Loading...
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
131#define CREATE_TRACE_POINTS
132#include <trace/events/filelock.h>
133
134#include <linux/uaccess.h>
135
136#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
137#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
138#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
139#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
140
141static inline bool is_remote_lock(struct file *filp)
142{
143 return likely(!(filp->f_path.dentry->d_sb->s_flags & MS_NOREMOTELOCK));
144}
145
146static bool lease_breaking(struct file_lock *fl)
147{
148 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
149}
150
151static int target_leasetype(struct file_lock *fl)
152{
153 if (fl->fl_flags & FL_UNLOCK_PENDING)
154 return F_UNLCK;
155 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
156 return F_RDLCK;
157 return fl->fl_type;
158}
159
160int leases_enable = 1;
161int lease_break_time = 45;
162
163/*
164 * The global file_lock_list is only used for displaying /proc/locks, so we
165 * keep a list on each CPU, with each list protected by its own spinlock.
166 * Global serialization is done using file_rwsem.
167 *
168 * Note that alterations to the list also require that the relevant flc_lock is
169 * held.
170 */
171struct file_lock_list_struct {
172 spinlock_t lock;
173 struct hlist_head hlist;
174};
175static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
176DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
177
178/*
179 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
180 * It is protected by blocked_lock_lock.
181 *
182 * We hash locks by lockowner in order to optimize searching for the lock a
183 * particular lockowner is waiting on.
184 *
185 * FIXME: make this value scale via some heuristic? We generally will want more
186 * buckets when we have more lockowners holding locks, but that's a little
187 * difficult to determine without knowing what the workload will look like.
188 */
189#define BLOCKED_HASH_BITS 7
190static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
191
192/*
193 * This lock protects the blocked_hash. Generally, if you're accessing it, you
194 * want to be holding this lock.
195 *
196 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
197 * pointer for file_lock structures that are acting as lock requests (in
198 * contrast to those that are acting as records of acquired locks).
199 *
200 * Note that when we acquire this lock in order to change the above fields,
201 * we often hold the flc_lock as well. In certain cases, when reading the fields
202 * protected by this lock, we can skip acquiring it iff we already hold the
203 * flc_lock.
204 *
205 * In particular, adding an entry to the fl_block list requires that you hold
206 * both the flc_lock and the blocked_lock_lock (acquired in that order).
207 * Deleting an entry from the list however only requires the file_lock_lock.
208 */
209static DEFINE_SPINLOCK(blocked_lock_lock);
210
211static struct kmem_cache *flctx_cache __read_mostly;
212static struct kmem_cache *filelock_cache __read_mostly;
213
214static struct file_lock_context *
215locks_get_lock_context(struct inode *inode, int type)
216{
217 struct file_lock_context *ctx;
218
219 /* paired with cmpxchg() below */
220 ctx = smp_load_acquire(&inode->i_flctx);
221 if (likely(ctx) || type == F_UNLCK)
222 goto out;
223
224 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
225 if (!ctx)
226 goto out;
227
228 spin_lock_init(&ctx->flc_lock);
229 INIT_LIST_HEAD(&ctx->flc_flock);
230 INIT_LIST_HEAD(&ctx->flc_posix);
231 INIT_LIST_HEAD(&ctx->flc_lease);
232
233 /*
234 * Assign the pointer if it's not already assigned. If it is, then
235 * free the context we just allocated.
236 */
237 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
238 kmem_cache_free(flctx_cache, ctx);
239 ctx = smp_load_acquire(&inode->i_flctx);
240 }
241out:
242 trace_locks_get_lock_context(inode, type, ctx);
243 return ctx;
244}
245
246static void
247locks_dump_ctx_list(struct list_head *list, char *list_type)
248{
249 struct file_lock *fl;
250
251 list_for_each_entry(fl, list, fl_list) {
252 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);
253 }
254}
255
256static void
257locks_check_ctx_lists(struct inode *inode)
258{
259 struct file_lock_context *ctx = inode->i_flctx;
260
261 if (unlikely(!list_empty(&ctx->flc_flock) ||
262 !list_empty(&ctx->flc_posix) ||
263 !list_empty(&ctx->flc_lease))) {
264 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
265 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
266 inode->i_ino);
267 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
268 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
269 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
270 }
271}
272
273void
274locks_free_lock_context(struct inode *inode)
275{
276 struct file_lock_context *ctx = inode->i_flctx;
277
278 if (unlikely(ctx)) {
279 locks_check_ctx_lists(inode);
280 kmem_cache_free(flctx_cache, ctx);
281 }
282}
283
284static void locks_init_lock_heads(struct file_lock *fl)
285{
286 INIT_HLIST_NODE(&fl->fl_link);
287 INIT_LIST_HEAD(&fl->fl_list);
288 INIT_LIST_HEAD(&fl->fl_block);
289 init_waitqueue_head(&fl->fl_wait);
290}
291
292/* Allocate an empty lock structure. */
293struct file_lock *locks_alloc_lock(void)
294{
295 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
296
297 if (fl)
298 locks_init_lock_heads(fl);
299
300 return fl;
301}
302EXPORT_SYMBOL_GPL(locks_alloc_lock);
303
304void locks_release_private(struct file_lock *fl)
305{
306 if (fl->fl_ops) {
307 if (fl->fl_ops->fl_release_private)
308 fl->fl_ops->fl_release_private(fl);
309 fl->fl_ops = NULL;
310 }
311
312 if (fl->fl_lmops) {
313 if (fl->fl_lmops->lm_put_owner) {
314 fl->fl_lmops->lm_put_owner(fl->fl_owner);
315 fl->fl_owner = NULL;
316 }
317 fl->fl_lmops = NULL;
318 }
319}
320EXPORT_SYMBOL_GPL(locks_release_private);
321
322/* Free a lock which is not in use. */
323void locks_free_lock(struct file_lock *fl)
324{
325 BUG_ON(waitqueue_active(&fl->fl_wait));
326 BUG_ON(!list_empty(&fl->fl_list));
327 BUG_ON(!list_empty(&fl->fl_block));
328 BUG_ON(!hlist_unhashed(&fl->fl_link));
329
330 locks_release_private(fl);
331 kmem_cache_free(filelock_cache, fl);
332}
333EXPORT_SYMBOL(locks_free_lock);
334
335static void
336locks_dispose_list(struct list_head *dispose)
337{
338 struct file_lock *fl;
339
340 while (!list_empty(dispose)) {
341 fl = list_first_entry(dispose, struct file_lock, fl_list);
342 list_del_init(&fl->fl_list);
343 locks_free_lock(fl);
344 }
345}
346
347void locks_init_lock(struct file_lock *fl)
348{
349 memset(fl, 0, sizeof(struct file_lock));
350 locks_init_lock_heads(fl);
351}
352
353EXPORT_SYMBOL(locks_init_lock);
354
355/*
356 * Initialize a new lock from an existing file_lock structure.
357 */
358void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
359{
360 new->fl_owner = fl->fl_owner;
361 new->fl_pid = fl->fl_pid;
362 new->fl_file = NULL;
363 new->fl_flags = fl->fl_flags;
364 new->fl_type = fl->fl_type;
365 new->fl_start = fl->fl_start;
366 new->fl_end = fl->fl_end;
367 new->fl_lmops = fl->fl_lmops;
368 new->fl_ops = NULL;
369
370 if (fl->fl_lmops) {
371 if (fl->fl_lmops->lm_get_owner)
372 fl->fl_lmops->lm_get_owner(fl->fl_owner);
373 }
374}
375EXPORT_SYMBOL(locks_copy_conflock);
376
377void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
378{
379 /* "new" must be a freshly-initialized lock */
380 WARN_ON_ONCE(new->fl_ops);
381
382 locks_copy_conflock(new, fl);
383
384 new->fl_file = fl->fl_file;
385 new->fl_ops = fl->fl_ops;
386
387 if (fl->fl_ops) {
388 if (fl->fl_ops->fl_copy_lock)
389 fl->fl_ops->fl_copy_lock(new, fl);
390 }
391}
392
393EXPORT_SYMBOL(locks_copy_lock);
394
395static inline int flock_translate_cmd(int cmd) {
396 if (cmd & LOCK_MAND)
397 return cmd & (LOCK_MAND | LOCK_RW);
398 switch (cmd) {
399 case LOCK_SH:
400 return F_RDLCK;
401 case LOCK_EX:
402 return F_WRLCK;
403 case LOCK_UN:
404 return F_UNLCK;
405 }
406 return -EINVAL;
407}
408
409/* Fill in a file_lock structure with an appropriate FLOCK lock. */
410static struct file_lock *
411flock_make_lock(struct file *filp, unsigned int cmd)
412{
413 struct file_lock *fl;
414 int type = flock_translate_cmd(cmd);
415
416 if (type < 0)
417 return ERR_PTR(type);
418
419 fl = locks_alloc_lock();
420 if (fl == NULL)
421 return ERR_PTR(-ENOMEM);
422
423 fl->fl_file = filp;
424 fl->fl_owner = filp;
425 fl->fl_pid = current->tgid;
426 fl->fl_flags = FL_FLOCK;
427 fl->fl_type = type;
428 fl->fl_end = OFFSET_MAX;
429
430 return fl;
431}
432
433static int assign_type(struct file_lock *fl, long type)
434{
435 switch (type) {
436 case F_RDLCK:
437 case F_WRLCK:
438 case F_UNLCK:
439 fl->fl_type = type;
440 break;
441 default:
442 return -EINVAL;
443 }
444 return 0;
445}
446
447static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
448 struct flock64 *l)
449{
450 switch (l->l_whence) {
451 case SEEK_SET:
452 fl->fl_start = 0;
453 break;
454 case SEEK_CUR:
455 fl->fl_start = filp->f_pos;
456 break;
457 case SEEK_END:
458 fl->fl_start = i_size_read(file_inode(filp));
459 break;
460 default:
461 return -EINVAL;
462 }
463 if (l->l_start > OFFSET_MAX - fl->fl_start)
464 return -EOVERFLOW;
465 fl->fl_start += l->l_start;
466 if (fl->fl_start < 0)
467 return -EINVAL;
468
469 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
470 POSIX-2001 defines it. */
471 if (l->l_len > 0) {
472 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
473 return -EOVERFLOW;
474 fl->fl_end = fl->fl_start + l->l_len - 1;
475
476 } else if (l->l_len < 0) {
477 if (fl->fl_start + l->l_len < 0)
478 return -EINVAL;
479 fl->fl_end = fl->fl_start - 1;
480 fl->fl_start += l->l_len;
481 } else
482 fl->fl_end = OFFSET_MAX;
483
484 fl->fl_owner = current->files;
485 fl->fl_pid = current->tgid;
486 fl->fl_file = filp;
487 fl->fl_flags = FL_POSIX;
488 fl->fl_ops = NULL;
489 fl->fl_lmops = NULL;
490
491 return assign_type(fl, l->l_type);
492}
493
494/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
495 * style lock.
