xref: /openbmc/linux/fs/locks.c (revision 4f205687)
1 /*
2  *  linux/fs/locks.c
3  *
4  *  Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5  *  Doug Evans (dje@spiff.uucp), August 07, 1992
6  *
7  *  Deadlock detection added.
8  *  FIXME: one thing isn't handled yet:
9  *	- mandatory locks (requires lots of changes elsewhere)
10  *  Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11  *
12  *  Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13  *  Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14  *
15  *  Converted file_lock_table to a linked list from an array, which eliminates
16  *  the limits on how many active file locks are open.
17  *  Chad Page (pageone@netcom.com), November 27, 1994
18  *
19  *  Removed dependency on file descriptors. dup()'ed file descriptors now
20  *  get the same locks as the original file descriptors, and a close() on
21  *  any file descriptor removes ALL the locks on the file for the current
22  *  process. Since locks still depend on the process id, locks are inherited
23  *  after an exec() but not after a fork(). This agrees with POSIX, and both
24  *  BSD and SVR4 practice.
25  *  Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26  *
27  *  Scrapped free list which is redundant now that we allocate locks
28  *  dynamically with kmalloc()/kfree().
29  *  Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30  *
31  *  Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32  *
33  *  FL_POSIX locks are created with calls to fcntl() and lockf() through the
34  *  fcntl() system call. They have the semantics described above.
35  *
36  *  FL_FLOCK locks are created with calls to flock(), through the flock()
37  *  system call, which is new. Old C libraries implement flock() via fcntl()
38  *  and will continue to use the old, broken implementation.
39  *
40  *  FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41  *  with a file pointer (filp). As a result they can be shared by a parent
42  *  process and its children after a fork(). They are removed when the last
43  *  file descriptor referring to the file pointer is closed (unless explicitly
44  *  unlocked).
45  *
46  *  FL_FLOCK locks never deadlock, an existing lock is always removed before
47  *  upgrading from shared to exclusive (or vice versa). When this happens
48  *  any processes blocked by the current lock are woken up and allowed to
49  *  run before the new lock is applied.
50  *  Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51  *
52  *  Removed some race conditions in flock_lock_file(), marked other possible
53  *  races. Just grep for FIXME to see them.
54  *  Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55  *
56  *  Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57  *  Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58  *  once we've checked for blocking and deadlocking.
59  *  Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60  *
61  *  Initial implementation of mandatory locks. SunOS turned out to be
62  *  a rotten model, so I implemented the "obvious" semantics.
63  *  See 'Documentation/filesystems/mandatory-locking.txt' for details.
64  *  Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65  *
66  *  Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67  *  check if a file has mandatory locks, used by mmap(), open() and creat() to
68  *  see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69  *  Manual, Section 2.
70  *  Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71  *
72  *  Tidied up block list handling. Added '/proc/locks' interface.
73  *  Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74  *
75  *  Fixed deadlock condition for pathological code that mixes calls to
76  *  flock() and fcntl().
77  *  Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78  *
79  *  Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80  *  for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81  *  guarantee sensible behaviour in the case where file system modules might
82  *  be compiled with different options than the kernel itself.
83  *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84  *
85  *  Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86  *  (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87  *  Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88  *
89  *  Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90  *  locks. Changed process synchronisation to avoid dereferencing locks that
91  *  have already been freed.
92  *  Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93  *
94  *  Made the block list a circular list to minimise searching in the list.
95  *  Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96  *
97  *  Made mandatory locking a mount option. Default is not to allow mandatory
98  *  locking.
99  *  Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100  *
101  *  Some adaptations for NFS support.
102  *  Olaf Kirch (okir@monad.swb.de), Dec 1996,
103  *
104  *  Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105  *  Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106  *
107  *  Use slab allocator instead of kmalloc/kfree.
108  *  Use generic list implementation from <linux/list.h>.
109  *  Sped up posix_locks_deadlock by only considering blocked locks.
110  *  Matthew Wilcox <willy@debian.org>, March, 2000.
111  *
112  *  Leases and LOCK_MAND
113  *  Matthew Wilcox <willy@debian.org>, June, 2000.
114  *  Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115  */
116 
117 #include <linux/capability.h>
118 #include <linux/file.h>
119 #include <linux/fdtable.h>
120 #include <linux/fs.h>
121 #include <linux/init.h>
122 #include <linux/security.h>
123 #include <linux/slab.h>
124 #include <linux/syscalls.h>
125 #include <linux/time.h>
126 #include <linux/rcupdate.h>
127 #include <linux/pid_namespace.h>
128 #include <linux/hashtable.h>
129 #include <linux/percpu.h>
130 #include <linux/lglock.h>
131 
132 #define CREATE_TRACE_POINTS
133 #include <trace/events/filelock.h>
134 
135 #include <asm/uaccess.h>
136 
137 #define IS_POSIX(fl)	(fl->fl_flags & FL_POSIX)
138 #define IS_FLOCK(fl)	(fl->fl_flags & FL_FLOCK)
139 #define IS_LEASE(fl)	(fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
140 #define IS_OFDLCK(fl)	(fl->fl_flags & FL_OFDLCK)
141 
142 static bool lease_breaking(struct file_lock *fl)
143 {
144 	return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
145 }
146 
147 static int target_leasetype(struct file_lock *fl)
148 {
149 	if (fl->fl_flags & FL_UNLOCK_PENDING)
150 		return F_UNLCK;
151 	if (fl->fl_flags & FL_DOWNGRADE_PENDING)
152 		return F_RDLCK;
153 	return fl->fl_type;
154 }
155 
156 int leases_enable = 1;
157 int lease_break_time = 45;
158 
159 /*
160  * The global file_lock_list is only used for displaying /proc/locks, so we
161  * keep a list on each CPU, with each list protected by its own spinlock via
162  * the file_lock_lglock. Note that alterations to the list also require that
163  * the relevant flc_lock is held.
164  */
165 DEFINE_STATIC_LGLOCK(file_lock_lglock);
166 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
167 
168 /*
169  * The blocked_hash is used to find POSIX lock loops for deadlock detection.
170  * It is protected by blocked_lock_lock.
171  *
172  * We hash locks by lockowner in order to optimize searching for the lock a
173  * particular lockowner is waiting on.
174  *
175  * FIXME: make this value scale via some heuristic? We generally will want more
176  * buckets when we have more lockowners holding locks, but that's a little
177  * difficult to determine without knowing what the workload will look like.
178  */
179 #define BLOCKED_HASH_BITS	7
180 static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
181 
182 /*
183  * This lock protects the blocked_hash. Generally, if you're accessing it, you
184  * want to be holding this lock.
185  *
186  * In addition, it also protects the fl->fl_block list, and the fl->fl_next
187  * pointer for file_lock structures that are acting as lock requests (in
188  * contrast to those that are acting as records of acquired locks).
189  *
190  * Note that when we acquire this lock in order to change the above fields,
191  * we often hold the flc_lock as well. In certain cases, when reading the fields
192  * protected by this lock, we can skip acquiring it iff we already hold the
193  * flc_lock.
194  *
195  * In particular, adding an entry to the fl_block list requires that you hold
196  * both the flc_lock and the blocked_lock_lock (acquired in that order).
197  * Deleting an entry from the list however only requires the file_lock_lock.
198  */
199 static DEFINE_SPINLOCK(blocked_lock_lock);
200 
201 static struct kmem_cache *flctx_cache __read_mostly;
202 static struct kmem_cache *filelock_cache __read_mostly;
203 
204 static struct file_lock_context *
205 locks_get_lock_context(struct inode *inode, int type)
206 {
207 	struct file_lock_context *ctx;
208 
209 	/* paired with cmpxchg() below */
210 	ctx = smp_load_acquire(&inode->i_flctx);
211 	if (likely(ctx) || type == F_UNLCK)
212 		goto out;
213 
214 	ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
215 	if (!ctx)
216 		goto out;
217 
218 	spin_lock_init(&ctx->flc_lock);
219 	INIT_LIST_HEAD(&ctx->flc_flock);
220 	INIT_LIST_HEAD(&ctx->flc_posix);
221 	INIT_LIST_HEAD(&ctx->flc_lease);
222 
223 	/*
224 	 * Assign the pointer if it's not already assigned. If it is, then
225 	 * free the context we just allocated.
226 	 */
227 	if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
228 		kmem_cache_free(flctx_cache, ctx);
229 		ctx = smp_load_acquire(&inode->i_flctx);
230 	}
231 out:
232 	trace_locks_get_lock_context(inode, type, ctx);
233 	return ctx;
234 }
235 
236 static void
237 locks_dump_ctx_list(struct list_head *list, char *list_type)
238 {
239 	struct file_lock *fl;
240 
241 	list_for_each_entry(fl, list, fl_list) {
242 		pr_warn("%s: fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n", list_type, fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
243 	}
244 }
245 
246 static void
247 locks_check_ctx_lists(struct inode *inode)
248 {
249 	struct file_lock_context *ctx = inode->i_flctx;
250 
251 	if (unlikely(!list_empty(&ctx->flc_flock) ||
252 		     !list_empty(&ctx->flc_posix) ||
253 		     !list_empty(&ctx->flc_lease))) {
254 		pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
255 			MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
256 			inode->i_ino);
257 		locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
258 		locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
259 		locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
260 	}
261 }
262 
263 void
264 locks_free_lock_context(struct inode *inode)
265 {
266 	struct file_lock_context *ctx = inode->i_flctx;
267 
268 	if (unlikely(ctx)) {
269 		locks_check_ctx_lists(inode);
270 		kmem_cache_free(flctx_cache, ctx);
271 	}
272 }
273 
274 static void locks_init_lock_heads(struct file_lock *fl)
275 {
276 	INIT_HLIST_NODE(&fl->fl_link);
277 	INIT_LIST_HEAD(&fl->fl_list);
278 	INIT_LIST_HEAD(&fl->fl_block);
279 	init_waitqueue_head(&fl->fl_wait);
280 }
281 
282 /* Allocate an empty lock structure. */
283 struct file_lock *locks_alloc_lock(void)
284 {
285 	struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
286 
287 	if (fl)
288 		locks_init_lock_heads(fl);
289 
290 	return fl;
291 }
292 EXPORT_SYMBOL_GPL(locks_alloc_lock);
293 
294 void locks_release_private(struct file_lock *fl)
295 {
296 	if (fl->fl_ops) {
297 		if (fl->fl_ops->fl_release_private)
298 			fl->fl_ops->fl_release_private(fl);
299 		fl->fl_ops = NULL;
300 	}
301 
302 	if (fl->fl_lmops) {
303 		if (fl->fl_lmops->lm_put_owner) {
304 			fl->fl_lmops->lm_put_owner(fl->fl_owner);
305 			fl->fl_owner = NULL;
306 		}
307 		fl->fl_lmops = NULL;
308 	}
309 }
310 EXPORT_SYMBOL_GPL(locks_release_private);
311 
312 /* Free a lock which is not in use. */
313 void locks_free_lock(struct file_lock *fl)
314 {
315 	BUG_ON(waitqueue_active(&fl->fl_wait));
316 	BUG_ON(!list_empty(&fl->fl_list));
317 	BUG_ON(!list_empty(&fl->fl_block));
318 	BUG_ON(!hlist_unhashed(&fl->fl_link));
319 
320 	locks_release_private(fl);
321 	kmem_cache_free(filelock_cache, fl);
322 }
323 EXPORT_SYMBOL(locks_free_lock);
324 
325 static void
326 locks_dispose_list(struct list_head *dispose)
327 {
328 	struct file_lock *fl;
329 
330 	while (!list_empty(dispose)) {
331 		fl = list_first_entry(dispose, struct file_lock, fl_list);
332 		list_del_init(&fl->fl_list);
333 		locks_free_lock(fl);
334 	}
335 }
336 
337 void locks_init_lock(struct file_lock *fl)
338 {
339 	memset(fl, 0, sizeof(struct file_lock));
340 	locks_init_lock_heads(fl);
341 }
342 
343 EXPORT_SYMBOL(locks_init_lock);
344 
345 /*
346  * Initialize a new lock from an existing file_lock structure.
