xref: /openbmc/linux/fs/locks.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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 LIST_HEAD(file_lock_list);
143 
144 EXPORT_SYMBOL(file_lock_list);
145 
146 static LIST_HEAD(blocked_list);
147 
148 static kmem_cache_t *filelock_cache;
149 
150 /* Allocate an empty lock structure. */
151 static struct file_lock *locks_alloc_lock(void)
152 {
153 	return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
154 }
155 
156 /* Free a lock which is not in use. */
157 static inline void locks_free_lock(struct file_lock *fl)
158 {
159 	if (fl == NULL) {
160 		BUG();
161 		return;
162 	}
163 	if (waitqueue_active(&fl->fl_wait))
164 		panic("Attempting to free lock with active wait queue");
165 
166 	if (!list_empty(&fl->fl_block))
167 		panic("Attempting to free lock with active block list");
168 
169 	if (!list_empty(&fl->fl_link))
170 		panic("Attempting to free lock on active lock list");
171 
172 	if (fl->fl_ops) {
173 		if (fl->fl_ops->fl_release_private)
174 			fl->fl_ops->fl_release_private(fl);
175 		fl->fl_ops = NULL;
176 	}
177 
178 	if (fl->fl_lmops) {
179 		if (fl->fl_lmops->fl_release_private)
180 			fl->fl_lmops->fl_release_private(fl);
181 		fl->fl_lmops = NULL;
182 	}
183 
184 	kmem_cache_free(filelock_cache, fl);
185 }
186 
187 void locks_init_lock(struct file_lock *fl)
188 {
189 	INIT_LIST_HEAD(&fl->fl_link);
190 	INIT_LIST_HEAD(&fl->fl_block);
191 	init_waitqueue_head(&fl->fl_wait);
192 	fl->fl_next = NULL;
193 	fl->fl_fasync = NULL;
194 	fl->fl_owner = NULL;
195 	fl->fl_pid = 0;
196 	fl->fl_file = NULL;
197 	fl->fl_flags = 0;
198 	fl->fl_type = 0;
199 	fl->fl_start = fl->fl_end = 0;
200 	fl->fl_ops = NULL;
201 	fl->fl_lmops = NULL;
202 }
203 
204 EXPORT_SYMBOL(locks_init_lock);
205 
206 /*
207  * Initialises the fields of the file lock which are invariant for
208  * free file_locks.
209  */
210 static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
211 {
212 	struct file_lock *lock = (struct file_lock *) foo;
213 
214 	if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
215 					SLAB_CTOR_CONSTRUCTOR)
216 		return;
217 
218 	locks_init_lock(lock);
219 }
220 
221 /*
222  * Initialize a new lock from an existing file_lock structure.
223  */
224 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
225 {
226 	new->fl_owner = fl->fl_owner;
227 	new->fl_pid = fl->fl_pid;
228 	new->fl_file = fl->fl_file;
229 	new->fl_flags = fl->fl_flags;
230 	new->fl_type = fl->fl_type;
231 	new->fl_start = fl->fl_start;
232 	new->fl_end = fl->fl_end;
233 	new->fl_ops = fl->fl_ops;
234 	new->fl_lmops = fl->fl_lmops;
235 	if (fl->fl_ops && fl->fl_ops->fl_copy_lock)
236 		fl->fl_ops->fl_copy_lock(new, fl);
237 	if (fl->fl_lmops && fl->fl_lmops->fl_copy_lock)
238 		fl->fl_lmops->fl_copy_lock(new, fl);
239 }
240 
241 EXPORT_SYMBOL(locks_copy_lock);
242 
243 static inline int flock_translate_cmd(int cmd) {
244 	if (cmd & LOCK_MAND)
245 		return cmd & (LOCK_MAND | LOCK_RW);
246 	switch (cmd) {
247 	case LOCK_SH:
248 		return F_RDLCK;
249 	case LOCK_EX:
250 		return F_WRLCK;
251 	case LOCK_UN:
252 		return F_UNLCK;
253 	}
254 	return -EINVAL;
255 }
256 
257 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
258 static int flock_make_lock(struct file *filp, struct file_lock **lock,
259 		unsigned int cmd)
260 {
261 	struct file_lock *fl;
262 	int type = flock_translate_cmd(cmd);
263 	if (type < 0)
264 		return type;
265 
266 	fl = locks_alloc_lock();
267 	if (fl == NULL)
268 		return -ENOMEM;
269 
270 	fl->fl_file = filp;
271 	fl->fl_pid = current->tgid;
272 	fl->fl_flags = FL_FLOCK;
273 	fl->fl_type = type;
274 	fl->fl_end = OFFSET_MAX;
275 
276 	*lock = fl;
277 	return 0;
278 }
279 
280 static int assign_type(struct file_lock *fl, int type)
281 {
282 	switch (type) {
283 	case F_RDLCK:
284 	case F_WRLCK:
285 	case F_UNLCK:
286 		fl->fl_type = type;
287 		break;
288 	default:
289 		return -EINVAL;
290 	}
291 	return 0;
292 }
293 
294 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
295  * style lock.
296  */
297 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
298 			       struct flock *l)
299 {
300 	off_t start, end;
301 
302 	switch (l->l_whence) {
303 	case 0: /*SEEK_SET*/
304 		start = 0;
305 		break;
306 	case 1: /*SEEK_CUR*/
307 		start = filp->f_pos;
308 		break;
309 	case 2: /*SEEK_END*/
310 		start = i_size_read(filp->f_dentry->d_inode);
311 		break;
312 	default:
313 		return -EINVAL;
314 	}
315 
316 	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
317 	   POSIX-2001 defines it. */
318 	start += l->l_start;
319 	if (start < 0)
320 		return -EINVAL;
321 	fl->fl_end = OFFSET_MAX;
322 	if (l->l_len > 0) {
323 		end = start + l->l_len - 1;
324 		fl->fl_end = end;
325 	} else if (l->l_len < 0) {
326 		end = start - 1;
327 		fl->fl_end = end;
328 		start += l->l_len;
329 		if (start < 0)
330 			return -EINVAL;
331 	}
332 	fl->fl_start = start;	/* we record the absolute position */
333 	if (fl->fl_end < fl->fl_start)
334 		return -EOVERFLOW;
335 
336 	fl->fl_owner = current->files;
337 	fl->fl_pid = current->tgid;
338 	fl->fl_file = filp;
339 	fl->fl_flags = FL_POSIX;
340 	fl->fl_ops = NULL;
341 	fl->fl_lmops = NULL;
342 
343 	return assign_type(fl, l->l_type);
344 }
345 
346 #if BITS_PER_LONG == 32
347 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
348 				 struct flock64 *l)
349 {
350 	loff_t start;
351 
352 	switch (l->l_whence) {
353 	case 0: /*SEEK_SET*/
354 		start = 0;
355 		break;
356 	case 1: /*SEEK_CUR*/
357 		start = filp->f_pos;
358 		break;
359 	case 2: /*SEEK_END*/
360 		start = i_size_read(filp->f_dentry->d_inode);
361 		break;
362 	default:
363 		return -EINVAL;
364 	}
365 
366 	start += l->l_start;
367 	if (start < 0)
368 		return -EINVAL;
369 	fl->fl_end = OFFSET_MAX;
370 	if (l->l_len > 0) {
371 		fl->fl_end = start + l->l_len - 1;
372 	} else if (l->l_len < 0) {
373 		fl->fl_end = start - 1;
374 		start += l->l_len;
375 		if (start < 0)
376 			return -EINVAL;
377 	}
378 	fl->fl_start = start;	/* we record the absolute position */
379 	if (fl->fl_end < fl->fl_start)
380 		return -EOVERFLOW;
381 
382 	fl->fl_owner = current->files;
383 	fl->fl_pid = current->tgid;
384 	fl->fl_file = filp;
385 	fl->fl_flags = FL_POSIX;
386 	fl->fl_ops = NULL;
387 	fl->fl_lmops = NULL;
388 
389 	switch (l->l_type) {
390 	case F_RDLCK:
391 	case F_WRLCK:
392 	case F_UNLCK:
393 		fl->fl_type = l->l_type;
394 		break;
395 	default:
396 		return -EINVAL;
397 	}
398 
399 	return (0);
400 }
401 #endif
402 
403 /* default lease lock manager operations */
404 static void lease_break_callback(struct file_lock *fl)
405 {
406 	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
407 }
408 
409 static void lease_release_private_callback(struct file_lock *fl)
410 {
411 	if (!fl->fl_file)
412 		return;
413 
414 	f_delown(fl->fl_file);
415 	fl->fl_file->f_owner.signum = 0;
416 }
417 
418 static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
419 {
420 	return fl->fl_file == try->fl_file;
421 }
422 
423 static struct lock_manager_operations lease_manager_ops = {
424 	.fl_break = lease_break_callback,
425 	.fl_release_private = lease_release_private_callback,
426 	.fl_mylease = lease_mylease_callback,
427 	.fl_change = lease_modify,
428 };
429 
430 /*
431  * Initialize a lease, use the default lock manager operations
432  */
433 static int lease_init(struct file *filp, int type, struct file_lock *fl)
434  {
435 	fl->fl_owner = current->files;
436 	fl->fl_pid = current->tgid;
437 
438 	fl->fl_file = filp;
439 	fl->fl_flags = FL_LEASE;
440 	if (assign_type(fl, type) != 0) {
441 		locks_free_lock(fl);
442 		return -EINVAL;
443 	}
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 int lease_alloc(struct file *filp, int type, struct file_lock **flp)
453 {
454 	struct file_lock *fl = locks_alloc_lock();
455 	int error;
456 
457 	if (fl == NULL)
458 		return -ENOMEM;
459 
460 	error = lease_init(filp, type, fl);
461 	if (error)
462 		return error;
463 	*flp = fl;
464 	return 0;
465 }
466 
467 /* Check if two locks overlap each other.
