xref: /openbmc/linux/fs/locks.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
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/mandatory.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/fs.h>
120 #include <linux/init.h>
121 #include <linux/module.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/smp_lock.h>
125 #include <linux/syscalls.h>
126 #include <linux/time.h>
127 #include <linux/rcupdate.h>
128 
129 #include <asm/semaphore.h>
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 int leases_enable = 1;
137 int lease_break_time = 45;
138 
139 #define for_each_lock(inode, lockp) \
140 	for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141 
142 static LIST_HEAD(file_lock_list);
143 static LIST_HEAD(blocked_list);
144 
145 static struct kmem_cache *filelock_cache __read_mostly;
146 
147 /* Allocate an empty lock structure. */
148 static struct file_lock *locks_alloc_lock(void)
149 {
150 	return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
151 }
152 
153 static void locks_release_private(struct file_lock *fl)
154 {
155 	if (fl->fl_ops) {
156 		if (fl->fl_ops->fl_release_private)
157 			fl->fl_ops->fl_release_private(fl);
158 		fl->fl_ops = NULL;
159 	}
160 	if (fl->fl_lmops) {
161 		if (fl->fl_lmops->fl_release_private)
162 			fl->fl_lmops->fl_release_private(fl);
163 		fl->fl_lmops = NULL;
164 	}
165 
166 }
167 
168 /* Free a lock which is not in use. */
169 static void locks_free_lock(struct file_lock *fl)
170 {
171 	BUG_ON(waitqueue_active(&fl->fl_wait));
172 	BUG_ON(!list_empty(&fl->fl_block));
173 	BUG_ON(!list_empty(&fl->fl_link));
174 
175 	locks_release_private(fl);
176 	kmem_cache_free(filelock_cache, fl);
177 }
178 
179 void locks_init_lock(struct file_lock *fl)
180 {
181 	INIT_LIST_HEAD(&fl->fl_link);
182 	INIT_LIST_HEAD(&fl->fl_block);
183 	init_waitqueue_head(&fl->fl_wait);
184 	fl->fl_next = NULL;
185 	fl->fl_fasync = NULL;
186 	fl->fl_owner = NULL;
187 	fl->fl_pid = 0;
188 	fl->fl_file = NULL;
189 	fl->fl_flags = 0;
190 	fl->fl_type = 0;
191 	fl->fl_start = fl->fl_end = 0;
192 	fl->fl_ops = NULL;
193 	fl->fl_lmops = NULL;
194 }
195 
196 EXPORT_SYMBOL(locks_init_lock);
197 
198 /*
199  * Initialises the fields of the file lock which are invariant for
200  * free file_locks.
201  */
202 static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
203 {
204 	struct file_lock *lock = (struct file_lock *) foo;
205 
206 	if (!(flags & SLAB_CTOR_CONSTRUCTOR))
207 		return;
208 
209 	locks_init_lock(lock);
210 }
211 
212 static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
213 {
214 	if (fl->fl_ops) {
215 		if (fl->fl_ops->fl_copy_lock)
216 			fl->fl_ops->fl_copy_lock(new, fl);
217 		new->fl_ops = fl->fl_ops;
218 	}
219 	if (fl->fl_lmops) {
220 		if (fl->fl_lmops->fl_copy_lock)
221 			fl->fl_lmops->fl_copy_lock(new, fl);
222 		new->fl_lmops = fl->fl_lmops;
223 	}
224 }
225 
226 /*
227  * Initialize a new lock from an existing file_lock structure.
228  */
229 static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
230 {
231 	new->fl_owner = fl->fl_owner;
232 	new->fl_pid = fl->fl_pid;
233 	new->fl_file = NULL;
234 	new->fl_flags = fl->fl_flags;
235 	new->fl_type = fl->fl_type;
236 	new->fl_start = fl->fl_start;
237 	new->fl_end = fl->fl_end;
238 	new->fl_ops = NULL;
239 	new->fl_lmops = NULL;
240 }
241 
242 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
243 {
244 	locks_release_private(new);
245 
246 	__locks_copy_lock(new, fl);
247 	new->fl_file = fl->fl_file;
248 	new->fl_ops = fl->fl_ops;
249 	new->fl_lmops = fl->fl_lmops;
250 
251 	locks_copy_private(new, fl);
252 }
253 
254 EXPORT_SYMBOL(locks_copy_lock);
255 
256 static inline int flock_translate_cmd(int cmd) {
257 	if (cmd & LOCK_MAND)
258 		return cmd & (LOCK_MAND | LOCK_RW);
259 	switch (cmd) {
260 	case LOCK_SH:
261 		return F_RDLCK;
262 	case LOCK_EX:
263 		return F_WRLCK;
264 	case LOCK_UN:
265 		return F_UNLCK;
266 	}
267 	return -EINVAL;
268 }
269 
270 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
271 static int flock_make_lock(struct file *filp, struct file_lock **lock,
272 		unsigned int cmd)
273 {
274 	struct file_lock *fl;
275 	int type = flock_translate_cmd(cmd);
276 	if (type < 0)
277 		return type;
278 
279 	fl = locks_alloc_lock();
280 	if (fl == NULL)
281 		return -ENOMEM;
282 
283 	fl->fl_file = filp;
284 	fl->fl_pid = current->tgid;
285 	fl->fl_flags = FL_FLOCK;
286 	fl->fl_type = type;
287 	fl->fl_end = OFFSET_MAX;
288 
289 	*lock = fl;
290 	return 0;
291 }
292 
293 static int assign_type(struct file_lock *fl, int type)
294 {
295 	switch (type) {
296 	case F_RDLCK:
297 	case F_WRLCK:
298 	case F_UNLCK:
299 		fl->fl_type = type;
300 		break;
301 	default:
302 		return -EINVAL;
303 	}
304 	return 0;
305 }
306 
307 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
308  * style lock.
309  */
310 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
311 			       struct flock *l)
312 {
313 	off_t start, end;
314 
315 	switch (l->l_whence) {
316 	case SEEK_SET:
317 		start = 0;
318 		break;
319 	case SEEK_CUR:
320 		start = filp->f_pos;
321 		break;
322 	case SEEK_END:
323 		start = i_size_read(filp->f_path.dentry->d_inode);
324 		break;
325 	default:
326 		return -EINVAL;
327 	}
328 
329 	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
330 	   POSIX-2001 defines it. */
331 	start += l->l_start;
332 	if (start < 0)
333 		return -EINVAL;
334 	fl->fl_end = OFFSET_MAX;
335 	if (l->l_len > 0) {
336 		end = start + l->l_len - 1;
337 		fl->fl_end = end;
338 	} else if (l->l_len < 0) {
339 		end = start - 1;
340 		fl->fl_end = end;
341 		start += l->l_len;
342 		if (start < 0)
343 			return -EINVAL;
344 	}
345 	fl->fl_start = start;	/* we record the absolute position */
346 	if (fl->fl_end < fl->fl_start)
347 		return -EOVERFLOW;
348 
349 	fl->fl_owner = current->files;
350 	fl->fl_pid = current->tgid;
351 	fl->fl_file = filp;
352 	fl->fl_flags = FL_POSIX;
353 	fl->fl_ops = NULL;
354 	fl->fl_lmops = NULL;
355 
356 	return assign_type(fl, l->l_type);
357 }
358 
359 #if BITS_PER_LONG == 32
360 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
361 				 struct flock64 *l)
362 {
363 	loff_t start;
364 
365 	switch (l->l_whence) {
366 	case SEEK_SET:
367 		start = 0;
368 		break;
369 	case SEEK_CUR:
370 		start = filp->f_pos;
371 		break;
372 	case SEEK_END:
373 		start = i_size_read(filp->f_path.dentry->d_inode);
374 		break;
375 	default:
376 		return -EINVAL;
377 	}
378 
379 	start += l->l_start;
380 	if (start < 0)
381 		return -EINVAL;
382 	fl->fl_end = OFFSET_MAX;
383 	if (l->l_len > 0) {
384 		fl->fl_end = start + l->l_len - 1;
385 	} else if (l->l_len < 0) {
386 		fl->fl_end = start - 1;
387 		start += l->l_len;
388 		if (start < 0)
389 			return -EINVAL;
390 	}
391 	fl->fl_start = start;	/* we record the absolute position */
392 	if (fl->fl_end < fl->fl_start)
393 		return -EOVERFLOW;
394 
395 	fl->fl_owner = current->files;
396 	fl->fl_pid = current->tgid;
397 	fl->fl_file = filp;
398 	fl->fl_flags = FL_POSIX;
399 	fl->fl_ops = NULL;
400 	fl->fl_lmops = NULL;
401 
402 	switch (l->l_type) {
403 	case F_RDLCK:
404 	case F_WRLCK:
405 	case F_UNLCK:
406 		fl->fl_type = l->l_type;
407 		break;
408 	default:
409 		return -EINVAL;
410 	}
411 
412 	return (0);
413 }
414 #endif
415 
416 /* default lease lock manager operations */
417 static void lease_break_callback(struct file_lock *fl)
418 {
419 	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
420 }
421 
422 static void lease_release_private_callback(struct file_lock *fl)
423 {
424 	if (!fl->fl_file)
425 		return;
426 
427 	f_delown(fl->fl_file);
428 	fl->fl_file->f_owner.signum = 0;
429 }
430 
431 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
432 {
433 	return fl->fl_file == try->fl_file;
434 }
435 
436 static struct lock_manager_operations lease_manager_ops = {
437 	.fl_break = lease_break_callback,
438 	.fl_release_private = lease_release_private_callback,
439 	.fl_mylease = lease_mylease_callback,
440 	.fl_change = lease_modify,
441 };
442 
443 /*
444  * Initialize a lease, use the default lock manager operations
445  */
446 static int lease_init(struct file *filp, int type, struct file_lock *fl)
447  {
448 	if (assign_type(fl, type) != 0)
449 		return -EINVAL;
450 
451 	fl->fl_owner = current->files;
452 	fl->fl_pid = current->tgid;
453 
454 	fl->fl_file = filp;
455 	fl->fl_flags = FL_LEASE;
456 	fl->fl_start = 0;
457 	fl->fl_end = OFFSET_MAX;
458 	fl->fl_ops = NULL;
459 	fl->fl_lmops = &lease_manager_ops;
460 	return 0;
461 }
462 
463 /* Allocate a file_lock initialised to this type of lease */
464 static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
465 {
466 	struct file_lock *fl = locks_alloc_lock();
467 	int error = -ENOMEM;
468 
469 	if (fl == NULL)
470 		goto out;
471 
472 	error = lease_init(filp, type, fl);
473 	if (error) {
474 		locks_free_lock(fl);
475 		fl = NULL;
476 	}
477 out:
478 	*flp = fl;
479 	return error;
480 }
481 
482 /* Check if two locks overlap each other.
