fcntl_delete_lease(filp);
1696 	return do_fcntl_add_lease(fd, filp, arg);
1697 }
1698 
1699 /**
1700  * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1701  * @filp: The file to apply the lock to
1702  * @fl: The lock to be applied
1703  *
1704  * Add a FLOCK style lock to a file.
1705  */
1706 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1707 {
1708 	int error;
1709 	might_sleep();
1710 	for (;;) {
1711 		error = flock_lock_file(filp, fl);
1712 		if (error != FILE_LOCK_DEFERRED)
1713 			break;
1714 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1715 		if (!error)
1716 			continue;
1717 
1718 		locks_delete_block(fl);
1719 		break;
1720 	}
1721 	return error;
1722 }
1723 
1724 EXPORT_SYMBOL(flock_lock_file_wait);
1725 
1726 /**
1727  *	sys_flock: - flock() system call.
1728  *	@fd: the file descriptor to lock.
1729  *	@cmd: the type of lock to apply.
1730  *
1731  *	Apply a %FL_FLOCK style lock to an open file descriptor.
1732  *	The @cmd can be one of
1733  *
1734  *	%LOCK_SH -- a shared lock.
1735  *
1736  *	%LOCK_EX -- an exclusive lock.
1737  *
1738  *	%LOCK_UN -- remove an existing lock.
1739  *
1740  *	%LOCK_MAND -- a `mandatory' flock.  This exists to emulate Windows Share Modes.
1741  *
1742  *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1743  *	processes read and write access respectively.
1744  */
1745 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1746 {
1747 	struct fd f = fdget(fd);
1748 	struct file_lock *lock;
1749 	int can_sleep, unlock;
1750 	int error;
1751 
1752 	error = -EBADF;
1753 	if (!f.file)
1754 		goto out;
1755 
1756 	can_sleep = !(cmd & LOCK_NB);
1757 	cmd &= ~LOCK_NB;
1758 	unlock = (cmd == LOCK_UN);
1759 
1760 	if (!unlock && !(cmd & LOCK_MAND) &&
1761 	    !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1762 		goto out_putf;
1763 
1764 	error = flock_make_lock(f.file, &lock, cmd);
1765 	if (error)
1766 		goto out_putf;
1767 	if (can_sleep)
1768 		lock->fl_flags |= FL_SLEEP;
1769 
1770 	error = security_file_lock(f.file, lock->fl_type);
1771 	if (error)
1772 		goto out_free;
1773 
1774 	if (f.file->f_op && f.file->f_op->flock)
1775 		error = f.file->f_op->flock(f.file,
1776 					  (can_sleep) ? F_SETLKW : F_SETLK,
1777 					  lock);
1778 	else
1779 		error = flock_lock_file_wait(f.file, lock);
1780 
1781  out_free:
1782 	locks_free_lock(lock);
1783 
1784  out_putf:
1785 	fdput(f);
1786  out:
1787 	return error;
1788 }
1789 
1790 /**
1791  * vfs_test_lock - test file byte range lock
1792  * @filp: The file to test lock for
1793  * @fl: The lock to test; also used to hold result
1794  *
1795  * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
1796  * setting conf->fl_type to something other than F_UNLCK.
1797  */
1798 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1799 {
1800 	if (filp->f_op && filp->f_op->lock)
1801 		return filp->f_op->lock(filp, F_GETLK, fl);
1802 	posix_test_lock(filp, fl);
1803 	return 0;
1804 }
1805 EXPORT_SYMBOL_GPL(vfs_test_lock);
1806 
1807 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1808 {
1809 	flock->l_pid = fl->fl_pid;
1810 #if BITS_PER_LONG == 32
1811 	/*
1812 	 * Make sure we can represent the posix lock via
1813 	 * legacy 32bit flock.
1814 	 */
1815 	if (fl->fl_start > OFFT_OFFSET_MAX)
1816 		return -EOVERFLOW;
1817 	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1818 		return -EOVERFLOW;
1819 #endif
1820 	flock->l_start = fl->fl_start;
1821 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1822 		fl->fl_end - fl->fl_start + 1;
1823 	flock->l_whence = 0;
1824 	flock->l_type = fl->fl_type;
1825 	return 0;
1826 }
1827 
1828 #if BITS_PER_LONG == 32
1829 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1830 {
1831 	flock->l_pid = fl->fl_pid;
1832 	flock->l_start = fl->fl_start;
1833 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1834 		fl->fl_end - fl->fl_start + 1;
1835 	flock->l_whence = 0;
1836 	flock->l_type = fl->fl_type;
1837 }
1838 #endif
1839 
1840 /* Report the first existing lock that would conflict with l.
