xref: /linux/fs/locks.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 
130 #include <asm/uaccess.h>
131 
132 #define IS_POSIX(fl)	(fl->fl_flags & FL_POSIX)
133 #define IS_FLOCK(fl)	(fl->fl_flags & FL_FLOCK)
134 #define IS_LEASE(fl)	(fl->fl_flags & FL_LEASE)
135 
136 static bool lease_breaking(struct file_lock *fl)
137 {
138 	return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
139 }
140 
141 static int target_leasetype(struct file_lock *fl)
142 {
143 	if (fl->fl_flags & FL_UNLOCK_PENDING)
144 		return F_UNLCK;
145 	if (fl->fl_flags & FL_DOWNGRADE_PENDING)
146 		return F_RDLCK;
147 	return fl->fl_type;
148 }
149 
150 int leases_enable = 1;
151 int lease_break_time = 45;
152 
153 #define for_each_lock(inode, lockp) \
154 	for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
155 
156 static LIST_HEAD(file_lock_list);
157 static LIST_HEAD(blocked_list);
158 static DEFINE_SPINLOCK(file_lock_lock);
159 
160 /*
161  * Protects the two list heads above, plus the inode->i_flock list
162  */
163 void lock_flocks(void)
164 {
165 	spin_lock(&file_lock_lock);
166 }
167 EXPORT_SYMBOL_GPL(lock_flocks);
168 
169 void unlock_flocks(void)
170 {
171 	spin_unlock(&file_lock_lock);
172 }
173 EXPORT_SYMBOL_GPL(unlock_flocks);
174 
175 static struct kmem_cache *filelock_cache __read_mostly;
176 
177 static void locks_init_lock_heads(struct file_lock *fl)
178 {
179 	INIT_LIST_HEAD(&fl->fl_link);
180 	INIT_LIST_HEAD(&fl->fl_block);
181 	init_waitqueue_head(&fl->fl_wait);
182 }
183 
184 /* Allocate an empty lock structure. */
185 struct file_lock *locks_alloc_lock(void)
186 {
187 	struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
188 
189 	if (fl)
190 		locks_init_lock_heads(fl);
191 
192 	return fl;
193 }
194 EXPORT_SYMBOL_GPL(locks_alloc_lock);
195 
196 void locks_release_private(struct file_lock *fl)
197 {
198 	if (fl->fl_ops) {
199 		if (fl->fl_ops->fl_release_private)
200 			fl->fl_ops->fl_release_private(fl);
201 		fl->fl_ops = NULL;
202 	}
203 	fl->fl_lmops = NULL;
204 
205 }
206 EXPORT_SYMBOL_GPL(locks_release_private);
207 
208 /* Free a lock which is not in use. */
209 void locks_free_lock(struct file_lock *fl)
210 {
211 	BUG_ON(waitqueue_active(&fl->fl_wait));
212 	BUG_ON(!list_empty(&fl->fl_block));
213 	BUG_ON(!list_empty(&fl->fl_link));
214 
215 	locks_release_private(fl);
216 	kmem_cache_free(filelock_cache, fl);
217 }
218 EXPORT_SYMBOL(locks_free_lock);
219 
220 void locks_init_lock(struct file_lock *fl)
221 {
222 	memset(fl, 0, sizeof(struct file_lock));
223 	locks_init_lock_heads(fl);
224 }
225 
226 EXPORT_SYMBOL(locks_init_lock);
227 
228 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
229 {
230 	if (fl->fl_ops) {
231 		if (fl->fl_ops->fl_copy_lock)
232 			fl->fl_ops->fl_copy_lock(new, fl);
233 		new->fl_ops = fl->fl_ops;
234 	}
235 	if (fl->fl_lmops)
236 		new->fl_lmops = fl->fl_lmops;
237 }
238 
239 /*
240  * Initialize a new lock from an existing file_lock structure.
241  */
242 void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
243 {
244 	new->fl_owner = fl->fl_owner;
245 	new->fl_pid = fl->fl_pid;
246 	new->fl_file = NULL;
247 	new->fl_flags = fl->fl_flags;
248 	new->fl_type = fl->fl_type;
249 	new->fl_start = fl->fl_start;
250 	new->fl_end = fl->fl_end;
251 	new->fl_ops = NULL;
252 	new->fl_lmops = NULL;
253 }
254 EXPORT_SYMBOL(__locks_copy_lock);
255 
256 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
257 {
258 	locks_release_private(new);
259 
260 	__locks_copy_lock(new, fl);
261 	new->fl_file = fl->fl_file;
262 	new->fl_ops = fl->fl_ops;
263 	new->fl_lmops = fl->fl_lmops;
264 
265 	locks_copy_private(new, fl);
266 }
267 
268 EXPORT_SYMBOL(locks_copy_lock);
269 
270 static inline int flock_translate_cmd(int cmd) {
271 	if (cmd & LOCK_MAND)
272 		return cmd & (LOCK_MAND | LOCK_RW);
273 	switch (cmd) {
274 	case LOCK_SH:
275 		return F_RDLCK;
276 	case LOCK_EX:
277 		return F_WRLCK;
278 	case LOCK_UN:
279 		return F_UNLCK;
280 	}
281 	return -EINVAL;
282 }
283 
284 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
285 static int flock_make_lock(struct file *filp, struct file_lock **lock,
286 		unsigned int cmd)
287 {
288 	struct file_lock *fl;
289 	int type = flock_translate_cmd(cmd);
290 	if (type < 0)
291 		return type;
292 
293 	fl = locks_alloc_lock();
294 	if (fl == NULL)
295 		return -ENOMEM;
296 
297 	fl->fl_file = filp;
298 	fl->fl_pid = current->tgid;
299 	fl->fl_flags = FL_FLOCK;
300 	fl->fl_type = type;
301 	fl->fl_end = OFFSET_MAX;
302 
303 	*lock = fl;
304 	return 0;
305 }
306 
307 static int assign_type(struct file_lock *fl, long type)
308 {
309 	switch (type) {
310 	case F_RDLCK:
311 	case F_WRLCK:
312 	case F_UNLCK:
313 		fl->fl_type = type;
314 		break;
315 	default:
316 		return -EINVAL;
317 	}
318 	return 0;
319 }
320 
321 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
322  * style lock.
323  */
324 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
325 			       struct flock *l)
326 {
327 	off_t start, end;
328 
329 	switch (l->l_whence) {
330 	case SEEK_SET:
331 		start = 0;
332 		break;
333 	case SEEK_CUR:
334 		start = filp->f_pos;
335 		break;
336 	case SEEK_END:
337 		start = i_size_read(filp->f_path.dentry->d_inode);
338 		break;
339 	default:
340 		return -EINVAL;
341 	}
342 
343 	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
344 	   POSIX-2001 defines it. */
345 	start += l->l_start;
346 	if (start < 0)
347 		return -EINVAL;
348 	fl->fl_end = OFFSET_MAX;
349 	if (l->l_len > 0) {
350 		end = start + l->l_len - 1;
351 		fl->fl_end = end;
352 	} else if (l->l_len < 0) {
353 		end = start - 1;
354 		fl->fl_end = end;
355 		start += l->l_len;
356 		if (start < 0)
357 			return -EINVAL;
358 	}
359 	fl->fl_start = start;	/* we record the absolute position */
360 	if (fl->fl_end < fl->fl_start)
361 		return -EOVERFLOW;
362 
363 	fl->fl_owner = current->files;
364 	fl->fl_pid = current->tgid;
365 	fl->fl_file = filp;
366 	fl->fl_flags = FL_POSIX;
367 	fl->fl_ops = NULL;
368 	fl->fl_lmops = NULL;
369 
370 	return assign_type(fl, l->l_type);
371 }
372 
373 #if BITS_PER_LONG == 32
374 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
375 				 struct flock64 *l)
376 {
377 	loff_t start;
378 
379 	switch (l->l_whence) {
380 	case SEEK_SET:
381 		start = 0;
382 		break;
383 	case SEEK_CUR:
384 		start = filp->f_pos;
385 		break;
386 	case SEEK_END:
387 		start = i_size_read(filp->f_path.dentry->d_inode);
388 		break;
389 	default:
390 		return -EINVAL;
391 	}
392 
393 	start += l->l_start;
394 	if (start < 0)
395 		return -EINVAL;
396 	fl->fl_end = OFFSET_MAX;
397 	if (l->l_len > 0) {
398 		fl->fl_end = start + l->l_len - 1;
399 	} else if (l->l_len < 0) {
400 		fl->fl_end = start - 1;
401 		start += l->l_len;
402 		if (start < 0)
403 			return -EINVAL;
404 	}
405 	fl->fl_start = start;	/* we record the absolute position */
406 	if (fl->fl_end < fl->fl_start)
407 		return -EOVERFLOW;
408 
409 	fl->fl_owner = current->files;
410 	fl->fl_pid = current->tgid;
411 	fl->fl_file = filp;
412 	fl->fl_flags = FL_POSIX;
413 	fl->fl_ops = NULL;
414 	fl->fl_lmops = NULL;
415 
416 	return assign_type(fl, l->l_type);
417 }
418 #endif
419 
420 /* default lease lock manager operations */
421 static void lease_break_callback(struct file_lock *fl)
422 {
423 	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
424 }
425 
426 static const struct lock_manager_operations lease_manager_ops = {
427 	.lm_break = lease_break_callback,
428 	.lm_change = lease_modify,
429 };
430 
431 /*
432  * Initialize a lease, use the default lock manager operations
433  */
434 static int lease_init(struct file *filp, long type, struct file_lock *fl)
435  {
436 	if (assign_type(fl, type) != 0)
437 		return -EINVAL;
438 
439 	fl->fl_owner = current->files;
440 	fl->fl_pid = current->tgid;
441 
442 	fl->fl_file = filp;
443 	fl->fl_flags = FL_LEASE;
444 	fl->fl_start = 0;
445 	fl->fl_end = OFFSET_MAX;
446 	fl->fl_ops = NULL;
447 	fl->fl_lmops = &lease_manager_ops;
448 	return 0;
449 }
450 
451 /* Allocate a file_lock initialised to this type of lease */
452 static struct file_lock *lease_alloc(struct file *filp, long type)
453 {
454 	struct file_lock *fl = locks_alloc_lock();
455 	int error = -ENOMEM;
456 
457 	if (fl == NULL)
458 		return ERR_PTR(error);
459 
460 	error = lease_init(filp, type, fl);
461 	if (error) {
462 		locks_free_lock(fl);
463 		return ERR_PTR(error);
464 	}
465 	return fl;
466 }
467 
468 /* Check if two locks overlap each other.