483  */
484 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
485 {
486 	return ((fl1->fl_end >= fl2->fl_start) &&
487 		(fl2->fl_end >= fl1->fl_start));
488 }
489 
490 /*
491  * Check whether two locks have the same owner.
492  */
493 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
494 {
495 	if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
496 		return fl2->fl_lmops == fl1->fl_lmops &&
497 			fl1->fl_lmops->fl_compare_owner(fl1, fl2);
498 	return fl1->fl_owner == fl2->fl_owner;
499 }
500 
501 /* Remove waiter from blocker's block list.
502  * When blocker ends up pointing to itself then the list is empty.
503  */
504 static void __locks_delete_block(struct file_lock *waiter)
505 {
506 	list_del_init(&waiter->fl_block);
507 	list_del_init(&waiter->fl_link);
508 	waiter->fl_next = NULL;
509 }
510 
511 /*
512  */
513 static void locks_delete_block(struct file_lock *waiter)
514 {
515 	lock_kernel();
516 	__locks_delete_block(waiter);
517 	unlock_kernel();
518 }
519 
520 /* Insert waiter into blocker's block list.
521  * We use a circular list so that processes can be easily woken up in
522  * the order they blocked. The documentation doesn't require this but
523  * it seems like the reasonable thing to do.
524  */
525 static void locks_insert_block(struct file_lock *blocker,
526 			       struct file_lock *waiter)
527 {
528 	BUG_ON(!list_empty(&waiter->fl_block));
529 	list_add_tail(&waiter->fl_block, &blocker->fl_block);
530 	waiter->fl_next = blocker;
531 	if (IS_POSIX(blocker))
532 		list_add(&waiter->fl_link, &blocked_list);
533 }
534 
535 /* Wake up processes blocked waiting for blocker.
536  * If told to wait then schedule the processes until the block list
537  * is empty, otherwise empty the block list ourselves.
538  */
539 static void locks_wake_up_blocks(struct file_lock *blocker)
540 {
541 	while (!list_empty(&blocker->fl_block)) {
542 		struct file_lock *waiter = list_entry(blocker->fl_block.next,
543 				struct file_lock, fl_block);
544 		__locks_delete_block(waiter);
545 		if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
546 			waiter->fl_lmops->fl_notify(waiter);
547 		else
548 			wake_up(&waiter->fl_wait);
549 	}
550 }
551 
552 /* Insert file lock fl into an inode's lock list at the position indicated
553  * by pos. At the same time add the lock to the global file lock list.
554  */
555 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
556 {
557 	list_add(&fl->fl_link, &file_lock_list);
558 
559 	/* insert into file's list */
560 	fl->fl_next = *pos;
561 	*pos = fl;
562 
563 	if (fl->fl_ops && fl->fl_ops->fl_insert)
564 		fl->fl_ops->fl_insert(fl);
565 }
566 
567 /*
568  * Delete a lock and then free it.
569  * Wake up processes that are blocked waiting for this lock,
570  * notify the FS that the lock has been cleared and
571  * finally free the lock.
572  */
573 static void locks_delete_lock(struct file_lock **thisfl_p)
574 {
575 	struct file_lock *fl = *thisfl_p;
576 
577 	*thisfl_p = fl->fl_next;
578 	fl->fl_next = NULL;
579 	list_del_init(&fl->fl_link);
580 
581 	fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
582 	if (fl->fl_fasync != NULL) {
583 		printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
584 		fl->fl_fasync = NULL;
585 	}
586 
587 	if (fl->fl_ops && fl->fl_ops->fl_remove)
588 		fl->fl_ops->fl_remove(fl);
589 
590 	locks_wake_up_blocks(fl);
591 	locks_free_lock(fl);
592 }
593 
594 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
595  * checks for shared/exclusive status of overlapping locks.
596  */
597 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
598 {
599 	if (sys_fl->fl_type == F_WRLCK)
600 		return 1;
601 	if (caller_fl->fl_type == F_WRLCK)
602 		return 1;
603 	return 0;
604 }
605 
606 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
607  * checking before calling the locks_conflict().
608  */
609 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
610 {
611 	/* POSIX locks owned by the same process do not conflict with
612 	 * each other.
613 	 */
614 	if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
615 		return (0);
616 
617 	/* Check whether they overlap */
618 	if (!locks_overlap(caller_fl, sys_fl))
619 		return 0;
620 
621 	return (locks_conflict(caller_fl, sys_fl));
622 }
623 
624 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
625  * checking before calling the locks_conflict().
626  */
627 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
628 {
629 	/* FLOCK locks referring to the same filp do not conflict with
630 	 * each other.
631 	 */
632 	if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
633 		return (0);
634 	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
635 		return 0;
636 
637 	return (locks_conflict(caller_fl, sys_fl));
638 }
639 
640 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
641 {
642 	int result = 0;
643 	DECLARE_WAITQUEUE(wait, current);
644 
645 	__set_current_state(TASK_INTERRUPTIBLE);
646 	add_wait_queue(fl_wait, &wait);
647 	if (timeout == 0)
648 		schedule();
649 	else
650 		result = schedule_timeout(timeout);
651 	if (signal_pending(current))
652 		result = -ERESTARTSYS;
653 	remove_wait_queue(fl_wait, &wait);
654 	__set_current_state(TASK_RUNNING);
655 	return result;
656 }
657 
658 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
659 {
660 	int result;
661 	locks_insert_block(blocker, waiter);
662 	result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
663 	__locks_delete_block(waiter);
664 	return result;
665 }
666 
667 int
668 posix_test_lock(struct file *filp, struct file_lock *fl)
669 {
670 	struct file_lock *cfl;
671 
672 	lock_kernel();
673 	for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
674 		if (!IS_POSIX(cfl))
675 			continue;
676 		if (posix_locks_conflict(cfl, fl))
677 			break;
678 	}
679 	if (cfl) {
680 		__locks_copy_lock(fl, cfl);
681 		unlock_kernel();
682 		return 1;
683 	} else
684 		fl->fl_type = F_UNLCK;
685 	unlock_kernel();
686 	return 0;
687 }
688 
689 EXPORT_SYMBOL(posix_test_lock);
690 
691 /* This function tests for deadlock condition before putting a process to
692  * sleep. The detection scheme is no longer recursive. Recursive was neat,
693  * but dangerous - we risked stack corruption if the lock data was bad, or
694  * if the recursion was too deep for any other reason.