468  */
469 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
470 {
471 	return ((fl1->fl_end >= fl2->fl_start) &&
472 		(fl2->fl_end >= fl1->fl_start));
473 }
474 
475 /*
476  * Check whether two locks have the same owner.
477  */
478 static inline int
479 posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
480 {
481 	if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
482 		return fl2->fl_lmops == fl1->fl_lmops &&
483 			fl1->fl_lmops->fl_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 inline 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 static void locks_delete_block(struct file_lock *waiter)
500 {
501 	lock_kernel();
502 	__locks_delete_block(waiter);
503 	unlock_kernel();
504 }
505 
506 /* Insert waiter into blocker's block list.
507  * We use a circular list so that processes can be easily woken up in
508  * the order they blocked. The documentation doesn't require this but
509  * it seems like the reasonable thing to do.
510  */
511 static void locks_insert_block(struct file_lock *blocker,
512 			       struct file_lock *waiter)
513 {
514 	if (!list_empty(&waiter->fl_block)) {
515 		printk(KERN_ERR "locks_insert_block: removing duplicated lock "
516 			"(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
517 			waiter->fl_start, waiter->fl_end, waiter->fl_type);
518 		__locks_delete_block(waiter);
519 	}
520 	list_add_tail(&waiter->fl_block, &blocker->fl_block);
521 	waiter->fl_next = blocker;
522 	if (IS_POSIX(blocker))
523 		list_add(&waiter->fl_link, &blocked_list);
524 }
525 
526 /* Wake up processes blocked waiting for blocker.
527  * If told to wait then schedule the processes until the block list
528  * is empty, otherwise empty the block list ourselves.
529  */
530 static void locks_wake_up_blocks(struct file_lock *blocker)
531 {
532 	while (!list_empty(&blocker->fl_block)) {
533 		struct file_lock *waiter = list_entry(blocker->fl_block.next,
534 				struct file_lock, fl_block);
535 		__locks_delete_block(waiter);
536 		if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
537 			waiter->fl_lmops->fl_notify(waiter);
538 		else
539 			wake_up(&waiter->fl_wait);
540 	}
541 }
542 
543 /* Insert file lock fl into an inode's lock list at the position indicated
544  * by pos. At the same time add the lock to the global file lock list.
545  */
546 static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
547 {
548 	list_add(&fl->fl_link, &file_lock_list);
549 
550 	/* insert into file's list */
551 	fl->fl_next = *pos;
552 	*pos = fl;
553 
554 	if (fl->fl_ops && fl->fl_ops->fl_insert)
555 		fl->fl_ops->fl_insert(fl);
556 }
557 
558 /*
559  * Delete a lock and then free it.
560  * Wake up processes that are blocked waiting for this lock,
561  * notify the FS that the lock has been cleared and
562  * finally free the lock.
563  */
564 static void locks_delete_lock(struct file_lock **thisfl_p)
565 {
566 	struct file_lock *fl = *thisfl_p;
567 
568 	*thisfl_p = fl->fl_next;
569 	fl->fl_next = NULL;
570 	list_del_init(&fl->fl_link);
571 
572 	fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
573 	if (fl->fl_fasync != NULL) {
574 		printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
575 		fl->fl_fasync = NULL;
576 	}
577 
578 	if (fl->fl_ops && fl->fl_ops->fl_remove)
579 		fl->fl_ops->fl_remove(fl);
580 
581 	locks_wake_up_blocks(fl);
582 	locks_free_lock(fl);
583 }
584 
585 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
586  * checks for shared/exclusive status of overlapping locks.
587  */
588 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
589 {
590 	if (sys_fl->fl_type == F_WRLCK)
591 		return 1;
592 	if (caller_fl->fl_type == F_WRLCK)
593 		return 1;
594 	return 0;
595 }
596 
597 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
598  * checking before calling the locks_conflict().
599  */
600 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
601 {
602 	/* POSIX locks owned by the same process do not conflict with
603 	 * each other.
604 	 */
605 	if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
606 		return (0);
607 
608 	/* Check whether they overlap */
609 	if (!locks_overlap(caller_fl, sys_fl))
610 		return 0;
611 
612 	return (locks_conflict(caller_fl, sys_fl));
613 }
614 
615 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
616  * checking before calling the locks_conflict().
617  */
618 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
619 {
620 	/* FLOCK locks referring to the same filp do not conflict with
621 	 * each other.
622 	 */
623 	if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
624 		return (0);
625 	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
626 		return 0;
627 
628 	return (locks_conflict(caller_fl, sys_fl));
629 }
630 
631 static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
632 {
633 	int result = 0;
634 	DECLARE_WAITQUEUE(wait, current);
635 
636 	__set_current_state(TASK_INTERRUPTIBLE);
637 	add_wait_queue(fl_wait, &wait);
638 	if (timeout == 0)
639 		schedule();
640 	else
641 		result = schedule_timeout(timeout);
642 	if (signal_pending(current))
643 		result = -ERESTARTSYS;
644 	remove_wait_queue(fl_wait, &wait);
645 	__set_current_state(TASK_RUNNING);
646 	return result;
647 }
648 
649 static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
650 {
651 	int result;
652 	locks_insert_block(blocker, waiter);
653 	result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
654 	__locks_delete_block(waiter);
655 	return result;
656 }
657 
658 struct file_lock *
659 posix_test_lock(struct file *filp, struct file_lock *fl)
660 {
661 	struct file_lock *cfl;
662 
663 	lock_kernel();
664 	for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
665 		if (!IS_POSIX(cfl))
666 			continue;
667 		if (posix_locks_conflict(cfl, fl))
668 			break;
669 	}
670 	unlock_kernel();
671 
672 	return (cfl);
673 }
674 
675 EXPORT_SYMBOL(posix_test_lock);
676 
677 /* This function tests for deadlock condition before putting a process to
678  * sleep. The detection scheme is no longer recursive. Recursive was neat,
679  * but dangerous - we risked stack corruption if the lock data was bad, or
680  * if the recursion was too deep for any other reason.
681  *
682  * We rely on the fact that a task can only be on one lock's wait queue
683  * at a time. When we find blocked_task on a wait queue we can re-search
684  * with blocked_task equal to that queue's owner, until either blocked_task
685  * isn't found, or blocked_task is found on a queue owned by my_task.
686  *
687  * Note: the above assumption may not be true when handling lock requests
688  * from a broken NFS client. But broken NFS clients have a lot more to
689  * worry about than proper deadlock detection anyway... --okir
690  */
691 int posix_locks_deadlock(struct file_lock *caller_fl,
692 				struct file_lock *block_fl)
693 {
694 	struct list_head *tmp;
695 
696 next_task:
697 	if (posix_same_owner(caller_fl, block_fl))
698 		return 1;
699 	list_for_each(tmp, &blocked_list) {
700 		struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
701 		if (posix_same_owner(fl, block_fl)) {
702 			fl = fl->fl_next;
703 			block_fl = fl;
704 			goto next_task;
705 		}
706 	}
707 	return 0;
708 }
709 
710 EXPORT_SYMBOL(posix_locks_deadlock);
711 
712 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
713  * at the head of the list, but that's secret knowledge known only to
714  * flock_lock_file and posix_lock_file.