1841  * This implements the F_GETLK command of fcntl().
1842  */
1843 int fcntl_getlk(struct file *filp, struct flock __user *l)
1844 {
1845 	struct file_lock file_lock;
1846 	struct flock flock;
1847 	int error;
1848 
1849 	error = -EFAULT;
1850 	if (copy_from_user(&flock, l, sizeof(flock)))
1851 		goto out;
1852 	error = -EINVAL;
1853 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1854 		goto out;
1855 
1856 	error = flock_to_posix_lock(filp, &file_lock, &flock);
1857 	if (error)
1858 		goto out;
1859 
1860 	error = vfs_test_lock(filp, &file_lock);
1861 	if (error)
1862 		goto out;
1863 
1864 	flock.l_type = file_lock.fl_type;
1865 	if (file_lock.fl_type != F_UNLCK) {
1866 		error = posix_lock_to_flock(&flock, &file_lock);
1867 		if (error)
1868 			goto out;
1869 	}
1870 	error = -EFAULT;
1871 	if (!copy_to_user(l, &flock, sizeof(flock)))
1872 		error = 0;
1873 out:
1874 	return error;
1875 }
1876 
1877 /**
1878  * vfs_lock_file - file byte range lock
1879  * @filp: The file to apply the lock to
1880  * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1881  * @fl: The lock to be applied
1882  * @conf: Place to return a copy of the conflicting lock, if found.
1883  *
1884  * A caller that doesn't care about the conflicting lock may pass NULL
1885  * as the final argument.
1886  *
1887  * If the filesystem defines a private ->lock() method, then @conf will
1888  * be left unchanged; so a caller that cares should initialize it to
1889  * some acceptable default.
1890  *
1891  * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1892  * locks, the ->lock() interface may return asynchronously, before the lock has
1893  * been granted or denied by the underlying filesystem, if (and only if)
1894  * lm_grant is set. Callers expecting ->lock() to return asynchronously
1895  * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1896  * the request is for a blocking lock. When ->lock() does return asynchronously,
1897  * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1898  * request completes.
1899  * If the request is for non-blocking lock the file system should return
1900  * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1901  * with the result. If the request timed out the callback routine will return a
1902  * nonzero return code and the file system should release the lock. The file
1903  * system is also responsible to keep a corresponding posix lock when it
1904  * grants a lock so the VFS can find out which locks are locally held and do
1905  * the correct lock cleanup when required.
1906  * The underlying filesystem must not drop the kernel lock or call
1907  * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1908  * return code.
1909  */
1910 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1911 {
1912 	if (filp->f_op && filp->f_op->lock)
1913 		return filp->f_op->lock(filp, cmd, fl);
1914 	else
1915 		return posix_lock_file(filp, fl, conf);
1916 }
1917 EXPORT_SYMBOL_GPL(vfs_lock_file);
1918 
1919 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1920 			     struct file_lock *fl)
1921 {
1922 	int error;
1923 
1924 	error = security_file_lock(filp, fl->fl_type);
1925 	if (error)
1926 		return error;
1927 
1928 	for (;;) {
1929 		error = vfs_lock_file(filp, cmd, fl, NULL);
1930 		if (error != FILE_LOCK_DEFERRED)
1931 			break;
1932 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1933 		if (!error)
1934 			continue;
1935 
1936 		locks_delete_block(fl);
1937 		break;
1938 	}
1939 
1940 	return error;
1941 }
1942 
1943 /* Apply the lock described by l to an open file descriptor.
1944  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1945  */
1946 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1947 		struct flock __user *l)
1948 {
1949 	struct file_lock *file_lock = locks_alloc_lock();
1950 	struct flock flock;
1951 	struct inode *inode;
1952 	struct file *f;
1953 	int error;
1954 
1955 	if (file_lock == NULL)
1956 		return -ENOLCK;
1957 
1958 	/*
1959 	 * This might block, so we do it before checking the inode.
1960 	 */
1961 	error = -EFAULT;
1962 	if (copy_from_user(&flock, l, sizeof(flock)))
1963 		goto out;
1964 
1965 	inode = file_inode(filp);
1966 
1967 	/* Don't allow mandatory locks on files that may be memory mapped
1968 	 * and shared.