496 */
497static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
498 struct flock *l)
499{
500 struct flock64 ll = {
501 .l_type = l->l_type,
502 .l_whence = l->l_whence,
503 .l_start = l->l_start,
504 .l_len = l->l_len,
505 };
506
507 return flock64_to_posix_lock(filp, fl, &ll);
508}
509
510/* default lease lock manager operations */
511static bool
512lease_break_callback(struct file_lock *fl)
513{
514 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
515 return false;
516}
517
518static void
519lease_setup(struct file_lock *fl, void **priv)
520{
521 struct file *filp = fl->fl_file;
522 struct fasync_struct *fa = *priv;
523
524 /*
525 * fasync_insert_entry() returns the old entry if any. If there was no
526 * old entry, then it used "priv" and inserted it into the fasync list.
527 * Clear the pointer to indicate that it shouldn't be freed.
528 */
529 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
530 *priv = NULL;
531
532 __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
533}
534
535static const struct lock_manager_operations lease_manager_ops = {
536 .lm_break = lease_break_callback,
537 .lm_change = lease_modify,
538 .lm_setup = lease_setup,
539};
540
541/*
542 * Initialize a lease, use the default lock manager operations
543 */
544static int lease_init(struct file *filp, long type, struct file_lock *fl)
545 {
546 if (assign_type(fl, type) != 0)
547 return -EINVAL;
548
549 fl->fl_owner = filp;
550 fl->fl_pid = current->tgid;
551
552 fl->fl_file = filp;
553 fl->fl_flags = FL_LEASE;
554 fl->fl_start = 0;
555 fl->fl_end = OFFSET_MAX;
556 fl->fl_ops = NULL;
557 fl->fl_lmops = &lease_manager_ops;
558 return 0;
559}
560
561/* Allocate a file_lock initialised to this type of lease */
562static struct file_lock *lease_alloc(struct file *filp, long type)
563{
564 struct file_lock *fl = locks_alloc_lock();
565 int error = -ENOMEM;
566
567 if (fl == NULL)
568 return ERR_PTR(error);
569
570 error = lease_init(filp, type, fl);
571 if (error) {
572 locks_free_lock(fl);
573 return ERR_PTR(error);
574 }
575 return fl;
576}
577
578/* Check if two locks overlap each other.
579 */
580static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
581{
582 return ((fl1->fl_end >= fl2->fl_start) &&
583 (fl2->fl_end >= fl1->fl_start));
584}
585
586/*
587 * Check whether two locks have the same owner.
588 */
589static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
590{
591 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
592 return fl2->fl_lmops == fl1->fl_lmops &&
593 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
594 return fl1->fl_owner == fl2->fl_owner;
595}
596
597/* Must be called with the flc_lock held! */
598static void locks_insert_global_locks(struct file_lock *fl)
599{
600 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
601
602 percpu_rwsem_assert_held(&file_rwsem);
603
604 spin_lock(&fll->lock);
605 fl->fl_link_cpu = smp_processor_id();
606 hlist_add_head(&fl->fl_link, &fll->hlist);
607 spin_unlock(&fll->lock);
608}
609
610/* Must be called with the flc_lock held! */
611static void locks_delete_global_locks(struct file_lock *fl)
612{
613 struct file_lock_list_struct *fll;
614
615 percpu_rwsem_assert_held(&file_rwsem);
616
617 /*
618 * Avoid taking lock if already unhashed. This is safe since this check
619 * is done while holding the flc_lock, and new insertions into the list
620 * also require that it be held.
621 */
622 if (hlist_unhashed(&fl->fl_link))
623 return;
624
625 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
626 spin_lock(&fll->lock);
627 hlist_del_init(&fl->fl_link);
628 spin_unlock(&fll->lock);
629}
630
631static unsigned long
632posix_owner_key(struct file_lock *fl)
633{
634 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
635 return fl->fl_lmops->lm_owner_key(fl);
636 return (unsigned long)fl->fl_owner;
637}
638
639static void locks_insert_global_blocked(struct file_lock *waiter)
640{
641 lockdep_assert_held(&blocked_lock_lock);
642
643 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
644}
645
646static void locks_delete_global_blocked(struct file_lock *waiter)
647{
648 lockdep_assert_held(&blocked_lock_lock);
649
650 hash_del(&waiter->fl_link);
651}
652
653/* Remove waiter from blocker's block list.
654 * When blocker ends up pointing to itself then the list is empty.
655 *
656 * Must be called with blocked_lock_lock held.
657 */
658static void __locks_delete_block(struct file_lock *waiter)
659{
660 locks_delete_global_blocked(waiter);
661 list_del_init(&waiter->fl_block);
662 waiter->fl_next = NULL;
663}
664
665static void locks_delete_block(struct file_lock *waiter)
666{
667 spin_lock(&blocked_lock_lock);
668 __locks_delete_block(waiter);
669 spin_unlock(&blocked_lock_lock);
670}
671
672/* Insert waiter into blocker's block list.
673 * We use a circular list so that processes can be easily woken up in
674 * the order they blocked. The documentation doesn't require this but
675 * it seems like the reasonable thing to do.
676 *
677 * Must be called with both the flc_lock and blocked_lock_lock held. The
678 * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
679 * that the flc_lock is also held on insertions we can avoid taking the
680 * blocked_lock_lock in some cases when we see that the fl_block list is empty.
681 */
682static void __locks_insert_block(struct file_lock *blocker,
683 struct file_lock *waiter)
684{
685 BUG_ON(!list_empty(&waiter->fl_block));
686 waiter->fl_next = blocker;
687 list_add_tail(&waiter->fl_block, &blocker->fl_block);
688 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
689 locks_insert_global_blocked(waiter);
690}
691
692/* Must be called with flc_lock held. */
693static void locks_insert_block(struct file_lock *blocker,
694 struct file_lock *waiter)
695{
696 spin_lock(&blocked_lock_lock);
697 __locks_insert_block(blocker, waiter);
698 spin_unlock(&blocked_lock_lock);
699}
700
701/*
702 * Wake up processes blocked waiting for blocker.
703 *
704 * Must be called with the inode->flc_lock held!
705 */
706static void locks_wake_up_blocks(struct file_lock *blocker)
707{
708 /*
709 * Avoid taking global lock if list is empty. This is safe since new
710 * blocked requests are only added to the list under the flc_lock, and
711 * the flc_lock is always held here. Note that removal from the fl_block
712 * list does not require the flc_lock, so we must recheck list_empty()
713 * after acquiring the blocked_lock_lock.
714 */
715 if (list_empty(&blocker->fl_block))
716 return;
717
718 spin_lock(&blocked_lock_lock);
719 while (!list_empty(&blocker->fl_block)) {
720 struct file_lock *waiter;
721
722 waiter = list_first_entry(&blocker->fl_block,
723 struct file_lock, fl_block);
724 __locks_delete_block(waiter);
725 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
726 waiter->fl_lmops->lm_notify(waiter);
727 else
728 wake_up(&waiter->fl_wait);
729 }
730 spin_unlock(&blocked_lock_lock);
731}
732
733static void
734locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
735{
736 fl->fl_nspid = get_pid(task_tgid(current));
737 list_add_tail(&fl->fl_list, before);
738 locks_insert_global_locks(fl);
739}
740
741static void
742locks_unlink_lock_ctx(struct file_lock *fl)
743{
744 locks_delete_global_locks(fl);
745 list_del_init(&fl->fl_list);
746 if (fl->fl_nspid) {
747 put_pid(fl->fl_nspid);
748 fl->fl_nspid = NULL;
749 }
750 locks_wake_up_blocks(fl);
751}
752
753static void
754locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
755{
756 locks_unlink_lock_ctx(fl);
757 if (dispose)
758 list_add(&fl->fl_list, dispose);
759 else
760 locks_free_lock(fl);
761}
762
763/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
764 * checks for shared/exclusive status of overlapping locks.
765 */
766static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
767{
768 if (sys_fl->fl_type == F_WRLCK)
769 return 1;
770 if (caller_fl->fl_type == F_WRLCK)
771 return 1;
772 return 0;
773}
774
775/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
776 * checking before calling the locks_conflict().
777 */
778static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
779{
780 /* POSIX locks owned by the same process do not conflict with
781 * each other.