695  *
696  * We rely on the fact that a task can only be on one lock's wait queue
697  * at a time. When we find blocked_task on a wait queue we can re-search
698  * with blocked_task equal to that queue's owner, until either blocked_task
699  * isn't found, or blocked_task is found on a queue owned by my_task.
700  *
701  * Note: the above assumption may not be true when handling lock requests
702  * from a broken NFS client. But broken NFS clients have a lot more to
703  * worry about than proper deadlock detection anyway... --okir
704  */
705 static int posix_locks_deadlock(struct file_lock *caller_fl,
706 				struct file_lock *block_fl)
707 {
708 	struct list_head *tmp;
709 
710 next_task:
711 	if (posix_same_owner(caller_fl, block_fl))
712 		return 1;
713 	list_for_each(tmp, &blocked_list) {
714 		struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
715 		if (posix_same_owner(fl, block_fl)) {
716 			fl = fl->fl_next;
717 			block_fl = fl;
718 			goto next_task;
719 		}
720 	}
721 	return 0;
722 }
723 
724 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
725  * at the head of the list, but that's secret knowledge known only to
726  * flock_lock_file and posix_lock_file.
727  *
728  * Note that if called with an FL_EXISTS argument, the caller may determine
729  * whether or not a lock was successfully freed by testing the return
730  * value for -ENOENT.
731  */
732 static int flock_lock_file(struct file *filp, struct file_lock *request)
733 {
734 	struct file_lock *new_fl = NULL;
735 	struct file_lock **before;
736 	struct inode * inode = filp->f_path.dentry->d_inode;
737 	int error = 0;
738 	int found = 0;
739 
740 	lock_kernel();
741 	if (request->fl_flags & FL_ACCESS)
742 		goto find_conflict;
743 	for_each_lock(inode, before) {
744 		struct file_lock *fl = *before;
745 		if (IS_POSIX(fl))
746 			break;
747 		if (IS_LEASE(fl))
748 			continue;
749 		if (filp != fl->fl_file)
750 			continue;
751 		if (request->fl_type == fl->fl_type)
752 			goto out;
753 		found = 1;
754 		locks_delete_lock(before);
755 		break;
756 	}
757 
758 	if (request->fl_type == F_UNLCK) {
759 		if ((request->fl_flags & FL_EXISTS) && !found)
760 			error = -ENOENT;
761 		goto out;
762 	}
763 
764 	error = -ENOMEM;
765 	new_fl = locks_alloc_lock();
766 	if (new_fl == NULL)
767 		goto out;
768 	/*
769 	 * If a higher-priority process was blocked on the old file lock,
770 	 * give it the opportunity to lock the file.
771 	 */
772 	if (found)
773 		cond_resched();
774 
775 find_conflict:
776 	for_each_lock(inode, before) {
777 		struct file_lock *fl = *before;
778 		if (IS_POSIX(fl))
779 			break;
780 		if (IS_LEASE(fl))
781 			continue;
782 		if (!flock_locks_conflict(request, fl))
783 			continue;
784 		error = -EAGAIN;
785 		if (request->fl_flags & FL_SLEEP)
786 			locks_insert_block(fl, request);
787 		goto out;
788 	}
789 	if (request->fl_flags & FL_ACCESS)
790 		goto out;
791 	locks_copy_lock(new_fl, request);
792 	locks_insert_lock(&inode->i_flock, new_fl);
793 	new_fl = NULL;
794 	error = 0;
795 
796 out:
797 	unlock_kernel();
798 	if (new_fl)
799 		locks_free_lock(new_fl);
800 	return error;
801 }
802 
803 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
804 {
805 	struct file_lock *fl;
806 	struct file_lock *new_fl = NULL;
807 	struct file_lock *new_fl2 = NULL;
808 	struct file_lock *left = NULL;
809 	struct file_lock *right = NULL;
810 	struct file_lock **before;
811 	int error, added = 0;
812 
813 	/*
814 	 * We may need two file_lock structures for this operation,
815 	 * so we get them in advance to avoid races.
816 	 *
817 	 * In some cases we can be sure, that no new locks will be needed
818 	 */
819 	if (!(request->fl_flags & FL_ACCESS) &&
820 	    (request->fl_type != F_UNLCK ||
821 	     request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
822 		new_fl = locks_alloc_lock();
823 		new_fl2 = locks_alloc_lock();
824 	}
825 
826 	lock_kernel();
827 	if (request->fl_type != F_UNLCK) {
828 		for_each_lock(inode, before) {
829 			struct file_lock *fl = *before;
830 			if (!IS_POSIX(fl))
831 				continue;
832 			if (!posix_locks_conflict(request, fl))
833 				continue;
834 			if (conflock)
835 				locks_copy_lock(conflock, fl);
836 			error = -EAGAIN;
837 			if (!(request->fl_flags & FL_SLEEP))
838 				goto out;
839 			error = -EDEADLK;
840 			if (posix_locks_deadlock(request, fl))
841 				goto out;
842 			error = -EAGAIN;
843 			locks_insert_block(fl, request);
844 			goto out;
845   		}
846   	}
847 
848 	/* If we're just looking for a conflict, we're done. */
849 	error = 0;
850 	if (request->fl_flags & FL_ACCESS)
851 		goto out;
852 
853 	/*
854 	 * Find the first old lock with the same owner as the new lock.
855 	 */
856 
857 	before = &inode->i_flock;
858 
859 	/* First skip locks owned by other processes.  */
860 	while ((fl = *before) && (!IS_POSIX(fl) ||
861 				  !posix_same_owner(request, fl))) {
862 		before = &fl->fl_next;
863 	}
864 
865 	/* Process locks with this owner.  */
866 	while ((fl = *before) && posix_same_owner(request, fl)) {
867 		/* Detect adjacent or overlapping regions (if same lock type)
868 		 */
869 		if (request->fl_type == fl->fl_type) {
870 			/* In all comparisons of start vs end, use
871 			 * "start - 1" rather than "end + 1". If end
872 			 * is OFFSET_MAX, end + 1 will become negative.
873 			 */
874 			if (fl->fl_end < request->fl_start - 1)
875 				goto next_lock;
876 			/* If the next lock in the list has entirely bigger
877 			 * addresses than the new one, insert the lock here.
878 			 */
879 			if (fl->fl_start - 1 > request->fl_end)
880 				break;
881 
882 			/* If we come here, the new and old lock are of the
883 			 * same type and adjacent or overlapping. Make one
884 			 * lock yielding from the lower start address of both
885 			 * locks to the higher end address.
886 			 */
887 			if (fl->fl_start > request->fl_start)
888 				fl->fl_start = request->fl_start;
889 			else
890 				request->fl_start = fl->fl_start;
891 			if (fl->fl_end < request->fl_end)
892 				fl->fl_end = request->fl_end;
893 			else
894 				request->fl_end = fl->fl_end;
895 			if (added) {
896 				locks_delete_lock(before);
897 				continue;
898 			}
899 			request = fl;
900 			added = 1;
901 		}
902 		else {
903 			/* Processing for different lock types is a bit
904 			 * more complex.
905 			 */
906 			if (fl->fl_end < request->fl_start)
907 				goto next_lock;
908 			if (fl->fl_start > request->fl_end)
909 				break;
910 			if (request->fl_type == F_UNLCK)
911 				added = 1;
912 			if (fl->fl_start < request->fl_start)
913 				left = fl;
914 			/* If the next lock in the list has a higher end
915 			 * address than the new one, insert the new one here.
916 			 */
917 			if (fl->fl_end > request->fl_end) {
918 				right = fl;
919 				break;
920 			}
921 			if (fl->fl_start >= request->fl_start) {
922 				/* The new lock completely replaces an old
923 				 * one (This may happen several times).
924 				 */
925 				if (added) {
926 					locks_delete_lock(before);
927 					continue;
928 				}
929 				/* Replace the old lock with the new one.
930 				 * Wake up anybody waiting for the old one,
931 				 * as the change in lock type might satisfy
932 				 * their needs.
933 				 */
934 				locks_wake_up_blocks(fl);
935 				fl->fl_start = request->fl_start;
936 				fl->fl_end = request->fl_end;
937 				fl->fl_type = request->fl_type;
938 				locks_release_private(fl);
939 				locks_copy_private(fl, request);
940 				request = fl;
941 				added = 1;
942 			}
943 		}
944 		/* Go on to next lock.
945 		 */
946 	next_lock:
947 		before = &fl->fl_next;
948 	}
949 
950 	/*
951 	 * The above code only modifies existing locks in case of
952 	 * merging or replacing.  If new lock(s) need to be inserted
953 	 * all modifications are done bellow this, so it's safe yet to
954 	 * bail out.