469  */
470 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
471 {
472 	return ((fl1->fl_end >= fl2->fl_start) &&
473 		(fl2->fl_end >= fl1->fl_start));
474 }
475 
476 /*
477  * Check whether two locks have the same owner.
478  */
479 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
480 {
481 	if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
482 		return fl2->fl_lmops == fl1->fl_lmops &&
483 			fl1->fl_lmops->lm_compare_owner(fl1, fl2);
484 	return fl1->fl_owner == fl2->fl_owner;
485 }
486 
487 /* Remove waiter from blocker's block list.
488  * When blocker ends up pointing to itself then the list is empty.
489  */
490 static void __locks_delete_block(struct file_lock *waiter)
491 {
492 	list_del_init(&waiter->fl_block);
493 	list_del_init(&waiter->fl_link);
494 	waiter->fl_next = NULL;
495 }
496 
497 /*
498  */
499 void locks_delete_block(struct file_lock *waiter)
500 {
501 	lock_flocks();
502 	__locks_delete_block(waiter);
503 	unlock_flocks();
504 }
505 EXPORT_SYMBOL(locks_delete_block);
506 
507 /* Insert waiter into blocker's block list.
508  * We use a circular list so that processes can be easily woken up in
509  * the order they blocked. The documentation doesn't require this but
510  * it seems like the reasonable thing to do.
511  */
512 static void locks_insert_block(struct file_lock *blocker,
513 			       struct file_lock *waiter)
514 {
515 	BUG_ON(!list_empty(&waiter->fl_block));
516 	list_add_tail(&waiter->fl_block, &blocker->fl_block);
517 	waiter->fl_next = blocker;
518 	if (IS_POSIX(blocker))
519 		list_add(&waiter->fl_link, &blocked_list);
520 }
521 
522 /* Wake up processes blocked waiting for blocker.
523  * If told to wait then schedule the processes until the block list
524  * is empty, otherwise empty the block list ourselves.
525  */
526 static void locks_wake_up_blocks(struct file_lock *blocker)
527 {
528 	while (!list_empty(&blocker->fl_block)) {
529 		struct file_lock *waiter;
530 
531 		waiter = list_first_entry(&blocker->fl_block,
532 				struct file_lock, fl_block);
533 		__locks_delete_block(waiter);
534 		if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
535 			waiter->fl_lmops->lm_notify(waiter);
536 		else
537 			wake_up(&waiter->fl_wait);
538 	}
539 }
540 
541 /* Insert file lock fl into an inode's lock list at the position indicated
542  * by pos. At the same time add the lock to the global file lock list.
543  */
544 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
545 {
546 	list_add(&fl->fl_link, &file_lock_list);
547 
548 	fl->fl_nspid = get_pid(task_tgid(current));
549 
550 	/* insert into file's list */
551 	fl->fl_next = *pos;
552 	*pos = fl;
553 }
554 
555 /*
556  * Delete a lock and then free it.
557  * Wake up processes that are blocked waiting for this lock,
558  * notify the FS that the lock has been cleared and
559  * finally free the lock.
560  */
561 static void locks_delete_lock(struct file_lock **thisfl_p)
562 {
563 	struct file_lock *fl = *thisfl_p;
564 
565 	*thisfl_p = fl->fl_next;
566 	fl->fl_next = NULL;
567 	list_del_init(&fl->fl_link);
568 
569 	if (fl->fl_nspid) {
570 		put_pid(fl->fl_nspid);
571 		fl->fl_nspid = NULL;
572 	}
573 
574 	locks_wake_up_blocks(fl);
575 	locks_free_lock(fl);
576 }
577 
578 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
579  * checks for shared/exclusive status of overlapping locks.
580  */
581 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
582 {
583 	if (sys_fl->fl_type == F_WRLCK)
584 		return 1;
585 	if (caller_fl->fl_type == F_WRLCK)
586 		return 1;
587 	return 0;
588 }
589 
590 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
591  * checking before calling the locks_conflict().
592  */
593 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
594 {
595 	/* POSIX locks owned by the same process do not conflict with
596 	 * each other.
597 	 */
598 	if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
599 		return (0);
600 
601 	/* Check whether they overlap */
602 	if (!locks_overlap(caller_fl, sys_fl))
603 		return 0;
604 
605 	return (locks_conflict(caller_fl, sys_fl));
606 }
607 
608 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
609  * checking before calling the locks_conflict().
610  */
611 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
612 {
613 	/* FLOCK locks referring to the same filp do not conflict with
614 	 * each other.
615 	 */
616 	if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
617 		return (0);
618 	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
619 		return 0;
620 
621 	return (locks_conflict(caller_fl, sys_fl));
622 }
623 
624 void
625 posix_test_lock(struct file *filp, struct file_lock *fl)
626 {
627 	struct file_lock *cfl;
628 
629 	lock_flocks();
630 	for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
631 		if (!IS_POSIX(cfl))
632 			continue;
633 		if (posix_locks_conflict(fl, cfl))
634 			break;
635 	}
636 	if (cfl) {
637 		__locks_copy_lock(fl, cfl);
638 		if (cfl->fl_nspid)
639 			fl->fl_pid = pid_vnr(cfl->fl_nspid);
640 	} else
641 		fl->fl_type = F_UNLCK;
642 	unlock_flocks();
643 	return;
644 }
645 EXPORT_SYMBOL(posix_test_lock);
646 
647 /*
648  * Deadlock detection:
649  *
650  * We attempt to detect deadlocks that are due purely to posix file
651  * locks.
652  *
653  * We assume that a task can be waiting for at most one lock at a time.
654  * So for any acquired lock, the process holding that lock may be
655  * waiting on at most one other lock.  That lock in turns may be held by
656  * someone waiting for at most one other lock.  Given a requested lock
657  * caller_fl which is about to wait for a conflicting lock block_fl, we
658  * follow this chain of waiters to ensure we are not about to create a
659  * cycle.
660  *
661  * Since we do this before we ever put a process to sleep on a lock, we
662  * are ensured that there is never a cycle; that is what guarantees that
663  * the while() loop in posix_locks_deadlock() eventually completes.
664  *
665  * Note: the above assumption may not be true when handling lock
666  * requests from a broken NFS client. It may also fail in the presence
667  * of tasks (such as posix threads) sharing the same open file table.
668  *
669  * To handle those cases, we just bail out after a few iterations.
670  */
671 
672 #define MAX_DEADLK_ITERATIONS 10
673 
674 /* Find a lock that the owner of the given block_fl is blocking on. */
675 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
676 {
677 	struct file_lock *fl;
678 
679 	list_for_each_entry(fl, &blocked_list, fl_link) {
680 		if (posix_same_owner(fl, block_fl))
681 			return fl->fl_next;
682 	}
683 	return NULL;
684 }
685 
686 static int posix_locks_deadlock(struct file_lock *caller_fl,
687 				struct file_lock *block_fl)
688 {
689 	int i = 0;
690 
691 	while ((block_fl = what_owner_is_waiting_for(block_fl))) {
692 		if (i++ > MAX_DEADLK_ITERATIONS)
693 			return 0;
694 		if (posix_same_owner(caller_fl, block_fl))
695 			return 1;
696 	}
697 	return 0;
698 }
699 
700 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
701  * after any leases, but before any posix locks.
702  *
703  * Note that if called with an FL_EXISTS argument, the caller may determine
704  * whether or not a lock was successfully freed by testing the return
705  * value for -ENOENT.