782 */
783 if (posix_same_owner(caller_fl, sys_fl))
784 return (0);
785
786 /* Check whether they overlap */
787 if (!locks_overlap(caller_fl, sys_fl))
788 return 0;
789
790 return (locks_conflict(caller_fl, sys_fl));
791}
792
793/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
794 * checking before calling the locks_conflict().
795 */
796static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
797{
798 /* FLOCK locks referring to the same filp do not conflict with
799 * each other.
800 */
801 if (caller_fl->fl_file == sys_fl->fl_file)
802 return (0);
803 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
804 return 0;
805
806 return (locks_conflict(caller_fl, sys_fl));
807}
808
809void
810posix_test_lock(struct file *filp, struct file_lock *fl)
811{
812 struct file_lock *cfl;
813 struct file_lock_context *ctx;
814 struct inode *inode = locks_inode(filp);
815
816 ctx = smp_load_acquire(&inode->i_flctx);
817 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
818 fl->fl_type = F_UNLCK;
819 return;
820 }
821
822 spin_lock(&ctx->flc_lock);
823 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
824 if (posix_locks_conflict(fl, cfl)) {
825 locks_copy_conflock(fl, cfl);
826 if (cfl->fl_nspid)
827 fl->fl_pid = pid_vnr(cfl->fl_nspid);
828 goto out;
829 }
830 }
831 fl->fl_type = F_UNLCK;
832out:
833 spin_unlock(&ctx->flc_lock);
834 return;
835}
836EXPORT_SYMBOL(posix_test_lock);
837
838/*
839 * Deadlock detection:
840 *
841 * We attempt to detect deadlocks that are due purely to posix file
842 * locks.
843 *
844 * We assume that a task can be waiting for at most one lock at a time.
845 * So for any acquired lock, the process holding that lock may be
846 * waiting on at most one other lock. That lock in turns may be held by
847 * someone waiting for at most one other lock. Given a requested lock
848 * caller_fl which is about to wait for a conflicting lock block_fl, we
849 * follow this chain of waiters to ensure we are not about to create a
850 * cycle.
851 *
852 * Since we do this before we ever put a process to sleep on a lock, we
853 * are ensured that there is never a cycle; that is what guarantees that
854 * the while() loop in posix_locks_deadlock() eventually completes.
855 *
856 * Note: the above assumption may not be true when handling lock
857 * requests from a broken NFS client. It may also fail in the presence
858 * of tasks (such as posix threads) sharing the same open file table.
859 * To handle those cases, we just bail out after a few iterations.
860 *
861 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
862 * Because the owner is not even nominally tied to a thread of
863 * execution, the deadlock detection below can't reasonably work well. Just
864 * skip it for those.
865 *
866 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
867 * locks that just checks for the case where two tasks are attempting to
868 * upgrade from read to write locks on the same inode.
869 */
870
871#define MAX_DEADLK_ITERATIONS 10
872
873/* Find a lock that the owner of the given block_fl is blocking on. */
874static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
875{
876 struct file_lock *fl;
877
878 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
879 if (posix_same_owner(fl, block_fl))
880 return fl->fl_next;
881 }
882 return NULL;
883}
884
885/* Must be called with the blocked_lock_lock held! */
886static int posix_locks_deadlock(struct file_lock *caller_fl,
887 struct file_lock *block_fl)
888{
889 int i = 0;
890
891 lockdep_assert_held(&blocked_lock_lock);
892
893 /*
894 * This deadlock detector can't reasonably detect deadlocks with
895 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
896 */
897 if (IS_OFDLCK(caller_fl))
898 return 0;
899
900 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
901 if (i++ > MAX_DEADLK_ITERATIONS)
902 return 0;
903 if (posix_same_owner(caller_fl, block_fl))
904 return 1;
905 }
906 return 0;
907}
908
909/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
910 * after any leases, but before any posix locks.
911 *
912 * Note that if called with an FL_EXISTS argument, the caller may determine
913 * whether or not a lock was successfully freed by testing the return
914 * value for -ENOENT.
915 */
916static int flock_lock_inode(struct inode *inode, struct file_lock *request)
917{
918 struct file_lock *new_fl = NULL;
919 struct file_lock *fl;
920 struct file_lock_context *ctx;
921 int error = 0;
922 bool found = false;
923 LIST_HEAD(dispose);
924
925 ctx = locks_get_lock_context(inode, request->fl_type);
926 if (!ctx) {
927 if (request->fl_type != F_UNLCK)
928 return -ENOMEM;
929 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
930 }
931
932 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
933 new_fl = locks_alloc_lock();
934 if (!new_fl)
935 return -ENOMEM;
936 }
937
938 percpu_down_read_preempt_disable(&file_rwsem);
939 spin_lock(&ctx->flc_lock);
940 if (request->fl_flags & FL_ACCESS)
941 goto find_conflict;
942
943 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
944 if (request->fl_file != fl->fl_file)
945 continue;
946 if (request->fl_type == fl->fl_type)
947 goto out;
948 found = true;
949 locks_delete_lock_ctx(fl, &dispose);
950 break;
951 }
952
953 if (request->fl_type == F_UNLCK) {
954 if ((request->fl_flags & FL_EXISTS) && !found)
955 error = -ENOENT;
956 goto out;
957 }
958
959find_conflict:
960 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
961 if (!flock_locks_conflict(request, fl))
962 continue;
963 error = -EAGAIN;
964 if (!(request->fl_flags & FL_SLEEP))
965 goto out;
966 error = FILE_LOCK_DEFERRED;
967 locks_insert_block(fl, request);
968 goto out;
969 }
970 if (request->fl_flags & FL_ACCESS)
971 goto out;
972 locks_copy_lock(new_fl, request);
973 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
974 new_fl = NULL;
975 error = 0;
976
977out:
978 spin_unlock(&ctx->flc_lock);
979 percpu_up_read_preempt_enable(&file_rwsem);
980 if (new_fl)
981 locks_free_lock(new_fl);
982 locks_dispose_list(&dispose);
983 return error;
984}
985
986static int posix_lock_inode(struct inode *inode, struct file_lock *request,
987 struct file_lock *conflock)
988{
989 struct file_lock *fl, *tmp;
990 struct file_lock *new_fl = NULL;
991 struct file_lock *new_fl2 = NULL;
992 struct file_lock *left = NULL;
993 struct file_lock *right = NULL;
994 struct file_lock_context *ctx;
995 int error;
996 bool added = false;
997 LIST_HEAD(dispose);
998
999 ctx = locks_get_lock_context(inode, request->fl_type);
1000 if (!ctx)
1001 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1002
1003 /*
1004 * We may need two file_lock structures for this operation,
1005 * so we get them in advance to avoid races.
1006 *
1007 * In some cases we can be sure, that no new locks will be needed
1008 */
1009 if (!(request->fl_flags & FL_ACCESS) &&
1010 (request->fl_type != F_UNLCK ||
1011 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1012 new_fl = locks_alloc_lock();
1013 new_fl2 = locks_alloc_lock();
1014 }
1015
1016 percpu_down_read_preempt_disable(&file_rwsem);
1017 spin_lock(&ctx->flc_lock);
1018 /*
1019 * New lock request. Walk all POSIX locks and look for conflicts. If
1020 * there are any, either return error or put the request on the
1021 * blocker's list of waiters and the global blocked_hash.
1022 */
1023 if (request->fl_type != F_UNLCK) {
1024 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1025 if (!posix_locks_conflict(request, fl))
1026 continue;
1027 if (conflock)
1028 locks_copy_conflock(conflock, fl);
1029 error = -EAGAIN;
1030 if (!(request->fl_flags & FL_SLEEP))
1031 goto out;
1032 /*
1033 * Deadlock detection and insertion into the blocked
1034 * locks list must be done while holding the same lock!
1035 */
1036 error = -EDEADLK;
1037 spin_lock(&blocked_lock_lock);
1038 if (likely(!posix_locks_deadlock(request, fl))) {
1039 error = FILE_LOCK_DEFERRED;
1040 __locks_insert_block(fl, request);
1041 }
1042 spin_unlock(&blocked_lock_lock);
1043 goto out;
1044 }
1045 }
1046
1047 /* If we're just looking for a conflict, we're done. */
1048 error = 0;
1049 if (request->fl_flags & FL_ACCESS)
1050 goto out;
1051
1052 /* Find the first old lock with the same owner as the new lock */
1053 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1054 if (posix_same_owner(request, fl))
1055 break;
1056 }
1057
1058 /* Process locks with this owner. */
1059 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1060 if (!posix_same_owner(request, fl))
1061 break;
1062
1063 /* Detect adjacent or overlapping regions (if same lock type) */
1064 if (request->fl_type == fl->fl_type) {
1065 /* In all comparisons of start vs end, use
1066 * "start - 1" rather than "end + 1". If end
1067 * is OFFSET_MAX, end + 1 will become negative.
1068 */
1069 if (fl->fl_end < request->fl_start - 1)
1070 continue;
1071 /* If the next lock in the list has entirely bigger
1072 * addresses than the new one, insert the lock here.
1073 */
1074 if (fl->fl_start - 1 > request->fl_end)
1075 break;
1076
1077 /* If we come here, the new and old lock are of the
1078 * same type and adjacent or overlapping. Make one
1079 * lock yielding from the lower start address of both
1080 * locks to the higher end address.
1081 */
1082 if (fl->fl_start > request->fl_start)
1083 fl->fl_start = request->fl_start;
1084 else
1085 request->fl_start = fl->fl_start;
1086 if (fl->fl_end < request->fl_end)
1087 fl->fl_end = request->fl_end;
1088 else
1089 request->fl_end = fl->fl_end;
1090 if (added) {
1091 locks_delete_lock_ctx(fl, &dispose);
1092 continue;
1093 }
1094 request = fl;
1095 added = true;
1096 } else {
1097 /* Processing for different lock types is a bit
1098 * more complex.