347  */
348 void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
349 {
350 	new->fl_owner = fl->fl_owner;
351 	new->fl_pid = fl->fl_pid;
352 	new->fl_file = NULL;
353 	new->fl_flags = fl->fl_flags;
354 	new->fl_type = fl->fl_type;
355 	new->fl_start = fl->fl_start;
356 	new->fl_end = fl->fl_end;
357 	new->fl_lmops = fl->fl_lmops;
358 	new->fl_ops = NULL;
359 
360 	if (fl->fl_lmops) {
361 		if (fl->fl_lmops->lm_get_owner)
362 			fl->fl_lmops->lm_get_owner(fl->fl_owner);
363 	}
364 }
365 EXPORT_SYMBOL(locks_copy_conflock);
366 
367 void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
368 {
369 	/* "new" must be a freshly-initialized lock */
370 	WARN_ON_ONCE(new->fl_ops);
371 
372 	locks_copy_conflock(new, fl);
373 
374 	new->fl_file = fl->fl_file;
375 	new->fl_ops = fl->fl_ops;
376 
377 	if (fl->fl_ops) {
378 		if (fl->fl_ops->fl_copy_lock)
379 			fl->fl_ops->fl_copy_lock(new, fl);
380 	}
381 }
382 
383 EXPORT_SYMBOL(locks_copy_lock);
384 
385 static inline int flock_translate_cmd(int cmd) {
386 	if (cmd & LOCK_MAND)
387 		return cmd & (LOCK_MAND | LOCK_RW);
388 	switch (cmd) {
389 	case LOCK_SH:
390 		return F_RDLCK;
391 	case LOCK_EX:
392 		return F_WRLCK;
393 	case LOCK_UN:
394 		return F_UNLCK;
395 	}
396 	return -EINVAL;
397 }
398 
399 /* Fill in a file_lock structure with an appropriate FLOCK lock. */
400 static struct file_lock *
401 flock_make_lock(struct file *filp, unsigned int cmd)
402 {
403 	struct file_lock *fl;
404 	int type = flock_translate_cmd(cmd);
405 
406 	if (type < 0)
407 		return ERR_PTR(type);
408 
409 	fl = locks_alloc_lock();
410 	if (fl == NULL)
411 		return ERR_PTR(-ENOMEM);
412 
413 	fl->fl_file = filp;
414 	fl->fl_owner = filp;
415 	fl->fl_pid = current->tgid;
416 	fl->fl_flags = FL_FLOCK;
417 	fl->fl_type = type;
418 	fl->fl_end = OFFSET_MAX;
419 
420 	return fl;
421 }
422 
423 static int assign_type(struct file_lock *fl, long type)
424 {
425 	switch (type) {
426 	case F_RDLCK:
427 	case F_WRLCK:
428 	case F_UNLCK:
429 		fl->fl_type = type;
430 		break;
431 	default:
432 		return -EINVAL;
433 	}
434 	return 0;
435 }
436 
437 static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
438 				 struct flock64 *l)
439 {
440 	switch (l->l_whence) {
441 	case SEEK_SET:
442 		fl->fl_start = 0;
443 		break;
444 	case SEEK_CUR:
445 		fl->fl_start = filp->f_pos;
446 		break;
447 	case SEEK_END:
448 		fl->fl_start = i_size_read(file_inode(filp));
449 		break;
450 	default:
451 		return -EINVAL;
452 	}
453 	if (l->l_start > OFFSET_MAX - fl->fl_start)
454 		return -EOVERFLOW;
455 	fl->fl_start += l->l_start;
456 	if (fl->fl_start < 0)
457 		return -EINVAL;
458 
459 	/* POSIX-1996 leaves the case l->l_len < 0 undefined;
460 	   POSIX-2001 defines it. */
461 	if (l->l_len > 0) {
462 		if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
463 			return -EOVERFLOW;
464 		fl->fl_end = fl->fl_start + l->l_len - 1;
465 
466 	} else if (l->l_len < 0) {
467 		if (fl->fl_start + l->l_len < 0)
468 			return -EINVAL;
469 		fl->fl_end = fl->fl_start - 1;
470 		fl->fl_start += l->l_len;
471 	} else
472 		fl->fl_end = OFFSET_MAX;
473 
474 	fl->fl_owner = current->files;
475 	fl->fl_pid = current->tgid;
476 	fl->fl_file = filp;
477 	fl->fl_flags = FL_POSIX;
478 	fl->fl_ops = NULL;
479 	fl->fl_lmops = NULL;
480 
481 	return assign_type(fl, l->l_type);
482 }
483 
484 /* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
485  * style lock.
486  */
487 static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
488 			       struct flock *l)
489 {
490 	struct flock64 ll = {
491 		.l_type = l->l_type,
492 		.l_whence = l->l_whence,
493 		.l_start = l->l_start,
494 		.l_len = l->l_len,
495 	};
496 
497 	return flock64_to_posix_lock(filp, fl, &ll);
498 }
499 
500 /* default lease lock manager operations */
501 static bool
502 lease_break_callback(struct file_lock *fl)
503 {
504 	kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
505 	return false;
506 }
507 
508 static void
509 lease_setup(struct file_lock *fl, void **priv)
510 {
511 	struct file *filp = fl->fl_file;
512 	struct fasync_struct *fa = *priv;
513 
514 	/*
515 	 * fasync_insert_entry() returns the old entry if any. If there was no
516 	 * old entry, then it used "priv" and inserted it into the fasync list.
517 	 * Clear the pointer to indicate that it shouldn't be freed.
518 	 */
519 	if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
520 		*priv = NULL;
521 
522 	__f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
523 }
524 
525 static const struct lock_manager_operations lease_manager_ops = {
526 	.lm_break = lease_break_callback,
527 	.lm_change = lease_modify,
528 	.lm_setup = lease_setup,
529 };
530 
531 /*
532  * Initialize a lease, use the default lock manager operations
533  */
534 static int lease_init(struct file *filp, long type, struct file_lock *fl)
535  {
536 	if (assign_type(fl, type) != 0)
537 		return -EINVAL;
538 
539 	fl->fl_owner = filp;
540 	fl->fl_pid = current->tgid;
541 
542 	fl->fl_file = filp;
543 	fl->fl_flags = FL_LEASE;
544 	fl->fl_start = 0;
545 	fl->fl_end = OFFSET_MAX;
546 	fl->fl_ops = NULL;
547 	fl->fl_lmops = &lease_manager_ops;
548 	return 0;
549 }
550 
551 /* Allocate a file_lock initialised to this type of lease */
552 static struct file_lock *lease_alloc(struct file *filp, long type)
553 {
554 	struct file_lock *fl = locks_alloc_lock();
555 	int error = -ENOMEM;
556 
557 	if (fl == NULL)
558 		return ERR_PTR(error);
559 
560 	error = lease_init(filp, type, fl);
561 	if (error) {
562 		locks_free_lock(fl);
563 		return ERR_PTR(error);
564 	}
565 	return fl;
566 }
567 
568 /* Check if two locks overlap each other.
569  */
570 static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
571 {
572 	return ((fl1->fl_end >= fl2->fl_start) &&
573 		(fl2->fl_end >= fl1->fl_start));
574 }
575 
576 /*
577  * Check whether two locks have the same owner.
578  */
579 static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
580 {
581 	if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
582 		return fl2->fl_lmops == fl1->fl_lmops &&
583 			fl1->fl_lmops->lm_compare_owner(fl1, fl2);
584 	return fl1->fl_owner == fl2->fl_owner;
585 }
586 
587 /* Must be called with the flc_lock held! */
588 static void locks_insert_global_locks(struct file_lock *fl)
589 {
590 	lg_local_lock(&file_lock_lglock);
591 	fl->fl_link_cpu = smp_processor_id();
592 	hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
593 	lg_local_unlock(&file_lock_lglock);
594 }
595 
596 /* Must be called with the flc_lock held! */
597 static void locks_delete_global_locks(struct file_lock *fl)
598 {
599 	/*
600 	 * Avoid taking lock if already unhashed. This is safe since this check
601 	 * is done while holding the flc_lock, and new insertions into the list
602 	 * also require that it be held.
603 	 */
604 	if (hlist_unhashed(&fl->fl_link))
605 		return;
606 	lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
607 	hlist_del_init(&fl->fl_link);
608 	lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
609 }
610 
611 static unsigned long
612 posix_owner_key(struct file_lock *fl)
613 {
614 	if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
615 		return fl->fl_lmops->lm_owner_key(fl);
616 	return (unsigned long)fl->fl_owner;
617 }
618 
619 static void locks_insert_global_blocked(struct file_lock *waiter)
620 {
621 	lockdep_assert_held(&blocked_lock_lock);
622 
623 	hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
624 }
625 
626 static void locks_delete_global_blocked(struct file_lock *waiter)
627 {
628 	lockdep_assert_held(&blocked_lock_lock);
629 
630 	hash_del(&waiter->fl_link);
631 }
632 
633 /* Remove waiter from blocker's block list.
634  * When blocker ends up pointing to itself then the list is empty.
635  *
636  * Must be called with blocked_lock_lock held.
637  */
638 static void __locks_delete_block(struct file_lock *waiter)
639 {
640 	locks_delete_global_blocked(waiter);
641 	list_del_init(&waiter->fl_block);
642 	waiter->fl_next = NULL;
643 }
644 
645 static void locks_delete_block(struct file_lock *waiter)
646 {
647 	spin_lock(&blocked_lock_lock);
648 	__locks_delete_block(waiter);
649 	spin_unlock(&blocked_lock_lock);
650 }
651 
652 /* Insert waiter into blocker's block list.
653  * We use a circular list so that processes can be easily woken up in
654  * the order they blocked. The documentation doesn't require this but
655  * it seems like the reasonable thing to do.
656  *
657  * Must be called with both the flc_lock and blocked_lock_lock held. The
658  * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
659  * that the flc_lock is also held on insertions we can avoid taking the
660  * blocked_lock_lock in some cases when we see that the fl_block list is empty.
661  */
662 static void __locks_insert_block(struct file_lock *blocker,
663 					struct file_lock *waiter)
664 {
665 	BUG_ON(!list_empty(&waiter->fl_block));
666 	waiter->fl_next = blocker;
667 	list_add_tail(&waiter->fl_block, &blocker->fl_block);
668 	if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
669 		locks_insert_global_blocked(waiter);
670 }
671 
672 /* Must be called with flc_lock held. */
673 static void locks_insert_block(struct file_lock *blocker,
674 					struct file_lock *waiter)
675 {
676 	spin_lock(&blocked_lock_lock);
677 	__locks_insert_block(blocker, waiter);
678 	spin_unlock(&blocked_lock_lock);
679 }
680 
681 /*
682  * Wake up processes blocked waiting for blocker.
683  *
684  * Must be called with the inode->flc_lock held!
685  */
686 static void locks_wake_up_blocks(struct file_lock *blocker)
687 {
688 	/*
689 	 * Avoid taking global lock if list is empty. This is safe since new
690 	 * blocked requests are only added to the list under the flc_lock, and
691 	 * the flc_lock is always held here. Note that removal from the fl_block
692 	 * list does not require the flc_lock, so we must recheck list_empty()
693 	 * after acquiring the blocked_lock_lock.