1969 	 */
1970 	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1971 		error = -EAGAIN;
1972 		goto out;
1973 	}
1974 
1975 again:
1976 	error = flock_to_posix_lock(filp, file_lock, &flock);
1977 	if (error)
1978 		goto out;
1979 	if (cmd == F_SETLKW) {
1980 		file_lock->fl_flags |= FL_SLEEP;
1981 	}
1982 
1983 	error = -EBADF;
1984 	switch (flock.l_type) {
1985 	case F_RDLCK:
1986 		if (!(filp->f_mode & FMODE_READ))
1987 			goto out;
1988 		break;
1989 	case F_WRLCK:
1990 		if (!(filp->f_mode & FMODE_WRITE))
1991 			goto out;
1992 		break;
1993 	case F_UNLCK:
1994 		break;
1995 	default:
1996 		error = -EINVAL;
1997 		goto out;
1998 	}
1999 
2000 	error = do_lock_file_wait(filp, cmd, file_lock);
2001 
2002 	/*
2003 	 * Attempt to detect a close/fcntl race and recover by
2004 	 * releasing the lock that was just acquired.
2005 	 */
2006 	/*
2007 	 * we need that spin_lock here - it prevents reordering between
2008 	 * update of inode->i_flock and check for it done in close().
2009 	 * rcu_read_lock() wouldn't do.
2010 	 */
2011 	spin_lock(&current->files->file_lock);
2012 	f = fcheck(fd);
2013 	spin_unlock(&current->files->file_lock);
2014 	if (!error && f != filp && flock.l_type != F_UNLCK) {
2015 		flock.l_type = F_UNLCK;
2016 		goto again;
2017 	}
2018 
2019 out:
2020 	locks_free_lock(file_lock);
2021 	return error;
2022 }
2023 
2024 #if BITS_PER_LONG == 32
2025 /* Report the first existing lock that would conflict with l.
2026  * This implements the F_GETLK command of fcntl().
2027  */
2028 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
2029 {
2030 	struct file_lock file_lock;
2031 	struct flock64 flock;
2032 	int error;
2033 
2034 	error = -EFAULT;
2035 	if (copy_from_user(&flock, l, sizeof(flock)))
2036 		goto out;
2037 	error = -EINVAL;
2038 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2039 		goto out;
2040 
2041 	error = flock64_to_posix_lock(filp, &file_lock, &flock);
2042 	if (error)
2043 		goto out;
2044 
2045 	error = vfs_test_lock(filp, &file_lock);
2046 	if (error)
2047 		goto out;
2048 
2049 	flock.l_type = file_lock.fl_type;
2050 	if (file_lock.fl_type != F_UNLCK)
2051 		posix_lock_to_flock64(&flock, &file_lock);
2052 
2053 	error = -EFAULT;
2054 	if (!copy_to_user(l, &flock, sizeof(flock)))
2055 		error = 0;
2056 
2057 out:
2058 	return error;
2059 }
2060 
2061 /* Apply the lock described by l to an open file descriptor.
2062  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2063  */
2064 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2065 		struct flock64 __user *l)
2066 {
2067 	struct file_lock *file_lock = locks_alloc_lock();
2068 	struct flock64 flock;
2069 	struct inode *inode;
2070 	struct file *f;
2071 	int error;
2072 
2073 	if (file_lock == NULL)
2074 		return -ENOLCK;
2075 
2076 	/*
2077 	 * This might block, so we do it before checking the inode.
2078 	 */
2079 	error = -EFAULT;
2080 	if (copy_from_user(&flock, l, sizeof(flock)))
2081 		goto out;
2082 
2083 	inode = file_inode(filp);
2084 
2085 	/* Don't allow mandatory locks on files that may be memory mapped
2086 	 * and shared.
2087 	 */
2088 	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2089 		error = -EAGAIN;
2090 		goto out;
2091 	}
2092 
2093 again:
2094 	error = flock64_to_posix_lock(filp, file_lock, &flock);
2095 	if (error)
2096 		goto out;
2097 	if (cmd == F_SETLKW64) {
2098 		file_lock->fl_flags |= FL_SLEEP;
2099 	}
2100 
2101 	error = -EBADF;
2102 	switch (flock.l_type) {
2103 	case F_RDLCK:
2104 		if (!(filp->f_mode & FMODE_READ))
2105 			goto out;
2106 		break;
2107 	case F_WRLCK:
2108 		if (!(filp->f_mode & FMODE_WRITE))
2109 			goto out;
2110 		break;
2111 	case F_UNLCK:
2112 		break;
2113 	default:
2114 		error = -EINVAL;
2115 		goto out;
2116 	}
2117 
2118 	error = do_lock_file_wait(filp, cmd, file_lock);
2119 
2120 	/*
2121 	 * Attempt to detect a close/fcntl race and recover by
2122 	 * releasing the lock that was just acquired.