706  */
707 static int flock_lock_file(struct file *filp, struct file_lock *request)
708 {
709 	struct file_lock *new_fl = NULL;
710 	struct file_lock **before;
711 	struct inode * inode = filp->f_path.dentry->d_inode;
712 	int error = 0;
713 	int found = 0;
714 
715 	if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
716 		new_fl = locks_alloc_lock();
717 		if (!new_fl)
718 			return -ENOMEM;
719 	}
720 
721 	lock_flocks();
722 	if (request->fl_flags & FL_ACCESS)
723 		goto find_conflict;
724 
725 	for_each_lock(inode, before) {
726 		struct file_lock *fl = *before;
727 		if (IS_POSIX(fl))
728 			break;
729 		if (IS_LEASE(fl))
730 			continue;
731 		if (filp != fl->fl_file)
732 			continue;
733 		if (request->fl_type == fl->fl_type)
734 			goto out;
735 		found = 1;
736 		locks_delete_lock(before);
737 		break;
738 	}
739 
740 	if (request->fl_type == F_UNLCK) {
741 		if ((request->fl_flags & FL_EXISTS) && !found)
742 			error = -ENOENT;
743 		goto out;
744 	}
745 
746 	/*
747 	 * If a higher-priority process was blocked on the old file lock,
748 	 * give it the opportunity to lock the file.
749 	 */
750 	if (found) {
751 		unlock_flocks();
752 		cond_resched();
753 		lock_flocks();
754 	}
755 
756 find_conflict:
757 	for_each_lock(inode, before) {
758 		struct file_lock *fl = *before;
759 		if (IS_POSIX(fl))
760 			break;
761 		if (IS_LEASE(fl))
762 			continue;
763 		if (!flock_locks_conflict(request, fl))
764 			continue;
765 		error = -EAGAIN;
766 		if (!(request->fl_flags & FL_SLEEP))
767 			goto out;
768 		error = FILE_LOCK_DEFERRED;
769 		locks_insert_block(fl, request);
770 		goto out;
771 	}
772 	if (request->fl_flags & FL_ACCESS)
773 		goto out;
774 	locks_copy_lock(new_fl, request);
775 	locks_insert_lock(before, new_fl);
776 	new_fl = NULL;
777 	error = 0;
778 
779 out:
780 	unlock_flocks();
781 	if (new_fl)
782 		locks_free_lock(new_fl);
783 	return error;
784 }
785 
786 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
787 {
788 	struct file_lock *fl;
789 	struct file_lock *new_fl = NULL;
790 	struct file_lock *new_fl2 = NULL;
791 	struct file_lock *left = NULL;
792 	struct file_lock *right = NULL;
793 	struct file_lock **before;
794 	int error, added = 0;
795 
796 	/*
797 	 * We may need two file_lock structures for this operation,
798 	 * so we get them in advance to avoid races.
799 	 *
800 	 * In some cases we can be sure, that no new locks will be needed
801 	 */
802 	if (!(request->fl_flags & FL_ACCESS) &&
803 	    (request->fl_type != F_UNLCK ||
804 	     request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
805 		new_fl = locks_alloc_lock();
806 		new_fl2 = locks_alloc_lock();
807 	}
808 
809 	lock_flocks();
810 	if (request->fl_type != F_UNLCK) {
811 		for_each_lock(inode, before) {
812 			fl = *before;
813 			if (!IS_POSIX(fl))
814 				continue;
815 			if (!posix_locks_conflict(request, fl))
816 				continue;
817 			if (conflock)
818 				__locks_copy_lock(conflock, fl);
819 			error = -EAGAIN;
820 			if (!(request->fl_flags & FL_SLEEP))
821 				goto out;
822 			error = -EDEADLK;
823 			if (posix_locks_deadlock(request, fl))
824 				goto out;
825 			error = FILE_LOCK_DEFERRED;
826 			locks_insert_block(fl, request);
827 			goto out;
828   		}
829   	}
830 
831 	/* If we're just looking for a conflict, we're done. */
832 	error = 0;
833 	if (request->fl_flags & FL_ACCESS)
834 		goto out;
835 
836 	/*
837 	 * Find the first old lock with the same owner as the new lock.
838 	 */
839 
840 	before = &inode->i_flock;
841 
842 	/* First skip locks owned by other processes.  */
843 	while ((fl = *before) && (!IS_POSIX(fl) ||
844 				  !posix_same_owner(request, fl))) {
845 		before = &fl->fl_next;
846 	}
847 
848 	/* Process locks with this owner.  */
849 	while ((fl = *before) && posix_same_owner(request, fl)) {
850 		/* Detect adjacent or overlapping regions (if same lock type)
851 		 */
852 		if (request->fl_type == fl->fl_type) {
853 			/* In all comparisons of start vs end, use
854 			 * "start - 1" rather than "end + 1". If end
855 			 * is OFFSET_MAX, end + 1 will become negative.
856 			 */
857 			if (fl->fl_end < request->fl_start - 1)
858 				goto next_lock;
859 			/* If the next lock in the list has entirely bigger
860 			 * addresses than the new one, insert the lock here.
861 			 */
862 			if (fl->fl_start - 1 > request->fl_end)
863 				break;
864 
865 			/* If we come here, the new and old lock are of the
866 			 * same type and adjacent or overlapping. Make one
867 			 * lock yielding from the lower start address of both
868 			 * locks to the higher end address.
869 			 */
870 			if (fl->fl_start > request->fl_start)
871 				fl->fl_start = request->fl_start;
872 			else
873 				request->fl_start = fl->fl_start;
874 			if (fl->fl_end < request->fl_end)
875 				fl->fl_end = request->fl_end;
876 			else
877 				request->fl_end = fl->fl_end;
878 			if (added) {
879 				locks_delete_lock(before);
880 				continue;
881 			}
882 			request = fl;
883 			added = 1;
884 		}
885 		else {
886 			/* Processing for different lock types is a bit
887 			 * more complex.
888 			 */
889 			if (fl->fl_end < request->fl_start)
890 				goto next_lock;
891 			if (fl->fl_start > request->fl_end)
892 				break;
893 			if (request->fl_type == F_UNLCK)
894 				added = 1;
895 			if (fl->fl_start < request->fl_start)
896 				left = fl;
897 			/* If the next lock in the list has a higher end
898 			 * address than the new one, insert the new one here.
899 			 */
900 			if (fl->fl_end > request->fl_end) {
901 				right = fl;
902 				break;
903 			}
904 			if (fl->fl_start >= request->fl_start) {
905 				/* The new lock completely replaces an old
906 				 * one (This may happen several times).
907 				 */
908 				if (added) {
909 					locks_delete_lock(before);
910 					continue;
911 				}
912 				/* Replace the old lock with the new one.
913 				 * Wake up anybody waiting for the old one,
914 				 * as the change in lock type might satisfy
915 				 * their needs.
916 				 */
917 				locks_wake_up_blocks(fl);
918 				fl->fl_start = request->fl_start;
919 				fl->fl_end = request->fl_end;
920 				fl->fl_type = request->fl_type;
921 				locks_release_private(fl);
922 				locks_copy_private(fl, request);
923 				request = fl;
924 				added = 1;
925 			}
926 		}
927 		/* Go on to next lock.
928 		 */
929 	next_lock:
930 		before = &fl->fl_next;
931 	}
932 
933 	/*
934 	 * The above code only modifies existing locks in case of
935 	 * merging or replacing.  If new lock(s) need to be inserted
936 	 * all modifications are done bellow this, so it's safe yet to
937 	 * bail out.
938 	 */
939 	error = -ENOLCK; /* "no luck" */
940 	if (right && left == right && !new_fl2)
941 		goto out;
942 
943 	error = 0;
944 	if (!added) {
945 		if (request->fl_type == F_UNLCK) {
946 			if (request->fl_flags & FL_EXISTS)
947 				error = -ENOENT;
948 			goto out;
949 		}
950 
951 		if (!new_fl) {
952 			error = -ENOLCK;
953 			goto out;
954 		}
955 		locks_copy_lock(new_fl, request);
956 		locks_insert_lock(before, new_fl);
957 		new_fl = NULL;
958 	}
959 	if (right) {
960 		if (left == right) {
961 			/* The new lock breaks the old one in two pieces,
962 			 * so we have to use the second new lock.
963 			 */
964 			left = new_fl2;
965 			new_fl2 = NULL;
966 			locks_copy_lock(left, right);
967 			locks_insert_lock(before, left);
968 		}
969 		right->fl_start = request->fl_end + 1;
970 		locks_wake_up_blocks(right);
971 	}
972 	if (left) {
973 		left->fl_end = request->fl_start - 1;
974 		locks_wake_up_blocks(left);
975 	}
976  out:
977 	unlock_flocks();
978 	/*
979 	 * Free any unused locks.
980 	 */
981 	if (new_fl)
982 		locks_free_lock(new_fl);
983 	if (new_fl2)
984 		locks_free_lock(new_fl2);
985 	return error;
986 }
987 
988 /**
989  * posix_lock_file - Apply a POSIX-style lock to a file
990  * @filp: The file to apply the lock to
991  * @fl: The lock to be applied
992  * @conflock: Place to return a copy of the conflicting lock, if found.