1099 */
1100 if (fl->fl_end < request->fl_start)
1101 continue;
1102 if (fl->fl_start > request->fl_end)
1103 break;
1104 if (request->fl_type == F_UNLCK)
1105 added = true;
1106 if (fl->fl_start < request->fl_start)
1107 left = fl;
1108 /* If the next lock in the list has a higher end
1109 * address than the new one, insert the new one here.
1110 */
1111 if (fl->fl_end > request->fl_end) {
1112 right = fl;
1113 break;
1114 }
1115 if (fl->fl_start >= request->fl_start) {
1116 /* The new lock completely replaces an old
1117 * one (This may happen several times).
1118 */
1119 if (added) {
1120 locks_delete_lock_ctx(fl, &dispose);
1121 continue;
1122 }
1123 /*
1124 * Replace the old lock with new_fl, and
1125 * remove the old one. It's safe to do the
1126 * insert here since we know that we won't be
1127 * using new_fl later, and that the lock is
1128 * just replacing an existing lock.
1129 */
1130 error = -ENOLCK;
1131 if (!new_fl)
1132 goto out;
1133 locks_copy_lock(new_fl, request);
1134 request = new_fl;
1135 new_fl = NULL;
1136 locks_insert_lock_ctx(request, &fl->fl_list);
1137 locks_delete_lock_ctx(fl, &dispose);
1138 added = true;
1139 }
1140 }
1141 }
1142
1143 /*
1144 * The above code only modifies existing locks in case of merging or
1145 * replacing. If new lock(s) need to be inserted all modifications are
1146 * done below this, so it's safe yet to bail out.
1147 */
1148 error = -ENOLCK; /* "no luck" */
1149 if (right && left == right && !new_fl2)
1150 goto out;
1151
1152 error = 0;
1153 if (!added) {
1154 if (request->fl_type == F_UNLCK) {
1155 if (request->fl_flags & FL_EXISTS)
1156 error = -ENOENT;
1157 goto out;
1158 }
1159
1160 if (!new_fl) {
1161 error = -ENOLCK;
1162 goto out;
1163 }
1164 locks_copy_lock(new_fl, request);
1165 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1166 fl = new_fl;
1167 new_fl = NULL;
1168 }
1169 if (right) {
1170 if (left == right) {
1171 /* The new lock breaks the old one in two pieces,
1172 * so we have to use the second new lock.
1173 */
1174 left = new_fl2;
1175 new_fl2 = NULL;
1176 locks_copy_lock(left, right);
1177 locks_insert_lock_ctx(left, &fl->fl_list);
1178 }
1179 right->fl_start = request->fl_end + 1;
1180 locks_wake_up_blocks(right);
1181 }
1182 if (left) {
1183 left->fl_end = request->fl_start - 1;
1184 locks_wake_up_blocks(left);
1185 }
1186 out:
1187 spin_unlock(&ctx->flc_lock);
1188 percpu_up_read_preempt_enable(&file_rwsem);
1189 /*
1190 * Free any unused locks.
1191 */
1192 if (new_fl)
1193 locks_free_lock(new_fl);
1194 if (new_fl2)
1195 locks_free_lock(new_fl2);
1196 locks_dispose_list(&dispose);
1197 trace_posix_lock_inode(inode, request, error);
1198
1199 return error;
1200}
1201
1202/**
1203 * posix_lock_file - Apply a POSIX-style lock to a file
1204 * @filp: The file to apply the lock to
1205 * @fl: The lock to be applied
1206 * @conflock: Place to return a copy of the conflicting lock, if found.
1207 *
1208 * Add a POSIX style lock to a file.
1209 * We merge adjacent & overlapping locks whenever possible.
1210 * POSIX locks are sorted by owner task, then by starting address
1211 *
1212 * Note that if called with an FL_EXISTS argument, the caller may determine
1213 * whether or not a lock was successfully freed by testing the return
1214 * value for -ENOENT.
1215 */
1216int posix_lock_file(struct file *filp, struct file_lock *fl,
1217 struct file_lock *conflock)
1218{
1219 return posix_lock_inode(locks_inode(filp), fl, conflock);
1220}
1221EXPORT_SYMBOL(posix_lock_file);
1222
1223/**
1224 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1225 * @inode: inode of file to which lock request should be applied
1226 * @fl: The lock to be applied
1227 *
1228 * Apply a POSIX style lock request to an inode.
1229 */
1230static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1231{
1232 int error;
1233 might_sleep ();
1234 for (;;) {
1235 error = posix_lock_inode(inode, fl, NULL);
1236 if (error != FILE_LOCK_DEFERRED)
1237 break;
1238 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1239 if (!error)
1240 continue;
1241
1242 locks_delete_block(fl);
1243 break;
1244 }
1245 return error;
1246}
1247
1248#ifdef CONFIG_MANDATORY_FILE_LOCKING
1249/**
1250 * locks_mandatory_locked - Check for an active lock
1251 * @file: the file to check
1252 *
1253 * Searches the inode's list of locks to find any POSIX locks which conflict.
1254 * This function is called from locks_verify_locked() only.
1255 */
1256int locks_mandatory_locked(struct file *file)
1257{
1258 int ret;
1259 struct inode *inode = locks_inode(file);
1260 struct file_lock_context *ctx;
1261 struct file_lock *fl;
1262
1263 ctx = smp_load_acquire(&inode->i_flctx);
1264 if (!ctx || list_empty_careful(&ctx->flc_posix))
1265 return 0;
1266
1267 /*
1268 * Search the lock list for this inode for any POSIX locks.
1269 */
1270 spin_lock(&ctx->flc_lock);
1271 ret = 0;
1272 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1273 if (fl->fl_owner != current->files &&
1274 fl->fl_owner != file) {
1275 ret = -EAGAIN;
1276 break;
1277 }
1278 }
1279 spin_unlock(&ctx->flc_lock);
1280 return ret;
1281}
1282
1283/**
1284 * locks_mandatory_area - Check for a conflicting lock
1285 * @inode: the file to check
1286 * @filp: how the file was opened (if it was)
1287 * @start: first byte in the file to check
1288 * @end: lastbyte in the file to check
1289 * @type: %F_WRLCK for a write lock, else %F_RDLCK
1290 *
1291 * Searches the inode's list of locks to find any POSIX locks which conflict.
1292 */
1293int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1294 loff_t end, unsigned char type)
1295{
1296 struct file_lock fl;
1297 int error;
1298 bool sleep = false;
1299
1300 locks_init_lock(&fl);
1301 fl.fl_pid = current->tgid;
1302 fl.fl_file = filp;
1303 fl.fl_flags = FL_POSIX | FL_ACCESS;
1304 if (filp && !(filp->f_flags & O_NONBLOCK))
1305 sleep = true;
1306 fl.fl_type = type;
1307 fl.fl_start = start;
1308 fl.fl_end = end;
1309
1310 for (;;) {
1311 if (filp) {
1312 fl.fl_owner = filp;
1313 fl.fl_flags &= ~FL_SLEEP;
1314 error = posix_lock_inode(inode, &fl, NULL);
1315 if (!error)
1316 break;
1317 }
1318
1319 if (sleep)
1320 fl.fl_flags |= FL_SLEEP;
1321 fl.fl_owner = current->files;
1322 error = posix_lock_inode(inode, &fl, NULL);
1323 if (error != FILE_LOCK_DEFERRED)
1324 break;
1325 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1326 if (!error) {
1327 /*
1328 * If we've been sleeping someone might have
1329 * changed the permissions behind our back.
1330 */
1331 if (__mandatory_lock(inode))
1332 continue;
1333 }
1334
1335 locks_delete_block(&fl);
1336 break;
1337 }
1338
1339 return error;
1340}
1341
1342EXPORT_SYMBOL(locks_mandatory_area);
1343#endif /* CONFIG_MANDATORY_FILE_LOCKING */
1344
1345static void lease_clear_pending(struct file_lock *fl, int arg)
1346{
1347 switch (arg) {
1348 case F_UNLCK:
1349 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1350 /* fall through: */
1351 case F_RDLCK:
1352 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1353 }
1354}
1355
1356/* We already had a lease on this file; just change its type */
1357int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1358{
1359 int error = assign_type(fl, arg);
1360
1361 if (error)
1362 return error;
1363 lease_clear_pending(fl, arg);
1364 locks_wake_up_blocks(fl);
1365 if (arg == F_UNLCK) {
1366 struct file *filp = fl->fl_file;
1367
1368 f_delown(filp);
1369 filp->f_owner.signum = 0;
1370 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1371 if (fl->fl_fasync != NULL) {
1372 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1373 fl->fl_fasync = NULL;
1374 }
1375 locks_delete_lock_ctx(fl, dispose);
1376 }
1377 return 0;
1378}
1379EXPORT_SYMBOL(lease_modify);
1380
1381static bool past_time(unsigned long then)
1382{
1383 if (!then)
1384 /* 0 is a special value meaning "this never expires": */
1385 return false;
1386 return time_after(jiffies, then);
1387}
1388
1389static void time_out_leases(struct inode *inode, struct list_head *dispose)
1390{
1391 struct file_lock_context *ctx = inode->i_flctx;
1392 struct file_lock *fl, *tmp;
1393
1394 lockdep_assert_held(&ctx->flc_lock);
1395
1396 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1397 trace_time_out_leases(inode, fl);
1398 if (past_time(fl->fl_downgrade_time))
1399 lease_modify(fl, F_RDLCK, dispose);
1400 if (past_time(fl->fl_break_time))
1401 lease_modify(fl, F_UNLCK, dispose);
1402 }
1403}
1404
1405static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1406{
1407 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1408 return false;
1409 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1410 return false;
1411 return locks_conflict(breaker, lease);
1412}
1413
1414static bool
1415any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1416{
1417 struct file_lock_context *ctx = inode->i_flctx;
1418 struct file_lock *fl;
1419
1420 lockdep_assert_held(&ctx->flc_lock);
1421
1422 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1423 if (leases_conflict(fl, breaker))
1424 return true;
1425 }
1426 return false;
1427}
1428
1429/**
1430 * __break_lease - revoke all outstanding leases on file
1431 * @inode: the inode of the file to return
1432 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1433 * break all leases
1434 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1435 * only delegations
1436 *
1437 * break_lease (inlined for speed) has checked there already is at least
1438 * some kind of lock (maybe a lease) on this file. Leases are broken on
1439 * a call to open() or truncate(). This function can sleep unless you
1440 * specified %O_NONBLOCK to your open().