2123 	 */
2124 	spin_lock(&current->files->file_lock);
2125 	f = fcheck(fd);
2126 	spin_unlock(&current->files->file_lock);
2127 	if (!error && f != filp && flock.l_type != F_UNLCK) {
2128 		flock.l_type = F_UNLCK;
2129 		goto again;
2130 	}
2131 
2132 out:
2133 	locks_free_lock(file_lock);
2134 	return error;
2135 }
2136 #endif /* BITS_PER_LONG == 32 */
2137 
2138 /*
2139  * This function is called when the file is being removed
2140  * from the task's fd array.  POSIX locks belonging to this task
2141  * are deleted at this time.
2142  */
2143 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2144 {
2145 	struct file_lock lock;
2146 
2147 	/*
2148 	 * If there are no locks held on this file, we don't need to call
2149 	 * posix_lock_file().  Another process could be setting a lock on this
2150 	 * file at the same time, but we wouldn't remove that lock anyway.
2151 	 */
2152 	if (!file_inode(filp)->i_flock)
2153 		return;
2154 
2155 	lock.fl_type = F_UNLCK;
2156 	lock.fl_flags = FL_POSIX | FL_CLOSE;
2157 	lock.fl_start = 0;
2158 	lock.fl_end = OFFSET_MAX;
2159 	lock.fl_owner = owner;
2160 	lock.fl_pid = current->tgid;
2161 	lock.fl_file = filp;
2162 	lock.fl_ops = NULL;
2163 	lock.fl_lmops = NULL;
2164 
2165 	vfs_lock_file(filp, F_SETLK, &lock, NULL);
2166 
2167 	if (lock.fl_ops && lock.fl_ops->fl_release_private)
2168 		lock.fl_ops->fl_release_private(&lock);
2169 }
2170 
2171 EXPORT_SYMBOL(locks_remove_posix);
2172 
2173 /*
2174  * This function is called on the last close of an open file.
2175  */
2176 void locks_remove_flock(struct file *filp)
2177 {
2178 	struct inode * inode = file_inode(filp);
2179 	struct file_lock *fl;
2180 	struct file_lock **before;
2181 
2182 	if (!inode->i_flock)
2183 		return;
2184 
2185 	if (filp->f_op && filp->f_op->flock) {
2186 		struct file_lock fl = {
2187 			.fl_pid = current->tgid,
2188 			.fl_file = filp,
2189 			.fl_flags = FL_FLOCK,
2190 			.fl_type = F_UNLCK,
2191 			.fl_end = OFFSET_MAX,
2192 		};
2193 		filp->f_op->flock(filp, F_SETLKW, &fl);
2194 		if (fl.fl_ops && fl.fl_ops->fl_release_private)
2195 			fl.fl_ops->fl_release_private(&fl);
2196 	}
2197 
2198 	spin_lock(&inode->i_lock);
2199 	before = &inode->i_flock;
2200 
2201 	while ((fl = *before) != NULL) {
2202 		if (fl->fl_file == filp) {
2203 			if (IS_FLOCK(fl)) {
2204 				locks_delete_lock(before);
2205 				continue;
2206 			}
2207 			if (IS_LEASE(fl)) {
2208 				lease_modify(before, F_UNLCK);
2209 				continue;
2210 			}
2211 			/* What? */
2212 			BUG();
2213  		}
2214 		before = &fl->fl_next;
2215 	}
2216 	spin_unlock(&inode->i_lock);
2217 }
2218 
2219 /**
2220  *	posix_unblock_lock - stop waiting for a file lock
2221  *	@waiter: the lock which was waiting
2222  *
2223  *	lockd needs to block waiting for locks.