993  *
994  * Add a POSIX style lock to a file.
995  * We merge adjacent & overlapping locks whenever possible.
996  * POSIX locks are sorted by owner task, then by starting address
997  *
998  * Note that if called with an FL_EXISTS argument, the caller may determine
999  * whether or not a lock was successfully freed by testing the return
1000  * value for -ENOENT.
1001  */
1002 int posix_lock_file(struct file *filp, struct file_lock *fl,
1003 			struct file_lock *conflock)
1004 {
1005 	return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1006 }
1007 EXPORT_SYMBOL(posix_lock_file);
1008 
1009 /**
1010  * posix_lock_file_wait - Apply a POSIX-style lock to a file
1011  * @filp: The file to apply the lock to
1012  * @fl: The lock to be applied
1013  *
1014  * Add a POSIX style lock to a file.
1015  * We merge adjacent & overlapping locks whenever possible.
1016  * POSIX locks are sorted by owner task, then by starting address
1017  */
1018 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1019 {
1020 	int error;
1021 	might_sleep ();
1022 	for (;;) {
1023 		error = posix_lock_file(filp, fl, NULL);
1024 		if (error != FILE_LOCK_DEFERRED)
1025 			break;
1026 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1027 		if (!error)
1028 			continue;
1029 
1030 		locks_delete_block(fl);
1031 		break;
1032 	}
1033 	return error;
1034 }
1035 EXPORT_SYMBOL(posix_lock_file_wait);
1036 
1037 /**
1038  * locks_mandatory_locked - Check for an active lock
1039  * @inode: the file to check
1040  *
1041  * Searches the inode's list of locks to find any POSIX locks which conflict.
1042  * This function is called from locks_verify_locked() only.
1043  */
1044 int locks_mandatory_locked(struct inode *inode)
1045 {
1046 	fl_owner_t owner = current->files;
1047 	struct file_lock *fl;
1048 
1049 	/*
1050 	 * Search the lock list for this inode for any POSIX locks.
1051 	 */
1052 	lock_flocks();
1053 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1054 		if (!IS_POSIX(fl))
1055 			continue;
1056 		if (fl->fl_owner != owner)
1057 			break;
1058 	}
1059 	unlock_flocks();
1060 	return fl ? -EAGAIN : 0;
1061 }
1062 
1063 /**
1064  * locks_mandatory_area - Check for a conflicting lock
1065  * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1066  *		for shared
1067  * @inode:      the file to check
1068  * @filp:       how the file was opened (if it was)
1069  * @offset:     start of area to check
1070  * @count:      length of area to check
1071  *
1072  * Searches the inode's list of locks to find any POSIX locks which conflict.
1073  * This function is called from rw_verify_area() and
1074  * locks_verify_truncate().
1075  */
1076 int locks_mandatory_area(int read_write, struct inode *inode,
1077 			 struct file *filp, loff_t offset,
1078 			 size_t count)
1079 {
1080 	struct file_lock fl;
1081 	int error;
1082 
1083 	locks_init_lock(&fl);
1084 	fl.fl_owner = current->files;
1085 	fl.fl_pid = current->tgid;
1086 	fl.fl_file = filp;
1087 	fl.fl_flags = FL_POSIX | FL_ACCESS;
1088 	if (filp && !(filp->f_flags & O_NONBLOCK))
1089 		fl.fl_flags |= FL_SLEEP;
1090 	fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1091 	fl.fl_start = offset;
1092 	fl.fl_end = offset + count - 1;
1093 
1094 	for (;;) {
1095 		error = __posix_lock_file(inode, &fl, NULL);
1096 		if (error != FILE_LOCK_DEFERRED)
1097 			break;
1098 		error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1099 		if (!error) {
1100 			/*
1101 			 * If we've been sleeping someone might have
1102 			 * changed the permissions behind our back.
1103 			 */
1104 			if (__mandatory_lock(inode))
1105 				continue;
1106 		}
1107 
1108 		locks_delete_block(&fl);
1109 		break;
1110 	}
1111 
1112 	return error;
1113 }
1114 
1115 EXPORT_SYMBOL(locks_mandatory_area);
1116 
1117 static void lease_clear_pending(struct file_lock *fl, int arg)
1118 {
1119 	switch (arg) {
1120 	case F_UNLCK:
1121 		fl->fl_flags &= ~FL_UNLOCK_PENDING;
1122 		/* fall through: */
1123 	case F_RDLCK:
1124 		fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1125 	}
1126 }
1127 
1128 /* We already had a lease on this file; just change its type */
1129 int lease_modify(struct file_lock **before, int arg)
1130 {
1131 	struct file_lock *fl = *before;
1132 	int error = assign_type(fl, arg);
1133 
1134 	if (error)
1135 		return error;
1136 	lease_clear_pending(fl, arg);
1137 	locks_wake_up_blocks(fl);
1138 	if (arg == F_UNLCK) {
1139 		struct file *filp = fl->fl_file;
1140 
1141 		f_delown(filp);
1142 		filp->f_owner.signum = 0;
1143 		fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1144 		if (fl->fl_fasync != NULL) {
1145 			printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1146 			fl->fl_fasync = NULL;
1147 		}
1148 		locks_delete_lock(before);
1149 	}
1150 	return 0;
1151 }
1152 
1153 EXPORT_SYMBOL(lease_modify);
1154 
1155 static bool past_time(unsigned long then)
1156 {
1157 	if (!then)
1158 		/* 0 is a special value meaning "this never expires": */
1159 		return false;
1160 	return time_after(jiffies, then);
1161 }
1162 
1163 static void time_out_leases(struct inode *inode)
1164 {
1165 	struct file_lock **before;
1166 	struct file_lock *fl;
1167 
1168 	before = &inode->i_flock;
1169 	while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
1170 		if (past_time(fl->fl_downgrade_time))
1171 			lease_modify(before, F_RDLCK);
1172 		if (past_time(fl->fl_break_time))
1173 			lease_modify(before, F_UNLCK);
1174 		if (fl == *before)	/* lease_modify may have freed fl */
1175 			before = &fl->fl_next;
1176 	}
1177 }
1178 
1179 /**
1180  *	__break_lease	-	revoke all outstanding leases on file
1181  *	@inode: the inode of the file to return
1182  *	@mode: the open mode (read or write)
1183  *
1184  *	break_lease (inlined for speed) has checked there already is at least
1185  *	some kind of lock (maybe a lease) on this file.  Leases are broken on
1186  *	a call to open() or truncate().  This function can sleep unless you
1187  *	specified %O_NONBLOCK to your open().
1188  */
1189 int __break_lease(struct inode *inode, unsigned int mode)
1190 {
1191 	int error = 0;
1192 	struct file_lock *new_fl, *flock;
1193 	struct file_lock *fl;
1194 	unsigned long break_time;
1195 	int i_have_this_lease = 0;
1196 	int want_write = (mode & O_ACCMODE) != O_RDONLY;
1197 
1198 	new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1199 	if (IS_ERR(new_fl))
1200 		return PTR_ERR(new_fl);
1201 
1202 	lock_flocks();
1203 
1204 	time_out_leases(inode);
1205 
1206 	flock = inode->i_flock;
1207 	if ((flock == NULL) || !IS_LEASE(flock))
1208 		goto out;
1209 
1210 	if (!locks_conflict(flock, new_fl))
1211 		goto out;
1212 
1213 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1214 		if (fl->fl_owner == current->files)
1215 			i_have_this_lease = 1;
1216 
1217 	break_time = 0;
1218 	if (lease_break_time > 0) {
1219 		break_time = jiffies + lease_break_time * HZ;
1220 		if (break_time == 0)
1221 			break_time++;	/* so that 0 means no break time */
1222 	}
1223 
1224 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1225 		if (want_write) {
1226 			if (fl->fl_flags & FL_UNLOCK_PENDING)
1227 				continue;
1228 			fl->fl_flags |= FL_UNLOCK_PENDING;
1229 			fl->fl_break_time = break_time;
1230 		} else {
1231 			if (lease_breaking(flock))
1232 				continue;
1233 			fl->fl_flags |= FL_DOWNGRADE_PENDING;
1234 			fl->fl_downgrade_time = break_time;
1235 		}
1236 		fl->fl_lmops->lm_break(fl);
1237 	}
1238 
1239 	if (i_have_this_lease || (mode & O_NONBLOCK)) {
1240 		error = -EWOULDBLOCK;
1241 		goto out;
1242 	}
1243 
1244 restart:
1245 	break_time = flock->fl_break_time;
1246 	if (break_time != 0) {
1247 		break_time -= jiffies;
1248 		if (break_time == 0)
1249 			break_time++;
1250 	}
1251 	locks_insert_block(flock, new_fl);
1252 	unlock_flocks();
1253 	error = wait_event_interruptible_timeout(new_fl->fl_wait,
1254 						!new_fl->fl_next, break_time);
1255 	lock_flocks();
1256 	__locks_delete_block(new_fl);
1257 	if (error >= 0) {
1258 		if (error == 0)
1259 			time_out_leases(inode);
1260 		/*
1261 		 * Wait for the next conflicting lease that has not been
1262 		 * broken yet
1263 		 */
1264 		for (flock = inode->i_flock; flock && IS_LEASE(flock);
1265 				flock = flock->fl_next) {
1266 			if (locks_conflict(new_fl, flock))
1267 				goto restart;
1268 		}
1269 		error = 0;
1270 	}
1271 
1272 out:
1273 	unlock_flocks();
1274 	locks_free_lock(new_fl);
1275 	return error;
1276 }
1277 
1278 EXPORT_SYMBOL(__break_lease);
1279 
1280 /**
1281  *	lease_get_mtime - get the last modified time of an inode
1282  *	@inode: the inode
1283  *      @time:  pointer to a timespec which will contain the last modified time
1284  *
1285  * This is to force NFS clients to flush their caches for files with
1286  * exclusive leases.  The justification is that if someone has an
1287  * exclusive lease, then they could be modifying it.