955 	 */
956 	error = -ENOLCK; /* "no luck" */
957 	if (right && left == right && !new_fl2)
958 		goto out;
959 
960 	error = 0;
961 	if (!added) {
962 		if (request->fl_type == F_UNLCK) {
963 			if (request->fl_flags & FL_EXISTS)
964 				error = -ENOENT;
965 			goto out;
966 		}
967 
968 		if (!new_fl) {
969 			error = -ENOLCK;
970 			goto out;
971 		}
972 		locks_copy_lock(new_fl, request);
973 		locks_insert_lock(before, new_fl);
974 		new_fl = NULL;
975 	}
976 	if (right) {
977 		if (left == right) {
978 			/* The new lock breaks the old one in two pieces,
979 			 * so we have to use the second new lock.
980 			 */
981 			left = new_fl2;
982 			new_fl2 = NULL;
983 			locks_copy_lock(left, right);
984 			locks_insert_lock(before, left);
985 		}
986 		right->fl_start = request->fl_end + 1;
987 		locks_wake_up_blocks(right);
988 	}
989 	if (left) {
990 		left->fl_end = request->fl_start - 1;
991 		locks_wake_up_blocks(left);
992 	}
993  out:
994 	unlock_kernel();
995 	/*
996 	 * Free any unused locks.
997 	 */
998 	if (new_fl)
999 		locks_free_lock(new_fl);
1000 	if (new_fl2)
1001 		locks_free_lock(new_fl2);
1002 	return error;
1003 }
1004 
1005 /**
1006  * posix_lock_file - Apply a POSIX-style lock to a file
1007  * @filp: The file to apply the lock to
1008  * @fl: The lock to be applied
1009  * @conflock: Place to return a copy of the conflicting lock, if found.
1010  *
1011  * Add a POSIX style lock to a file.
1012  * We merge adjacent & overlapping locks whenever possible.
1013  * POSIX locks are sorted by owner task, then by starting address
1014  *
1015  * Note that if called with an FL_EXISTS argument, the caller may determine
1016  * whether or not a lock was successfully freed by testing the return
1017  * value for -ENOENT.
1018  */
1019 int posix_lock_file(struct file *filp, struct file_lock *fl,
1020 			struct file_lock *conflock)
1021 {
1022 	return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
1023 }
1024 EXPORT_SYMBOL(posix_lock_file);
1025 
1026 /**
1027  * posix_lock_file_wait - Apply a POSIX-style lock to a file
1028  * @filp: The file to apply the lock to
1029  * @fl: The lock to be applied
1030  *
1031  * Add a POSIX style lock to a file.
1032  * We merge adjacent & overlapping locks whenever possible.
1033  * POSIX locks are sorted by owner task, then by starting address
1034  */
1035 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1036 {
1037 	int error;
1038 	might_sleep ();
1039 	for (;;) {
1040 		error = posix_lock_file(filp, fl, NULL);
1041 		if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1042 			break;
1043 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1044 		if (!error)
1045 			continue;
1046 
1047 		locks_delete_block(fl);
1048 		break;
1049 	}
1050 	return error;
1051 }
1052 EXPORT_SYMBOL(posix_lock_file_wait);
1053 
1054 /**
1055  * locks_mandatory_locked - Check for an active lock
1056  * @inode: the file to check
1057  *
1058  * Searches the inode's list of locks to find any POSIX locks which conflict.
1059  * This function is called from locks_verify_locked() only.
1060  */
1061 int locks_mandatory_locked(struct inode *inode)
1062 {
1063 	fl_owner_t owner = current->files;
1064 	struct file_lock *fl;
1065 
1066 	/*
1067 	 * Search the lock list for this inode for any POSIX locks.
1068 	 */
1069 	lock_kernel();
1070 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1071 		if (!IS_POSIX(fl))
1072 			continue;
1073 		if (fl->fl_owner != owner)
1074 			break;
1075 	}
1076 	unlock_kernel();
1077 	return fl ? -EAGAIN : 0;
1078 }
1079 
1080 /**
1081  * locks_mandatory_area - Check for a conflicting lock
1082  * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1083  *		for shared
1084  * @inode:      the file to check
1085  * @filp:       how the file was opened (if it was)
1086  * @offset:     start of area to check
1087  * @count:      length of area to check
1088  *
1089  * Searches the inode's list of locks to find any POSIX locks which conflict.
1090  * This function is called from rw_verify_area() and
1091  * locks_verify_truncate().
1092  */
1093 int locks_mandatory_area(int read_write, struct inode *inode,
1094 			 struct file *filp, loff_t offset,
1095 			 size_t count)
1096 {
1097 	struct file_lock fl;
1098 	int error;
1099 
1100 	locks_init_lock(&fl);
1101 	fl.fl_owner = current->files;
1102 	fl.fl_pid = current->tgid;
1103 	fl.fl_file = filp;
1104 	fl.fl_flags = FL_POSIX | FL_ACCESS;
1105 	if (filp && !(filp->f_flags & O_NONBLOCK))
1106 		fl.fl_flags |= FL_SLEEP;
1107 	fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1108 	fl.fl_start = offset;
1109 	fl.fl_end = offset + count - 1;
1110 
1111 	for (;;) {
1112 		error = __posix_lock_file(inode, &fl, NULL);
1113 		if (error != -EAGAIN)
1114 			break;
1115 		if (!(fl.fl_flags & FL_SLEEP))
1116 			break;
1117 		error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1118 		if (!error) {
1119 			/*
1120 			 * If we've been sleeping someone might have
1121 			 * changed the permissions behind our back.
1122 			 */
1123 			if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1124 				continue;
1125 		}
1126 
1127 		locks_delete_block(&fl);
1128 		break;
1129 	}
1130 
1131 	return error;
1132 }
1133 
1134 EXPORT_SYMBOL(locks_mandatory_area);
1135 
1136 /* We already had a lease on this file; just change its type */
1137 int lease_modify(struct file_lock **before, int arg)
1138 {
1139 	struct file_lock *fl = *before;
1140 	int error = assign_type(fl, arg);
1141 
1142 	if (error)
1143 		return error;
1144 	locks_wake_up_blocks(fl);
1145 	if (arg == F_UNLCK)
1146 		locks_delete_lock(before);
1147 	return 0;
1148 }
1149 
1150 EXPORT_SYMBOL(lease_modify);
1151 
1152 static void time_out_leases(struct inode *inode)
1153 {
1154 	struct file_lock **before;
1155 	struct file_lock *fl;
1156 
1157 	before = &inode->i_flock;
1158 	while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1159 		if ((fl->fl_break_time == 0)
1160 				|| time_before(jiffies, fl->fl_break_time)) {
1161 			before = &fl->fl_next;
1162 			continue;
1163 		}
1164 		lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1165 		if (fl == *before)	/* lease_modify may have freed fl */
1166 			before = &fl->fl_next;
1167 	}
1168 }
1169 
1170 /**
1171  *	__break_lease	-	revoke all outstanding leases on file
1172  *	@inode: the inode of the file to return
1173  *	@mode: the open mode (read or write)
1174  *
1175  *	break_lease (inlined for speed) has checked there already
1176  *	is a lease on this file.  Leases are broken on a call to open()
1177  *	or truncate().  This function can sleep unless you
1178  *	specified %O_NONBLOCK to your open().