715  */
716 static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
717 {
718 	struct file_lock **before;
719 	struct inode * inode = filp->f_dentry->d_inode;
720 	int error = 0;
721 	int found = 0;
722 
723 	lock_kernel();
724 	for_each_lock(inode, before) {
725 		struct file_lock *fl = *before;
726 		if (IS_POSIX(fl))
727 			break;
728 		if (IS_LEASE(fl))
729 			continue;
730 		if (filp != fl->fl_file)
731 			continue;
732 		if (new_fl->fl_type == fl->fl_type)
733 			goto out;
734 		found = 1;
735 		locks_delete_lock(before);
736 		break;
737 	}
738 	unlock_kernel();
739 
740 	if (new_fl->fl_type == F_UNLCK)
741 		return 0;
742 
743 	/*
744 	 * If a higher-priority process was blocked on the old file lock,
745 	 * give it the opportunity to lock the file.
746 	 */
747 	if (found)
748 		cond_resched();
749 
750 	lock_kernel();
751 	for_each_lock(inode, before) {
752 		struct file_lock *fl = *before;
753 		if (IS_POSIX(fl))
754 			break;
755 		if (IS_LEASE(fl))
756 			continue;
757 		if (!flock_locks_conflict(new_fl, fl))
758 			continue;
759 		error = -EAGAIN;
760 		if (new_fl->fl_flags & FL_SLEEP) {
761 			locks_insert_block(fl, new_fl);
762 		}
763 		goto out;
764 	}
765 	locks_insert_lock(&inode->i_flock, new_fl);
766 	error = 0;
767 
768 out:
769 	unlock_kernel();
770 	return error;
771 }
772 
773 EXPORT_SYMBOL(posix_lock_file);
774 
775 static int __posix_lock_file(struct inode *inode, struct file_lock *request)
776 {
777 	struct file_lock *fl;
778 	struct file_lock *new_fl, *new_fl2;
779 	struct file_lock *left = NULL;
780 	struct file_lock *right = NULL;
781 	struct file_lock **before;
782 	int error, added = 0;
783 
784 	/*
785 	 * We may need two file_lock structures for this operation,
786 	 * so we get them in advance to avoid races.
787 	 */
788 	new_fl = locks_alloc_lock();
789 	new_fl2 = locks_alloc_lock();
790 
791 	lock_kernel();
792 	if (request->fl_type != F_UNLCK) {
793 		for_each_lock(inode, before) {
794 			struct file_lock *fl = *before;
795 			if (!IS_POSIX(fl))
796 				continue;
797 			if (!posix_locks_conflict(request, fl))
798 				continue;
799 			error = -EAGAIN;
800 			if (!(request->fl_flags & FL_SLEEP))
801 				goto out;
802 			error = -EDEADLK;
803 			if (posix_locks_deadlock(request, fl))
804 				goto out;
805 			error = -EAGAIN;
806 			locks_insert_block(fl, request);
807 			goto out;
808   		}
809   	}
810 
811 	/* If we're just looking for a conflict, we're done. */
812 	error = 0;
813 	if (request->fl_flags & FL_ACCESS)
814 		goto out;
815 
816 	error = -ENOLCK; /* "no luck" */
817 	if (!(new_fl && new_fl2))
818 		goto out;
819 
820 	/*
821 	 * We've allocated the new locks in advance, so there are no
822 	 * errors possible (and no blocking operations) from here on.
823 	 *
824 	 * Find the first old lock with the same owner as the new lock.
825 	 */
826 
827 	before = &inode->i_flock;
828 
829 	/* First skip locks owned by other processes.  */
830 	while ((fl = *before) && (!IS_POSIX(fl) ||
831 				  !posix_same_owner(request, fl))) {
832 		before = &fl->fl_next;
833 	}
834 
835 	/* Process locks with this owner.  */
836 	while ((fl = *before) && posix_same_owner(request, fl)) {
837 		/* Detect adjacent or overlapping regions (if same lock type)
838 		 */
839 		if (request->fl_type == fl->fl_type) {
840 			/* In all comparisons of start vs end, use
841 			 * "start - 1" rather than "end + 1". If end
842 			 * is OFFSET_MAX, end + 1 will become negative.
843 			 */
844 			if (fl->fl_end < request->fl_start - 1)
845 				goto next_lock;
846 			/* If the next lock in the list has entirely bigger
847 			 * addresses than the new one, insert the lock here.
848 			 */
849 			if (fl->fl_start - 1 > request->fl_end)
850 				break;
851 
852 			/* If we come here, the new and old lock are of the
853 			 * same type and adjacent or overlapping. Make one
854 			 * lock yielding from the lower start address of both
855 			 * locks to the higher end address.
856 			 */
857 			if (fl->fl_start > request->fl_start)
858 				fl->fl_start = request->fl_start;
859 			else
860 				request->fl_start = fl->fl_start;
861 			if (fl->fl_end < request->fl_end)
862 				fl->fl_end = request->fl_end;
863 			else
864 				request->fl_end = fl->fl_end;
865 			if (added) {
866 				locks_delete_lock(before);
867 				continue;
868 			}
869 			request = fl;
870 			added = 1;
871 		}
872 		else {
873 			/* Processing for different lock types is a bit
874 			 * more complex.
875 			 */
876 			if (fl->fl_end < request->fl_start)
877 				goto next_lock;
878 			if (fl->fl_start > request->fl_end)
879 				break;
880 			if (request->fl_type == F_UNLCK)
881 				added = 1;
882 			if (fl->fl_start < request->fl_start)
883 				left = fl;
884 			/* If the next lock in the list has a higher end
885 			 * address than the new one, insert the new one here.
886 			 */
887 			if (fl->fl_end > request->fl_end) {
888 				right = fl;
889 				break;
890 			}
891 			if (fl->fl_start >= request->fl_start) {
892 				/* The new lock completely replaces an old
893 				 * one (This may happen several times).
894 				 */
895 				if (added) {
896 					locks_delete_lock(before);
897 					continue;
898 				}
899 				/* Replace the old lock with the new one.
900 				 * Wake up anybody waiting for the old one,
901 				 * as the change in lock type might satisfy
902 				 * their needs.
903 				 */
904 				locks_wake_up_blocks(fl);
905 				fl->fl_start = request->fl_start;
906 				fl->fl_end = request->fl_end;
907 				fl->fl_type = request->fl_type;
908 				fl->fl_u = request->fl_u;
909 				request = fl;
910 				added = 1;
911 			}
912 		}
913 		/* Go on to next lock.
914 		 */
915 	next_lock:
916 		before = &fl->fl_next;
917 	}
918 
919 	error = 0;
920 	if (!added) {
921 		if (request->fl_type == F_UNLCK)
922 			goto out;
923 		locks_copy_lock(new_fl, request);
924 		locks_insert_lock(before, new_fl);
925 		new_fl = NULL;
926 	}
927 	if (right) {
928 		if (left == right) {
929 			/* The new lock breaks the old one in two pieces,
930 			 * so we have to use the second new lock.
931 			 */
932 			left = new_fl2;
933 			new_fl2 = NULL;
934 			locks_copy_lock(left, right);
935 			locks_insert_lock(before, left);
936 		}
937 		right->fl_start = request->fl_end + 1;
938 		locks_wake_up_blocks(right);
939 	}
940 	if (left) {
941 		left->fl_end = request->fl_start - 1;
942 		locks_wake_up_blocks(left);
943 	}
944  out:
945 	unlock_kernel();
946 	/*
947 	 * Free any unused locks.
948 	 */
949 	if (new_fl)
950 		locks_free_lock(new_fl);
951 	if (new_fl2)
952 		locks_free_lock(new_fl2);
953 	return error;
954 }
955 
956 /**
957  * posix_lock_file - Apply a POSIX-style lock to a file
958  * @filp: The file to apply the lock to
959  * @fl: The lock to be applied
960  *
961  * Add a POSIX style lock to a file.
962  * We merge adjacent & overlapping locks whenever possible.