1441 */
1442int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1443{
1444 int error = 0;
1445 struct file_lock_context *ctx;
1446 struct file_lock *new_fl, *fl, *tmp;
1447 unsigned long break_time;
1448 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1449 LIST_HEAD(dispose);
1450
1451 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1452 if (IS_ERR(new_fl))
1453 return PTR_ERR(new_fl);
1454 new_fl->fl_flags = type;
1455
1456 /* typically we will check that ctx is non-NULL before calling */
1457 ctx = smp_load_acquire(&inode->i_flctx);
1458 if (!ctx) {
1459 WARN_ON_ONCE(1);
1460 return error;
1461 }
1462
1463 percpu_down_read_preempt_disable(&file_rwsem);
1464 spin_lock(&ctx->flc_lock);
1465
1466 time_out_leases(inode, &dispose);
1467
1468 if (!any_leases_conflict(inode, new_fl))
1469 goto out;
1470
1471 break_time = 0;
1472 if (lease_break_time > 0) {
1473 break_time = jiffies + lease_break_time * HZ;
1474 if (break_time == 0)
1475 break_time++; /* so that 0 means no break time */
1476 }
1477
1478 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1479 if (!leases_conflict(fl, new_fl))
1480 continue;
1481 if (want_write) {
1482 if (fl->fl_flags & FL_UNLOCK_PENDING)
1483 continue;
1484 fl->fl_flags |= FL_UNLOCK_PENDING;
1485 fl->fl_break_time = break_time;
1486 } else {
1487 if (lease_breaking(fl))
1488 continue;
1489 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1490 fl->fl_downgrade_time = break_time;
1491 }
1492 if (fl->fl_lmops->lm_break(fl))
1493 locks_delete_lock_ctx(fl, &dispose);
1494 }
1495
1496 if (list_empty(&ctx->flc_lease))
1497 goto out;
1498
1499 if (mode & O_NONBLOCK) {
1500 trace_break_lease_noblock(inode, new_fl);
1501 error = -EWOULDBLOCK;
1502 goto out;
1503 }
1504
1505restart:
1506 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1507 break_time = fl->fl_break_time;
1508 if (break_time != 0)
1509 break_time -= jiffies;
1510 if (break_time == 0)
1511 break_time++;
1512 locks_insert_block(fl, new_fl);
1513 trace_break_lease_block(inode, new_fl);
1514 spin_unlock(&ctx->flc_lock);
1515 percpu_up_read_preempt_enable(&file_rwsem);
1516
1517 locks_dispose_list(&dispose);
1518 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1519 !new_fl->fl_next, break_time);
1520
1521 percpu_down_read_preempt_disable(&file_rwsem);
1522 spin_lock(&ctx->flc_lock);
1523 trace_break_lease_unblock(inode, new_fl);
1524 locks_delete_block(new_fl);
1525 if (error >= 0) {
1526 /*
1527 * Wait for the next conflicting lease that has not been
1528 * broken yet
1529 */
1530 if (error == 0)
1531 time_out_leases(inode, &dispose);
1532 if (any_leases_conflict(inode, new_fl))
1533 goto restart;
1534 error = 0;
1535 }
1536out:
1537 spin_unlock(&ctx->flc_lock);
1538 percpu_up_read_preempt_enable(&file_rwsem);
1539 locks_dispose_list(&dispose);
1540 locks_free_lock(new_fl);
1541 return error;
1542}
1543
1544EXPORT_SYMBOL(__break_lease);
1545
1546/**
1547 * lease_get_mtime - get the last modified time of an inode
1548 * @inode: the inode
1549 * @time: pointer to a timespec which will contain the last modified time
1550 *
1551 * This is to force NFS clients to flush their caches for files with
1552 * exclusive leases. The justification is that if someone has an
1553 * exclusive lease, then they could be modifying it.
1554 */
1555void lease_get_mtime(struct inode *inode, struct timespec *time)
1556{
1557 bool has_lease = false;
1558 struct file_lock_context *ctx;
1559 struct file_lock *fl;
1560
1561 ctx = smp_load_acquire(&inode->i_flctx);
1562 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1563 spin_lock(&ctx->flc_lock);
1564 fl = list_first_entry_or_null(&ctx->flc_lease,
1565 struct file_lock, fl_list);
1566 if (fl && (fl->fl_type == F_WRLCK))
1567 has_lease = true;
1568 spin_unlock(&ctx->flc_lock);
1569 }
1570
1571 if (has_lease)
1572 *time = current_time(inode);
1573 else
1574 *time = inode->i_mtime;
1575}
1576
1577EXPORT_SYMBOL(lease_get_mtime);
1578
1579/**
1580 * fcntl_getlease - Enquire what lease is currently active
1581 * @filp: the file
1582 *
1583 * The value returned by this function will be one of
1584 * (if no lease break is pending):
1585 *
1586 * %F_RDLCK to indicate a shared lease is held.
1587 *
1588 * %F_WRLCK to indicate an exclusive lease is held.
1589 *
1590 * %F_UNLCK to indicate no lease is held.
1591 *
1592 * (if a lease break is pending):
1593 *
1594 * %F_RDLCK to indicate an exclusive lease needs to be
1595 * changed to a shared lease (or removed).
1596 *
1597 * %F_UNLCK to indicate the lease needs to be removed.
1598 *
1599 * XXX: sfr & willy disagree over whether F_INPROGRESS
1600 * should be returned to userspace.
1601 */
1602int fcntl_getlease(struct file *filp)
1603{
1604 struct file_lock *fl;
1605 struct inode *inode = locks_inode(filp);
1606 struct file_lock_context *ctx;
1607 int type = F_UNLCK;
1608 LIST_HEAD(dispose);
1609
1610 ctx = smp_load_acquire(&inode->i_flctx);
1611 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1612 percpu_down_read_preempt_disable(&file_rwsem);
1613 spin_lock(&ctx->flc_lock);
1614 time_out_leases(inode, &dispose);
1615 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1616 if (fl->fl_file != filp)
1617 continue;
1618 type = target_leasetype(fl);
1619 break;
1620 }
1621 spin_unlock(&ctx->flc_lock);
1622 percpu_up_read_preempt_enable(&file_rwsem);
1623
1624 locks_dispose_list(&dispose);
1625 }
1626 return type;
1627}
1628
1629/**
1630 * check_conflicting_open - see if the given dentry points to a file that has
1631 * an existing open that would conflict with the
1632 * desired lease.
1633 * @dentry: dentry to check
1634 * @arg: type of lease that we're trying to acquire
1635 * @flags: current lock flags
1636 *
1637 * Check to see if there's an existing open fd on this file that would
1638 * conflict with the lease we're trying to set.
1639 */
1640static int
1641check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
1642{
1643 int ret = 0;
1644 struct inode *inode = dentry->d_inode;
1645
1646 if (flags & FL_LAYOUT)
1647 return 0;
1648
1649 if ((arg == F_RDLCK) &&
1650 (atomic_read(&d_real_inode(dentry)->i_writecount) > 0))
1651 return -EAGAIN;
1652
1653 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1654 (atomic_read(&inode->i_count) > 1)))
1655 ret = -EAGAIN;
1656
1657 return ret;
1658}
1659
1660static int
1661generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1662{
1663 struct file_lock *fl, *my_fl = NULL, *lease;
1664 struct dentry *dentry = filp->f_path.dentry;
1665 struct inode *inode = dentry->d_inode;
1666 struct file_lock_context *ctx;
1667 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1668 int error;
1669 LIST_HEAD(dispose);
1670
1671 lease = *flp;
1672 trace_generic_add_lease(inode, lease);
1673
1674 /* Note that arg is never F_UNLCK here */
1675 ctx = locks_get_lock_context(inode, arg);
1676 if (!ctx)
1677 return -ENOMEM;
1678
1679 /*
1680 * In the delegation case we need mutual exclusion with
1681 * a number of operations that take the i_mutex. We trylock
1682 * because delegations are an optional optimization, and if
1683 * there's some chance of a conflict--we'd rather not
1684 * bother, maybe that's a sign this just isn't a good file to
1685 * hand out a delegation on.
1686 */
1687 if (is_deleg && !inode_trylock(inode))
1688 return -EAGAIN;
1689
1690 if (is_deleg && arg == F_WRLCK) {
1691 /* Write delegations are not currently supported: */
1692 inode_unlock(inode);
1693 WARN_ON_ONCE(1);
1694 return -EINVAL;
1695 }
1696
1697 percpu_down_read_preempt_disable(&file_rwsem);
1698 spin_lock(&ctx->flc_lock);
1699 time_out_leases(inode, &dispose);
1700 error = check_conflicting_open(dentry, arg, lease->fl_flags);
1701 if (error)
1702 goto out;
1703
1704 /*
1705 * At this point, we know that if there is an exclusive
1706 * lease on this file, then we hold it on this filp
1707 * (otherwise our open of this file would have blocked).