1179  */
1180 int __break_lease(struct inode *inode, unsigned int mode)
1181 {
1182 	int error = 0, future;
1183 	struct file_lock *new_fl, *flock;
1184 	struct file_lock *fl;
1185 	int alloc_err;
1186 	unsigned long break_time;
1187 	int i_have_this_lease = 0;
1188 
1189 	alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1190 			&new_fl);
1191 
1192 	lock_kernel();
1193 
1194 	time_out_leases(inode);
1195 
1196 	flock = inode->i_flock;
1197 	if ((flock == NULL) || !IS_LEASE(flock))
1198 		goto out;
1199 
1200 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1201 		if (fl->fl_owner == current->files)
1202 			i_have_this_lease = 1;
1203 
1204 	if (mode & FMODE_WRITE) {
1205 		/* If we want write access, we have to revoke any lease. */
1206 		future = F_UNLCK | F_INPROGRESS;
1207 	} else if (flock->fl_type & F_INPROGRESS) {
1208 		/* If the lease is already being broken, we just leave it */
1209 		future = flock->fl_type;
1210 	} else if (flock->fl_type & F_WRLCK) {
1211 		/* Downgrade the exclusive lease to a read-only lease. */
1212 		future = F_RDLCK | F_INPROGRESS;
1213 	} else {
1214 		/* the existing lease was read-only, so we can read too. */
1215 		goto out;
1216 	}
1217 
1218 	if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1219 		error = alloc_err;
1220 		goto out;
1221 	}
1222 
1223 	break_time = 0;
1224 	if (lease_break_time > 0) {
1225 		break_time = jiffies + lease_break_time * HZ;
1226 		if (break_time == 0)
1227 			break_time++;	/* so that 0 means no break time */
1228 	}
1229 
1230 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1231 		if (fl->fl_type != future) {
1232 			fl->fl_type = future;
1233 			fl->fl_break_time = break_time;
1234 			/* lease must have lmops break callback */
1235 			fl->fl_lmops->fl_break(fl);
1236 		}
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 	error = locks_block_on_timeout(flock, new_fl, break_time);
1252 	if (error >= 0) {
1253 		if (error == 0)
1254 			time_out_leases(inode);
1255 		/* Wait for the next lease that has not been broken yet */
1256 		for (flock = inode->i_flock; flock && IS_LEASE(flock);
1257 				flock = flock->fl_next) {
1258 			if (flock->fl_type & F_INPROGRESS)
1259 				goto restart;
1260 		}
1261 		error = 0;
1262 	}
1263 
1264 out:
1265 	unlock_kernel();
1266 	if (!alloc_err)
1267 		locks_free_lock(new_fl);
1268 	return error;
1269 }
1270 
1271 EXPORT_SYMBOL(__break_lease);
1272 
1273 /**
1274  *	lease_get_mtime
1275  *	@inode: the inode
1276  *      @time:  pointer to a timespec which will contain the last modified time
1277  *
1278  * This is to force NFS clients to flush their caches for files with
1279  * exclusive leases.  The justification is that if someone has an
1280  * exclusive lease, then they could be modifiying it.
1281  */
1282 void lease_get_mtime(struct inode *inode, struct timespec *time)
1283 {
1284 	struct file_lock *flock = inode->i_flock;
1285 	if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1286 		*time = current_fs_time(inode->i_sb);
1287 	else
1288 		*time = inode->i_mtime;
1289 }
1290 
1291 EXPORT_SYMBOL(lease_get_mtime);
1292 
1293 /**
1294  *	fcntl_getlease - Enquire what lease is currently active
1295  *	@filp: the file
1296  *
1297  *	The value returned by this function will be one of
1298  *	(if no lease break is pending):
1299  *
1300  *	%F_RDLCK to indicate a shared lease is held.
1301  *
1302  *	%F_WRLCK to indicate an exclusive lease is held.
1303  *
1304  *	%F_UNLCK to indicate no lease is held.
1305  *
1306  *	(if a lease break is pending):
1307  *
1308  *	%F_RDLCK to indicate an exclusive lease needs to be
1309  *		changed to a shared lease (or removed).
1310  *
1311  *	%F_UNLCK to indicate the lease needs to be removed.
1312  *
1313  *	XXX: sfr & willy disagree over whether F_INPROGRESS
1314  *	should be returned to userspace.
1315  */
1316 int fcntl_getlease(struct file *filp)
1317 {
1318 	struct file_lock *fl;
1319 	int type = F_UNLCK;
1320 
1321 	lock_kernel();
1322 	time_out_leases(filp->f_path.dentry->d_inode);
1323 	for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1324 			fl = fl->fl_next) {
1325 		if (fl->fl_file == filp) {
1326 			type = fl->fl_type & ~F_INPROGRESS;
1327 			break;
1328 		}
1329 	}
1330 	unlock_kernel();
1331 	return type;
1332 }
1333 
1334 /**
1335  *	__setlease	-	sets a lease on an open file
1336  *	@filp: file pointer
1337  *	@arg: type of lease to obtain
1338  *	@flp: input - file_lock to use, output - file_lock inserted
1339  *
1340  *	The (input) flp->fl_lmops->fl_break function is required
1341  *	by break_lease().
1342  *
1343  *	Called with kernel lock held.
1344  */
1345 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1346 {
1347 	struct file_lock *fl, **before, **my_before = NULL, *lease;
1348 	struct dentry *dentry = filp->f_path.dentry;
1349 	struct inode *inode = dentry->d_inode;
1350 	int error, rdlease_count = 0, wrlease_count = 0;
1351 
1352 	time_out_leases(inode);
1353 
1354 	error = -EINVAL;
1355 	if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1356 		goto out;
1357 
1358 	lease = *flp;
1359 
1360 	error = -EAGAIN;
1361 	if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1362 		goto out;
1363 	if ((arg == F_WRLCK)
1364 	    && ((atomic_read(&dentry->d_count) > 1)
1365 		|| (atomic_read(&inode->i_count) > 1)))
1366 		goto out;
1367 
1368 	/*
1369 	 * At this point, we know that if there is an exclusive
1370 	 * lease on this file, then we hold it on this filp
1371 	 * (otherwise our open of this file would have blocked).
1372 	 * And if we are trying to acquire an exclusive lease,
1373 	 * then the file is not open by anyone (including us)
1374 	 * except for this filp.
1375 	 */
1376 	for (before = &inode->i_flock;
1377 			((fl = *before) != NULL) && IS_LEASE(fl);
1378 			before = &fl->fl_next) {
1379 		if (lease->fl_lmops->fl_mylease(fl, lease))
1380 			my_before = before;
1381 		else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1382 			/*
1383 			 * Someone is in the process of opening this
1384 			 * file for writing so we may not take an
1385 			 * exclusive lease on it.
1386 			 */
1387 			wrlease_count++;
1388 		else
1389 			rdlease_count++;
1390 	}
1391 
1392 	if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1393 	    (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1394 		goto out;
1395 
1396 	if (my_before != NULL) {
1397 		*flp = *my_before;
1398 		error = lease->fl_lmops->fl_change(my_before, arg);
1399 		goto out;
1400 	}
1401 
1402 	error = 0;
1403 	if (arg == F_UNLCK)
1404 		goto out;
1405 
1406 	error = -EINVAL;
1407 	if (!leases_enable)
1408 		goto out;
1409 
1410 	error = -ENOMEM;
1411 	fl = locks_alloc_lock();
1412 	if (fl == NULL)
1413 		goto out;
1414 
1415 	locks_copy_lock(fl, lease);
1416 
1417 	locks_insert_lock(before, fl);
1418 
1419 	*flp = fl;
1420 	error = 0;
1421 out:
1422 	return error;
1423 }
1424 
1425  /**
1426  *	setlease        -       sets a lease on an open file
1427  *	@filp: file pointer
1428  *	@arg: type of lease to obtain
1429  *	@lease: file_lock to use
1430  *
1431  *	Call this to establish a lease on the file.
1432  *	The fl_lmops fl_break function is required by break_lease
1433  */
1434 
1435 int setlease(struct file *filp, long arg, struct file_lock **lease)
1436 {
1437 	struct dentry *dentry = filp->f_path.dentry;
1438 	struct inode *inode = dentry->d_inode;
1439 	int error;
1440 
1441 	if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1442 		return -EACCES;
1443 	if (!S_ISREG(inode->i_mode))
1444 		return -EINVAL;
1445 	error = security_file_lock(filp, arg);
1446 	if (error)
1447 		return error;
1448 
1449 	lock_kernel();
1450 	error = __setlease(filp, arg, lease);
1451 	unlock_kernel();
1452 
1453 	return error;
1454 }
1455 
1456 EXPORT_SYMBOL(setlease);
1457 
1458 /**
1459  *	fcntl_setlease	-	sets a lease on an open file
1460  *	@fd: open file descriptor
1461  *	@filp: file pointer
1462  *	@arg: type of lease to obtain
1463  *
1464  *	Call this fcntl to establish a lease on the file.
1465  *	Note that you also need to call %F_SETSIG to
1466  *	receive a signal when the lease is broken.
1467  */
1468 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1469 {
1470 	struct file_lock fl, *flp = &fl;
1471 	struct dentry *dentry = filp->f_path.dentry;
1472 	struct inode *inode = dentry->d_inode;
1473 	int error;
1474 
1475 	if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1476 		return -EACCES;
1477 	if (!S_ISREG(inode->i_mode))
1478 		return -EINVAL;
1479 	error = security_file_lock(filp, arg);
1480 	if (error)
1481 		return error;
1482 
1483 	locks_init_lock(&fl);
1484 	error = lease_init(filp, arg, &fl);
1485 	if (error)
1486 		return error;
1487 
1488 	lock_kernel();
1489 
1490 	error = __setlease(filp, arg, &flp);
1491 	if (error || arg == F_UNLCK)
1492 		goto out_unlock;
1493 
1494 	error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1495 	if (error < 0) {
1496 		/* remove lease just inserted by __setlease */
1497 		flp->fl_type = F_UNLCK | F_INPROGRESS;
1498 		flp->fl_break_time = jiffies- 10;
1499 		time_out_leases(inode);
1500 		goto out_unlock;
1501 	}
1502 
1503 	error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1504 out_unlock:
1505 	unlock_kernel();
1506 	return error;
1507 }
1508 
1509 /**
1510  * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1511  * @filp: The file to apply the lock to
1512  * @fl: The lock to be applied
1513  *
1514  * Add a FLOCK style lock to a file.