963  * POSIX locks are sorted by owner task, then by starting address
964  */
965 int posix_lock_file(struct file *filp, struct file_lock *fl)
966 {
967 	return __posix_lock_file(filp->f_dentry->d_inode, fl);
968 }
969 
970 /**
971  * posix_lock_file_wait - Apply a POSIX-style lock to a file
972  * @filp: The file to apply the lock to
973  * @fl: The lock to be applied
974  *
975  * Add a POSIX style lock to a file.
976  * We merge adjacent & overlapping locks whenever possible.
977  * POSIX locks are sorted by owner task, then by starting address
978  */
979 int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
980 {
981 	int error;
982 	might_sleep ();
983 	for (;;) {
984 		error = __posix_lock_file(filp->f_dentry->d_inode, fl);
985 		if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
986 			break;
987 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
988 		if (!error)
989 			continue;
990 
991 		locks_delete_block(fl);
992 		break;
993 	}
994 	return error;
995 }
996 EXPORT_SYMBOL(posix_lock_file_wait);
997 
998 /**
999  * locks_mandatory_locked - Check for an active lock
1000  * @inode: the file to check
1001  *
1002  * Searches the inode's list of locks to find any POSIX locks which conflict.
1003  * This function is called from locks_verify_locked() only.
1004  */
1005 int locks_mandatory_locked(struct inode *inode)
1006 {
1007 	fl_owner_t owner = current->files;
1008 	struct file_lock *fl;
1009 
1010 	/*
1011 	 * Search the lock list for this inode for any POSIX locks.
1012 	 */
1013 	lock_kernel();
1014 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1015 		if (!IS_POSIX(fl))
1016 			continue;
1017 		if (fl->fl_owner != owner)
1018 			break;
1019 	}
1020 	unlock_kernel();
1021 	return fl ? -EAGAIN : 0;
1022 }
1023 
1024 /**
1025  * locks_mandatory_area - Check for a conflicting lock
1026  * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1027  *		for shared
1028  * @inode:      the file to check
1029  * @filp:       how the file was opened (if it was)
1030  * @offset:     start of area to check
1031  * @count:      length of area to check
1032  *
1033  * Searches the inode's list of locks to find any POSIX locks which conflict.
1034  * This function is called from rw_verify_area() and
1035  * locks_verify_truncate().
1036  */
1037 int locks_mandatory_area(int read_write, struct inode *inode,
1038 			 struct file *filp, loff_t offset,
1039 			 size_t count)
1040 {
1041 	struct file_lock fl;
1042 	int error;
1043 
1044 	locks_init_lock(&fl);
1045 	fl.fl_owner = current->files;
1046 	fl.fl_pid = current->tgid;
1047 	fl.fl_file = filp;
1048 	fl.fl_flags = FL_POSIX | FL_ACCESS;
1049 	if (filp && !(filp->f_flags & O_NONBLOCK))
1050 		fl.fl_flags |= FL_SLEEP;
1051 	fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1052 	fl.fl_start = offset;
1053 	fl.fl_end = offset + count - 1;
1054 
1055 	for (;;) {
1056 		error = __posix_lock_file(inode, &fl);
1057 		if (error != -EAGAIN)
1058 			break;
1059 		if (!(fl.fl_flags & FL_SLEEP))
1060 			break;
1061 		error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1062 		if (!error) {
1063 			/*
1064 			 * If we've been sleeping someone might have
1065 			 * changed the permissions behind our back.
1066 			 */
1067 			if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1068 				continue;
1069 		}
1070 
1071 		locks_delete_block(&fl);
1072 		break;
1073 	}
1074 
1075 	return error;
1076 }
1077 
1078 EXPORT_SYMBOL(locks_mandatory_area);
1079 
1080 /* We already had a lease on this file; just change its type */
1081 int lease_modify(struct file_lock **before, int arg)
1082 {
1083 	struct file_lock *fl = *before;
1084 	int error = assign_type(fl, arg);
1085 
1086 	if (error)
1087 		return error;
1088 	locks_wake_up_blocks(fl);
1089 	if (arg == F_UNLCK)
1090 		locks_delete_lock(before);
1091 	return 0;
1092 }
1093 
1094 EXPORT_SYMBOL(lease_modify);
1095 
1096 static void time_out_leases(struct inode *inode)
1097 {
1098 	struct file_lock **before;
1099 	struct file_lock *fl;
1100 
1101 	before = &inode->i_flock;
1102 	while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1103 		if ((fl->fl_break_time == 0)
1104 				|| time_before(jiffies, fl->fl_break_time)) {
1105 			before = &fl->fl_next;
1106 			continue;
1107 		}
1108 		printk(KERN_INFO "lease broken - owner pid = %d\n", fl->fl_pid);
1109 		lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1110 		if (fl == *before)	/* lease_modify may have freed fl */
1111 			before = &fl->fl_next;
1112 	}
1113 }
1114 
1115 /**
1116  *	__break_lease	-	revoke all outstanding leases on file
1117  *	@inode: the inode of the file to return
1118  *	@mode: the open mode (read or write)
1119  *
1120  *	break_lease (inlined for speed) has checked there already
1121  *	is a lease on this file.  Leases are broken on a call to open()
1122  *	or truncate().  This function can sleep unless you
1123  *	specified %O_NONBLOCK to your open().
1124  */
1125 int __break_lease(struct inode *inode, unsigned int mode)
1126 {
1127 	int error = 0, future;
1128 	struct file_lock *new_fl, *flock;
1129 	struct file_lock *fl;
1130 	int alloc_err;
1131 	unsigned long break_time;
1132 	int i_have_this_lease = 0;
1133 
1134 	alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1135 			&new_fl);
1136 
1137 	lock_kernel();
1138 
1139 	time_out_leases(inode);
1140 
1141 	flock = inode->i_flock;
1142 	if ((flock == NULL) || !IS_LEASE(flock))
1143 		goto out;
1144 
1145 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1146 		if (fl->fl_owner == current->files)
1147 			i_have_this_lease = 1;
1148 
1149 	if (mode & FMODE_WRITE) {
1150 		/* If we want write access, we have to revoke any lease. */
1151 		future = F_UNLCK | F_INPROGRESS;
1152 	} else if (flock->fl_type & F_INPROGRESS) {
1153 		/* If the lease is already being broken, we just leave it */
1154 		future = flock->fl_type;
1155 	} else if (flock->fl_type & F_WRLCK) {
1156 		/* Downgrade the exclusive lease to a read-only lease. */
1157 		future = F_RDLCK | F_INPROGRESS;
1158 	} else {
1159 		/* the existing lease was read-only, so we can read too. */
1160 		goto out;
1161 	}
1162 
1163 	if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1164 		error = alloc_err;
1165 		goto out;
1166 	}
1167 
1168 	break_time = 0;
1169 	if (lease_break_time > 0) {
1170 		break_time = jiffies + lease_break_time * HZ;
1171 		if (break_time == 0)
1172 			break_time++;	/* so that 0 means no break time */
1173 	}
1174 
1175 	for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1176 		if (fl->fl_type != future) {
1177 			fl->fl_type = future;
1178 			fl->fl_break_time = break_time;
1179 			/* lease must have lmops break callback */
1180 			fl->fl_lmops->fl_break(fl);
1181 		}
1182 	}
1183 
1184 	if (i_have_this_lease || (mode & O_NONBLOCK)) {
1185 		error = -EWOULDBLOCK;
1186 		goto out;
1187 	}
1188 
1189 restart:
1190 	break_time = flock->fl_break_time;
1191 	if (break_time != 0) {
1192 		break_time -= jiffies;
1193 		if (break_time == 0)
1194 			break_time++;
1195 	}
1196 	error = locks_block_on_timeout(flock, new_fl, break_time);
1197 	if (error >= 0) {
1198 		if (error == 0)
1199 			time_out_leases(inode);
1200 		/* Wait for the next lease that has not been broken yet */
1201 		for (flock = inode->i_flock; flock && IS_LEASE(flock);
1202 				flock = flock->fl_next) {
1203 			if (flock->fl_type & F_INPROGRESS)
1204 				goto restart;
1205 		}
1206 		error = 0;
1207 	}
1208 
1209 out:
1210 	unlock_kernel();
1211 	if (!alloc_err)
1212 		locks_free_lock(new_fl);
1213 	return error;
1214 }
1215 
1216 EXPORT_SYMBOL(__break_lease);
1217 
1218 /**
1219  *	lease_get_mtime
1220  *	@inode: the inode
1221  *      @time:  pointer to a timespec which will contain the last modified time
1222  *
1223  * This is to force NFS clients to flush their caches for files with
1224  * exclusive leases.  The justification is that if someone has an
1225  * exclusive lease, then they could be modifiying it.