1708 * And if we are trying to acquire an exclusive lease,
1709 * then the file is not open by anyone (including us)
1710 * except for this filp.
1711 */
1712 error = -EAGAIN;
1713 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1714 if (fl->fl_file == filp &&
1715 fl->fl_owner == lease->fl_owner) {
1716 my_fl = fl;
1717 continue;
1718 }
1719
1720 /*
1721 * No exclusive leases if someone else has a lease on
1722 * this file:
1723 */
1724 if (arg == F_WRLCK)
1725 goto out;
1726 /*
1727 * Modifying our existing lease is OK, but no getting a
1728 * new lease if someone else is opening for write:
1729 */
1730 if (fl->fl_flags & FL_UNLOCK_PENDING)
1731 goto out;
1732 }
1733
1734 if (my_fl != NULL) {
1735 lease = my_fl;
1736 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1737 if (error)
1738 goto out;
1739 goto out_setup;
1740 }
1741
1742 error = -EINVAL;
1743 if (!leases_enable)
1744 goto out;
1745
1746 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1747 /*
1748 * The check in break_lease() is lockless. It's possible for another
1749 * open to race in after we did the earlier check for a conflicting
1750 * open but before the lease was inserted. Check again for a
1751 * conflicting open and cancel the lease if there is one.
1752 *
1753 * We also add a barrier here to ensure that the insertion of the lock
1754 * precedes these checks.
1755 */
1756 smp_mb();
1757 error = check_conflicting_open(dentry, arg, lease->fl_flags);
1758 if (error) {
1759 locks_unlink_lock_ctx(lease);
1760 goto out;
1761 }
1762
1763out_setup:
1764 if (lease->fl_lmops->lm_setup)
1765 lease->fl_lmops->lm_setup(lease, priv);
1766out:
1767 spin_unlock(&ctx->flc_lock);
1768 percpu_up_read_preempt_enable(&file_rwsem);
1769 locks_dispose_list(&dispose);
1770 if (is_deleg)
1771 inode_unlock(inode);
1772 if (!error && !my_fl)
1773 *flp = NULL;
1774 return error;
1775}
1776
1777static int generic_delete_lease(struct file *filp, void *owner)
1778{
1779 int error = -EAGAIN;
1780 struct file_lock *fl, *victim = NULL;
1781 struct inode *inode = locks_inode(filp);
1782 struct file_lock_context *ctx;
1783 LIST_HEAD(dispose);
1784
1785 ctx = smp_load_acquire(&inode->i_flctx);
1786 if (!ctx) {
1787 trace_generic_delete_lease(inode, NULL);
1788 return error;
1789 }
1790
1791 percpu_down_read_preempt_disable(&file_rwsem);
1792 spin_lock(&ctx->flc_lock);
1793 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1794 if (fl->fl_file == filp &&
1795 fl->fl_owner == owner) {
1796 victim = fl;
1797 break;
1798 }
1799 }
1800 trace_generic_delete_lease(inode, victim);
1801 if (victim)
1802 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1803 spin_unlock(&ctx->flc_lock);
1804 percpu_up_read_preempt_enable(&file_rwsem);
1805 locks_dispose_list(&dispose);
1806 return error;
1807}
1808
1809/**
1810 * generic_setlease - sets a lease on an open file
1811 * @filp: file pointer
1812 * @arg: type of lease to obtain
1813 * @flp: input - file_lock to use, output - file_lock inserted
1814 * @priv: private data for lm_setup (may be NULL if lm_setup
1815 * doesn't require it)
1816 *
1817 * The (input) flp->fl_lmops->lm_break function is required
1818 * by break_lease().
1819 */
1820int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1821 void **priv)
1822{
1823 struct inode *inode = locks_inode(filp);
1824 int error;
1825
1826 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1827 return -EACCES;
1828 if (!S_ISREG(inode->i_mode))
1829 return -EINVAL;
1830 error = security_file_lock(filp, arg);
1831 if (error)
1832 return error;
1833
1834 switch (arg) {
1835 case F_UNLCK:
1836 return generic_delete_lease(filp, *priv);
1837 case F_RDLCK:
1838 case F_WRLCK:
1839 if (!(*flp)->fl_lmops->lm_break) {
1840 WARN_ON_ONCE(1);
1841 return -ENOLCK;
1842 }
1843
1844 return generic_add_lease(filp, arg, flp, priv);
1845 default:
1846 return -EINVAL;
1847 }
1848}
1849EXPORT_SYMBOL(generic_setlease);
1850
1851/**
1852 * vfs_setlease - sets a lease on an open file
1853 * @filp: file pointer
1854 * @arg: type of lease to obtain
1855 * @lease: file_lock to use when adding a lease
1856 * @priv: private info for lm_setup when adding a lease (may be
1857 * NULL if lm_setup doesn't require it)
1858 *
1859 * Call this to establish a lease on the file. The "lease" argument is not
1860 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1861 * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1862 * if not, this function will return -ENOLCK (and generate a scary-looking
1863 * stack trace).
1864 *
1865 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1866 * may be NULL if the lm_setup operation doesn't require it.
1867 */
1868int
1869vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1870{
1871 if (filp->f_op->setlease && is_remote_lock(filp))
1872 return filp->f_op->setlease(filp, arg, lease, priv);
1873 else
1874 return generic_setlease(filp, arg, lease, priv);
1875}
1876EXPORT_SYMBOL_GPL(vfs_setlease);
1877
1878static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1879{
1880 struct file_lock *fl;
1881 struct fasync_struct *new;
1882 int error;
1883
1884 fl = lease_alloc(filp, arg);
1885 if (IS_ERR(fl))
1886 return PTR_ERR(fl);
1887
1888 new = fasync_alloc();
1889 if (!new) {
1890 locks_free_lock(fl);
1891 return -ENOMEM;
1892 }
1893 new->fa_fd = fd;
1894
1895 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1896 if (fl)
1897 locks_free_lock(fl);
1898 if (new)
1899 fasync_free(new);
1900 return error;
1901}
1902
1903/**
1904 * fcntl_setlease - sets a lease on an open file
1905 * @fd: open file descriptor
1906 * @filp: file pointer
1907 * @arg: type of lease to obtain
1908 *
1909 * Call this fcntl to establish a lease on the file.
1910 * Note that you also need to call %F_SETSIG to
1911 * receive a signal when the lease is broken.
1912 */
1913int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1914{
1915 if (arg == F_UNLCK)
1916 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
1917 return do_fcntl_add_lease(fd, filp, arg);
1918}
1919
1920/**
1921 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
1922 * @inode: inode of the file to apply to
1923 * @fl: The lock to be applied
1924 *
1925 * Apply a FLOCK style lock request to an inode.
1926 */
1927static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1928{
1929 int error;
1930 might_sleep();
1931 for (;;) {
1932 error = flock_lock_inode(inode, fl);
1933 if (error != FILE_LOCK_DEFERRED)
1934 break;
1935 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1936 if (!error)
1937 continue;
1938
1939 locks_delete_block(fl);
1940 break;
1941 }
1942 return error;
1943}
1944
1945/**
1946 * locks_lock_inode_wait - Apply a lock to an inode
1947 * @inode: inode of the file to apply to
1948 * @fl: The lock to be applied
1949 *
1950 * Apply a POSIX or FLOCK style lock request to an inode.
1951 */
1952int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1953{
1954 int res = 0;
1955 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
1956 case FL_POSIX:
1957 res = posix_lock_inode_wait(inode, fl);
1958 break;
1959 case FL_FLOCK:
1960 res = flock_lock_inode_wait(inode, fl);
1961 break;
1962 default:
1963 BUG();
1964 }
1965 return res;
1966}
1967EXPORT_SYMBOL(locks_lock_inode_wait);
1968
1969/**
1970 * sys_flock: - flock() system call.
1971 * @fd: the file descriptor to lock.
1972 * @cmd: the type of lock to apply.
1973 *
1974 * Apply a %FL_FLOCK style lock to an open file descriptor.
1975 * The @cmd can be one of
1976 *
1977 * %LOCK_SH -- a shared lock.
1978 *
1979 * %LOCK_EX -- an exclusive lock.
1980 *
1981 * %LOCK_UN -- remove an existing lock.
1982 *
1983 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1984 *
1985 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1986 * processes read and write access respectively.
1987 */
1988SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1989{
1990 struct fd f = fdget(fd);
1991 struct file_lock *lock;
1992 int can_sleep, unlock;
1993 int error;
1994
1995 error = -EBADF;
1996 if (!f.file)
1997 goto out;
1998
1999 can_sleep = !(cmd & LOCK_NB);
2000 cmd &= ~LOCK_NB;
2001 unlock = (cmd == LOCK_UN);
2002
2003 if (!unlock && !(cmd & LOCK_MAND) &&
2004 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2005 goto out_putf;
2006
2007 lock = flock_make_lock(f.file, cmd);
2008 if (IS_ERR(lock)) {
2009 error = PTR_ERR(lock);
2010 goto out_putf;
2011 }
2012
2013 if (can_sleep)
2014 lock->fl_flags |= FL_SLEEP;
2015
2016 error = security_file_lock(f.file, lock->fl_type);
2017 if (error)
2018 goto out_free;
2019
2020 if (f.file->f_op->flock && is_remote_lock(f.file))
2021 error = f.file->f_op->flock(f.file,
2022 (can_sleep) ? F_SETLKW : F_SETLK,
2023 lock);
2024 else
2025 error = locks_lock_file_wait(f.file, lock);
2026
2027 out_free:
2028 locks_free_lock(lock);
2029
2030 out_putf:
2031 fdput(f);
2032 out:
2033 return error;
2034}
2035
2036/**
2037 * vfs_test_lock - test file byte range lock
2038 * @filp: The file to test lock for
2039 * @fl: The lock to test; also used to hold result
2040 *
2041 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2042 * setting conf->fl_type to something other than F_UNLCK.