1288  */
1289 void lease_get_mtime(struct inode *inode, struct timespec *time)
1290 {
1291 	struct file_lock *flock = inode->i_flock;
1292 	if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1293 		*time = current_fs_time(inode->i_sb);
1294 	else
1295 		*time = inode->i_mtime;
1296 }
1297 
1298 EXPORT_SYMBOL(lease_get_mtime);
1299 
1300 /**
1301  *	fcntl_getlease - Enquire what lease is currently active
1302  *	@filp: the file
1303  *
1304  *	The value returned by this function will be one of
1305  *	(if no lease break is pending):
1306  *
1307  *	%F_RDLCK to indicate a shared lease is held.
1308  *
1309  *	%F_WRLCK to indicate an exclusive lease is held.
1310  *
1311  *	%F_UNLCK to indicate no lease is held.
1312  *
1313  *	(if a lease break is pending):
1314  *
1315  *	%F_RDLCK to indicate an exclusive lease needs to be
1316  *		changed to a shared lease (or removed).
1317  *
1318  *	%F_UNLCK to indicate the lease needs to be removed.
1319  *
1320  *	XXX: sfr & willy disagree over whether F_INPROGRESS
1321  *	should be returned to userspace.
1322  */
1323 int fcntl_getlease(struct file *filp)
1324 {
1325 	struct file_lock *fl;
1326 	int type = F_UNLCK;
1327 
1328 	lock_flocks();
1329 	time_out_leases(filp->f_path.dentry->d_inode);
1330 	for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1331 			fl = fl->fl_next) {
1332 		if (fl->fl_file == filp) {
1333 			type = target_leasetype(fl);
1334 			break;
1335 		}
1336 	}
1337 	unlock_flocks();
1338 	return type;
1339 }
1340 
1341 int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1342 {
1343 	struct file_lock *fl, **before, **my_before = NULL, *lease;
1344 	struct dentry *dentry = filp->f_path.dentry;
1345 	struct inode *inode = dentry->d_inode;
1346 	int error;
1347 
1348 	lease = *flp;
1349 
1350 	error = -EAGAIN;
1351 	if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1352 		goto out;
1353 	if ((arg == F_WRLCK)
1354 	    && ((dentry->d_count > 1)
1355 		|| (atomic_read(&inode->i_count) > 1)))
1356 		goto out;
1357 
1358 	/*
1359 	 * At this point, we know that if there is an exclusive
1360 	 * lease on this file, then we hold it on this filp
1361 	 * (otherwise our open of this file would have blocked).
1362 	 * And if we are trying to acquire an exclusive lease,
1363 	 * then the file is not open by anyone (including us)
1364 	 * except for this filp.
1365 	 */
1366 	error = -EAGAIN;
1367 	for (before = &inode->i_flock;
1368 			((fl = *before) != NULL) && IS_LEASE(fl);
1369 			before = &fl->fl_next) {
1370 		if (fl->fl_file == filp) {
1371 			my_before = before;
1372 			continue;
1373 		}
1374 		/*
1375 		 * No exclusive leases if someone else has a lease on
1376 		 * this file:
1377 		 */
1378 		if (arg == F_WRLCK)
1379 			goto out;
1380 		/*
1381 		 * Modifying our existing lease is OK, but no getting a
1382 		 * new lease if someone else is opening for write:
1383 		 */
1384 		if (fl->fl_flags & FL_UNLOCK_PENDING)
1385 			goto out;
1386 	}
1387 
1388 	if (my_before != NULL) {
1389 		error = lease->fl_lmops->lm_change(my_before, arg);
1390 		if (!error)
1391 			*flp = *my_before;
1392 		goto out;
1393 	}
1394 
1395 	error = -EINVAL;
1396 	if (!leases_enable)
1397 		goto out;
1398 
1399 	locks_insert_lock(before, lease);
1400 	return 0;
1401 
1402 out:
1403 	return error;
1404 }
1405 
1406 int generic_delete_lease(struct file *filp, struct file_lock **flp)
1407 {
1408 	struct file_lock *fl, **before;
1409 	struct dentry *dentry = filp->f_path.dentry;
1410 	struct inode *inode = dentry->d_inode;
1411 
1412 	for (before = &inode->i_flock;
1413 			((fl = *before) != NULL) && IS_LEASE(fl);
1414 			before = &fl->fl_next) {
1415 		if (fl->fl_file != filp)
1416 			continue;
1417 		return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1418 	}
1419 	return -EAGAIN;
1420 }
1421 
1422 /**
1423  *	generic_setlease	-	sets a lease on an open file
1424  *	@filp: file pointer
1425  *	@arg: type of lease to obtain
1426  *	@flp: input - file_lock to use, output - file_lock inserted
1427  *
1428  *	The (input) flp->fl_lmops->lm_break function is required
1429  *	by break_lease().
1430  *
1431  *	Called with file_lock_lock held.
1432  */
1433 int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1434 {
1435 	struct dentry *dentry = filp->f_path.dentry;
1436 	struct inode *inode = dentry->d_inode;
1437 	int error;
1438 
1439 	if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1440 		return -EACCES;
1441 	if (!S_ISREG(inode->i_mode))
1442 		return -EINVAL;
1443 	error = security_file_lock(filp, arg);
1444 	if (error)
1445 		return error;
1446 
1447 	time_out_leases(inode);
1448 
1449 	BUG_ON(!(*flp)->fl_lmops->lm_break);
1450 
1451 	switch (arg) {
1452 	case F_UNLCK:
1453 		return generic_delete_lease(filp, flp);
1454 	case F_RDLCK:
1455 	case F_WRLCK:
1456 		return generic_add_lease(filp, arg, flp);
1457 	default:
1458 		return -EINVAL;
1459 	}
1460 }
1461 EXPORT_SYMBOL(generic_setlease);
1462 
1463 static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1464 {
1465 	if (filp->f_op && filp->f_op->setlease)
1466 		return filp->f_op->setlease(filp, arg, lease);
1467 	else
1468 		return generic_setlease(filp, arg, lease);
1469 }
1470 
1471 /**
1472  *	vfs_setlease        -       sets a lease on an open file
1473  *	@filp: file pointer
1474  *	@arg: type of lease to obtain
1475  *	@lease: file_lock to use
1476  *
1477  *	Call this to establish a lease on the file.
1478  *	The (*lease)->fl_lmops->lm_break operation must be set; if not,
1479  *	break_lease will oops!
1480  *
1481  *	This will call the filesystem's setlease file method, if
1482  *	defined.  Note that there is no getlease method; instead, the
1483  *	filesystem setlease method should call back to setlease() to
1484  *	add a lease to the inode's lease list, where fcntl_getlease() can
1485  *	find it.  Since fcntl_getlease() only reports whether the current
1486  *	task holds a lease, a cluster filesystem need only do this for
1487  *	leases held by processes on this node.
1488  *
1489  *	There is also no break_lease method; filesystems that
1490  *	handle their own leases should break leases themselves from the
1491  *	filesystem's open, create, and (on truncate) setattr methods.
1492  *
1493  *	Warning: the only current setlease methods exist only to disable
1494  *	leases in certain cases.  More vfs changes may be required to
1495  *	allow a full filesystem lease implementation.
1496  */
1497 
1498 int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1499 {
1500 	int error;
1501 
1502 	lock_flocks();
1503 	error = __vfs_setlease(filp, arg, lease);
1504 	unlock_flocks();
1505 
1506 	return error;
1507 }
1508 EXPORT_SYMBOL_GPL(vfs_setlease);
1509 
1510 static int do_fcntl_delete_lease(struct file *filp)
1511 {
1512 	struct file_lock fl, *flp = &fl;
1513 
1514 	lease_init(filp, F_UNLCK, flp);
1515 
1516 	return vfs_setlease(filp, F_UNLCK, &flp);
1517 }
1518 
1519 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1520 {
1521 	struct file_lock *fl, *ret;
1522 	struct fasync_struct *new;
1523 	int error;
1524 
1525 	fl = lease_alloc(filp, arg);
1526 	if (IS_ERR(fl))
1527 		return PTR_ERR(fl);
1528 
1529 	new = fasync_alloc();
1530 	if (!new) {
1531 		locks_free_lock(fl);
1532 		return -ENOMEM;
1533 	}
1534 	ret = fl;
1535 	lock_flocks();
1536 	error = __vfs_setlease(filp, arg, &ret);
1537 	if (error) {
1538 		unlock_flocks();
1539 		locks_free_lock(fl);
1540 		goto out_free_fasync;
1541 	}
1542 	if (ret != fl)
1543 		locks_free_lock(fl);
1544 
1545 	/*
1546 	 * fasync_insert_entry() returns the old entry if any.