1226  */
1227 void lease_get_mtime(struct inode *inode, struct timespec *time)
1228 {
1229 	struct file_lock *flock = inode->i_flock;
1230 	if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1231 		*time = current_fs_time(inode->i_sb);
1232 	else
1233 		*time = inode->i_mtime;
1234 }
1235 
1236 EXPORT_SYMBOL(lease_get_mtime);
1237 
1238 /**
1239  *	fcntl_getlease - Enquire what lease is currently active
1240  *	@filp: the file
1241  *
1242  *	The value returned by this function will be one of
1243  *	(if no lease break is pending):
1244  *
1245  *	%F_RDLCK to indicate a shared lease is held.
1246  *
1247  *	%F_WRLCK to indicate an exclusive lease is held.
1248  *
1249  *	%F_UNLCK to indicate no lease is held.
1250  *
1251  *	(if a lease break is pending):
1252  *
1253  *	%F_RDLCK to indicate an exclusive lease needs to be
1254  *		changed to a shared lease (or removed).
1255  *
1256  *	%F_UNLCK to indicate the lease needs to be removed.
1257  *
1258  *	XXX: sfr & willy disagree over whether F_INPROGRESS
1259  *	should be returned to userspace.
1260  */
1261 int fcntl_getlease(struct file *filp)
1262 {
1263 	struct file_lock *fl;
1264 	int type = F_UNLCK;
1265 
1266 	lock_kernel();
1267 	time_out_leases(filp->f_dentry->d_inode);
1268 	for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1269 			fl = fl->fl_next) {
1270 		if (fl->fl_file == filp) {
1271 			type = fl->fl_type & ~F_INPROGRESS;
1272 			break;
1273 		}
1274 	}
1275 	unlock_kernel();
1276 	return type;
1277 }
1278 
1279 /**
1280  *	__setlease	-	sets a lease on an open file
1281  *	@filp: file pointer
1282  *	@arg: type of lease to obtain
1283  *	@flp: input - file_lock to use, output - file_lock inserted
1284  *
1285  *	The (input) flp->fl_lmops->fl_break function is required
1286  *	by break_lease().
1287  *
1288  *	Called with kernel lock held.
1289  */
1290 static int __setlease(struct file *filp, long arg, struct file_lock **flp)
1291 {
1292 	struct file_lock *fl, **before, **my_before = NULL, *lease;
1293 	struct dentry *dentry = filp->f_dentry;
1294 	struct inode *inode = dentry->d_inode;
1295 	int error, rdlease_count = 0, wrlease_count = 0;
1296 
1297 	time_out_leases(inode);
1298 
1299 	error = -EINVAL;
1300 	if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1301 		goto out;
1302 
1303 	lease = *flp;
1304 
1305 	error = -EAGAIN;
1306 	if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1307 		goto out;
1308 	if ((arg == F_WRLCK)
1309 	    && ((atomic_read(&dentry->d_count) > 1)
1310 		|| (atomic_read(&inode->i_count) > 1)))
1311 		goto out;
1312 
1313 	/*
1314 	 * At this point, we know that if there is an exclusive
1315 	 * lease on this file, then we hold it on this filp
1316 	 * (otherwise our open of this file would have blocked).
1317 	 * And if we are trying to acquire an exclusive lease,
1318 	 * then the file is not open by anyone (including us)
1319 	 * except for this filp.
1320 	 */
1321 	for (before = &inode->i_flock;
1322 			((fl = *before) != NULL) && IS_LEASE(fl);
1323 			before = &fl->fl_next) {
1324 		if (lease->fl_lmops->fl_mylease(fl, lease))
1325 			my_before = before;
1326 		else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1327 			/*
1328 			 * Someone is in the process of opening this
1329 			 * file for writing so we may not take an
1330 			 * exclusive lease on it.
1331 			 */
1332 			wrlease_count++;
1333 		else
1334 			rdlease_count++;
1335 	}
1336 
1337 	if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1338 	    (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1339 		goto out;
1340 
1341 	if (my_before != NULL) {
1342 		error = lease->fl_lmops->fl_change(my_before, arg);
1343 		goto out;
1344 	}
1345 
1346 	error = 0;
1347 	if (arg == F_UNLCK)
1348 		goto out;
1349 
1350 	error = -EINVAL;
1351 	if (!leases_enable)
1352 		goto out;
1353 
1354 	error = lease_alloc(filp, arg, &fl);
1355 	if (error)
1356 		goto out;
1357 
1358 	locks_copy_lock(fl, lease);
1359 
1360 	locks_insert_lock(before, fl);
1361 
1362 	*flp = fl;
1363 out:
1364 	return error;
1365 }
1366 
1367  /**
1368  *	setlease        -       sets a lease on an open file
1369  *	@filp: file pointer
1370  *	@arg: type of lease to obtain
1371  *	@lease: file_lock to use
1372  *
1373  *	Call this to establish a lease on the file.
1374  *	The fl_lmops fl_break function is required by break_lease
1375  */
1376 
1377 int setlease(struct file *filp, long arg, struct file_lock **lease)
1378 {
1379 	struct dentry *dentry = filp->f_dentry;
1380 	struct inode *inode = dentry->d_inode;
1381 	int error;
1382 
1383 	if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1384 		return -EACCES;
1385 	if (!S_ISREG(inode->i_mode))
1386 		return -EINVAL;
1387 	error = security_file_lock(filp, arg);
1388 	if (error)
1389 		return error;
1390 
1391 	lock_kernel();
1392 	error = __setlease(filp, arg, lease);
1393 	unlock_kernel();
1394 
1395 	return error;
1396 }
1397 
1398 EXPORT_SYMBOL(setlease);
1399 
1400 /**
1401  *	fcntl_setlease	-	sets a lease on an open file
1402  *	@fd: open file descriptor
1403  *	@filp: file pointer
1404  *	@arg: type of lease to obtain
1405  *
1406  *	Call this fcntl to establish a lease on the file.
1407  *	Note that you also need to call %F_SETSIG to
1408  *	receive a signal when the lease is broken.
1409  */
1410 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1411 {
1412 	struct file_lock fl, *flp = &fl;
1413 	struct dentry *dentry = filp->f_dentry;
1414 	struct inode *inode = dentry->d_inode;
1415 	int error;
1416 
1417 	if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1418 		return -EACCES;
1419 	if (!S_ISREG(inode->i_mode))
1420 		return -EINVAL;
1421 	error = security_file_lock(filp, arg);
1422 	if (error)
1423 		return error;
1424 
1425 	locks_init_lock(&fl);
1426 	error = lease_init(filp, arg, &fl);
1427 	if (error)
1428 		return error;
1429 
1430 	lock_kernel();
1431 
1432 	error = __setlease(filp, arg, &flp);
1433 	if (error)
1434 		goto out_unlock;
1435 
1436 	error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1437 	if (error < 0) {
1438 		/* remove lease just inserted by __setlease */
1439 		flp->fl_type = F_UNLCK | F_INPROGRESS;
1440 		flp->fl_break_time = jiffies- 10;
1441 		time_out_leases(inode);
1442 		goto out_unlock;
1443 	}
1444 
1445 	error = f_setown(filp, current->pid, 0);
1446 out_unlock:
1447 	unlock_kernel();
1448 	return error;
1449 }
1450 
1451 /**
1452  * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1453  * @filp: The file to apply the lock to
1454  * @fl: The lock to be applied
1455  *
1456  * Add a FLOCK style lock to a file.
1457  */
1458 int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1459 {
1460 	int error;
1461 	might_sleep();
1462 	for (;;) {
1463 		error = flock_lock_file(filp, fl);
1464 		if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1465 			break;
1466 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1467 		if (!error)
1468 			continue;
1469 
1470 		locks_delete_block(fl);
1471 		break;
1472 	}
1473 	return error;
1474 }
1475 
1476 EXPORT_SYMBOL(flock_lock_file_wait);
1477 
1478 /**
1479  *	sys_flock: - flock() system call.
1480  *	@fd: the file descriptor to lock.
1481  *	@cmd: the type of lock to apply.
1482  *
1483  *	Apply a %FL_FLOCK style lock to an open file descriptor.
1484  *	The @cmd can be one of
1485  *
1486  *	%LOCK_SH -- a shared lock.