2043 */
2044int vfs_test_lock(struct file *filp, struct file_lock *fl)
2045{
2046 if (filp->f_op->lock && is_remote_lock(filp))
2047 return filp->f_op->lock(filp, F_GETLK, fl);
2048 posix_test_lock(filp, fl);
2049 return 0;
2050}
2051EXPORT_SYMBOL_GPL(vfs_test_lock);
2052
2053static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2054{
2055 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
2056#if BITS_PER_LONG == 32
2057 /*
2058 * Make sure we can represent the posix lock via
2059 * legacy 32bit flock.
2060 */
2061 if (fl->fl_start > OFFT_OFFSET_MAX)
2062 return -EOVERFLOW;
2063 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2064 return -EOVERFLOW;
2065#endif
2066 flock->l_start = fl->fl_start;
2067 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2068 fl->fl_end - fl->fl_start + 1;
2069 flock->l_whence = 0;
2070 flock->l_type = fl->fl_type;
2071 return 0;
2072}
2073
2074#if BITS_PER_LONG == 32
2075static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2076{
2077 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
2078 flock->l_start = fl->fl_start;
2079 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2080 fl->fl_end - fl->fl_start + 1;
2081 flock->l_whence = 0;
2082 flock->l_type = fl->fl_type;
2083}
2084#endif
2085
2086/* Report the first existing lock that would conflict with l.
2087 * This implements the F_GETLK command of fcntl().
2088 */
2089int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
2090{
2091 struct file_lock file_lock;
2092 struct flock flock;
2093 int error;
2094
2095 error = -EFAULT;
2096 if (copy_from_user(&flock, l, sizeof(flock)))
2097 goto out;
2098 error = -EINVAL;
2099 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2100 goto out;
2101
2102 error = flock_to_posix_lock(filp, &file_lock, &flock);
2103 if (error)
2104 goto out;
2105
2106 if (cmd == F_OFD_GETLK) {
2107 error = -EINVAL;
2108 if (flock.l_pid != 0)
2109 goto out;
2110
2111 cmd = F_GETLK;
2112 file_lock.fl_flags |= FL_OFDLCK;
2113 file_lock.fl_owner = filp;
2114 }
2115
2116 error = vfs_test_lock(filp, &file_lock);
2117 if (error)
2118 goto out;
2119
2120 flock.l_type = file_lock.fl_type;
2121 if (file_lock.fl_type != F_UNLCK) {
2122 error = posix_lock_to_flock(&flock, &file_lock);
2123 if (error)
2124 goto rel_priv;
2125 }
2126 error = -EFAULT;
2127 if (!copy_to_user(l, &flock, sizeof(flock)))
2128 error = 0;
2129rel_priv:
2130 locks_release_private(&file_lock);
2131out:
2132 return error;
2133}
2134
2135/**
2136 * vfs_lock_file - file byte range lock
2137 * @filp: The file to apply the lock to
2138 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2139 * @fl: The lock to be applied
2140 * @conf: Place to return a copy of the conflicting lock, if found.
2141 *
2142 * A caller that doesn't care about the conflicting lock may pass NULL
2143 * as the final argument.
2144 *
2145 * If the filesystem defines a private ->lock() method, then @conf will
2146 * be left unchanged; so a caller that cares should initialize it to
2147 * some acceptable default.
2148 *
2149 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2150 * locks, the ->lock() interface may return asynchronously, before the lock has
2151 * been granted or denied by the underlying filesystem, if (and only if)
2152 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2153 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2154 * the request is for a blocking lock. When ->lock() does return asynchronously,
2155 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2156 * request completes.
2157 * If the request is for non-blocking lock the file system should return
2158 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2159 * with the result. If the request timed out the callback routine will return a
2160 * nonzero return code and the file system should release the lock. The file
2161 * system is also responsible to keep a corresponding posix lock when it
2162 * grants a lock so the VFS can find out which locks are locally held and do
2163 * the correct lock cleanup when required.
2164 * The underlying filesystem must not drop the kernel lock or call
2165 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2166 * return code.
2167 */
2168int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2169{
2170 if (filp->f_op->lock && is_remote_lock(filp))
2171 return filp->f_op->lock(filp, cmd, fl);
2172 else
2173 return posix_lock_file(filp, fl, conf);
2174}
2175EXPORT_SYMBOL_GPL(vfs_lock_file);
2176
2177static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2178 struct file_lock *fl)
2179{
2180 int error;
2181
2182 error = security_file_lock(filp, fl->fl_type);
2183 if (error)
2184 return error;
2185
2186 for (;;) {
2187 error = vfs_lock_file(filp, cmd, fl, NULL);
2188 if (error != FILE_LOCK_DEFERRED)
2189 break;
2190 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2191 if (!error)
2192 continue;
2193
2194 locks_delete_block(fl);
2195 break;
2196 }
2197
2198 return error;
2199}
2200
2201/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2202static int
2203check_fmode_for_setlk(struct file_lock *fl)
2204{
2205 switch (fl->fl_type) {
2206 case F_RDLCK:
2207 if (!(fl->fl_file->f_mode & FMODE_READ))
2208 return -EBADF;
2209 break;
2210 case F_WRLCK:
2211 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2212 return -EBADF;
2213 }
2214 return 0;
2215}
2216
2217/* Apply the lock described by l to an open file descriptor.
2218 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2219 */
2220int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2221 struct flock __user *l)
2222{
2223 struct file_lock *file_lock = locks_alloc_lock();
2224 struct flock flock;
2225 struct inode *inode;
2226 struct file *f;
2227 int error;
2228
2229 if (file_lock == NULL)
2230 return -ENOLCK;
2231
2232 inode = locks_inode(filp);
2233
2234 /*
2235 * This might block, so we do it before checking the inode.
2236 */
2237 error = -EFAULT;
2238 if (copy_from_user(&flock, l, sizeof(flock)))
2239 goto out;
2240
2241 /* Don't allow mandatory locks on files that may be memory mapped
2242 * and shared.
2243 */
2244 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2245 error = -EAGAIN;
2246 goto out;
2247 }
2248
2249 error = flock_to_posix_lock(filp, file_lock, &flock);
2250 if (error)
2251 goto out;
2252
2253 error = check_fmode_for_setlk(file_lock);
2254 if (error)
2255 goto out;
2256
2257 /*
2258 * If the cmd is requesting file-private locks, then set the
2259 * FL_OFDLCK flag and override the owner.
2260 */
2261 switch (cmd) {
2262 case F_OFD_SETLK:
2263 error = -EINVAL;
2264 if (flock.l_pid != 0)
2265 goto out;
2266
2267 cmd = F_SETLK;
2268 file_lock->fl_flags |= FL_OFDLCK;
2269 file_lock->fl_owner = filp;
2270 break;
2271 case F_OFD_SETLKW:
2272 error = -EINVAL;
2273 if (flock.l_pid != 0)
2274 goto out;
2275
2276 cmd = F_SETLKW;
2277 file_lock->fl_flags |= FL_OFDLCK;
2278 file_lock->fl_owner = filp;
2279 /* Fallthrough */
2280 case F_SETLKW:
2281 file_lock->fl_flags |= FL_SLEEP;
2282 }
2283
2284 error = do_lock_file_wait(filp, cmd, file_lock);
2285
2286 /*
2287 * Attempt to detect a close/fcntl race and recover by releasing the
2288 * lock that was just acquired. There is no need to do that when we're
2289 * unlocking though, or for OFD locks.
2290 */
2291 if (!error && file_lock->fl_type != F_UNLCK &&
2292 !(file_lock->fl_flags & FL_OFDLCK)) {
2293 /*
2294 * We need that spin_lock here - it prevents reordering between
2295 * update of i_flctx->flc_posix and check for it done in
2296 * close(). rcu_read_lock() wouldn't do.
2297 */
2298 spin_lock(¤t->files->file_lock);
2299 f = fcheck(fd);
2300 spin_unlock(¤t->files->file_lock);
2301 if (f != filp) {
2302 file_lock->fl_type = F_UNLCK;
2303 error = do_lock_file_wait(filp, cmd, file_lock);
2304 WARN_ON_ONCE(error);
2305 error = -EBADF;
2306 }
2307 }
2308out:
2309 trace_fcntl_setlk(inode, file_lock, error);
2310 locks_free_lock(file_lock);
2311 return error;
2312}
2313
2314#if BITS_PER_LONG == 32
2315/* Report the first existing lock that would conflict with l.
2316 * This implements the F_GETLK command of fcntl().
2317 */
2318int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
2319{
2320 struct file_lock file_lock;
2321 struct flock64 flock;
2322 int error;
2323
2324 error = -EFAULT;
2325 if (copy_from_user(&flock, l, sizeof(flock)))
2326 goto out;
2327 error = -EINVAL;
2328 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2329 goto out;
2330
2331 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2332 if (error)
2333 goto out;
2334
2335 if (cmd == F_OFD_GETLK) {
2336 error = -EINVAL;
2337 if (flock.l_pid != 0)
2338 goto out;
2339
2340 cmd = F_GETLK64;
2341 file_lock.fl_flags |= FL_OFDLCK;
2342 file_lock.fl_owner = filp;
2343 }
2344
2345 error = vfs_test_lock(filp, &file_lock);
2346 if (error)
2347 goto out;
2348
2349 flock.l_type = file_lock.fl_type;
2350 if (file_lock.fl_type != F_UNLCK)
2351 posix_lock_to_flock64(&flock, &file_lock);
2352
2353 error = -EFAULT;
2354 if (!copy_to_user(l, &flock, sizeof(flock)))
2355 error = 0;
2356
2357 locks_release_private(&file_lock);
2358out:
2359 return error;
2360}
2361
2362/* Apply the lock described by l to an open file descriptor.