1515  */
1516 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1517 {
1518 	int error;
1519 	might_sleep();
1520 	for (;;) {
1521 		error = flock_lock_file(filp, fl);
1522 		if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1523 			break;
1524 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1525 		if (!error)
1526 			continue;
1527 
1528 		locks_delete_block(fl);
1529 		break;
1530 	}
1531 	return error;
1532 }
1533 
1534 EXPORT_SYMBOL(flock_lock_file_wait);
1535 
1536 /**
1537  *	sys_flock: - flock() system call.
1538  *	@fd: the file descriptor to lock.
1539  *	@cmd: the type of lock to apply.
1540  *
1541  *	Apply a %FL_FLOCK style lock to an open file descriptor.
1542  *	The @cmd can be one of
1543  *
1544  *	%LOCK_SH -- a shared lock.
1545  *
1546  *	%LOCK_EX -- an exclusive lock.
1547  *
1548  *	%LOCK_UN -- remove an existing lock.
1549  *
1550  *	%LOCK_MAND -- a `mandatory' flock.  This exists to emulate Windows Share Modes.
1551  *
1552  *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1553  *	processes read and write access respectively.
1554  */
1555 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1556 {
1557 	struct file *filp;
1558 	struct file_lock *lock;
1559 	int can_sleep, unlock;
1560 	int error;
1561 
1562 	error = -EBADF;
1563 	filp = fget(fd);
1564 	if (!filp)
1565 		goto out;
1566 
1567 	can_sleep = !(cmd & LOCK_NB);
1568 	cmd &= ~LOCK_NB;
1569 	unlock = (cmd == LOCK_UN);
1570 
1571 	if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1572 		goto out_putf;
1573 
1574 	error = flock_make_lock(filp, &lock, cmd);
1575 	if (error)
1576 		goto out_putf;
1577 	if (can_sleep)
1578 		lock->fl_flags |= FL_SLEEP;
1579 
1580 	error = security_file_lock(filp, cmd);
1581 	if (error)
1582 		goto out_free;
1583 
1584 	if (filp->f_op && filp->f_op->flock)
1585 		error = filp->f_op->flock(filp,
1586 					  (can_sleep) ? F_SETLKW : F_SETLK,
1587 					  lock);
1588 	else
1589 		error = flock_lock_file_wait(filp, lock);
1590 
1591  out_free:
1592 	locks_free_lock(lock);
1593 
1594  out_putf:
1595 	fput(filp);
1596  out:
1597 	return error;
1598 }
1599 
1600 /**
1601  * vfs_test_lock - test file byte range lock
1602  * @filp: The file to test lock for
1603  * @fl: The lock to test
1604  * @conf: Place to return a copy of the conflicting lock, if found
1605  *
1606  * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
1607  * setting conf->fl_type to something other than F_UNLCK.
1608  */
1609 int vfs_test_lock(struct file *filp, struct file_lock *fl)
1610 {
1611 	if (filp->f_op && filp->f_op->lock)
1612 		return filp->f_op->lock(filp, F_GETLK, fl);
1613 	posix_test_lock(filp, fl);
1614 	return 0;
1615 }
1616 EXPORT_SYMBOL_GPL(vfs_test_lock);
1617 
1618 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1619 {
1620 	flock->l_pid = fl->fl_pid;
1621 #if BITS_PER_LONG == 32
1622 	/*
1623 	 * Make sure we can represent the posix lock via
1624 	 * legacy 32bit flock.
1625 	 */
1626 	if (fl->fl_start > OFFT_OFFSET_MAX)
1627 		return -EOVERFLOW;
1628 	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1629 		return -EOVERFLOW;
1630 #endif
1631 	flock->l_start = fl->fl_start;
1632 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1633 		fl->fl_end - fl->fl_start + 1;
1634 	flock->l_whence = 0;
1635 	flock->l_type = fl->fl_type;
1636 	return 0;
1637 }
1638 
1639 #if BITS_PER_LONG == 32
1640 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1641 {
1642 	flock->l_pid = fl->fl_pid;
1643 	flock->l_start = fl->fl_start;
1644 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1645 		fl->fl_end - fl->fl_start + 1;
1646 	flock->l_whence = 0;
1647 	flock->l_type = fl->fl_type;
1648 }
1649 #endif
1650 
1651 /* Report the first existing lock that would conflict with l.
1652  * This implements the F_GETLK command of fcntl().
1653  */
1654 int fcntl_getlk(struct file *filp, struct flock __user *l)
1655 {
1656 	struct file_lock file_lock;
1657 	struct flock flock;
1658 	int error;
1659 
1660 	error = -EFAULT;
1661 	if (copy_from_user(&flock, l, sizeof(flock)))
1662 		goto out;
1663 	error = -EINVAL;
1664 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1665 		goto out;
1666 
1667 	error = flock_to_posix_lock(filp, &file_lock, &flock);
1668 	if (error)
1669 		goto out;
1670 
1671 	error = vfs_test_lock(filp, &file_lock);
1672 	if (error)
1673 		goto out;
1674 
1675 	flock.l_type = file_lock.fl_type;
1676 	if (file_lock.fl_type != F_UNLCK) {
1677 		error = posix_lock_to_flock(&flock, &file_lock);
1678 		if (error)
1679 			goto out;
1680 	}
1681 	error = -EFAULT;
1682 	if (!copy_to_user(l, &flock, sizeof(flock)))
1683 		error = 0;
1684 out:
1685 	return error;
1686 }
1687 
1688 /**
1689  * vfs_lock_file - file byte range lock
1690  * @filp: The file to apply the lock to
1691  * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1692  * @fl: The lock to be applied
1693  * @conf: Place to return a copy of the conflicting lock, if found.
1694  *
1695  * A caller that doesn't care about the conflicting lock may pass NULL
1696  * as the final argument.
1697  *
1698  * If the filesystem defines a private ->lock() method, then @conf will
1699  * be left unchanged; so a caller that cares should initialize it to
1700  * some acceptable default.
1701  *
1702  * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1703  * locks, the ->lock() interface may return asynchronously, before the lock has
1704  * been granted or denied by the underlying filesystem, if (and only if)
1705  * fl_grant is set. Callers expecting ->lock() to return asynchronously
1706  * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1707  * the request is for a blocking lock. When ->lock() does return asynchronously,
1708  * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1709  * request completes.
1710  * If the request is for non-blocking lock the file system should return
1711  * -EINPROGRESS then try to get the lock and call the callback routine with
1712  * the result. If the request timed out the callback routine will return a
1713  * nonzero return code and the file system should release the lock. The file
1714  * system is also responsible to keep a corresponding posix lock when it
1715  * grants a lock so the VFS can find out which locks are locally held and do
1716  * the correct lock cleanup when required.
1717  * The underlying filesystem must not drop the kernel lock or call
1718  * ->fl_grant() before returning to the caller with a -EINPROGRESS
1719  * return code.
1720  */
1721 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
1722 {
1723 	if (filp->f_op && filp->f_op->lock)
1724 		return filp->f_op->lock(filp, cmd, fl);
1725 	else
1726 		return posix_lock_file(filp, fl, conf);
1727 }
1728 EXPORT_SYMBOL_GPL(vfs_lock_file);
1729 
1730 /* Apply the lock described by l to an open file descriptor.
1731  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1732  */
1733 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1734 		struct flock __user *l)
1735 {
1736 	struct file_lock *file_lock = locks_alloc_lock();
1737 	struct flock flock;
1738 	struct inode *inode;
1739 	int error;
1740 
1741 	if (file_lock == NULL)
1742 		return -ENOLCK;
1743 
1744 	/*
1745 	 * This might block, so we do it before checking the inode.
1746 	 */
1747 	error = -EFAULT;
1748 	if (copy_from_user(&flock, l, sizeof(flock)))
1749 		goto out;
1750 
1751 	inode = filp->f_path.dentry->d_inode;
1752 
1753 	/* Don't allow mandatory locks on files that may be memory mapped
1754 	 * and shared.