1487  *
1488  *	%LOCK_EX -- an exclusive lock.
1489  *
1490  *	%LOCK_UN -- remove an existing lock.
1491  *
1492  *	%LOCK_MAND -- a `mandatory' flock.  This exists to emulate Windows Share Modes.
1493  *
1494  *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1495  *	processes read and write access respectively.
1496  */
1497 asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1498 {
1499 	struct file *filp;
1500 	struct file_lock *lock;
1501 	int can_sleep, unlock;
1502 	int error;
1503 
1504 	error = -EBADF;
1505 	filp = fget(fd);
1506 	if (!filp)
1507 		goto out;
1508 
1509 	can_sleep = !(cmd & LOCK_NB);
1510 	cmd &= ~LOCK_NB;
1511 	unlock = (cmd == LOCK_UN);
1512 
1513 	if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1514 		goto out_putf;
1515 
1516 	error = flock_make_lock(filp, &lock, cmd);
1517 	if (error)
1518 		goto out_putf;
1519 	if (can_sleep)
1520 		lock->fl_flags |= FL_SLEEP;
1521 
1522 	error = security_file_lock(filp, cmd);
1523 	if (error)
1524 		goto out_free;
1525 
1526 	if (filp->f_op && filp->f_op->flock)
1527 		error = filp->f_op->flock(filp,
1528 					  (can_sleep) ? F_SETLKW : F_SETLK,
1529 					  lock);
1530 	else
1531 		error = flock_lock_file_wait(filp, lock);
1532 
1533  out_free:
1534 	if (list_empty(&lock->fl_link)) {
1535 		locks_free_lock(lock);
1536 	}
1537 
1538  out_putf:
1539 	fput(filp);
1540  out:
1541 	return error;
1542 }
1543 
1544 /* Report the first existing lock that would conflict with l.
1545  * This implements the F_GETLK command of fcntl().
1546  */
1547 int fcntl_getlk(struct file *filp, struct flock __user *l)
1548 {
1549 	struct file_lock *fl, file_lock;
1550 	struct flock flock;
1551 	int error;
1552 
1553 	error = -EFAULT;
1554 	if (copy_from_user(&flock, l, sizeof(flock)))
1555 		goto out;
1556 	error = -EINVAL;
1557 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1558 		goto out;
1559 
1560 	error = flock_to_posix_lock(filp, &file_lock, &flock);
1561 	if (error)
1562 		goto out;
1563 
1564 	if (filp->f_op && filp->f_op->lock) {
1565 		error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1566 		if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1567 			file_lock.fl_ops->fl_release_private(&file_lock);
1568 		if (error < 0)
1569 			goto out;
1570 		else
1571 		  fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1572 	} else {
1573 		fl = posix_test_lock(filp, &file_lock);
1574 	}
1575 
1576 	flock.l_type = F_UNLCK;
1577 	if (fl != NULL) {
1578 		flock.l_pid = fl->fl_pid;
1579 #if BITS_PER_LONG == 32
1580 		/*
1581 		 * Make sure we can represent the posix lock via
1582 		 * legacy 32bit flock.
1583 		 */
1584 		error = -EOVERFLOW;
1585 		if (fl->fl_start > OFFT_OFFSET_MAX)
1586 			goto out;
1587 		if ((fl->fl_end != OFFSET_MAX)
1588 		    && (fl->fl_end > OFFT_OFFSET_MAX))
1589 			goto out;
1590 #endif
1591 		flock.l_start = fl->fl_start;
1592 		flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1593 			fl->fl_end - fl->fl_start + 1;
1594 		flock.l_whence = 0;
1595 		flock.l_type = fl->fl_type;
1596 	}
1597 	error = -EFAULT;
1598 	if (!copy_to_user(l, &flock, sizeof(flock)))
1599 		error = 0;
1600 out:
1601 	return error;
1602 }
1603 
1604 /* Apply the lock described by l to an open file descriptor.
1605  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1606  */
1607 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1608 		struct flock __user *l)
1609 {
1610 	struct file_lock *file_lock = locks_alloc_lock();
1611 	struct flock flock;
1612 	struct inode *inode;
1613 	int error;
1614 
1615 	if (file_lock == NULL)
1616 		return -ENOLCK;
1617 
1618 	/*
1619 	 * This might block, so we do it before checking the inode.
1620 	 */
1621 	error = -EFAULT;
1622 	if (copy_from_user(&flock, l, sizeof(flock)))
1623 		goto out;
1624 
1625 	inode = filp->f_dentry->d_inode;
1626 
1627 	/* Don't allow mandatory locks on files that may be memory mapped
1628 	 * and shared.
1629 	 */
1630 	if (IS_MANDLOCK(inode) &&
1631 	    (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1632 	    mapping_writably_mapped(filp->f_mapping)) {
1633 		error = -EAGAIN;
1634 		goto out;
1635 	}
1636 
1637 again:
1638 	error = flock_to_posix_lock(filp, file_lock, &flock);
1639 	if (error)
1640 		goto out;
1641 	if (cmd == F_SETLKW) {
1642 		file_lock->fl_flags |= FL_SLEEP;
1643 	}
1644 
1645 	error = -EBADF;
1646 	switch (flock.l_type) {
1647 	case F_RDLCK:
1648 		if (!(filp->f_mode & FMODE_READ))
1649 			goto out;
1650 		break;
1651 	case F_WRLCK:
1652 		if (!(filp->f_mode & FMODE_WRITE))
1653 			goto out;
1654 		break;
1655 	case F_UNLCK:
1656 		break;
1657 	default:
1658 		error = -EINVAL;
1659 		goto out;
1660 	}
1661 
1662 	error = security_file_lock(filp, file_lock->fl_type);
1663 	if (error)
1664 		goto out;
1665 
1666 	if (filp->f_op && filp->f_op->lock != NULL)
1667 		error = filp->f_op->lock(filp, cmd, file_lock);
1668 	else {
1669 		for (;;) {
1670 			error = __posix_lock_file(inode, file_lock);
1671 			if ((error != -EAGAIN) || (cmd == F_SETLK))
1672 				break;
1673 			error = wait_event_interruptible(file_lock->fl_wait,
1674 					!file_lock->fl_next);
1675 			if (!error)
1676 				continue;
1677 
1678 			locks_delete_block(file_lock);
1679 			break;
1680 		}
1681 	}
1682 
1683 	/*
1684 	 * Attempt to detect a close/fcntl race and recover by
1685 	 * releasing the lock that was just acquired.
1686 	 */
1687 	if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1688 		flock.l_type = F_UNLCK;
1689 		goto again;
1690 	}
1691 
1692 out:
1693 	locks_free_lock(file_lock);
1694 	return error;
1695 }
1696 
1697 #if BITS_PER_LONG == 32
1698 /* Report the first existing lock that would conflict with l.
1699  * This implements the F_GETLK command of fcntl().
1700  */
1701 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1702 {
1703 	struct file_lock *fl, file_lock;
1704 	struct flock64 flock;
1705 	int error;
1706 
1707 	error = -EFAULT;
1708 	if (copy_from_user(&flock, l, sizeof(flock)))
1709 		goto out;
1710 	error = -EINVAL;
1711 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1712 		goto out;
1713 
1714 	error = flock64_to_posix_lock(filp, &file_lock, &flock);
1715 	if (error)
1716 		goto out;
1717 
1718 	if (filp->f_op && filp->f_op->lock) {
1719 		error = filp->f_op->lock(filp, F_GETLK, &file_lock);
1720 		if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1721 			file_lock.fl_ops->fl_release_private(&file_lock);
1722 		if (error < 0)
1723 			goto out;
1724 		else
1725 		  fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1726 	} else {
1727 		fl = posix_test_lock(filp, &file_lock);
1728 	}
1729 
1730 	flock.l_type = F_UNLCK;
1731 	if (fl != NULL) {
1732 		flock.l_pid = fl->fl_pid;
1733 		flock.l_start = fl->fl_start;
1734 		flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1735 			fl->fl_end - fl->fl_start + 1;
1736 		flock.l_whence = 0;
1737 		flock.l_type = fl->fl_type;
1738 	}
1739 	error = -EFAULT;
1740 	if (!copy_to_user(l, &flock, sizeof(flock)))
1741 		error = 0;
1742 
1743 out:
1744 	return error;
1745 }
1746 
1747 /* Apply the lock described by l to an open file descriptor.