694 	 */
695 	if (list_empty(&blocker->fl_block))
696 		return;
697 
698 	spin_lock(&blocked_lock_lock);
699 	while (!list_empty(&blocker->fl_block)) {
700 		struct file_lock *waiter;
701 
702 		waiter = list_first_entry(&blocker->fl_block,
703 				struct file_lock, fl_block);
704 		__locks_delete_block(waiter);
705 		if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
706 			waiter->fl_lmops->lm_notify(waiter);
707 		else
708 			wake_up(&waiter->fl_wait);
709 	}
710 	spin_unlock(&blocked_lock_lock);
711 }
712 
713 static void
714 locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
715 {
716 	fl->fl_nspid = get_pid(task_tgid(current));
717 	list_add_tail(&fl->fl_list, before);
718 	locks_insert_global_locks(fl);
719 }
720 
721 static void
722 locks_unlink_lock_ctx(struct file_lock *fl)
723 {
724 	locks_delete_global_locks(fl);
725 	list_del_init(&fl->fl_list);
726 	if (fl->fl_nspid) {
727 		put_pid(fl->fl_nspid);
728 		fl->fl_nspid = NULL;
729 	}
730 	locks_wake_up_blocks(fl);
731 }
732 
733 static void
734 locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
735 {
736 	locks_unlink_lock_ctx(fl);
737 	if (dispose)
738 		list_add(&fl->fl_list, dispose);
739 	else
740 		locks_free_lock(fl);
741 }
742 
743 /* Determine if lock sys_fl blocks lock caller_fl. Common functionality
744  * checks for shared/exclusive status of overlapping locks.
745  */
746 static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
747 {
748 	if (sys_fl->fl_type == F_WRLCK)
749 		return 1;
750 	if (caller_fl->fl_type == F_WRLCK)
751 		return 1;
752 	return 0;
753 }
754 
755 /* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
756  * checking before calling the locks_conflict().
757  */
758 static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
759 {
760 	/* POSIX locks owned by the same process do not conflict with
761 	 * each other.
762 	 */
763 	if (posix_same_owner(caller_fl, sys_fl))
764 		return (0);
765 
766 	/* Check whether they overlap */
767 	if (!locks_overlap(caller_fl, sys_fl))
768 		return 0;
769 
770 	return (locks_conflict(caller_fl, sys_fl));
771 }
772 
773 /* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
774  * checking before calling the locks_conflict().
775  */
776 static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
777 {
778 	/* FLOCK locks referring to the same filp do not conflict with
779 	 * each other.
780 	 */
781 	if (caller_fl->fl_file == sys_fl->fl_file)
782 		return (0);
783 	if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
784 		return 0;
785 
786 	return (locks_conflict(caller_fl, sys_fl));
787 }
788 
789 void
790 posix_test_lock(struct file *filp, struct file_lock *fl)
791 {
792 	struct file_lock *cfl;
793 	struct file_lock_context *ctx;
794 	struct inode *inode = file_inode(filp);
795 
796 	ctx = smp_load_acquire(&inode->i_flctx);
797 	if (!ctx || list_empty_careful(&ctx->flc_posix)) {
798 		fl->fl_type = F_UNLCK;
799 		return;
800 	}
801 
802 	spin_lock(&ctx->flc_lock);
803 	list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
804 		if (posix_locks_conflict(fl, cfl)) {
805 			locks_copy_conflock(fl, cfl);
806 			if (cfl->fl_nspid)
807 				fl->fl_pid = pid_vnr(cfl->fl_nspid);
808 			goto out;
809 		}
810 	}
811 	fl->fl_type = F_UNLCK;
812 out:
813 	spin_unlock(&ctx->flc_lock);
814 	return;
815 }
816 EXPORT_SYMBOL(posix_test_lock);
817 
818 /*
819  * Deadlock detection:
820  *
821  * We attempt to detect deadlocks that are due purely to posix file
822  * locks.
823  *
824  * We assume that a task can be waiting for at most one lock at a time.
825  * So for any acquired lock, the process holding that lock may be
826  * waiting on at most one other lock.  That lock in turns may be held by
827  * someone waiting for at most one other lock.  Given a requested lock
828  * caller_fl which is about to wait for a conflicting lock block_fl, we
829  * follow this chain of waiters to ensure we are not about to create a
830  * cycle.
831  *
832  * Since we do this before we ever put a process to sleep on a lock, we
833  * are ensured that there is never a cycle; that is what guarantees that
834  * the while() loop in posix_locks_deadlock() eventually completes.
835  *
836  * Note: the above assumption may not be true when handling lock
837  * requests from a broken NFS client. It may also fail in the presence
838  * of tasks (such as posix threads) sharing the same open file table.
839  * To handle those cases, we just bail out after a few iterations.
840  *
841  * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
842  * Because the owner is not even nominally tied to a thread of
843  * execution, the deadlock detection below can't reasonably work well. Just
844  * skip it for those.
845  *
846  * In principle, we could do a more limited deadlock detection on FL_OFDLCK
847  * locks that just checks for the case where two tasks are attempting to
848  * upgrade from read to write locks on the same inode.
849  */
850 
851 #define MAX_DEADLK_ITERATIONS 10
852 
853 /* Find a lock that the owner of the given block_fl is blocking on. */
854 static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
855 {
856 	struct file_lock *fl;
857 
858 	hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
859 		if (posix_same_owner(fl, block_fl))
860 			return fl->fl_next;
861 	}
862 	return NULL;
863 }
864 
865 /* Must be called with the blocked_lock_lock held! */
866 static int posix_locks_deadlock(struct file_lock *caller_fl,
867 				struct file_lock *block_fl)
868 {
869 	int i = 0;
870 
871 	lockdep_assert_held(&blocked_lock_lock);
872 
873 	/*
874 	 * This deadlock detector can't reasonably detect deadlocks with
875 	 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
876 	 */
877 	if (IS_OFDLCK(caller_fl))
878 		return 0;
879 
880 	while ((block_fl = what_owner_is_waiting_for(block_fl))) {
881 		if (i++ > MAX_DEADLK_ITERATIONS)
882 			return 0;
883 		if (posix_same_owner(caller_fl, block_fl))
884 			return 1;
885 	}
886 	return 0;
887 }
888 
889 /* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
890  * after any leases, but before any posix locks.
891  *
892  * Note that if called with an FL_EXISTS argument, the caller may determine
893  * whether or not a lock was successfully freed by testing the return
894  * value for -ENOENT.
895  */
896 static int flock_lock_inode(struct inode *inode, struct file_lock *request)
897 {
898 	struct file_lock *new_fl = NULL;
899 	struct file_lock *fl;
900 	struct file_lock_context *ctx;
901 	int error = 0;
902 	bool found = false;
903 	LIST_HEAD(dispose);
904 
905 	ctx = locks_get_lock_context(inode, request->fl_type);
906 	if (!ctx) {
907 		if (request->fl_type != F_UNLCK)
908 			return -ENOMEM;
909 		return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
910 	}
911 
912 	if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
913 		new_fl = locks_alloc_lock();
914 		if (!new_fl)
915 			return -ENOMEM;
916 	}
917 
918 	spin_lock(&ctx->flc_lock);
919 	if (request->fl_flags & FL_ACCESS)
920 		goto find_conflict;
921 
922 	list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
923 		if (request->fl_file != fl->fl_file)
924 			continue;
925 		if (request->fl_type == fl->fl_type)
926 			goto out;
927 		found = true;
928 		locks_delete_lock_ctx(fl, &dispose);
929 		break;
930 	}
931 
932 	if (request->fl_type == F_UNLCK) {
933 		if ((request->fl_flags & FL_EXISTS) && !found)
934 			error = -ENOENT;
935 		goto out;
936 	}
937 
938 find_conflict:
939 	list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
940 		if (!flock_locks_conflict(request, fl))
941 			continue;
942 		error = -EAGAIN;
943 		if (!(request->fl_flags & FL_SLEEP))
944 			goto out;
945 		error = FILE_LOCK_DEFERRED;
946 		locks_insert_block(fl, request);
947 		goto out;
948 	}
949 	if (request->fl_flags & FL_ACCESS)
950 		goto out;
951 	locks_copy_lock(new_fl, request);
952 	locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
953 	new_fl = NULL;
954 	error = 0;
955 
956 out:
957 	spin_unlock(&ctx->flc_lock);
958 	if (new_fl)
959 		locks_free_lock(new_fl);
960 	locks_dispose_list(&dispose);
961 	return error;
962 }
963 
964 static int posix_lock_inode(struct inode *inode, struct file_lock *request,
965 			    struct file_lock *conflock)
966 {
967 	struct file_lock *fl, *tmp;
968 	struct file_lock *new_fl = NULL;
969 	struct file_lock *new_fl2 = NULL;
970 	struct file_lock *left = NULL;
971 	struct file_lock *right = NULL;
972 	struct file_lock_context *ctx;
973 	int error;
974 	bool added = false;
975 	LIST_HEAD(dispose);
976 
977 	ctx = locks_get_lock_context(inode, request->fl_type);
978 	if (!ctx)
979 		return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
980 
981 	/*
982 	 * We may need two file_lock structures for this operation,
983 	 * so we get them in advance to avoid races.
984 	 *
985 	 * In some cases we can be sure, that no new locks will be needed
986 	 */
987 	if (!(request->fl_flags & FL_ACCESS) &&
988 	    (request->fl_type != F_UNLCK ||
989 	     request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
990 		new_fl = locks_alloc_lock();
991 		new_fl2 = locks_alloc_lock();
992 	}
993 
994 	spin_lock(&ctx->flc_lock);
995 	/*
996 	 * New lock request. Walk all POSIX locks and look for conflicts. If
997 	 * there are any, either return error or put the request on the
998 	 * blocker's list of waiters and the global blocked_hash.
999 	 */
1000 	if (request->fl_type != F_UNLCK) {
1001 		list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1002 			if (!posix_locks_conflict(request, fl))
1003 				continue;
1004 			if (conflock)
1005 				locks_copy_conflock(conflock, fl);
1006 			error = -EAGAIN;
1007 			if (!(request->fl_flags & FL_SLEEP))
1008 				goto out;
1009 			/*
1010 			 * Deadlock detection and insertion into the blocked
1011 			 * locks list must be done while holding the same lock!
1012 			 */
1013 			error = -EDEADLK;
1014 			spin_lock(&blocked_lock_lock);
1015 			if (likely(!posix_locks_deadlock(request, fl))) {
1016 				error = FILE_LOCK_DEFERRED;
1017 				__locks_insert_block(fl, request);
1018 			}
1019 			spin_unlock(&blocked_lock_lock);
1020 			goto out;
1021   		}
1022   	}
1023 
1024 	/* If we're just looking for a conflict, we're done. */
1025 	error = 0;
1026 	if (request->fl_flags & FL_ACCESS)
1027 		goto out;
1028 
1029 	/* Find the first old lock with the same owner as the new lock */
1030 	list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1031 		if (posix_same_owner(request, fl))
1032 			break;
1033 	}
1034 
1035 	/* Process locks with this owner. */
1036 	list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1037 		if (!posix_same_owner(request, fl))
1038 			break;
1039 
1040 		/* Detect adjacent or overlapping regions (if same lock type) */
1041 		if (request->fl_type == fl->fl_type) {
1042 			/* In all comparisons of start vs end, use
1043 			 * "start - 1" rather than "end + 1". If end
1044 			 * is OFFSET_MAX, end + 1 will become negative.
1045 			 */
1046 			if (fl->fl_end < request->fl_start - 1)
1047 				continue;
1048 			/* If the next lock in the list has entirely bigger
1049 			 * addresses than the new one, insert the lock here.