1547 	 * If there was no old entry, then it used 'new' and
1548 	 * inserted it into the fasync list. Clear new so that
1549 	 * we don't release it here.
1550 	 */
1551 	if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
1552 		new = NULL;
1553 
1554 	error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1555 	unlock_flocks();
1556 
1557 out_free_fasync:
1558 	if (new)
1559 		fasync_free(new);
1560 	return error;
1561 }
1562 
1563 /**
1564  *	fcntl_setlease	-	sets a lease on an open file
1565  *	@fd: open file descriptor
1566  *	@filp: file pointer
1567  *	@arg: type of lease to obtain
1568  *
1569  *	Call this fcntl to establish a lease on the file.
1570  *	Note that you also need to call %F_SETSIG to
1571  *	receive a signal when the lease is broken.
1572  */
1573 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1574 {
1575 	if (arg == F_UNLCK)
1576 		return do_fcntl_delete_lease(filp);
1577 	return do_fcntl_add_lease(fd, filp, arg);
1578 }
1579 
1580 /**
1581  * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1582  * @filp: The file to apply the lock to
1583  * @fl: The lock to be applied
1584  *
1585  * Add a FLOCK style lock to a file.
1586  */
1587 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1588 {
1589 	int error;
1590 	might_sleep();
1591 	for (;;) {
1592 		error = flock_lock_file(filp, fl);
1593 		if (error != FILE_LOCK_DEFERRED)
1594 			break;
1595 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1596 		if (!error)
1597 			continue;
1598 
1599 		locks_delete_block(fl);
1600 		break;
1601 	}
1602 	return error;
1603 }
1604 
1605 EXPORT_SYMBOL(flock_lock_file_wait);
1606 
1607 /**
1608  *	sys_flock: - flock() system call.
1609  *	@fd: the file descriptor to lock.
1610  *	@cmd: the type of lock to apply.
1611  *
1612  *	Apply a %FL_FLOCK style lock to an open file descriptor.
1613  *	The @cmd can be one of
1614  *
1615  *	%LOCK_SH -- a shared lock.
1616  *
1617  *	%LOCK_EX -- an exclusive lock.
1618  *
1619  *	%LOCK_UN -- remove an existing lock.
1620  *
1621  *	%LOCK_MAND -- a `mandatory' flock.  This exists to emulate Windows Share Modes.
1622  *
1623  *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1624  *	processes read and write access respectively.
1625  */
1626 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1627 {
1628 	struct fd f = fdget(fd);
1629 	struct file_lock *lock;
1630 	int can_sleep, unlock;
1631 	int error;
1632 
1633 	error = -EBADF;
1634 	if (!f.file)
1635 		goto out;
1636 
1637 	can_sleep = !(cmd & LOCK_NB);
1638 	cmd &= ~LOCK_NB;
1639 	unlock = (cmd == LOCK_UN);
1640 
1641 	if (!unlock && !(cmd & LOCK_MAND) &&
1642 	    !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1643 		goto out_putf;
1644 
1645 	error = flock_make_lock(f.file, &lock, cmd);
1646 	if (error)
1647 		goto out_putf;
1648 	if (can_sleep)
1649 		lock->fl_flags |= FL_SLEEP;
1650 
1651 	error = security_file_lock(f.file, lock->fl_type);
1652 	if (error)
1653 		goto out_free;
1654 
1655 	if (f.file->f_op && f.file->f_op->flock)
1656 		error = f.file->f_op->flock(f.file,
1657 					  (can_sleep) ? F_SETLKW : F_SETLK,
1658 					  lock);
1659 	else
1660 		error = flock_lock_file_wait(f.file, lock);
1661 
1662  out_free:
1663 	locks_free_lock(lock);
1664 
1665  out_putf:
1666 	fdput(f);
1667  out:
1668 	return error;
1669 }
1670 
1671 /**
1672  * vfs_test_lock - test file byte range lock
1673  * @filp: The file to test lock for
1674  * @fl: The lock to test; also used to hold result
1675  *
1676  * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
1677  * setting conf->fl_type to something other than F_UNLCK.
1678  */
1679 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1680 {
1681 	if (filp->f_op && filp->f_op->lock)
1682 		return filp->f_op->lock(filp, F_GETLK, fl);
1683 	posix_test_lock(filp, fl);
1684 	return 0;
1685 }
1686 EXPORT_SYMBOL_GPL(vfs_test_lock);
1687 
1688 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1689 {
1690 	flock->l_pid = fl->fl_pid;
1691 #if BITS_PER_LONG == 32
1692 	/*
1693 	 * Make sure we can represent the posix lock via
1694 	 * legacy 32bit flock.
1695 	 */
1696 	if (fl->fl_start > OFFT_OFFSET_MAX)
1697 		return -EOVERFLOW;
1698 	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1699 		return -EOVERFLOW;
1700 #endif
1701 	flock->l_start = fl->fl_start;
1702 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1703 		fl->fl_end - fl->fl_start + 1;
1704 	flock->l_whence = 0;
1705 	flock->l_type = fl->fl_type;
1706 	return 0;
1707 }
1708 
1709 #if BITS_PER_LONG == 32
1710 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1711 {
1712 	flock->l_pid = fl->fl_pid;
1713 	flock->l_start = fl->fl_start;
1714 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1715 		fl->fl_end - fl->fl_start + 1;
1716 	flock->l_whence = 0;
1717 	flock->l_type = fl->fl_type;
1718 }
1719 #endif
1720 
1721 /* Report the first existing lock that would conflict with l.
1722  * This implements the F_GETLK command of fcntl().
1723  */
1724 int fcntl_getlk(struct file *filp, struct flock __user *l)
1725 {
1726 	struct file_lock file_lock;
1727 	struct flock flock;
1728 	int error;
1729 
1730 	error = -EFAULT;
1731 	if (copy_from_user(&flock, l, sizeof(flock)))
1732 		goto out;
1733 	error = -EINVAL;
1734 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1735 		goto out;
1736 
1737 	error = flock_to_posix_lock(filp, &file_lock, &flock);
1738 	if (error)
1739 		goto out;
1740 
1741 	error = vfs_test_lock(filp, &file_lock);
1742 	if (error)
1743 		goto out;
1744 
1745 	flock.l_type = file_lock.fl_type;
1746 	if (file_lock.fl_type != F_UNLCK) {
1747 		error = posix_lock_to_flock(&flock, &file_lock);
1748 		if (error)
1749 			goto out;
1750 	}
1751 	error = -EFAULT;
1752 	if (!copy_to_user(l, &flock, sizeof(flock)))
1753 		error = 0;
1754 out:
1755 	return error;
1756 }
1757 
1758 /**
1759  * vfs_lock_file - file byte range lock
1760  * @filp: The file to apply the lock to
1761  * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1762  * @fl: The lock to be applied
1763  * @conf: Place to return a copy of the conflicting lock, if found.
1764  *
1765  * A caller that doesn't care about the conflicting lock may pass NULL
1766  * as the final argument.
1767  *
1768  * If the filesystem defines a private ->lock() method, then @conf will
1769  * be left unchanged; so a caller that cares should initialize it to
1770  * some acceptable default.
1771  *
1772  * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1773  * locks, the ->lock() interface may return asynchronously, before the lock has
1774  * been granted or denied by the underlying filesystem, if (and only if)
1775  * lm_grant is set. Callers expecting ->lock() to return asynchronously
1776  * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1777  * the request is for a blocking lock. When ->lock() does return asynchronously,
1778  * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
1779  * request completes.
1780  * If the request is for non-blocking lock the file system should return
1781  * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1782  * with the result. If the request timed out the callback routine will return a
1783  * nonzero return code and the file system should release the lock. The file
1784  * system is also responsible to keep a corresponding posix lock when it
1785  * grants a lock so the VFS can find out which locks are locally held and do
1786  * the correct lock cleanup when required.
1787  * The underlying filesystem must not drop the kernel lock or call
1788  * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
1789  * return code.