1748  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1749  */
1750 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1751 		struct flock64 __user *l)
1752 {
1753 	struct file_lock *file_lock = locks_alloc_lock();
1754 	struct flock64 flock;
1755 	struct inode *inode;
1756 	int error;
1757 
1758 	if (file_lock == NULL)
1759 		return -ENOLCK;
1760 
1761 	/*
1762 	 * This might block, so we do it before checking the inode.
1763 	 */
1764 	error = -EFAULT;
1765 	if (copy_from_user(&flock, l, sizeof(flock)))
1766 		goto out;
1767 
1768 	inode = filp->f_dentry->d_inode;
1769 
1770 	/* Don't allow mandatory locks on files that may be memory mapped
1771 	 * and shared.
1772 	 */
1773 	if (IS_MANDLOCK(inode) &&
1774 	    (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1775 	    mapping_writably_mapped(filp->f_mapping)) {
1776 		error = -EAGAIN;
1777 		goto out;
1778 	}
1779 
1780 again:
1781 	error = flock64_to_posix_lock(filp, file_lock, &flock);
1782 	if (error)
1783 		goto out;
1784 	if (cmd == F_SETLKW64) {
1785 		file_lock->fl_flags |= FL_SLEEP;
1786 	}
1787 
1788 	error = -EBADF;
1789 	switch (flock.l_type) {
1790 	case F_RDLCK:
1791 		if (!(filp->f_mode & FMODE_READ))
1792 			goto out;
1793 		break;
1794 	case F_WRLCK:
1795 		if (!(filp->f_mode & FMODE_WRITE))
1796 			goto out;
1797 		break;
1798 	case F_UNLCK:
1799 		break;
1800 	default:
1801 		error = -EINVAL;
1802 		goto out;
1803 	}
1804 
1805 	error = security_file_lock(filp, file_lock->fl_type);
1806 	if (error)
1807 		goto out;
1808 
1809 	if (filp->f_op && filp->f_op->lock != NULL)
1810 		error = filp->f_op->lock(filp, cmd, file_lock);
1811 	else {
1812 		for (;;) {
1813 			error = __posix_lock_file(inode, file_lock);
1814 			if ((error != -EAGAIN) || (cmd == F_SETLK64))
1815 				break;
1816 			error = wait_event_interruptible(file_lock->fl_wait,
1817 					!file_lock->fl_next);
1818 			if (!error)
1819 				continue;
1820 
1821 			locks_delete_block(file_lock);
1822 			break;
1823 		}
1824 	}
1825 
1826 	/*
1827 	 * Attempt to detect a close/fcntl race and recover by
1828 	 * releasing the lock that was just acquired.
1829 	 */
1830 	if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1831 		flock.l_type = F_UNLCK;
1832 		goto again;
1833 	}
1834 
1835 out:
1836 	locks_free_lock(file_lock);
1837 	return error;
1838 }
1839 #endif /* BITS_PER_LONG == 32 */
1840 
1841 /*
1842  * This function is called when the file is being removed
1843  * from the task's fd array.  POSIX locks belonging to this task
1844  * are deleted at this time.
1845  */
1846 void locks_remove_posix(struct file *filp, fl_owner_t owner)
1847 {
1848 	struct file_lock lock, **before;
1849 
1850 	/*
1851 	 * If there are no locks held on this file, we don't need to call
1852 	 * posix_lock_file().  Another process could be setting a lock on this
1853 	 * file at the same time, but we wouldn't remove that lock anyway.
1854 	 */
1855 	before = &filp->f_dentry->d_inode->i_flock;
1856 	if (*before == NULL)
1857 		return;
1858 
1859 	lock.fl_type = F_UNLCK;
1860 	lock.fl_flags = FL_POSIX;
1861 	lock.fl_start = 0;
1862 	lock.fl_end = OFFSET_MAX;
1863 	lock.fl_owner = owner;
1864 	lock.fl_pid = current->tgid;
1865 	lock.fl_file = filp;
1866 	lock.fl_ops = NULL;
1867 	lock.fl_lmops = NULL;
1868 
1869 	if (filp->f_op && filp->f_op->lock != NULL) {
1870 		filp->f_op->lock(filp, F_SETLK, &lock);
1871 		goto out;
1872 	}
1873 
1874 	/* Can't use posix_lock_file here; we need to remove it no matter
1875 	 * which pid we have.
1876 	 */
1877 	lock_kernel();
1878 	while (*before != NULL) {
1879 		struct file_lock *fl = *before;
1880 		if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1881 			locks_delete_lock(before);
1882 			continue;
1883 		}
1884 		before = &fl->fl_next;
1885 	}
1886 	unlock_kernel();
1887 out:
1888 	if (lock.fl_ops && lock.fl_ops->fl_release_private)
1889 		lock.fl_ops->fl_release_private(&lock);
1890 }
1891 
1892 EXPORT_SYMBOL(locks_remove_posix);
1893 
1894 /*
1895  * This function is called on the last close of an open file.
1896  */
1897 void locks_remove_flock(struct file *filp)
1898 {
1899 	struct inode * inode = filp->f_dentry->d_inode;
1900 	struct file_lock *fl;
1901 	struct file_lock **before;
1902 
1903 	if (!inode->i_flock)
1904 		return;
1905 
1906 	if (filp->f_op && filp->f_op->flock) {
1907 		struct file_lock fl = {
1908 			.fl_pid = current->tgid,
1909 			.fl_file = filp,
1910 			.fl_flags = FL_FLOCK,
1911 			.fl_type = F_UNLCK,
1912 			.fl_end = OFFSET_MAX,
1913 		};
1914 		filp->f_op->flock(filp, F_SETLKW, &fl);
1915 		if (fl.fl_ops && fl.fl_ops->fl_release_private)
1916 			fl.fl_ops->fl_release_private(&fl);
1917 	}
1918 
1919 	lock_kernel();
1920 	before = &inode->i_flock;
1921 
1922 	while ((fl = *before) != NULL) {
1923 		if (fl->fl_file == filp) {
1924 			if (IS_FLOCK(fl)) {
1925 				locks_delete_lock(before);
1926 				continue;
1927 			}
1928 			if (IS_LEASE(fl)) {
1929 				lease_modify(before, F_UNLCK);
1930 				continue;
1931 			}
1932 			/* What? */
1933 			BUG();
1934  		}
1935 		before = &fl->fl_next;
1936 	}
1937 	unlock_kernel();
1938 }
1939 
1940 /**
1941  *	posix_block_lock - blocks waiting for a file lock
1942  *	@blocker: the lock which is blocking
1943  *	@waiter: the lock which conflicts and has to wait
1944  *
1945  * lockd needs to block waiting for locks.
1946  */
1947 void
1948 posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
1949 {
1950 	locks_insert_block(blocker, waiter);
1951 }
1952 
1953 EXPORT_SYMBOL(posix_block_lock);
1954 
1955 /**
1956  *	posix_unblock_lock - stop waiting for a file lock
1957  *      @filp:   how the file was opened
1958  *	@waiter: the lock which was waiting
1959  *
1960  *	lockd needs to block waiting for locks.
1961  */
1962 void
1963 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1964 {
1965 	/*
1966 	 * A remote machine may cancel the lock request after it's been
1967 	 * granted locally.  If that happens, we need to delete the lock.