1050 			 */
1051 			if (fl->fl_start - 1 > request->fl_end)
1052 				break;
1053 
1054 			/* If we come here, the new and old lock are of the
1055 			 * same type and adjacent or overlapping. Make one
1056 			 * lock yielding from the lower start address of both
1057 			 * locks to the higher end address.
1058 			 */
1059 			if (fl->fl_start > request->fl_start)
1060 				fl->fl_start = request->fl_start;
1061 			else
1062 				request->fl_start = fl->fl_start;
1063 			if (fl->fl_end < request->fl_end)
1064 				fl->fl_end = request->fl_end;
1065 			else
1066 				request->fl_end = fl->fl_end;
1067 			if (added) {
1068 				locks_delete_lock_ctx(fl, &dispose);
1069 				continue;
1070 			}
1071 			request = fl;
1072 			added = true;
1073 		} else {
1074 			/* Processing for different lock types is a bit
1075 			 * more complex.
1076 			 */
1077 			if (fl->fl_end < request->fl_start)
1078 				continue;
1079 			if (fl->fl_start > request->fl_end)
1080 				break;
1081 			if (request->fl_type == F_UNLCK)
1082 				added = true;
1083 			if (fl->fl_start < request->fl_start)
1084 				left = fl;
1085 			/* If the next lock in the list has a higher end
1086 			 * address than the new one, insert the new one here.
1087 			 */
1088 			if (fl->fl_end > request->fl_end) {
1089 				right = fl;
1090 				break;
1091 			}
1092 			if (fl->fl_start >= request->fl_start) {
1093 				/* The new lock completely replaces an old
1094 				 * one (This may happen several times).
1095 				 */
1096 				if (added) {
1097 					locks_delete_lock_ctx(fl, &dispose);
1098 					continue;
1099 				}
1100 				/*
1101 				 * Replace the old lock with new_fl, and
1102 				 * remove the old one. It's safe to do the
1103 				 * insert here since we know that we won't be
1104 				 * using new_fl later, and that the lock is
1105 				 * just replacing an existing lock.
1106 				 */
1107 				error = -ENOLCK;
1108 				if (!new_fl)
1109 					goto out;
1110 				locks_copy_lock(new_fl, request);
1111 				request = new_fl;
1112 				new_fl = NULL;
1113 				locks_insert_lock_ctx(request, &fl->fl_list);
1114 				locks_delete_lock_ctx(fl, &dispose);
1115 				added = true;
1116 			}
1117 		}
1118 	}
1119 
1120 	/*
1121 	 * The above code only modifies existing locks in case of merging or
1122 	 * replacing. If new lock(s) need to be inserted all modifications are
1123 	 * done below this, so it's safe yet to bail out.
1124 	 */
1125 	error = -ENOLCK; /* "no luck" */
1126 	if (right && left == right && !new_fl2)
1127 		goto out;
1128 
1129 	error = 0;
1130 	if (!added) {
1131 		if (request->fl_type == F_UNLCK) {
1132 			if (request->fl_flags & FL_EXISTS)
1133 				error = -ENOENT;
1134 			goto out;
1135 		}
1136 
1137 		if (!new_fl) {
1138 			error = -ENOLCK;
1139 			goto out;
1140 		}
1141 		locks_copy_lock(new_fl, request);
1142 		locks_insert_lock_ctx(new_fl, &fl->fl_list);
1143 		fl = new_fl;
1144 		new_fl = NULL;
1145 	}
1146 	if (right) {
1147 		if (left == right) {
1148 			/* The new lock breaks the old one in two pieces,
1149 			 * so we have to use the second new lock.
1150 			 */
1151 			left = new_fl2;
1152 			new_fl2 = NULL;
1153 			locks_copy_lock(left, right);
1154 			locks_insert_lock_ctx(left, &fl->fl_list);
1155 		}
1156 		right->fl_start = request->fl_end + 1;
1157 		locks_wake_up_blocks(right);
1158 	}
1159 	if (left) {
1160 		left->fl_end = request->fl_start - 1;
1161 		locks_wake_up_blocks(left);
1162 	}
1163  out:
1164 	spin_unlock(&ctx->flc_lock);
1165 	/*
1166 	 * Free any unused locks.
1167 	 */
1168 	if (new_fl)
1169 		locks_free_lock(new_fl);
1170 	if (new_fl2)
1171 		locks_free_lock(new_fl2);
1172 	locks_dispose_list(&dispose);
1173 	trace_posix_lock_inode(inode, request, error);
1174 
1175 	return error;
1176 }
1177 
1178 /**
1179  * posix_lock_file - Apply a POSIX-style lock to a file
1180  * @filp: The file to apply the lock to
1181  * @fl: The lock to be applied
1182  * @conflock: Place to return a copy of the conflicting lock, if found.
1183  *
1184  * Add a POSIX style lock to a file.
1185  * We merge adjacent & overlapping locks whenever possible.
1186  * POSIX locks are sorted by owner task, then by starting address
1187  *
1188  * Note that if called with an FL_EXISTS argument, the caller may determine
1189  * whether or not a lock was successfully freed by testing the return
1190  * value for -ENOENT.
1191  */
1192 int posix_lock_file(struct file *filp, struct file_lock *fl,
1193 			struct file_lock *conflock)
1194 {
1195 	return posix_lock_inode(file_inode(filp), fl, conflock);
1196 }
1197 EXPORT_SYMBOL(posix_lock_file);
1198 
1199 /**
1200  * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1201  * @inode: inode of file to which lock request should be applied
1202  * @fl: The lock to be applied
1203  *
1204  * Apply a POSIX style lock request to an inode.
1205  */
1206 static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1207 {
1208 	int error;
1209 	might_sleep ();
1210 	for (;;) {
1211 		error = posix_lock_inode(inode, fl, NULL);
1212 		if (error != FILE_LOCK_DEFERRED)
1213 			break;
1214 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1215 		if (!error)
1216 			continue;
1217 
1218 		locks_delete_block(fl);
1219 		break;
1220 	}
1221 	return error;
1222 }
1223 
1224 #ifdef CONFIG_MANDATORY_FILE_LOCKING
1225 /**
1226  * locks_mandatory_locked - Check for an active lock
1227  * @file: the file to check
1228  *
1229  * Searches the inode's list of locks to find any POSIX locks which conflict.
1230  * This function is called from locks_verify_locked() only.
1231  */
1232 int locks_mandatory_locked(struct file *file)
1233 {
1234 	int ret;
1235 	struct inode *inode = file_inode(file);
1236 	struct file_lock_context *ctx;
1237 	struct file_lock *fl;
1238 
1239 	ctx = smp_load_acquire(&inode->i_flctx);
1240 	if (!ctx || list_empty_careful(&ctx->flc_posix))
1241 		return 0;
1242 
1243 	/*
1244 	 * Search the lock list for this inode for any POSIX locks.
1245 	 */
1246 	spin_lock(&ctx->flc_lock);
1247 	ret = 0;
1248 	list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1249 		if (fl->fl_owner != current->files &&
1250 		    fl->fl_owner != file) {
1251 			ret = -EAGAIN;
1252 			break;
1253 		}
1254 	}
1255 	spin_unlock(&ctx->flc_lock);
1256 	return ret;
1257 }
1258 
1259 /**
1260  * locks_mandatory_area - Check for a conflicting lock
1261  * @inode:	the file to check
1262  * @filp:       how the file was opened (if it was)
1263  * @start:	first byte in the file to check
1264  * @end:	lastbyte in the file to check
1265  * @type:	%F_WRLCK for a write lock, else %F_RDLCK
1266  *
1267  * Searches the inode's list of locks to find any POSIX locks which conflict.
1268  */
1269 int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1270 			 loff_t end, unsigned char type)
1271 {
1272 	struct file_lock fl;
1273 	int error;
1274 	bool sleep = false;
1275 
1276 	locks_init_lock(&fl);
1277 	fl.fl_pid = current->tgid;
1278 	fl.fl_file = filp;
1279 	fl.fl_flags = FL_POSIX | FL_ACCESS;
1280 	if (filp && !(filp->f_flags & O_NONBLOCK))
1281 		sleep = true;
1282 	fl.fl_type = type;
1283 	fl.fl_start = start;
1284 	fl.fl_end = end;
1285 
1286 	for (;;) {
1287 		if (filp) {
1288 			fl.fl_owner = filp;
1289 			fl.fl_flags &= ~FL_SLEEP;
1290 			error = posix_lock_inode(inode, &fl, NULL);
1291 			if (!error)
1292 				break;
1293 		}
1294 
1295 		if (sleep)
1296 			fl.fl_flags |= FL_SLEEP;
1297 		fl.fl_owner = current->files;
1298 		error = posix_lock_inode(inode, &fl, NULL);
1299 		if (error != FILE_LOCK_DEFERRED)
1300 			break;
1301 		error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1302 		if (!error) {
1303 			/*
1304 			 * If we've been sleeping someone might have
1305 			 * changed the permissions behind our back.
1306 			 */
1307 			if (__mandatory_lock(inode))
1308 				continue;
1309 		}
1310 
1311 		locks_delete_block(&fl);
1312 		break;
1313 	}
1314 
1315 	return error;
1316 }
1317 
1318 EXPORT_SYMBOL(locks_mandatory_area);
1319 #endif /* CONFIG_MANDATORY_FILE_LOCKING */
1320 
1321 static void lease_clear_pending(struct file_lock *fl, int arg)
1322 {
1323 	switch (arg) {
1324 	case F_UNLCK:
1325 		fl->fl_flags &= ~FL_UNLOCK_PENDING;
1326 		/* fall through: */
1327 	case F_RDLCK:
1328 		fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1329 	}
1330 }
1331 
1332 /* We already had a lease on this file; just change its type */
1333 int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1334 {
1335 	int error = assign_type(fl, arg);
1336 
1337 	if (error)
1338 		return error;
1339 	lease_clear_pending(fl, arg);
1340 	locks_wake_up_blocks(fl);
1341 	if (arg == F_UNLCK) {
1342 		struct file *filp = fl->fl_file;
1343 
1344 		f_delown(filp);
1345 		filp->f_owner.signum = 0;
1346 		fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1347 		if (fl->fl_fasync != NULL) {
1348 			printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1349 			fl->fl_fasync = NULL;
1350 		}
1351 		locks_delete_lock_ctx(fl, dispose);
1352 	}
1353 	return 0;
1354 }
1355 EXPORT_SYMBOL(lease_modify);
1356 
1357 static bool past_time(unsigned long then)
1358 {
1359 	if (!then)
1360 		/* 0 is a special value meaning "this never expires": */
1361 		return false;
1362 	return time_after(jiffies, then);
1363 }
1364 
1365 static void time_out_leases(struct inode *inode, struct list_head *dispose)
1366 {
1367 	struct file_lock_context *ctx = inode->i_flctx;
1368 	struct file_lock *fl, *tmp;
1369 
1370 	lockdep_assert_held(&ctx->flc_lock);
1371 
1372 	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1373 		trace_time_out_leases(inode, fl);
1374 		if (past_time(fl->fl_downgrade_time))
1375 			lease_modify(fl, F_RDLCK, dispose);
1376 		if (past_time(fl->fl_break_time))
1377 			lease_modify(fl, F_UNLCK, dispose);
1378 	}
1379 }
1380 
1381 static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1382 {
1383 	if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1384 		return false;
1385 	if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1386 		return false;
1387 	return locks_conflict(breaker, lease);
1388 }
1389 
1390 static bool
1391 any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1392 {
1393 	struct file_lock_context *ctx = inode->i_flctx;
1394 	struct file_lock *fl;
1395 
1396 	lockdep_assert_held(&ctx->flc_lock);
1397 
1398 	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1399 		if (leases_conflict(fl, breaker))
1400 			return true;
1401 	}
1402 	return false;
1403 }
1404 
1405 /**
1406  *	__break_lease	-	revoke all outstanding leases on file
1407  *	@inode: the inode of the file to return
1408  *	@mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1409  *	    break all leases
1410  *	@type: FL_LEASE: break leases and delegations; FL_DELEG: break
1411  *	    only delegations
1412  *
1413  *	break_lease (inlined for speed) has checked there already is at least
1414  *	some kind of lock (maybe a lease) on this file.  Leases are broken on
1415  *	a call to open() or truncate().  This function can sleep unless you
1416  *	specified %O_NONBLOCK to your open().