1790  */
1791 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1792 {
1793 	if (filp->f_op && filp->f_op->lock)
1794 		return filp->f_op->lock(filp, cmd, fl);
1795 	else
1796 		return posix_lock_file(filp, fl, conf);
1797 }
1798 EXPORT_SYMBOL_GPL(vfs_lock_file);
1799 
1800 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1801 			     struct file_lock *fl)
1802 {
1803 	int error;
1804 
1805 	error = security_file_lock(filp, fl->fl_type);
1806 	if (error)
1807 		return error;
1808 
1809 	for (;;) {
1810 		error = vfs_lock_file(filp, cmd, fl, NULL);
1811 		if (error != FILE_LOCK_DEFERRED)
1812 			break;
1813 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1814 		if (!error)
1815 			continue;
1816 
1817 		locks_delete_block(fl);
1818 		break;
1819 	}
1820 
1821 	return error;
1822 }
1823 
1824 /* Apply the lock described by l to an open file descriptor.
1825  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1826  */
1827 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1828 		struct flock __user *l)
1829 {
1830 	struct file_lock *file_lock = locks_alloc_lock();
1831 	struct flock flock;
1832 	struct inode *inode;
1833 	struct file *f;
1834 	int error;
1835 
1836 	if (file_lock == NULL)
1837 		return -ENOLCK;
1838 
1839 	/*
1840 	 * This might block, so we do it before checking the inode.
1841 	 */
1842 	error = -EFAULT;
1843 	if (copy_from_user(&flock, l, sizeof(flock)))
1844 		goto out;
1845 
1846 	inode = filp->f_path.dentry->d_inode;
1847 
1848 	/* Don't allow mandatory locks on files that may be memory mapped
1849 	 * and shared.
1850 	 */
1851 	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1852 		error = -EAGAIN;
1853 		goto out;
1854 	}
1855 
1856 again:
1857 	error = flock_to_posix_lock(filp, file_lock, &flock);
1858 	if (error)
1859 		goto out;
1860 	if (cmd == F_SETLKW) {
1861 		file_lock->fl_flags |= FL_SLEEP;
1862 	}
1863 
1864 	error = -EBADF;
1865 	switch (flock.l_type) {
1866 	case F_RDLCK:
1867 		if (!(filp->f_mode & FMODE_READ))
1868 			goto out;
1869 		break;
1870 	case F_WRLCK:
1871 		if (!(filp->f_mode & FMODE_WRITE))
1872 			goto out;
1873 		break;
1874 	case F_UNLCK:
1875 		break;
1876 	default:
1877 		error = -EINVAL;
1878 		goto out;
1879 	}
1880 
1881 	error = do_lock_file_wait(filp, cmd, file_lock);
1882 
1883 	/*
1884 	 * Attempt to detect a close/fcntl race and recover by
1885 	 * releasing the lock that was just acquired.
1886 	 */
1887 	/*
1888 	 * we need that spin_lock here - it prevents reordering between
1889 	 * update of inode->i_flock and check for it done in close().
1890 	 * rcu_read_lock() wouldn't do.
1891 	 */
1892 	spin_lock(&current->files->file_lock);
1893 	f = fcheck(fd);
1894 	spin_unlock(&current->files->file_lock);
1895 	if (!error && f != filp && flock.l_type != F_UNLCK) {
1896 		flock.l_type = F_UNLCK;
1897 		goto again;
1898 	}
1899 
1900 out:
1901 	locks_free_lock(file_lock);
1902 	return error;
1903 }
1904 
1905 #if BITS_PER_LONG == 32
1906 /* Report the first existing lock that would conflict with l.
1907  * This implements the F_GETLK command of fcntl().
1908  */
1909 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1910 {
1911 	struct file_lock file_lock;
1912 	struct flock64 flock;
1913 	int error;
1914 
1915 	error = -EFAULT;
1916 	if (copy_from_user(&flock, l, sizeof(flock)))
1917 		goto out;
1918 	error = -EINVAL;
1919 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1920 		goto out;
1921 
1922 	error = flock64_to_posix_lock(filp, &file_lock, &flock);
1923 	if (error)
1924 		goto out;
1925 
1926 	error = vfs_test_lock(filp, &file_lock);
1927 	if (error)
1928 		goto out;
1929 
1930 	flock.l_type = file_lock.fl_type;
1931 	if (file_lock.fl_type != F_UNLCK)
1932 		posix_lock_to_flock64(&flock, &file_lock);
1933 
1934 	error = -EFAULT;
1935 	if (!copy_to_user(l, &flock, sizeof(flock)))
1936 		error = 0;
1937 
1938 out:
1939 	return error;
1940 }
1941 
1942 /* Apply the lock described by l to an open file descriptor.
1943  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1944  */
1945 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1946 		struct flock64 __user *l)
1947 {
1948 	struct file_lock *file_lock = locks_alloc_lock();
1949 	struct flock64 flock;
1950 	struct inode *inode;
1951 	struct file *f;
1952 	int error;
1953 
1954 	if (file_lock == NULL)
1955 		return -ENOLCK;
1956 
1957 	/*
1958 	 * This might block, so we do it before checking the inode.
1959 	 */
1960 	error = -EFAULT;
1961 	if (copy_from_user(&flock, l, sizeof(flock)))
1962 		goto out;
1963 
1964 	inode = filp->f_path.dentry->d_inode;
1965 
1966 	/* Don't allow mandatory locks on files that may be memory mapped
1967 	 * and shared.
1968 	 */
1969 	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1970 		error = -EAGAIN;
1971 		goto out;
1972 	}
1973 
1974 again:
1975 	error = flock64_to_posix_lock(filp, file_lock, &flock);
1976 	if (error)
1977 		goto out;
1978 	if (cmd == F_SETLKW64) {
1979 		file_lock->fl_flags |= FL_SLEEP;
1980 	}
1981 
1982 	error = -EBADF;
1983 	switch (flock.l_type) {
1984 	case F_RDLCK:
1985 		if (!(filp->f_mode & FMODE_READ))
1986 			goto out;
1987 		break;
1988 	case F_WRLCK:
1989 		if (!(filp->f_mode & FMODE_WRITE))
1990 			goto out;
1991 		break;
1992 	case F_UNLCK:
1993 		break;
1994 	default:
1995 		error = -EINVAL;
1996 		goto out;
1997 	}
1998 
1999 	error = do_lock_file_wait(filp, cmd, file_lock);
2000 
2001 	/*
2002 	 * Attempt to detect a close/fcntl race and recover by
2003 	 * releasing the lock that was just acquired.
2004 	 */
2005 	spin_lock(&current->files->file_lock);
2006 	f = fcheck(fd);
2007 	spin_unlock(&current->files->file_lock);
2008 	if (!error && f != filp && flock.l_type != F_UNLCK) {
2009 		flock.l_type = F_UNLCK;
2010 		goto again;
2011 	}
2012 
2013 out:
2014 	locks_free_lock(file_lock);
2015 	return error;
2016 }
2017 #endif /* BITS_PER_LONG == 32 */
2018 
2019 /*
2020  * This function is called when the file is being removed
2021  * from the task's fd array.  POSIX locks belonging to this task
2022  * are deleted at this time.
2023  */
2024 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2025 {
2026 	struct file_lock lock;
2027 
2028 	/*
2029 	 * If there are no locks held on this file, we don't need to call
2030 	 * posix_lock_file().  Another process could be setting a lock on this
2031 	 * file at the same time, but we wouldn't remove that lock anyway.
2032 	 */
2033 	if (!filp->f_path.dentry->d_inode->i_flock)
2034 		return;
2035 
2036 	lock.fl_type = F_UNLCK;
2037 	lock.fl_flags = FL_POSIX | FL_CLOSE;
2038 	lock.fl_start = 0;
2039 	lock.fl_end = OFFSET_MAX;
2040 	lock.fl_owner = owner;
2041 	lock.fl_pid = current->tgid;
2042 	lock.fl_file = filp;
2043 	lock.fl_ops = NULL;
2044 	lock.fl_lmops = NULL;
2045 
2046 	vfs_lock_file(filp, F_SETLK, &lock, NULL);
2047 
2048 	if (lock.fl_ops && lock.fl_ops->fl_release_private)
2049 		lock.fl_ops->fl_release_private(&lock);
2050 }
2051 
2052 EXPORT_SYMBOL(locks_remove_posix);
2053 
2054 /*
2055  * This function is called on the last close of an open file.
2056  */
2057 void locks_remove_flock(struct file *filp)
2058 {
2059 	struct inode * inode = filp->f_path.dentry->d_inode;
2060 	struct file_lock *fl;
2061 	struct file_lock **before;
2062 
2063 	if (!inode->i_flock)
2064 		return;
2065 
2066 	if (filp->f_op && filp->f_op->flock) {
2067 		struct file_lock fl = {
2068 			.fl_pid = current->tgid,
2069 			.fl_file = filp,
2070 			.fl_flags = FL_FLOCK,
2071 			.fl_type = F_UNLCK,
2072 			.fl_end = OFFSET_MAX,
2073 		};
2074 		filp->f_op->flock(filp, F_SETLKW, &fl);
2075 		if (fl.fl_ops && fl.fl_ops->fl_release_private)
2076 			fl.fl_ops->fl_release_private(&fl);
2077 	}
2078 
2079 	lock_flocks();
2080 	before = &inode->i_flock;
2081 
2082 	while ((fl = *before) != NULL) {
2083 		if (fl->fl_file == filp) {
2084 			if (IS_FLOCK(fl)) {
2085 				locks_delete_lock(before);
2086 				continue;
2087 			}
2088 			if (IS_LEASE(fl)) {
2089 				lease_modify(before, F_UNLCK);
2090 				continue;
2091 			}
2092 			/* What? */
2093 			BUG();
2094  		}
2095 		before = &fl->fl_next;
2096 	}
2097 	unlock_flocks();
2098 }
2099 
2100 /**
2101  *	posix_unblock_lock - stop waiting for a file lock
2102  *      @filp:   how the file was opened
2103  *	@waiter: the lock which was waiting
2104  *
2105  *	lockd needs to block waiting for locks.