1968 	 */
1969 	lock_kernel();
1970 	if (waiter->fl_next) {
1971 		__locks_delete_block(waiter);
1972 		unlock_kernel();
1973 	} else {
1974 		unlock_kernel();
1975 		waiter->fl_type = F_UNLCK;
1976 		posix_lock_file(filp, waiter);
1977 	}
1978 }
1979 
1980 EXPORT_SYMBOL(posix_unblock_lock);
1981 
1982 static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1983 {
1984 	struct inode *inode = NULL;
1985 
1986 	if (fl->fl_file != NULL)
1987 		inode = fl->fl_file->f_dentry->d_inode;
1988 
1989 	out += sprintf(out, "%d:%s ", id, pfx);
1990 	if (IS_POSIX(fl)) {
1991 		out += sprintf(out, "%6s %s ",
1992 			     (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
1993 			     (inode == NULL) ? "*NOINODE*" :
1994 			     (IS_MANDLOCK(inode) &&
1995 			      (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
1996 			     "MANDATORY" : "ADVISORY ");
1997 	} else if (IS_FLOCK(fl)) {
1998 		if (fl->fl_type & LOCK_MAND) {
1999 			out += sprintf(out, "FLOCK  MSNFS     ");
2000 		} else {
2001 			out += sprintf(out, "FLOCK  ADVISORY  ");
2002 		}
2003 	} else if (IS_LEASE(fl)) {
2004 		out += sprintf(out, "LEASE  ");
2005 		if (fl->fl_type & F_INPROGRESS)
2006 			out += sprintf(out, "BREAKING  ");
2007 		else if (fl->fl_file)
2008 			out += sprintf(out, "ACTIVE    ");
2009 		else
2010 			out += sprintf(out, "BREAKER   ");
2011 	} else {
2012 		out += sprintf(out, "UNKNOWN UNKNOWN  ");
2013 	}
2014 	if (fl->fl_type & LOCK_MAND) {
2015 		out += sprintf(out, "%s ",
2016 			       (fl->fl_type & LOCK_READ)
2017 			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2018 			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2019 	} else {
2020 		out += sprintf(out, "%s ",
2021 			       (fl->fl_type & F_INPROGRESS)
2022 			       ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2023 			       : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2024 	}
2025 	if (inode) {
2026 #ifdef WE_CAN_BREAK_LSLK_NOW
2027 		out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2028 				inode->i_sb->s_id, inode->i_ino);
2029 #else
2030 		/* userspace relies on this representation of dev_t ;-( */
2031 		out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2032 				MAJOR(inode->i_sb->s_dev),
2033 				MINOR(inode->i_sb->s_dev), inode->i_ino);
2034 #endif
2035 	} else {
2036 		out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2037 	}
2038 	if (IS_POSIX(fl)) {
2039 		if (fl->fl_end == OFFSET_MAX)
2040 			out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2041 		else
2042 			out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2043 					fl->fl_end);
2044 	} else {
2045 		out += sprintf(out, "0 EOF\n");
2046 	}
2047 }
2048 
2049 static void move_lock_status(char **p, off_t* pos, off_t offset)
2050 {
2051 	int len;
2052 	len = strlen(*p);
2053 	if(*pos >= offset) {
2054 		/* the complete line is valid */
2055 		*p += len;
2056 		*pos += len;
2057 		return;
2058 	}
2059 	if(*pos+len > offset) {
2060 		/* use the second part of the line */
2061 		int i = offset-*pos;
2062 		memmove(*p,*p+i,len-i);
2063 		*p += len-i;
2064 		*pos += len;
2065 		return;
2066 	}
2067 	/* discard the complete line */
2068 	*pos += len;
2069 }
2070 
2071 /**
2072  *	get_locks_status	-	reports lock usage in /proc/locks
2073  *	@buffer: address in userspace to write into
2074  *	@start: ?
2075  *	@offset: how far we are through the buffer
2076  *	@length: how much to read
2077  */
2078 
2079 int get_locks_status(char *buffer, char **start, off_t offset, int length)
2080 {
2081 	struct list_head *tmp;
2082 	char *q = buffer;
2083 	off_t pos = 0;
2084 	int i = 0;
2085 
2086 	lock_kernel();
2087 	list_for_each(tmp, &file_lock_list) {
2088 		struct list_head *btmp;
2089 		struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2090 		lock_get_status(q, fl, ++i, "");
2091 		move_lock_status(&q, &pos, offset);
2092 
2093 		if(pos >= offset+length)
2094 			goto done;
2095 
2096 		list_for_each(btmp, &fl->fl_block) {
2097 			struct file_lock *bfl = list_entry(btmp,
2098 					struct file_lock, fl_block);
2099 			lock_get_status(q, bfl, i, " ->");
2100 			move_lock_status(&q, &pos, offset);
2101 
2102 			if(pos >= offset+length)
2103 				goto done;
2104 		}
2105 	}
2106 done:
2107 	unlock_kernel();
2108 	*start = buffer;
2109 	if(q-buffer < length)
2110 		return (q-buffer);
2111 	return length;
2112 }
2113 
2114 /**
2115  *	lock_may_read - checks that the region is free of locks
2116  *	@inode: the inode that is being read
2117  *	@start: the first byte to read
2118  *	@len: the number of bytes to read
2119  *
2120  *	Emulates Windows locking requirements.  Whole-file
2121  *	mandatory locks (share modes) can prohibit a read and
2122  *	byte-range POSIX locks can prohibit a read if they overlap.
2123  *
2124  *	N.B. this function is only ever called
2125  *	from knfsd and ownership of locks is never checked.
2126  */
2127 int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2128 {
2129 	struct file_lock *fl;
2130 	int result = 1;
2131 	lock_kernel();
2132 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2133 		if (IS_POSIX(fl)) {
2134 			if (fl->fl_type == F_RDLCK)
2135 				continue;
2136 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2137 				continue;
2138 		} else if (IS_FLOCK(fl)) {
2139 			if (!(fl->fl_type & LOCK_MAND))
2140 				continue;
2141 			if (fl->fl_type & LOCK_READ)
2142 				continue;
2143 		} else
2144 			continue;
2145 		result = 0;
2146 		break;
2147 	}
2148 	unlock_kernel();
2149 	return result;
2150 }
2151 
2152 EXPORT_SYMBOL(lock_may_read);
2153 
2154 /**
2155  *	lock_may_write - checks that the region is free of locks
2156  *	@inode: the inode that is being written
2157  *	@start: the first byte to write
2158  *	@len: the number of bytes to write
2159  *
2160  *	Emulates Windows locking requirements.  Whole-file
2161  *	mandatory locks (share modes) can prohibit a write and
2162  *	byte-range POSIX locks can prohibit a write if they overlap.
2163  *
2164  *	N.B. this function is only ever called
2165  *	from knfsd and ownership of locks is never checked.
2166  */
2167 int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2168 {
2169 	struct file_lock *fl;
2170 	int result = 1;
2171 	lock_kernel();
2172 	for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2173 		if (IS_POSIX(fl)) {
2174 			if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2175 				continue;
2176 		} else if (IS_FLOCK(fl)) {
2177 			if (!(fl->fl_type & LOCK_MAND))
2178 				continue;
2179 			if (fl->fl_type & LOCK_WRITE)
2180 				continue;
2181 		} else
2182 			continue;
2183 		result = 0;
2184 		break;
2185 	}
2186 	unlock_kernel();
2187 	return result;
2188 }
2189 
2190 EXPORT_SYMBOL(lock_may_write);
2191 
2192 static inline void __steal_locks(struct file *file, fl_owner_t from)
2193 {
2194 	struct inode *inode = file->f_dentry->d_inode;
2195 	struct file_lock *fl = inode->i_flock;
2196 
2197 	while (fl) {
2198 		if (fl->fl_file == file && fl->fl_owner == from)
2199 			fl->fl_owner = current->files;
2200 		fl = fl->fl_next;
2201 	}
2202 }
2203 
2204 /* When getting ready for executing a binary, we make sure that current
2205  * has a files_struct on its own. Before dropping the old files_struct,
2206  * we take over ownership of all locks for all file descriptors we own.
2207  * Note that we may accidentally steal a lock for a file that a sibling
2208  * has created since the unshare_files() call.
2209  */
2210 void steal_locks(fl_owner_t from)
2211 {
2212 	struct files_struct *files = current->files;
2213 	int i, j;
2214 	struct fdtable *fdt;
2215 
2216 	if (from == files)
2217 		return;
2218 
2219 	lock_kernel();
2220 	j = 0;
2221 	rcu_read_lock();
2222 	fdt = files_fdtable(files);
2223 	for (;;) {
2224 		unsigned long set;
2225 		i = j * __NFDBITS;
2226 		if (i >= fdt->max_fdset || i >= fdt->max_fds)
2227 			break;
2228 		set = fdt->open_fds->fds_bits[j++];
2229 		while (set) {
2230 			if (set & 1) {
2231 				struct file *file = fdt->fd[i];
2232 				if (file)
2233 					__steal_locks(file, from);
2234 			}
2235 			i++;
2236 			set >>= 1;
2237 		}
2238 	}
2239 	rcu_read_unlock();
2240 	unlock_kernel();
2241 }
2242 EXPORT_SYMBOL(steal_locks);
2243 
2244 static int __init filelock_init(void)
2245 {
2246 	filelock_cache = kmem_cache_create("file_lock_cache",
2247 			sizeof(struct file_lock), 0, SLAB_PANIC,
2248 			init_once, NULL);
2249 	return 0;
2250 }
2251 
2252 core_initcall(filelock_init);
2253