1417  */
1418 int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1419 {
1420 	int error = 0;
1421 	struct file_lock_context *ctx;
1422 	struct file_lock *new_fl, *fl, *tmp;
1423 	unsigned long break_time;
1424 	int want_write = (mode & O_ACCMODE) != O_RDONLY;
1425 	LIST_HEAD(dispose);
1426 
1427 	new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1428 	if (IS_ERR(new_fl))
1429 		return PTR_ERR(new_fl);
1430 	new_fl->fl_flags = type;
1431 
1432 	/* typically we will check that ctx is non-NULL before calling */
1433 	ctx = smp_load_acquire(&inode->i_flctx);
1434 	if (!ctx) {
1435 		WARN_ON_ONCE(1);
1436 		return error;
1437 	}
1438 
1439 	spin_lock(&ctx->flc_lock);
1440 
1441 	time_out_leases(inode, &dispose);
1442 
1443 	if (!any_leases_conflict(inode, new_fl))
1444 		goto out;
1445 
1446 	break_time = 0;
1447 	if (lease_break_time > 0) {
1448 		break_time = jiffies + lease_break_time * HZ;
1449 		if (break_time == 0)
1450 			break_time++;	/* so that 0 means no break time */
1451 	}
1452 
1453 	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1454 		if (!leases_conflict(fl, new_fl))
1455 			continue;
1456 		if (want_write) {
1457 			if (fl->fl_flags & FL_UNLOCK_PENDING)
1458 				continue;
1459 			fl->fl_flags |= FL_UNLOCK_PENDING;
1460 			fl->fl_break_time = break_time;
1461 		} else {
1462 			if (lease_breaking(fl))
1463 				continue;
1464 			fl->fl_flags |= FL_DOWNGRADE_PENDING;
1465 			fl->fl_downgrade_time = break_time;
1466 		}
1467 		if (fl->fl_lmops->lm_break(fl))
1468 			locks_delete_lock_ctx(fl, &dispose);
1469 	}
1470 
1471 	if (list_empty(&ctx->flc_lease))
1472 		goto out;
1473 
1474 	if (mode & O_NONBLOCK) {
1475 		trace_break_lease_noblock(inode, new_fl);
1476 		error = -EWOULDBLOCK;
1477 		goto out;
1478 	}
1479 
1480 restart:
1481 	fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1482 	break_time = fl->fl_break_time;
1483 	if (break_time != 0)
1484 		break_time -= jiffies;
1485 	if (break_time == 0)
1486 		break_time++;
1487 	locks_insert_block(fl, new_fl);
1488 	trace_break_lease_block(inode, new_fl);
1489 	spin_unlock(&ctx->flc_lock);
1490 	locks_dispose_list(&dispose);
1491 	error = wait_event_interruptible_timeout(new_fl->fl_wait,
1492 						!new_fl->fl_next, break_time);
1493 	spin_lock(&ctx->flc_lock);
1494 	trace_break_lease_unblock(inode, new_fl);
1495 	locks_delete_block(new_fl);
1496 	if (error >= 0) {
1497 		/*
1498 		 * Wait for the next conflicting lease that has not been
1499 		 * broken yet
1500 		 */
1501 		if (error == 0)
1502 			time_out_leases(inode, &dispose);
1503 		if (any_leases_conflict(inode, new_fl))
1504 			goto restart;
1505 		error = 0;
1506 	}
1507 out:
1508 	spin_unlock(&ctx->flc_lock);
1509 	locks_dispose_list(&dispose);
1510 	locks_free_lock(new_fl);
1511 	return error;
1512 }
1513 
1514 EXPORT_SYMBOL(__break_lease);
1515 
1516 /**
1517  *	lease_get_mtime - get the last modified time of an inode
1518  *	@inode: the inode
1519  *      @time:  pointer to a timespec which will contain the last modified time
1520  *
1521  * This is to force NFS clients to flush their caches for files with
1522  * exclusive leases.  The justification is that if someone has an
1523  * exclusive lease, then they could be modifying it.
1524  */
1525 void lease_get_mtime(struct inode *inode, struct timespec *time)
1526 {
1527 	bool has_lease = false;
1528 	struct file_lock_context *ctx;
1529 	struct file_lock *fl;
1530 
1531 	ctx = smp_load_acquire(&inode->i_flctx);
1532 	if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1533 		spin_lock(&ctx->flc_lock);
1534 		fl = list_first_entry_or_null(&ctx->flc_lease,
1535 					      struct file_lock, fl_list);
1536 		if (fl && (fl->fl_type == F_WRLCK))
1537 			has_lease = true;
1538 		spin_unlock(&ctx->flc_lock);
1539 	}
1540 
1541 	if (has_lease)
1542 		*time = current_fs_time(inode->i_sb);
1543 	else
1544 		*time = inode->i_mtime;
1545 }
1546 
1547 EXPORT_SYMBOL(lease_get_mtime);
1548 
1549 /**
1550  *	fcntl_getlease - Enquire what lease is currently active
1551  *	@filp: the file
1552  *
1553  *	The value returned by this function will be one of
1554  *	(if no lease break is pending):
1555  *
1556  *	%F_RDLCK to indicate a shared lease is held.
1557  *
1558  *	%F_WRLCK to indicate an exclusive lease is held.
1559  *
1560  *	%F_UNLCK to indicate no lease is held.
1561  *
1562  *	(if a lease break is pending):
1563  *
1564  *	%F_RDLCK to indicate an exclusive lease needs to be
1565  *		changed to a shared lease (or removed).
1566  *
1567  *	%F_UNLCK to indicate the lease needs to be removed.
1568  *
1569  *	XXX: sfr & willy disagree over whether F_INPROGRESS
1570  *	should be returned to userspace.
1571  */
1572 int fcntl_getlease(struct file *filp)
1573 {
1574 	struct file_lock *fl;
1575 	struct inode *inode = file_inode(filp);
1576 	struct file_lock_context *ctx;
1577 	int type = F_UNLCK;
1578 	LIST_HEAD(dispose);
1579 
1580 	ctx = smp_load_acquire(&inode->i_flctx);
1581 	if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1582 		spin_lock(&ctx->flc_lock);
1583 		time_out_leases(file_inode(filp), &dispose);
1584 		list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1585 			if (fl->fl_file != filp)
1586 				continue;
1587 			type = target_leasetype(fl);
1588 			break;
1589 		}
1590 		spin_unlock(&ctx->flc_lock);
1591 		locks_dispose_list(&dispose);
1592 	}
1593 	return type;
1594 }
1595 
1596 /**
1597  * check_conflicting_open - see if the given dentry points to a file that has
1598  * 			    an existing open that would conflict with the
1599  * 			    desired lease.
1600  * @dentry:	dentry to check
1601  * @arg:	type of lease that we're trying to acquire
1602  * @flags:	current lock flags
1603  *
1604  * Check to see if there's an existing open fd on this file that would
1605  * conflict with the lease we're trying to set.
1606  */
1607 static int
1608 check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
1609 {
1610 	int ret = 0;
1611 	struct inode *inode = dentry->d_inode;
1612 
1613 	if (flags & FL_LAYOUT)
1614 		return 0;
1615 
1616 	if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1617 		return -EAGAIN;
1618 
1619 	if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1620 	    (atomic_read(&inode->i_count) > 1)))
1621 		ret = -EAGAIN;
1622 
1623 	return ret;
1624 }
1625 
1626 static int
1627 generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1628 {
1629 	struct file_lock *fl, *my_fl = NULL, *lease;
1630 	struct dentry *dentry = filp->f_path.dentry;
1631 	struct inode *inode = file_inode(filp);
1632 	struct file_lock_context *ctx;
1633 	bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1634 	int error;
1635 	LIST_HEAD(dispose);
1636 
1637 	lease = *flp;
1638 	trace_generic_add_lease(inode, lease);
1639 
1640 	/* Note that arg is never F_UNLCK here */
1641 	ctx = locks_get_lock_context(inode, arg);
1642 	if (!ctx)
1643 		return -ENOMEM;
1644 
1645 	/*
1646 	 * In the delegation case we need mutual exclusion with
1647 	 * a number of operations that take the i_mutex.  We trylock
1648 	 * because delegations are an optional optimization, and if
1649 	 * there's some chance of a conflict--we'd rather not
1650 	 * bother, maybe that's a sign this just isn't a good file to
1651 	 * hand out a delegation on.
1652 	 */
1653 	if (is_deleg && !inode_trylock(inode))
1654 		return -EAGAIN;
1655 
1656 	if (is_deleg && arg == F_WRLCK) {
1657 		/* Write delegations are not currently supported: */
1658 		inode_unlock(inode);
1659 		WARN_ON_ONCE(1);
1660 		return -EINVAL;
1661 	}
1662 
1663 	spin_lock(&ctx->flc_lock);
1664 	time_out_leases(inode, &dispose);
1665 	error = check_conflicting_open(dentry, arg, lease->fl_flags);
1666 	if (error)
1667 		goto out;
1668 
1669 	/*
1670 	 * At this point, we know that if there is an exclusive
1671 	 * lease on this file, then we hold it on this filp
1672 	 * (otherwise our open of this file would have blocked).
1673 	 * And if we are trying to acquire an exclusive lease,
1674 	 * then the file is not open by anyone (including us)
1675 	 * except for this filp.
1676 	 */
1677 	error = -EAGAIN;
1678 	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1679 		if (fl->fl_file == filp &&
1680 		    fl->fl_owner == lease->fl_owner) {
1681 			my_fl = fl;
1682 			continue;
1683 		}
1684 
1685 		/*
1686 		 * No exclusive leases if someone else has a lease on
1687 		 * this file:
1688 		 */
1689 		if (arg == F_WRLCK)
1690 			goto out;
1691 		/*
1692 		 * Modifying our existing lease is OK, but no getting a
1693 		 * new lease if someone else is opening for write:
1694 		 */
1695 		if (fl->fl_flags & FL_UNLOCK_PENDING)
1696 			goto out;
1697 	}
1698 
1699 	if (my_fl != NULL) {
1700 		lease = my_fl;
1701 		error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1702 		if (error)
1703 			goto out;
1704 		goto out_setup;
1705 	}
1706 
1707 	error = -EINVAL;
1708 	if (!leases_enable)
1709 		goto out;
1710 
1711 	locks_insert_lock_ctx(lease, &ctx->flc_lease);
1712 	/*
1713 	 * The check in break_lease() is lockless. It's possible for another
1714 	 * open to race in after we did the earlier check for a conflicting
1715 	 * open but before the lease was inserted. Check again for a
1716 	 * conflicting open and cancel the lease if there is one.
1717 	 *
1718 	 * We also add a barrier here to ensure that the insertion of the lock
1719 	 * precedes these checks.