1755 	 */
1756 	if (IS_MANDLOCK(inode) &&
1757 	    (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1758 	    mapping_writably_mapped(filp->f_mapping)) {
1759 		error = -EAGAIN;
1760 		goto out;
1761 	}
1762 
1763 again:
1764 	error = flock_to_posix_lock(filp, file_lock, &flock);
1765 	if (error)
1766 		goto out;
1767 	if (cmd == F_SETLKW) {
1768 		file_lock->fl_flags |= FL_SLEEP;
1769 	}
1770 
1771 	error = -EBADF;
1772 	switch (flock.l_type) {
1773 	case F_RDLCK:
1774 		if (!(filp->f_mode & FMODE_READ))
1775 			goto out;
1776 		break;
1777 	case F_WRLCK:
1778 		if (!(filp->f_mode & FMODE_WRITE))
1779 			goto out;
1780 		break;
1781 	case F_UNLCK:
1782 		break;
1783 	default:
1784 		error = -EINVAL;
1785 		goto out;
1786 	}
1787 
1788 	error = security_file_lock(filp, file_lock->fl_type);
1789 	if (error)
1790 		goto out;
1791 
1792 	for (;;) {
1793 		error = vfs_lock_file(filp, cmd, file_lock, NULL);
1794 		if (error != -EAGAIN || cmd == F_SETLK)
1795 			break;
1796 		error = wait_event_interruptible(file_lock->fl_wait,
1797 				!file_lock->fl_next);
1798 		if (!error)
1799 			continue;
1800 
1801 		locks_delete_block(file_lock);
1802 		break;
1803 	}
1804 
1805 	/*
1806 	 * Attempt to detect a close/fcntl race and recover by
1807 	 * releasing the lock that was just acquired.
1808 	 */
1809 	if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1810 		flock.l_type = F_UNLCK;
1811 		goto again;
1812 	}
1813 
1814 out:
1815 	locks_free_lock(file_lock);
1816 	return error;
1817 }
1818 
1819 #if BITS_PER_LONG == 32
1820 /* Report the first existing lock that would conflict with l.
1821  * This implements the F_GETLK command of fcntl().
1822  */
1823 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1824 {
1825 	struct file_lock file_lock;
1826 	struct flock64 flock;
1827 	int error;
1828 
1829 	error = -EFAULT;
1830 	if (copy_from_user(&flock, l, sizeof(flock)))
1831 		goto out;
1832 	error = -EINVAL;
1833 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1834 		goto out;
1835 
1836 	error = flock64_to_posix_lock(filp, &file_lock, &flock);
1837 	if (error)
1838 		goto out;
1839 
1840 	error = vfs_test_lock(filp, &file_lock);
1841 	if (error)
1842 		goto out;
1843 
1844 	flock.l_type = file_lock.fl_type;
1845 	if (file_lock.fl_type != F_UNLCK)
1846 		posix_lock_to_flock64(&flock, &file_lock);
1847 
1848 	error = -EFAULT;
1849 	if (!copy_to_user(l, &flock, sizeof(flock)))
1850 		error = 0;
1851 
1852 out:
1853 	return error;
1854 }
1855 
1856 /* Apply the lock described by l to an open file descriptor.
1857  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1858  */
1859 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1860 		struct flock64 __user *l)
1861 {
1862 	struct file_lock *file_lock = locks_alloc_lock();
1863 	struct flock64 flock;
1864 	struct inode *inode;
1865 	int error;
1866 
1867 	if (file_lock == NULL)
1868 		return -ENOLCK;
1869 
1870 	/*
1871 	 * This might block, so we do it before checking the inode.
1872 	 */
1873 	error = -EFAULT;
1874 	if (copy_from_user(&flock, l, sizeof(flock)))
1875 		goto out;
1876 
1877 	inode = filp->f_path.dentry->d_inode;
1878 
1879 	/* Don't allow mandatory locks on files that may be memory mapped
1880 	 * and shared.
1881 	 */
1882 	if (IS_MANDLOCK(inode) &&
1883 	    (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1884 	    mapping_writably_mapped(filp->f_mapping)) {
1885 		error = -EAGAIN;
1886 		goto out;
1887 	}
1888 
1889 again:
1890 	error = flock64_to_posix_lock(filp, file_lock, &flock);
1891 	if (error)
1892 		goto out;
1893 	if (cmd == F_SETLKW64) {
1894 		file_lock->fl_flags |= FL_SLEEP;
1895 	}
1896 
1897 	error = -EBADF;
1898 	switch (flock.l_type) {
1899 	case F_RDLCK:
1900 		if (!(filp->f_mode & FMODE_READ))
1901 			goto out;
1902 		break;
1903 	case F_WRLCK:
1904 		if (!(filp->f_mode & FMODE_WRITE))
1905 			goto out;
1906 		break;
1907 	case F_UNLCK:
1908 		break;
1909 	default:
1910 		error = -EINVAL;
1911 		goto out;
1912 	}
1913 
1914 	error = security_file_lock(filp, file_lock->fl_type);
1915 	if (error)
1916 		goto out;
1917 
1918 	for (;;) {
1919 		error = vfs_lock_file(filp, cmd, file_lock, NULL);
1920 		if (error != -EAGAIN || cmd == F_SETLK64)
1921 			break;
1922 		error = wait_event_interruptible(file_lock->fl_wait,
1923 				!file_lock->fl_next);
1924 		if (!error)
1925 			continue;
1926 
1927 		locks_delete_block(file_lock);
1928 		break;
1929 	}
1930 
1931 	/*
1932 	 * Attempt to detect a close/fcntl race and recover by
1933 	 * releasing the lock that was just acquired.
1934 	 */
1935 	if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1936 		flock.l_type = F_UNLCK;
1937 		goto again;
1938 	}
1939 
1940 out:
1941 	locks_free_lock(file_lock);
1942 	return error;
1943 }
1944 #endif /* BITS_PER_LONG == 32 */
1945 
1946 /*
1947  * This function is called when the file is being removed
1948  * from the task's fd array.  POSIX locks belonging to this task
1949  * are deleted at this time.
1950  */
1951 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1952 {
1953 	struct file_lock lock;
1954 
1955 	/*
1956 	 * If there are no locks held on this file, we don't need to call
1957 	 * posix_lock_file().  Another process could be setting a lock on this
1958 	 * file at the same time, but we wouldn't remove that lock anyway.
1959 	 */
1960 	if (!filp->f_path.dentry->d_inode->i_flock)
1961 		return;
1962 
1963 	lock.fl_type = F_UNLCK;
1964 	lock.fl_flags = FL_POSIX | FL_CLOSE;
1965 	lock.fl_start = 0;
1966 	lock.fl_end = OFFSET_MAX;
1967 	lock.fl_owner = owner;
1968 	lock.fl_pid = current->tgid;
1969 	lock.fl_file = filp;
1970 	lock.fl_ops = NULL;
1971 	lock.fl_lmops = NULL;
1972 
1973 	vfs_lock_file(filp, F_SETLK, &lock, NULL);
1974 
1975 	if (lock.fl_ops && lock.fl_ops->fl_release_private)
1976 		lock.fl_ops->fl_release_private(&lock);
1977 }
1978 
1979 EXPORT_SYMBOL(locks_remove_posix);
1980 
1981 /*
1982  * This function is called on the last close of an open file.
1983  */
1984 void locks_remove_flock(struct file *filp)
1985 {
1986 	struct inode * inode = filp->f_path.dentry->d_inode;
1987 	struct file_lock *fl;
1988 	struct file_lock **before;
1989 
1990 	if (!inode->i_flock)
1991 		return;
1992 
1993 	if (filp->f_op && filp->f_op->flock) {
1994 		struct file_lock fl = {
1995 			.fl_pid = current->tgid,
1996 			.fl_file = filp,
1997 			.fl_flags = FL_FLOCK,
1998 			.fl_type = F_UNLCK,
1999 			.fl_end = OFFSET_MAX,
2000 		};
2001 		filp->f_op->flock(filp, F_SETLKW, &fl);
2002 		if (fl.fl_ops && fl.fl_ops->fl_release_private)
2003 			fl.fl_ops->fl_release_private(&fl);
2004 	}
2005 
2006 	lock_kernel();
2007 	before = &inode->i_flock;
2008 
2009 	while ((fl = *before) != NULL) {
2010 		if (fl->fl_file == filp) {
2011 			if (IS_FLOCK(fl)) {
2012 				locks_delete_lock(before);
2013 				continue;
2014 			}
2015 			if (IS_LEASE(fl)) {
2016 				lease_modify(before, F_UNLCK);
2017 				continue;
2018 			}
2019 			/* What? */
2020 			BUG();
2021  		}
2022 		before = &fl->fl_next;
2023 	}
2024 	unlock_kernel();
2025 }
2026 
2027 /**
2028  *	posix_unblock_lock - stop waiting for a file lock
2029  *      @filp:   how the file was opened
2030  *	@waiter: the lock which was waiting
2031  *
2032  *	lockd needs to block waiting for locks.