2224  */
2225 int
2226 posix_unblock_lock(struct file_lock *waiter)
2227 {
2228 	int status = 0;
2229 
2230 	spin_lock(&blocked_lock_lock);
2231 	if (waiter->fl_next)
2232 		__locks_delete_block(waiter);
2233 	else
2234 		status = -ENOENT;
2235 	spin_unlock(&blocked_lock_lock);
2236 	return status;
2237 }
2238 EXPORT_SYMBOL(posix_unblock_lock);
2239 
2240 /**
2241  * vfs_cancel_lock - file byte range unblock lock
2242  * @filp: The file to apply the unblock to
2243  * @fl: The lock to be unblocked
2244  *
2245  * Used by lock managers to cancel blocked requests
2246  */
2247 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2248 {
2249 	if (filp->f_op && filp->f_op->lock)
2250 		return filp->f_op->lock(filp, F_CANCELLK, fl);
2251 	return 0;
2252 }
2253 
2254 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2255 
2256 #ifdef CONFIG_PROC_FS
2257 #include <linux/proc_fs.h>
2258 #include <linux/seq_file.h>
2259 
2260 struct locks_iterator {
2261 	int	li_cpu;
2262 	loff_t	li_pos;
2263 };
2264 
2265 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2266 			    loff_t id, char *pfx)
2267 {
2268 	struct inode *inode = NULL;
2269 	unsigned int fl_pid;
2270 
2271 	if (fl->fl_nspid)
2272 		fl_pid = pid_vnr(fl->fl_nspid);
2273 	else
2274 		fl_pid = fl->fl_pid;
2275 
2276 	if (fl->fl_file != NULL)
2277 		inode = file_inode(fl->fl_file);
2278 
2279 	seq_printf(f, "%lld:%s ", id, pfx);
2280 	if (IS_POSIX(fl)) {
2281 		seq_printf(f, "%6s %s ",
2282 			     (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2283 			     (inode == NULL) ? "*NOINODE*" :
2284 			     mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2285 	} else if (IS_FLOCK(fl)) {
2286 		if (fl->fl_type & LOCK_MAND) {
2287 			seq_printf(f, "FLOCK  MSNFS     ");
2288 		} else {
2289 			seq_printf(f, "FLOCK  ADVISORY  ");
2290 		}
2291 	} else if (IS_LEASE(fl)) {
2292 		seq_printf(f, "LEASE  ");
2293 		if (lease_breaking(fl))
2294 			seq_printf(f, "BREAKING  ");
2295 		else if (fl->fl_file)
2296 			seq_printf(f, "ACTIVE    ");
2297 		else
2298 			seq_printf(f, "BREAKER   ");
2299 	} else {
2300 		seq_printf(f, "UNKNOWN UNKNOWN  ");
2301 	}
2302 	if (fl->fl_type & LOCK_MAND) {
2303 		seq_printf(f, "%s ",
2304 			       (fl->fl_type & LOCK_READ)
2305 			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2306 			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2307 	} else {
2308 		seq_printf(f, "%s ",
2309 			       (lease_breaking(fl))
2310 			       ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2311 			       : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2312 	}
2313 	if (inode) {
2314 #ifdef WE_CAN_BREAK_LSLK_NOW
2315 		seq_printf(f, "%d %s:%ld ", fl_pid,
2316 				inode->i_sb->s_id, inode->i_ino);
2317 #else
2318 		/* userspace relies on this representation of dev_t ;-( */
2319 		seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2320 				MAJOR(inode->i_sb->s_dev),
2321 				MINOR(inode->i_sb->s_dev), inode->i_ino);
2322 #endif
2323 	} else {
2324 		seq_printf(f, "%d <none>:0 ", fl_pid);
2325 	}
2326 	if (IS_POSIX(fl)) {
2327 		if (fl->fl_end == OFFSET_MAX)
2328 			seq_printf(f, "%Ld EOF\n", fl->fl_start);
2329 		else
2330 			seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2331 	} else {
2332 		seq_printf(f, "0 EOF\n");
2333 	}
2334 }
2335 
2336 static int locks_show(struct seq_file *f, void *v)
2337 {
2338 	struct locks_iterator *iter = f->private;
2339 	struct file_lock *fl, *bfl;
2340 
2341 	fl = hlist_entry(v, struct file_lock, fl_link);
2342 
2343 	lock_get_status(f, fl, iter->li_pos, "");
2344 
2345 	list_for_each_entry(bfl, &fl->fl_block, fl_block)
2346 		lock_get_status(f, bfl, iter->li_pos, " ->");
2347 
2348 	return 0;
2349 }
2350 
2351 static void *locks_start(struct seq_file *f, loff_t *pos)
2352 {
2353 	struct locks_iterator *iter = f->private;
2354 
2355 	iter->li_pos = *pos + 1;
2356 	lg_global_lock(&file_lock_lglock);
2357 	spin_lock(&blocked_lock_lock);
2358 	return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2359 }