1720 	 */
1721 	smp_mb();
1722 	error = check_conflicting_open(dentry, arg, lease->fl_flags);
1723 	if (error) {
1724 		locks_unlink_lock_ctx(lease);
1725 		goto out;
1726 	}
1727 
1728 out_setup:
1729 	if (lease->fl_lmops->lm_setup)
1730 		lease->fl_lmops->lm_setup(lease, priv);
1731 out:
1732 	spin_unlock(&ctx->flc_lock);
1733 	locks_dispose_list(&dispose);
1734 	if (is_deleg)
1735 		inode_unlock(inode);
1736 	if (!error && !my_fl)
1737 		*flp = NULL;
1738 	return error;
1739 }
1740 
1741 static int generic_delete_lease(struct file *filp, void *owner)
1742 {
1743 	int error = -EAGAIN;
1744 	struct file_lock *fl, *victim = NULL;
1745 	struct inode *inode = file_inode(filp);
1746 	struct file_lock_context *ctx;
1747 	LIST_HEAD(dispose);
1748 
1749 	ctx = smp_load_acquire(&inode->i_flctx);
1750 	if (!ctx) {
1751 		trace_generic_delete_lease(inode, NULL);
1752 		return error;
1753 	}
1754 
1755 	spin_lock(&ctx->flc_lock);
1756 	list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1757 		if (fl->fl_file == filp &&
1758 		    fl->fl_owner == owner) {
1759 			victim = fl;
1760 			break;
1761 		}
1762 	}
1763 	trace_generic_delete_lease(inode, victim);
1764 	if (victim)
1765 		error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1766 	spin_unlock(&ctx->flc_lock);
1767 	locks_dispose_list(&dispose);
1768 	return error;
1769 }
1770 
1771 /**
1772  *	generic_setlease	-	sets a lease on an open file
1773  *	@filp:	file pointer
1774  *	@arg:	type of lease to obtain
1775  *	@flp:	input - file_lock to use, output - file_lock inserted
1776  *	@priv:	private data for lm_setup (may be NULL if lm_setup
1777  *		doesn't require it)
1778  *
1779  *	The (input) flp->fl_lmops->lm_break function is required
1780  *	by break_lease().
1781  */
1782 int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1783 			void **priv)
1784 {
1785 	struct inode *inode = file_inode(filp);
1786 	int error;
1787 
1788 	if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1789 		return -EACCES;
1790 	if (!S_ISREG(inode->i_mode))
1791 		return -EINVAL;
1792 	error = security_file_lock(filp, arg);
1793 	if (error)
1794 		return error;
1795 
1796 	switch (arg) {
1797 	case F_UNLCK:
1798 		return generic_delete_lease(filp, *priv);
1799 	case F_RDLCK:
1800 	case F_WRLCK:
1801 		if (!(*flp)->fl_lmops->lm_break) {
1802 			WARN_ON_ONCE(1);
1803 			return -ENOLCK;
1804 		}
1805 
1806 		return generic_add_lease(filp, arg, flp, priv);
1807 	default:
1808 		return -EINVAL;
1809 	}
1810 }
1811 EXPORT_SYMBOL(generic_setlease);
1812 
1813 /**
1814  * vfs_setlease        -       sets a lease on an open file
1815  * @filp:	file pointer
1816  * @arg:	type of lease to obtain
1817  * @lease:	file_lock to use when adding a lease
1818  * @priv:	private info for lm_setup when adding a lease (may be
1819  * 		NULL if lm_setup doesn't require it)
1820  *
1821  * Call this to establish a lease on the file. The "lease" argument is not
1822  * used for F_UNLCK requests and may be NULL. For commands that set or alter
1823  * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1824  * if not, this function will return -ENOLCK (and generate a scary-looking
1825  * stack trace).
1826  *
1827  * The "priv" pointer is passed directly to the lm_setup function as-is. It
1828  * may be NULL if the lm_setup operation doesn't require it.
1829  */
1830 int
1831 vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1832 {
1833 	if (filp->f_op->setlease)
1834 		return filp->f_op->setlease(filp, arg, lease, priv);
1835 	else
1836 		return generic_setlease(filp, arg, lease, priv);
1837 }
1838 EXPORT_SYMBOL_GPL(vfs_setlease);
1839 
1840 static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1841 {
1842 	struct file_lock *fl;
1843 	struct fasync_struct *new;
1844 	int error;
1845 
1846 	fl = lease_alloc(filp, arg);
1847 	if (IS_ERR(fl))
1848 		return PTR_ERR(fl);
1849 
1850 	new = fasync_alloc();
1851 	if (!new) {
1852 		locks_free_lock(fl);
1853 		return -ENOMEM;
1854 	}
1855 	new->fa_fd = fd;
1856 
1857 	error = vfs_setlease(filp, arg, &fl, (void **)&new);
1858 	if (fl)
1859 		locks_free_lock(fl);
1860 	if (new)
1861 		fasync_free(new);
1862 	return error;
1863 }
1864 
1865 /**
1866  *	fcntl_setlease	-	sets a lease on an open file
1867  *	@fd: open file descriptor
1868  *	@filp: file pointer
1869  *	@arg: type of lease to obtain
1870  *
1871  *	Call this fcntl to establish a lease on the file.
1872  *	Note that you also need to call %F_SETSIG to
1873  *	receive a signal when the lease is broken.
1874  */
1875 int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1876 {
1877 	if (arg == F_UNLCK)
1878 		return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
1879 	return do_fcntl_add_lease(fd, filp, arg);
1880 }
1881 
1882 /**
1883  * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
1884  * @inode: inode of the file to apply to
1885  * @fl: The lock to be applied
1886  *
1887  * Apply a FLOCK style lock request to an inode.
1888  */
1889 static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1890 {
1891 	int error;
1892 	might_sleep();
1893 	for (;;) {
1894 		error = flock_lock_inode(inode, fl);
1895 		if (error != FILE_LOCK_DEFERRED)
1896 			break;
1897 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1898 		if (!error)
1899 			continue;
1900 
1901 		locks_delete_block(fl);
1902 		break;
1903 	}
1904 	return error;
1905 }
1906 
1907 /**
1908  * locks_lock_inode_wait - Apply a lock to an inode
1909  * @inode: inode of the file to apply to
1910  * @fl: The lock to be applied
1911  *
1912  * Apply a POSIX or FLOCK style lock request to an inode.
1913  */
1914 int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1915 {
1916 	int res = 0;
1917 	switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
1918 		case FL_POSIX:
1919 			res = posix_lock_inode_wait(inode, fl);
1920 			break;
1921 		case FL_FLOCK:
1922 			res = flock_lock_inode_wait(inode, fl);
1923 			break;
1924 		default:
1925 			BUG();
1926 	}
1927 	return res;
1928 }
1929 EXPORT_SYMBOL(locks_lock_inode_wait);
1930 
1931 /**
1932  *	sys_flock: - flock() system call.
1933  *	@fd: the file descriptor to lock.
1934  *	@cmd: the type of lock to apply.
1935  *
1936  *	Apply a %FL_FLOCK style lock to an open file descriptor.
1937  *	The @cmd can be one of
1938  *
1939  *	%LOCK_SH -- a shared lock.
1940  *
1941  *	%LOCK_EX -- an exclusive lock.
1942  *
1943  *	%LOCK_UN -- remove an existing lock.
1944  *
1945  *	%LOCK_MAND -- a `mandatory' flock.  This exists to emulate Windows Share Modes.
1946  *
1947  *	%LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1948  *	processes read and write access respectively.
1949  */
1950 SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1951 {
1952 	struct fd f = fdget(fd);
1953 	struct file_lock *lock;
1954 	int can_sleep, unlock;
1955 	int error;
1956 
1957 	error = -EBADF;
1958 	if (!f.file)
1959 		goto out;
1960 
1961 	can_sleep = !(cmd & LOCK_NB);
1962 	cmd &= ~LOCK_NB;
1963 	unlock = (cmd == LOCK_UN);
1964 
1965 	if (!unlock && !(cmd & LOCK_MAND) &&
1966 	    !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1967 		goto out_putf;
1968 
1969 	lock = flock_make_lock(f.file, cmd);
1970 	if (IS_ERR(lock)) {
1971 		error = PTR_ERR(lock);
1972 		goto out_putf;
1973 	}
1974 
1975 	if (can_sleep)
1976 		lock->fl_flags |= FL_SLEEP;
1977 
1978 	error = security_file_lock(f.file, lock->fl_type);
1979 	if (error)
1980 		goto out_free;
1981 
1982 	if (f.file->f_op->flock)
1983 		error = f.file->f_op->flock(f.file,
1984 					  (can_sleep) ? F_SETLKW : F_SETLK,
1985 					  lock);
1986 	else
1987 		error = locks_lock_file_wait(f.file, lock);
1988 
1989  out_free:
1990 	locks_free_lock(lock);
1991 
1992  out_putf:
1993 	fdput(f);
1994  out:
1995 	return error;
1996 }
1997 
1998 /**
1999  * vfs_test_lock - test file byte range lock
2000  * @filp: The file to test lock for
2001  * @fl: The lock to test; also used to hold result
2002  *
2003  * Returns -ERRNO on failure.  Indicates presence of conflicting lock by
2004  * setting conf->fl_type to something other than F_UNLCK.
2005  */
2006 int vfs_test_lock(struct file *filp, struct file_lock *fl)
2007 {
2008 	if (filp->f_op->lock)
2009 		return filp->f_op->lock(filp, F_GETLK, fl);
2010 	posix_test_lock(filp, fl);
2011 	return 0;
2012 }
2013 EXPORT_SYMBOL_GPL(vfs_test_lock);
2014 
2015 static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2016 {
2017 	flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
2018 #if BITS_PER_LONG == 32
2019 	/*
2020 	 * Make sure we can represent the posix lock via
2021 	 * legacy 32bit flock.
2022 	 */
2023 	if (fl->fl_start > OFFT_OFFSET_MAX)
2024 		return -EOVERFLOW;
2025 	if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2026 		return -EOVERFLOW;
2027 #endif
2028 	flock->l_start = fl->fl_start;
2029 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2030 		fl->fl_end - fl->fl_start + 1;
2031 	flock->l_whence = 0;
2032 	flock->l_type = fl->fl_type;
2033 	return 0;
2034 }
2035 
2036 #if BITS_PER_LONG == 32
2037 static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2038 {
2039 	flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
2040 	flock->l_start = fl->fl_start;
2041 	flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2042 		fl->fl_end - fl->fl_start + 1;
2043 	flock->l_whence = 0;
2044 	flock->l_type = fl->fl_type;
2045 }
2046 #endif
2047 
2048 /* Report the first existing lock that would conflict with l.
2049  * This implements the F_GETLK command of fcntl().
2050  */
2051 int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
2052 {
2053 	struct file_lock file_lock;
2054 	struct flock flock;
2055 	int error;
2056 
2057 	error = -EFAULT;
2058 	if (copy_from_user(&flock, l, sizeof(flock)))
2059 		goto out;
2060 	error = -EINVAL;
2061 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2062 		goto out;
2063 
2064 	error = flock_to_posix_lock(filp, &file_lock, &flock);
2065 	if (error)
2066 		goto out;
2067 
2068 	if (cmd == F_OFD_GETLK) {
2069 		error = -EINVAL;
2070 		if (flock.l_pid != 0)
2071 			goto out;
2072 
2073 		cmd = F_GETLK;
2074 		file_lock.fl_flags |= FL_OFDLCK;
2075 		file_lock.fl_owner = filp;
2076 	}
2077 
2078 	error = vfs_test_lock(filp, &file_lock);
2079 	if (error)
2080 		goto out;
2081 
2082 	flock.l_type = file_lock.fl_type;
2083 	if (file_lock.fl_type != F_UNLCK) {
2084 		error = posix_lock_to_flock(&flock, &file_lock);
2085 		if (error)
2086 			goto rel_priv;
2087 	}
2088 	error = -EFAULT;
2089 	if (!copy_to_user(l, &flock, sizeof(flock)))
2090 		error = 0;
2091 rel_priv:
2092 	locks_release_private(&file_lock);
2093 out:
2094 	return error;
2095 }
2096 
2097 /**
2098  * vfs_lock_file - file byte range lock
2099  * @filp: The file to apply the lock to
2100  * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2101  * @fl: The lock to be applied
2102  * @conf: Place to return a copy of the conflicting lock, if found.