2033  */
2034 int
2035 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2036 {
2037 	int status = 0;
2038 
2039 	lock_kernel();
2040 	if (waiter->fl_next)
2041 		__locks_delete_block(waiter);
2042 	else
2043 		status = -ENOENT;
2044 	unlock_kernel();
2045 	return status;
2046 }
2047 
2048 EXPORT_SYMBOL(posix_unblock_lock);
2049 
2050 /**
2051  * vfs_cancel_lock - file byte range unblock lock
2052  * @filp: The file to apply the unblock to
2053  * @fl: The lock to be unblocked
2054  *
2055  * Used by lock managers to cancel blocked requests
2056  */
2057 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2058 {
2059 	if (filp->f_op && filp->f_op->lock)
2060 		return filp->f_op->lock(filp, F_CANCELLK, fl);
2061 	return 0;
2062 }
2063 
2064 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2065 
2066 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2067 {
2068 	struct inode *inode = NULL;
2069 
2070 	if (fl->fl_file != NULL)
2071 		inode = fl->fl_file->f_path.dentry->d_inode;
2072 
2073 	out += sprintf(out, "%d:%s ", id, pfx);
2074 	if (IS_POSIX(fl)) {
2075 		out += sprintf(out, "%6s %s ",
2076 			     (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2077 			     (inode == NULL) ? "*NOINODE*" :
2078 			     (IS_MANDLOCK(inode) &&
2079 			      (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2080 			     "MANDATORY" : "ADVISORY ");
2081 	} else if (IS_FLOCK(fl)) {
2082 		if (fl->fl_type & LOCK_MAND) {
2083 			out += sprintf(out, "FLOCK  MSNFS     ");
2084 		} else {
2085 			out += sprintf(out, "FLOCK  ADVISORY  ");
2086 		}
2087 	} else if (IS_LEASE(fl)) {
2088 		out += sprintf(out, "LEASE  ");
2089 		if (fl->fl_type & F_INPROGRESS)
2090 			out += sprintf(out, "BREAKING  ");
2091 		else if (fl->fl_file)
2092 			out += sprintf(out, "ACTIVE    ");
2093 		else
2094 			out += sprintf(out, "BREAKER   ");
2095 	} else {
2096 		out += sprintf(out, "UNKNOWN UNKNOWN  ");
2097 	}
2098 	if (fl->fl_type & LOCK_MAND) {
2099 		out += sprintf(out, "%s ",
2100 			       (fl->fl_type & LOCK_READ)
2101 			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2102 			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2103 	} else {
2104 		out += sprintf(out, "%s ",
2105 			       (fl->fl_type & F_INPROGRESS)
2106 			       ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2107 			       : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2108 	}
2109 	if (inode) {
2110 #ifdef WE_CAN_BREAK_LSLK_NOW
2111 		out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2112 				inode->i_sb->s_id, inode->i_ino);
2113 #else
2114 		/* userspace relies on this representation of dev_t ;-( */
2115 		out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2116 				MAJOR(inode->i_sb->s_dev),
2117 				MINOR(inode->i_sb->s_dev), inode->i_ino);
2118 #endif
2119 	} else {
2120 		out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2121 	}
2122 	if (IS_POSIX(fl)) {
2123 		if (fl->fl_end == OFFSET_MAX)
2124 			out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2125 		else
2126 			out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2127 					fl->fl_end);
2128 	} else {
2129 		out += sprintf(out, "0 EOF\n");
2130 	}
2131 }
2132 
2133 static void move_lock_status(char **p, off_t* pos, off_t offset)
2134 {
2135 	int len;
2136 	len = strlen(*p);
2137 	if(*pos >= offset) {
2138 		/* the complete line is valid */
2139 		*p += len;
2140 		*pos += len;
2141 		return;
2142 	}
2143 	if(*pos+len > offset) {
2144 		/* use the second part of the line */
2145 		int i = offset-*pos;
2146 		memmove(*p,*p+i,len-i);
2147 		*p += len-i;
2148 		*pos += len;
2149 		return;
2150 	}
2151 	/* discard the complete line */
2152 	*pos += len;
2153 }
2154 
2155 /**
2156  *	get_locks_status	-	reports lock usage in /proc/locks
2157  *	@buffer: address in userspace to write into
2158  *	@start: ?
2159  *	@offset: how far we are through the buffer
2160  *	@length: how much to read
2161  */
2162 
2163 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2164 {
2165 	struct list_head *tmp;
2166 	char *q = buffer;
2167 	off_t pos = 0;
2168 	int i = 0;
2169 
2170 	lock_kernel();
2171 	list_for_each(tmp, &file_lock_list) {
2172 		struct list_head *btmp;
2173 		struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2174 		lock_get_status(q, fl, ++i, "");
2175 		move_lock_status(&q, &pos, offset);
2176 
2177 		if(pos >= offset+length)
2178 			goto done;
2179 
2180 		list_for_each(btmp, &fl->fl_block) {
2181 			struct file_lock *bfl = list_entry(btmp,
2182 					struct file_lock, fl_block);
2183 			lock_get_status(q, bfl, i, " ->");
2184 			move_lock_status(&q, &pos, offset);
2185 
2186 			if(pos >= offset+length)
2187 				goto done;
2188 		}
2189 	}
2190 done:
2191 	unlock_kernel();
2192 	*start = buffer;
2193 	if(q-buffer < length)
2194 		return (q-buffer);
2195 	return length;
2196 }
2197 
2198 /**
2199  *	lock_may_read - checks that the region is free of locks
2200  *	@inode: the inode that is being read
2201  *	@start: the first byte to read
2202  *	@len: the number of bytes to read
2203  *
2204  *	Emulates Windows locking requirements.  Whole-file
2205  *	mandatory locks (share modes) can prohibit a read and
2206  *	byte-range POSIX locks can prohibit a read if they overlap.
2207  *
2208  *	N.B. this function is only ever called
2209  *	from knfsd and ownership of locks is never checked.
2210  */
2211 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2212 {
2213 	struct file_lock *fl;
2214 	int result = 1;
2215 	lock_kernel();
2216 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2217 		if (IS_POSIX(fl)) {
2218 			if (fl->fl_type == F_RDLCK)
2219 				continue;
2220 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2221 				continue;
2222 		} else if (IS_FLOCK(fl)) {
2223 			if (!(fl->fl_type & LOCK_MAND))
2224 				continue;
2225 			if (fl->fl_type & LOCK_READ)
2226 				continue;
2227 		} else
2228 			continue;
2229 		result = 0;
2230 		break;
2231 	}
2232 	unlock_kernel();
2233 	return result;
2234 }
2235 
2236 EXPORT_SYMBOL(lock_may_read);
2237 
2238 /**
2239  *	lock_may_write - checks that the region is free of locks
2240  *	@inode: the inode that is being written
2241  *	@start: the first byte to write
2242  *	@len: the number of bytes to write
2243  *
2244  *	Emulates Windows locking requirements.  Whole-file
2245  *	mandatory locks (share modes) can prohibit a write and
2246  *	byte-range POSIX locks can prohibit a write if they overlap.
2247  *
2248  *	N.B. this function is only ever called
2249  *	from knfsd and ownership of locks is never checked.
2250  */
2251 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2252 {
2253 	struct file_lock *fl;
2254 	int result = 1;
2255 	lock_kernel();
2256 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2257 		if (IS_POSIX(fl)) {
2258 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2259 				continue;
2260 		} else if (IS_FLOCK(fl)) {
2261 			if (!(fl->fl_type & LOCK_MAND))
2262 				continue;
2263 			if (fl->fl_type & LOCK_WRITE)
2264 				continue;
2265 		} else
2266 			continue;
2267 		result = 0;
2268 		break;
2269 	}
2270 	unlock_kernel();
2271 	return result;
2272 }
2273 
2274 EXPORT_SYMBOL(lock_may_write);
2275 
2276 static int __init filelock_init(void)
2277 {
2278 	filelock_cache = kmem_cache_create("file_lock_cache",
2279 			sizeof(struct file_lock), 0, SLAB_PANIC,
2280 			init_once, NULL);
2281 	return 0;
2282 }
2283 
2284 core_initcall(filelock_init);
2285