2106  */
2107 int
2108 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2109 {
2110 	int status = 0;
2111 
2112 	lock_flocks();
2113 	if (waiter->fl_next)
2114 		__locks_delete_block(waiter);
2115 	else
2116 		status = -ENOENT;
2117 	unlock_flocks();
2118 	return status;
2119 }
2120 
2121 EXPORT_SYMBOL(posix_unblock_lock);
2122 
2123 /**
2124  * vfs_cancel_lock - file byte range unblock lock
2125  * @filp: The file to apply the unblock to
2126  * @fl: The lock to be unblocked
2127  *
2128  * Used by lock managers to cancel blocked requests
2129  */
2130 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2131 {
2132 	if (filp->f_op && filp->f_op->lock)
2133 		return filp->f_op->lock(filp, F_CANCELLK, fl);
2134 	return 0;
2135 }
2136 
2137 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2138 
2139 #ifdef CONFIG_PROC_FS
2140 #include <linux/proc_fs.h>
2141 #include <linux/seq_file.h>
2142 
2143 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2144 			    loff_t id, char *pfx)
2145 {
2146 	struct inode *inode = NULL;
2147 	unsigned int fl_pid;
2148 
2149 	if (fl->fl_nspid)
2150 		fl_pid = pid_vnr(fl->fl_nspid);
2151 	else
2152 		fl_pid = fl->fl_pid;
2153 
2154 	if (fl->fl_file != NULL)
2155 		inode = fl->fl_file->f_path.dentry->d_inode;
2156 
2157 	seq_printf(f, "%lld:%s ", id, pfx);
2158 	if (IS_POSIX(fl)) {
2159 		seq_printf(f, "%6s %s ",
2160 			     (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2161 			     (inode == NULL) ? "*NOINODE*" :
2162 			     mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2163 	} else if (IS_FLOCK(fl)) {
2164 		if (fl->fl_type & LOCK_MAND) {
2165 			seq_printf(f, "FLOCK  MSNFS     ");
2166 		} else {
2167 			seq_printf(f, "FLOCK  ADVISORY  ");
2168 		}
2169 	} else if (IS_LEASE(fl)) {
2170 		seq_printf(f, "LEASE  ");
2171 		if (lease_breaking(fl))
2172 			seq_printf(f, "BREAKING  ");
2173 		else if (fl->fl_file)
2174 			seq_printf(f, "ACTIVE    ");
2175 		else
2176 			seq_printf(f, "BREAKER   ");
2177 	} else {
2178 		seq_printf(f, "UNKNOWN UNKNOWN  ");
2179 	}
2180 	if (fl->fl_type & LOCK_MAND) {
2181 		seq_printf(f, "%s ",
2182 			       (fl->fl_type & LOCK_READ)
2183 			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2184 			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2185 	} else {
2186 		seq_printf(f, "%s ",
2187 			       (lease_breaking(fl))
2188 			       ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2189 			       : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2190 	}
2191 	if (inode) {
2192 #ifdef WE_CAN_BREAK_LSLK_NOW
2193 		seq_printf(f, "%d %s:%ld ", fl_pid,
2194 				inode->i_sb->s_id, inode->i_ino);
2195 #else
2196 		/* userspace relies on this representation of dev_t ;-( */
2197 		seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2198 				MAJOR(inode->i_sb->s_dev),
2199 				MINOR(inode->i_sb->s_dev), inode->i_ino);
2200 #endif
2201 	} else {
2202 		seq_printf(f, "%d <none>:0 ", fl_pid);
2203 	}
2204 	if (IS_POSIX(fl)) {
2205 		if (fl->fl_end == OFFSET_MAX)
2206 			seq_printf(f, "%Ld EOF\n", fl->fl_start);
2207 		else
2208 			seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2209 	} else {
2210 		seq_printf(f, "0 EOF\n");
2211 	}
2212 }
2213 
2214 static int locks_show(struct seq_file *f, void *v)
2215 {
2216 	struct file_lock *fl, *bfl;
2217 
2218 	fl = list_entry(v, struct file_lock, fl_link);
2219 
2220 	lock_get_status(f, fl, *((loff_t *)f->private), "");
2221 
2222 	list_for_each_entry(bfl, &fl->fl_block, fl_block)
2223 		lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
2224 
2225 	return 0;
2226 }
2227 
2228 static void *locks_start(struct seq_file *f, loff_t *pos)
2229 {
2230 	loff_t *p = f->private;
2231 
2232 	lock_flocks();
2233 	*p = (*pos + 1);
2234 	return seq_list_start(&file_lock_list, *pos);
2235 }
2236 
2237 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2238 {
2239 	loff_t *p = f->private;
2240 	++*p;
2241 	return seq_list_next(v, &file_lock_list, pos);
2242 }
2243 
2244 static void locks_stop(struct seq_file *f, void *v)
2245 {
2246 	unlock_flocks();
2247 }
2248 
2249 static const struct seq_operations locks_seq_operations = {
2250 	.start	= locks_start,
2251 	.next	= locks_next,
2252 	.stop	= locks_stop,
2253 	.show	= locks_show,
2254 };
2255 
2256 static int locks_open(struct inode *inode, struct file *filp)
2257 {
2258 	return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
2259 }
2260 
2261 static const struct file_operations proc_locks_operations = {
2262 	.open		= locks_open,
2263 	.read		= seq_read,
2264 	.llseek		= seq_lseek,
2265 	.release	= seq_release_private,
2266 };
2267 
2268 static int __init proc_locks_init(void)
2269 {
2270 	proc_create("locks", 0, NULL, &proc_locks_operations);
2271 	return 0;
2272 }
2273 module_init(proc_locks_init);
2274 #endif
2275 
2276 /**
2277  *	lock_may_read - checks that the region is free of locks
2278  *	@inode: the inode that is being read
2279  *	@start: the first byte to read
2280  *	@len: the number of bytes to read
2281  *
2282  *	Emulates Windows locking requirements.  Whole-file
2283  *	mandatory locks (share modes) can prohibit a read and
2284  *	byte-range POSIX locks can prohibit a read if they overlap.
2285  *
2286  *	N.B. this function is only ever called
2287  *	from knfsd and ownership of locks is never checked.
2288  */
2289 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2290 {
2291 	struct file_lock *fl;
2292 	int result = 1;
2293 	lock_flocks();
2294 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2295 		if (IS_POSIX(fl)) {
2296 			if (fl->fl_type == F_RDLCK)
2297 				continue;
2298 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2299 				continue;
2300 		} else if (IS_FLOCK(fl)) {
2301 			if (!(fl->fl_type & LOCK_MAND))
2302 				continue;
2303 			if (fl->fl_type & LOCK_READ)
2304 				continue;
2305 		} else
2306 			continue;
2307 		result = 0;
2308 		break;
2309 	}
2310 	unlock_flocks();
2311 	return result;
2312 }
2313 
2314 EXPORT_SYMBOL(lock_may_read);
2315 
2316 /**
2317  *	lock_may_write - checks that the region is free of locks
2318  *	@inode: the inode that is being written
2319  *	@start: the first byte to write
2320  *	@len: the number of bytes to write
2321  *
2322  *	Emulates Windows locking requirements.  Whole-file
2323  *	mandatory locks (share modes) can prohibit a write and
2324  *	byte-range POSIX locks can prohibit a write if they overlap.
2325  *
2326  *	N.B. this function is only ever called
2327  *	from knfsd and ownership of locks is never checked.
2328  */
2329 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2330 {
2331 	struct file_lock *fl;
2332 	int result = 1;
2333 	lock_flocks();
2334 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2335 		if (IS_POSIX(fl)) {
2336 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2337 				continue;
2338 		} else if (IS_FLOCK(fl)) {
2339 			if (!(fl->fl_type & LOCK_MAND))
2340 				continue;
2341 			if (fl->fl_type & LOCK_WRITE)
2342 				continue;
2343 		} else
2344 			continue;
2345 		result = 0;
2346 		break;
2347 	}
2348 	unlock_flocks();
2349 	return result;
2350 }
2351 
2352 EXPORT_SYMBOL(lock_may_write);
2353 
2354 static int __init filelock_init(void)
2355 {
2356 	filelock_cache = kmem_cache_create("file_lock_cache",
2357 			sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2358 
2359 	return 0;
2360 }
2361 
2362 core_initcall(filelock_init);
2363