2103  *
2104  * A caller that doesn't care about the conflicting lock may pass NULL
2105  * as the final argument.
2106  *
2107  * If the filesystem defines a private ->lock() method, then @conf will
2108  * be left unchanged; so a caller that cares should initialize it to
2109  * some acceptable default.
2110  *
2111  * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2112  * locks, the ->lock() interface may return asynchronously, before the lock has
2113  * been granted or denied by the underlying filesystem, if (and only if)
2114  * lm_grant is set. Callers expecting ->lock() to return asynchronously
2115  * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2116  * the request is for a blocking lock. When ->lock() does return asynchronously,
2117  * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2118  * request completes.
2119  * If the request is for non-blocking lock the file system should return
2120  * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2121  * with the result. If the request timed out the callback routine will return a
2122  * nonzero return code and the file system should release the lock. The file
2123  * system is also responsible to keep a corresponding posix lock when it
2124  * grants a lock so the VFS can find out which locks are locally held and do
2125  * the correct lock cleanup when required.
2126  * The underlying filesystem must not drop the kernel lock or call
2127  * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2128  * return code.
2129  */
2130 int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2131 {
2132 	if (filp->f_op->lock)
2133 		return filp->f_op->lock(filp, cmd, fl);
2134 	else
2135 		return posix_lock_file(filp, fl, conf);
2136 }
2137 EXPORT_SYMBOL_GPL(vfs_lock_file);
2138 
2139 static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2140 			     struct file_lock *fl)
2141 {
2142 	int error;
2143 
2144 	error = security_file_lock(filp, fl->fl_type);
2145 	if (error)
2146 		return error;
2147 
2148 	for (;;) {
2149 		error = vfs_lock_file(filp, cmd, fl, NULL);
2150 		if (error != FILE_LOCK_DEFERRED)
2151 			break;
2152 		error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2153 		if (!error)
2154 			continue;
2155 
2156 		locks_delete_block(fl);
2157 		break;
2158 	}
2159 
2160 	return error;
2161 }
2162 
2163 /* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2164 static int
2165 check_fmode_for_setlk(struct file_lock *fl)
2166 {
2167 	switch (fl->fl_type) {
2168 	case F_RDLCK:
2169 		if (!(fl->fl_file->f_mode & FMODE_READ))
2170 			return -EBADF;
2171 		break;
2172 	case F_WRLCK:
2173 		if (!(fl->fl_file->f_mode & FMODE_WRITE))
2174 			return -EBADF;
2175 	}
2176 	return 0;
2177 }
2178 
2179 /* Apply the lock described by l to an open file descriptor.
2180  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2181  */
2182 int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2183 		struct flock __user *l)
2184 {
2185 	struct file_lock *file_lock = locks_alloc_lock();
2186 	struct flock flock;
2187 	struct inode *inode;
2188 	struct file *f;
2189 	int error;
2190 
2191 	if (file_lock == NULL)
2192 		return -ENOLCK;
2193 
2194 	inode = file_inode(filp);
2195 
2196 	/*
2197 	 * This might block, so we do it before checking the inode.
2198 	 */
2199 	error = -EFAULT;
2200 	if (copy_from_user(&flock, l, sizeof(flock)))
2201 		goto out;
2202 
2203 	/* Don't allow mandatory locks on files that may be memory mapped
2204 	 * and shared.
2205 	 */
2206 	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2207 		error = -EAGAIN;
2208 		goto out;
2209 	}
2210 
2211 	error = flock_to_posix_lock(filp, file_lock, &flock);
2212 	if (error)
2213 		goto out;
2214 
2215 	error = check_fmode_for_setlk(file_lock);
2216 	if (error)
2217 		goto out;
2218 
2219 	/*
2220 	 * If the cmd is requesting file-private locks, then set the
2221 	 * FL_OFDLCK flag and override the owner.
2222 	 */
2223 	switch (cmd) {
2224 	case F_OFD_SETLK:
2225 		error = -EINVAL;
2226 		if (flock.l_pid != 0)
2227 			goto out;
2228 
2229 		cmd = F_SETLK;
2230 		file_lock->fl_flags |= FL_OFDLCK;
2231 		file_lock->fl_owner = filp;
2232 		break;
2233 	case F_OFD_SETLKW:
2234 		error = -EINVAL;
2235 		if (flock.l_pid != 0)
2236 			goto out;
2237 
2238 		cmd = F_SETLKW;
2239 		file_lock->fl_flags |= FL_OFDLCK;
2240 		file_lock->fl_owner = filp;
2241 		/* Fallthrough */
2242 	case F_SETLKW:
2243 		file_lock->fl_flags |= FL_SLEEP;
2244 	}
2245 
2246 	error = do_lock_file_wait(filp, cmd, file_lock);
2247 
2248 	/*
2249 	 * Attempt to detect a close/fcntl race and recover by releasing the
2250 	 * lock that was just acquired. There is no need to do that when we're
2251 	 * unlocking though, or for OFD locks.
2252 	 */
2253 	if (!error && file_lock->fl_type != F_UNLCK &&
2254 	    !(file_lock->fl_flags & FL_OFDLCK)) {
2255 		/*
2256 		 * We need that spin_lock here - it prevents reordering between
2257 		 * update of i_flctx->flc_posix and check for it done in
2258 		 * close(). rcu_read_lock() wouldn't do.
2259 		 */
2260 		spin_lock(&current->files->file_lock);
2261 		f = fcheck(fd);
2262 		spin_unlock(&current->files->file_lock);
2263 		if (f != filp) {
2264 			file_lock->fl_type = F_UNLCK;
2265 			error = do_lock_file_wait(filp, cmd, file_lock);
2266 			WARN_ON_ONCE(error);
2267 			error = -EBADF;
2268 		}
2269 	}
2270 out:
2271 	trace_fcntl_setlk(inode, file_lock, error);
2272 	locks_free_lock(file_lock);
2273 	return error;
2274 }
2275 
2276 #if BITS_PER_LONG == 32
2277 /* Report the first existing lock that would conflict with l.
2278  * This implements the F_GETLK command of fcntl().
2279  */
2280 int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
2281 {
2282 	struct file_lock file_lock;
2283 	struct flock64 flock;
2284 	int error;
2285 
2286 	error = -EFAULT;
2287 	if (copy_from_user(&flock, l, sizeof(flock)))
2288 		goto out;
2289 	error = -EINVAL;
2290 	if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2291 		goto out;
2292 
2293 	error = flock64_to_posix_lock(filp, &file_lock, &flock);
2294 	if (error)
2295 		goto out;
2296 
2297 	if (cmd == F_OFD_GETLK) {
2298 		error = -EINVAL;
2299 		if (flock.l_pid != 0)
2300 			goto out;
2301 
2302 		cmd = F_GETLK64;
2303 		file_lock.fl_flags |= FL_OFDLCK;
2304 		file_lock.fl_owner = filp;
2305 	}
2306 
2307 	error = vfs_test_lock(filp, &file_lock);
2308 	if (error)
2309 		goto out;
2310 
2311 	flock.l_type = file_lock.fl_type;
2312 	if (file_lock.fl_type != F_UNLCK)
2313 		posix_lock_to_flock64(&flock, &file_lock);
2314 
2315 	error = -EFAULT;
2316 	if (!copy_to_user(l, &flock, sizeof(flock)))
2317 		error = 0;
2318 
2319 	locks_release_private(&file_lock);
2320 out:
2321 	return error;
2322 }
2323 
2324 /* Apply the lock described by l to an open file descriptor.
2325  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2326  */
2327 int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2328 		struct flock64 __user *l)
2329 {
2330 	struct file_lock *file_lock = locks_alloc_lock();
2331 	struct flock64 flock;
2332 	struct inode *inode;
2333 	struct file *f;
2334 	int error;
2335 
2336 	if (file_lock == NULL)
2337 		return -ENOLCK;
2338 
2339 	/*
2340 	 * This might block, so we do it before checking the inode.
2341 	 */
2342 	error = -EFAULT;
2343 	if (copy_from_user(&flock, l, sizeof(flock)))
2344 		goto out;
2345 
2346 	inode = file_inode(filp);
2347 
2348 	/* Don't allow mandatory locks on files that may be memory mapped
2349 	 * and shared.
2350 	 */
2351 	if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2352 		error = -EAGAIN;
2353 		goto out;
2354 	}
2355 
2356 	error = flock64_to_posix_lock(filp, file_lock, &flock);
2357 	if (error)
2358 		goto out;
2359 
2360 	error = check_fmode_for_setlk(file_lock);
2361 	if (error)
2362 		goto out;
2363 
2364 	/*
2365 	 * If the cmd is requesting file-private locks, then set the
2366 	 * FL_OFDLCK flag and override the owner.
2367 	 */
2368 	switch (cmd) {
2369 	case F_OFD_SETLK:
2370 		error = -EINVAL;
2371 		if (flock.l_pid != 0)
2372 			goto out;
2373 
2374 		cmd = F_SETLK64;
2375 		file_lock->fl_flags |= FL_OFDLCK;
2376 		file_lock->fl_owner = filp;
2377 		break;
2378 	case F_OFD_SETLKW:
2379 		error = -EINVAL;
2380 		if (flock.l_pid != 0)
2381 			goto out;
2382 
2383 		cmd = F_SETLKW64;
2384 		file_lock->fl_flags |= FL_OFDLCK;
2385 		file_lock->fl_owner = filp;
2386 		/* Fallthrough */
2387 	case F_SETLKW64:
2388 		file_lock->fl_flags |= FL_SLEEP;
2389 	}
2390 
2391 	error = do_lock_file_wait(filp, cmd, file_lock);
2392 
2393 	/*
2394 	 * Attempt to detect a close/fcntl race and recover by releasing the
2395 	 * lock that was just acquired. There is no need to do that when we're
2396 	 * unlocking though, or for OFD locks.
2397 	 */
2398 	if (!error && file_lock->fl_type != F_UNLCK &&
2399 	    !(file_lock->fl_flags & FL_OFDLCK)) {
2400 		/*
2401 		 * We need that spin_lock here - it prevents reordering between
2402 		 * update of i_flctx->flc_posix and check for it done in
2403 		 * close(). rcu_read_lock() wouldn't do.
2404 		 */
2405 		spin_lock(&current->files->file_lock);
2406 		f = fcheck(fd);
2407 		spin_unlock(&current->files->file_lock);
2408 		if (f != filp) {
2409 			file_lock->fl_type = F_UNLCK;
2410 			error = do_lock_file_wait(filp, cmd, file_lock);
2411 			WARN_ON_ONCE(error);
2412 			error = -EBADF;
2413 		}
2414 	}
2415 out:
2416 	locks_free_lock(file_lock);
2417 	return error;
2418 }
2419 #endif /* BITS_PER_LONG == 32 */
2420 
2421 /*
2422  * This function is called when the file is being removed
2423  * from the task's fd array.  POSIX locks belonging to this task
2424  * are deleted at this time.