2360 
2361 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2362 {
2363 	struct locks_iterator *iter = f->private;
2364 
2365 	++iter->li_pos;
2366 	return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2367 }
2368 
2369 static void locks_stop(struct seq_file *f, void *v)
2370 {
2371 	spin_unlock(&blocked_lock_lock);
2372 	lg_global_unlock(&file_lock_lglock);
2373 }
2374 
2375 static const struct seq_operations locks_seq_operations = {
2376 	.start	= locks_start,
2377 	.next	= locks_next,
2378 	.stop	= locks_stop,
2379 	.show	= locks_show,
2380 };
2381 
2382 static int locks_open(struct inode *inode, struct file *filp)
2383 {
2384 	return seq_open_private(filp, &locks_seq_operations,
2385 					sizeof(struct locks_iterator));
2386 }
2387 
2388 static const struct file_operations proc_locks_operations = {
2389 	.open		= locks_open,
2390 	.read		= seq_read,
2391 	.llseek		= seq_lseek,
2392 	.release	= seq_release_private,
2393 };
2394 
2395 static int __init proc_locks_init(void)
2396 {
2397 	proc_create("locks", 0, NULL, &proc_locks_operations);
2398 	return 0;
2399 }
2400 module_init(proc_locks_init);
2401 #endif
2402 
2403 /**
2404  *	lock_may_read - checks that the region is free of locks
2405  *	@inode: the inode that is being read
2406  *	@start: the first byte to read
2407  *	@len: the number of bytes to read
2408  *
2409  *	Emulates Windows locking requirements.  Whole-file
2410  *	mandatory locks (share modes) can prohibit a read and
2411  *	byte-range POSIX locks can prohibit a read if they overlap.
2412  *
2413  *	N.B. this function is only ever called
2414  *	from knfsd and ownership of locks is never checked.
2415  */
2416 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2417 {
2418 	struct file_lock *fl;
2419 	int result = 1;
2420 
2421 	spin_lock(&inode->i_lock);
2422 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2423 		if (IS_POSIX(fl)) {
2424 			if (fl->fl_type == F_RDLCK)
2425 				continue;
2426 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2427 				continue;
2428 		} else if (IS_FLOCK(fl)) {
2429 			if (!(fl->fl_type & LOCK_MAND))
2430 				continue;
2431 			if (fl->fl_type & LOCK_READ)
2432 				continue;
2433 		} else
2434 			continue;
2435 		result = 0;
2436 		break;
2437 	}
2438 	spin_unlock(&inode->i_lock);
2439 	return result;
2440 }
2441 
2442 EXPORT_SYMBOL(lock_may_read);
2443 
2444 /**
2445  *	lock_may_write - checks that the region is free of locks
2446  *	@inode: the inode that is being written
2447  *	@start: the first byte to write
2448  *	@len: the number of bytes to write
2449  *
2450  *	Emulates Windows locking requirements.  Whole-file
2451  *	mandatory locks (share modes) can prohibit a write and
2452  *	byte-range POSIX locks can prohibit a write if they overlap.
2453  *
2454  *	N.B. this function is only ever called
2455  *	from knfsd and ownership of locks is never checked.
2456  */
2457 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2458 {
2459 	struct file_lock *fl;
2460 	int result = 1;
2461 
2462 	spin_lock(&inode->i_lock);
2463 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2464 		if (IS_POSIX(fl)) {
2465 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2466 				continue;
2467 		} else if (IS_FLOCK(fl)) {
2468 			if (!(fl->fl_type & LOCK_MAND))
2469 				continue;
2470 			if (fl->fl_type & LOCK_WRITE)
2471 				continue;
2472 		} else
2473 			continue;
2474 		result = 0;
2475 		break;
2476 	}
2477 	spin_unlock(&inode->i_lock);
2478 	return result;
2479 }
2480 
2481 EXPORT_SYMBOL(lock_may_write);
2482 
2483 static int __init filelock_init(void)
2484 {
2485 	int i;
2486 
2487 	filelock_cache = kmem_cache_create("file_lock_cache",
2488 			sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2489 
2490 	lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2491 
2492 	for_each_possible_cpu(i)
2493 		INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2494 
2495 	return 0;
2496 }
2497 
2498 core_initcall(filelock_init);
2499