2425  */
2426 void locks_remove_posix(struct file *filp, fl_owner_t owner)
2427 {
2428 	int error;
2429 	struct file_lock lock;
2430 	struct file_lock_context *ctx;
2431 
2432 	/*
2433 	 * If there are no locks held on this file, we don't need to call
2434 	 * posix_lock_file().  Another process could be setting a lock on this
2435 	 * file at the same time, but we wouldn't remove that lock anyway.
2436 	 */
2437 	ctx =  smp_load_acquire(&file_inode(filp)->i_flctx);
2438 	if (!ctx || list_empty(&ctx->flc_posix))
2439 		return;
2440 
2441 	lock.fl_type = F_UNLCK;
2442 	lock.fl_flags = FL_POSIX | FL_CLOSE;
2443 	lock.fl_start = 0;
2444 	lock.fl_end = OFFSET_MAX;
2445 	lock.fl_owner = owner;
2446 	lock.fl_pid = current->tgid;
2447 	lock.fl_file = filp;
2448 	lock.fl_ops = NULL;
2449 	lock.fl_lmops = NULL;
2450 
2451 	error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2452 
2453 	if (lock.fl_ops && lock.fl_ops->fl_release_private)
2454 		lock.fl_ops->fl_release_private(&lock);
2455 	trace_locks_remove_posix(file_inode(filp), &lock, error);
2456 }
2457 
2458 EXPORT_SYMBOL(locks_remove_posix);
2459 
2460 /* The i_flctx must be valid when calling into here */
2461 static void
2462 locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2463 {
2464 	struct file_lock fl = {
2465 		.fl_owner = filp,
2466 		.fl_pid = current->tgid,
2467 		.fl_file = filp,
2468 		.fl_flags = FL_FLOCK,
2469 		.fl_type = F_UNLCK,
2470 		.fl_end = OFFSET_MAX,
2471 	};
2472 	struct inode *inode = file_inode(filp);
2473 
2474 	if (list_empty(&flctx->flc_flock))
2475 		return;
2476 
2477 	if (filp->f_op->flock)
2478 		filp->f_op->flock(filp, F_SETLKW, &fl);
2479 	else
2480 		flock_lock_inode(inode, &fl);
2481 
2482 	if (fl.fl_ops && fl.fl_ops->fl_release_private)
2483 		fl.fl_ops->fl_release_private(&fl);
2484 }
2485 
2486 /* The i_flctx must be valid when calling into here */
2487 static void
2488 locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2489 {
2490 	struct file_lock *fl, *tmp;
2491 	LIST_HEAD(dispose);
2492 
2493 	if (list_empty(&ctx->flc_lease))
2494 		return;
2495 
2496 	spin_lock(&ctx->flc_lock);
2497 	list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2498 		if (filp == fl->fl_file)
2499 			lease_modify(fl, F_UNLCK, &dispose);
2500 	spin_unlock(&ctx->flc_lock);
2501 	locks_dispose_list(&dispose);
2502 }
2503 
2504 /*
2505  * This function is called on the last close of an open file.
2506  */
2507 void locks_remove_file(struct file *filp)
2508 {
2509 	struct file_lock_context *ctx;
2510 
2511 	ctx = smp_load_acquire(&file_inode(filp)->i_flctx);
2512 	if (!ctx)
2513 		return;
2514 
2515 	/* remove any OFD locks */
2516 	locks_remove_posix(filp, filp);
2517 
2518 	/* remove flock locks */
2519 	locks_remove_flock(filp, ctx);
2520 
2521 	/* remove any leases */
2522 	locks_remove_lease(filp, ctx);
2523 }
2524 
2525 /**
2526  *	posix_unblock_lock - stop waiting for a file lock
2527  *	@waiter: the lock which was waiting
2528  *
2529  *	lockd needs to block waiting for locks.
2530  */
2531 int
2532 posix_unblock_lock(struct file_lock *waiter)
2533 {
2534 	int status = 0;
2535 
2536 	spin_lock(&blocked_lock_lock);
2537 	if (waiter->fl_next)
2538 		__locks_delete_block(waiter);
2539 	else
2540 		status = -ENOENT;
2541 	spin_unlock(&blocked_lock_lock);
2542 	return status;
2543 }
2544 EXPORT_SYMBOL(posix_unblock_lock);
2545 
2546 /**
2547  * vfs_cancel_lock - file byte range unblock lock
2548  * @filp: The file to apply the unblock to
2549  * @fl: The lock to be unblocked
2550  *
2551  * Used by lock managers to cancel blocked requests
2552  */
2553 int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2554 {
2555 	if (filp->f_op->lock)
2556 		return filp->f_op->lock(filp, F_CANCELLK, fl);
2557 	return 0;
2558 }
2559 
2560 EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2561 
2562 #ifdef CONFIG_PROC_FS
2563 #include <linux/proc_fs.h>
2564 #include <linux/seq_file.h>
2565 
2566 struct locks_iterator {
2567 	int	li_cpu;
2568 	loff_t	li_pos;
2569 };
2570 
2571 static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2572 			    loff_t id, char *pfx)
2573 {
2574 	struct inode *inode = NULL;
2575 	unsigned int fl_pid;
2576 
2577 	if (fl->fl_nspid)
2578 		fl_pid = pid_vnr(fl->fl_nspid);
2579 	else
2580 		fl_pid = fl->fl_pid;
2581 
2582 	if (fl->fl_file != NULL)
2583 		inode = file_inode(fl->fl_file);
2584 
2585 	seq_printf(f, "%lld:%s ", id, pfx);
2586 	if (IS_POSIX(fl)) {
2587 		if (fl->fl_flags & FL_ACCESS)
2588 			seq_puts(f, "ACCESS");
2589 		else if (IS_OFDLCK(fl))
2590 			seq_puts(f, "OFDLCK");
2591 		else
2592 			seq_puts(f, "POSIX ");
2593 
2594 		seq_printf(f, " %s ",
2595 			     (inode == NULL) ? "*NOINODE*" :
2596 			     mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2597 	} else if (IS_FLOCK(fl)) {
2598 		if (fl->fl_type & LOCK_MAND) {
2599 			seq_puts(f, "FLOCK  MSNFS     ");
2600 		} else {
2601 			seq_puts(f, "FLOCK  ADVISORY  ");
2602 		}
2603 	} else if (IS_LEASE(fl)) {
2604 		if (fl->fl_flags & FL_DELEG)
2605 			seq_puts(f, "DELEG  ");
2606 		else
2607 			seq_puts(f, "LEASE  ");
2608 
2609 		if (lease_breaking(fl))
2610 			seq_puts(f, "BREAKING  ");
2611 		else if (fl->fl_file)
2612 			seq_puts(f, "ACTIVE    ");
2613 		else
2614 			seq_puts(f, "BREAKER   ");
2615 	} else {
2616 		seq_puts(f, "UNKNOWN UNKNOWN  ");
2617 	}
2618 	if (fl->fl_type & LOCK_MAND) {
2619 		seq_printf(f, "%s ",
2620 			       (fl->fl_type & LOCK_READ)
2621 			       ? (fl->fl_type & LOCK_WRITE) ? "RW   " : "READ "
2622 			       : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2623 	} else {
2624 		seq_printf(f, "%s ",
2625 			       (lease_breaking(fl))
2626 			       ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2627 			       : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2628 	}
2629 	if (inode) {
2630 		/* userspace relies on this representation of dev_t */
2631 		seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2632 				MAJOR(inode->i_sb->s_dev),
2633 				MINOR(inode->i_sb->s_dev), inode->i_ino);
2634 	} else {
2635 		seq_printf(f, "%d <none>:0 ", fl_pid);
2636 	}
2637 	if (IS_POSIX(fl)) {
2638 		if (fl->fl_end == OFFSET_MAX)
2639 			seq_printf(f, "%Ld EOF\n", fl->fl_start);
2640 		else
2641 			seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2642 	} else {
2643 		seq_puts(f, "0 EOF\n");
2644 	}
2645 }
2646 
2647 static int locks_show(struct seq_file *f, void *v)
2648 {
2649 	struct locks_iterator *iter = f->private;
2650 	struct file_lock *fl, *bfl;
2651 
2652 	fl = hlist_entry(v, struct file_lock, fl_link);
2653 
2654 	lock_get_status(f, fl, iter->li_pos, "");
2655 
2656 	list_for_each_entry(bfl, &fl->fl_block, fl_block)
2657 		lock_get_status(f, bfl, iter->li_pos, " ->");
2658 
2659 	return 0;
2660 }
2661 
2662 static void __show_fd_locks(struct seq_file *f,
2663 			struct list_head *head, int *id,
2664 			struct file *filp, struct files_struct *files)
2665 {
2666 	struct file_lock *fl;
2667 
2668 	list_for_each_entry(fl, head, fl_list) {
2669 
2670 		if (filp != fl->fl_file)
2671 			continue;
2672 		if (fl->fl_owner != files &&
2673 		    fl->fl_owner != filp)
2674 			continue;
2675 
2676 		(*id)++;
2677 		seq_puts(f, "lock:\t");
2678 		lock_get_status(f, fl, *id, "");
2679 	}
2680 }
2681 
2682 void show_fd_locks(struct seq_file *f,
2683 		  struct file *filp, struct files_struct *files)
2684 {
2685 	struct inode *inode = file_inode(filp);
2686 	struct file_lock_context *ctx;
2687 	int id = 0;
2688 
2689 	ctx = smp_load_acquire(&inode->i_flctx);
2690 	if (!ctx)
2691 		return;
2692 
2693 	spin_lock(&ctx->flc_lock);
2694 	__show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2695 	__show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2696 	__show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2697 	spin_unlock(&ctx->flc_lock);
2698 }
2699 
2700 static void *locks_start(struct seq_file *f, loff_t *pos)
2701 	__acquires(&blocked_lock_lock)
2702 {
2703 	struct locks_iterator *iter = f->private;
2704 
2705 	iter->li_pos = *pos + 1;
2706 	lg_global_lock(&file_lock_lglock);
2707 	spin_lock(&blocked_lock_lock);
2708 	return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
2709 }
2710 
2711 static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2712 {
2713 	struct locks_iterator *iter = f->private;
2714 
2715 	++iter->li_pos;
2716 	return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
2717 }
2718 
2719 static void locks_stop(struct seq_file *f, void *v)
2720 	__releases(&blocked_lock_lock)
2721 {
2722 	spin_unlock(&blocked_lock_lock);
2723 	lg_global_unlock(&file_lock_lglock);
2724 }
2725 
2726 static const struct seq_operations locks_seq_operations = {
2727 	.start	= locks_start,
2728 	.next	= locks_next,
2729 	.stop	= locks_stop,
2730 	.show	= locks_show,
2731 };
2732 
2733 static int locks_open(struct inode *inode, struct file *filp)
2734 {
2735 	return seq_open_private(filp, &locks_seq_operations,
2736 					sizeof(struct locks_iterator));
2737 }
2738 
2739 static const struct file_operations proc_locks_operations = {
2740 	.open		= locks_open,
2741 	.read		= seq_read,
2742 	.llseek		= seq_lseek,
2743 	.release	= seq_release_private,
2744 };
2745 
2746 static int __init proc_locks_init(void)
2747 {
2748 	proc_create("locks", 0, NULL, &proc_locks_operations);
2749 	return 0;
2750 }
2751 fs_initcall(proc_locks_init);
2752 #endif
2753 
2754 static int __init filelock_init(void)
2755 {
2756 	int i;
2757 
2758 	flctx_cache = kmem_cache_create("file_lock_ctx",
2759 			sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2760 
2761 	filelock_cache = kmem_cache_create("file_lock_cache",
2762 			sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2763 
2764 	lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2765 
2766 	for_each_possible_cpu(i)
2767 		INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2768 
2769 	return 0;
2770 }
2771 
2772 core_initcall(filelock_init);
2773