2363 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2364 */
2365int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2366 struct flock64 __user *l)
2367{
2368 struct file_lock *file_lock = locks_alloc_lock();
2369 struct flock64 flock;
2370 struct inode *inode;
2371 struct file *f;
2372 int error;
2373
2374 if (file_lock == NULL)
2375 return -ENOLCK;
2376
2377 /*
2378 * This might block, so we do it before checking the inode.
2379 */
2380 error = -EFAULT;
2381 if (copy_from_user(&flock, l, sizeof(flock)))
2382 goto out;
2383
2384 inode = locks_inode(filp);
2385
2386 /* Don't allow mandatory locks on files that may be memory mapped
2387 * and shared.
2388 */
2389 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2390 error = -EAGAIN;
2391 goto out;
2392 }
2393
2394 error = flock64_to_posix_lock(filp, file_lock, &flock);
2395 if (error)
2396 goto out;
2397
2398 error = check_fmode_for_setlk(file_lock);
2399 if (error)
2400 goto out;
2401
2402 /*
2403 * If the cmd is requesting file-private locks, then set the
2404 * FL_OFDLCK flag and override the owner.
2405 */
2406 switch (cmd) {
2407 case F_OFD_SETLK:
2408 error = -EINVAL;
2409 if (flock.l_pid != 0)
2410 goto out;
2411
2412 cmd = F_SETLK64;
2413 file_lock->fl_flags |= FL_OFDLCK;
2414 file_lock->fl_owner = filp;
2415 break;
2416 case F_OFD_SETLKW:
2417 error = -EINVAL;
2418 if (flock.l_pid != 0)
2419 goto out;
2420
2421 cmd = F_SETLKW64;
2422 file_lock->fl_flags |= FL_OFDLCK;
2423 file_lock->fl_owner = filp;
2424 /* Fallthrough */
2425 case F_SETLKW64:
2426 file_lock->fl_flags |= FL_SLEEP;
2427 }
2428
2429 error = do_lock_file_wait(filp, cmd, file_lock);
2430
2431 /*
2432 * Attempt to detect a close/fcntl race and recover by releasing the
2433 * lock that was just acquired. There is no need to do that when we're
2434 * unlocking though, or for OFD locks.
2435 */
2436 if (!error && file_lock->fl_type != F_UNLCK &&
2437 !(file_lock->fl_flags & FL_OFDLCK)) {
2438 /*
2439 * We need that spin_lock here - it prevents reordering between
2440 * update of i_flctx->flc_posix and check for it done in
2441 * close(). rcu_read_lock() wouldn't do.
2442 */
2443 spin_lock(¤t->files->file_lock);
2444 f = fcheck(fd);
2445 spin_unlock(¤t->files->file_lock);
2446 if (f != filp) {
2447 file_lock->fl_type = F_UNLCK;
2448 error = do_lock_file_wait(filp, cmd, file_lock);
2449 WARN_ON_ONCE(error);
2450 error = -EBADF;
2451 }
2452 }
2453out:
2454 locks_free_lock(file_lock);
2455 return error;
2456}
2457#endif /* BITS_PER_LONG == 32 */
2458
2459/*
2460 * This function is called when the file is being removed
2461 * from the task's fd array. POSIX locks belonging to this task
2462 * are deleted at this time.
2463 */
2464void locks_remove_posix(struct file *filp, fl_owner_t owner)
2465{
2466 int error;
2467 struct inode *inode = locks_inode(filp);
2468 struct file_lock lock;
2469 struct file_lock_context *ctx;
2470
2471 /*
2472 * If there are no locks held on this file, we don't need to call
2473 * posix_lock_file(). Another process could be setting a lock on this
2474 * file at the same time, but we wouldn't remove that lock anyway.
2475 */
2476 ctx = smp_load_acquire(&inode->i_flctx);
2477 if (!ctx || list_empty(&ctx->flc_posix))
2478 return;
2479
2480 lock.fl_type = F_UNLCK;
2481 lock.fl_flags = FL_POSIX | FL_CLOSE;
2482 lock.fl_start = 0;
2483 lock.fl_end = OFFSET_MAX;
2484 lock.fl_owner = owner;
2485 lock.fl_pid = current->tgid;
2486 lock.fl_file = filp;
2487 lock.fl_ops = NULL;
2488 lock.fl_lmops = NULL;
2489
2490 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2491
2492 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2493 lock.fl_ops->fl_release_private(&lock);
2494 trace_locks_remove_posix(inode, &lock, error);
2495}
2496
2497EXPORT_SYMBOL(locks_remove_posix);
2498
2499/* The i_flctx must be valid when calling into here */
2500static void
2501locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2502{
2503 struct file_lock fl = {
2504 .fl_owner = filp,
2505 .fl_pid = current->tgid,
2506 .fl_file = filp,
2507 .fl_flags = FL_FLOCK,
2508 .fl_type = F_UNLCK,
2509 .fl_end = OFFSET_MAX,
2510 };
2511 struct inode *inode = locks_inode(filp);
2512
2513 if (list_empty(&flctx->flc_flock))
2514 return;
2515
2516 if (filp->f_op->flock && is_remote_lock(filp))
2517 filp->f_op->flock(filp, F_SETLKW, &fl);
2518 else
2519 flock_lock_inode(inode, &fl);
2520
2521 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2522 fl.fl_ops->fl_release_private(&fl);
2523}
2524
2525/* The i_flctx must be valid when calling into here */
2526static void
2527locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2528{
2529 struct file_lock *fl, *tmp;
2530 LIST_HEAD(dispose);
2531
2532 if (list_empty(&ctx->flc_lease))
2533 return;
2534
2535 percpu_down_read_preempt_disable(&file_rwsem);
2536 spin_lock(&ctx->flc_lock);
2537 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2538 if (filp == fl->fl_file)
2539 lease_modify(fl, F_UNLCK, &dispose);
2540 spin_unlock(&ctx->flc_lock);
2541 percpu_up_read_preempt_enable(&file_rwsem);
2542
2543 locks_dispose_list(&dispose);
2544}
2545
2546/*
2547 * This function is called on the last close of an open file.
2548 */
2549void locks_remove_file(struct file *filp)
2550{
2551 struct file_lock_context *ctx;
2552
2553 ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2554 if (!ctx)
2555 return;
2556
2557 /* remove any OFD locks */
2558 locks_remove_posix(filp, filp);
2559
2560 /* remove flock locks */
2561 locks_remove_flock(filp, ctx);
2562
2563 /* remove any leases */
2564 locks_remove_lease(filp, ctx);
2565}
2566
2567/**
2568 * posix_unblock_lock - stop waiting for a file lock
2569 * @waiter: the lock which was waiting
2570 *
2571 * lockd needs to block waiting for locks.
2572 */
2573int
2574posix_unblock_lock(struct file_lock *waiter)
2575{
2576 int status = 0;
2577
2578 spin_lock(&blocked_lock_lock);
2579 if (waiter->fl_next)
2580 __locks_delete_block(waiter);
2581 else
2582 status = -ENOENT;
2583 spin_unlock(&blocked_lock_lock);
2584 return status;
2585}
2586EXPORT_SYMBOL(posix_unblock_lock);
2587
2588/**
2589 * vfs_cancel_lock - file byte range unblock lock
2590 * @filp: The file to apply the unblock to
2591 * @fl: The lock to be unblocked
2592 *
2593 * Used by lock managers to cancel blocked requests
2594 */
2595int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2596{
2597 if (filp->f_op->lock && is_remote_lock(filp))
2598 return filp->f_op->lock(filp, F_CANCELLK, fl);
2599 return 0;
2600}
2601
2602EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2603
2604#ifdef CONFIG_PROC_FS
2605#include <linux/proc_fs.h>
2606#include <linux/seq_file.h>
2607
2608struct locks_iterator {
2609 int li_cpu;
2610 loff_t li_pos;
2611};
2612
2613static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2614 loff_t id, char *pfx)
2615{
2616 struct inode *inode = NULL;
2617 unsigned int fl_pid;
2618
2619 if (fl->fl_nspid) {
2620 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2621
2622 /* Don't let fl_pid change based on who is reading the file */
2623 fl_pid = pid_nr_ns(fl->fl_nspid, proc_pidns);
2624
2625 /*
2626 * If there isn't a fl_pid don't display who is waiting on
2627 * the lock if we are called from locks_show, or if we are
2628 * called from __show_fd_info - skip lock entirely
2629 */
2630 if (fl_pid == 0)
2631 return;
2632 } else
2633 fl_pid = fl->fl_pid;
2634
2635 if (fl->fl_file != NULL)
2636 inode = locks_inode(fl->fl_file);
2637
2638 seq_printf(f, "%lld:%s ", id, pfx);
2639 if (IS_POSIX(fl)) {
2640 if (fl->fl_flags & FL_ACCESS)
2641 seq_puts(f, "ACCESS");
2642 else if (IS_OFDLCK(fl))
2643 seq_puts(f, "OFDLCK");
2644 else
2645 seq_puts(f, "POSIX ");
2646
2647 seq_printf(f, " %s ",
2648 (inode == NULL) ? "*NOINODE*" :
2649 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2650 } else if (IS_FLOCK(fl)) {
2651 if (fl->fl_type & LOCK_MAND) {
2652 seq_puts(f, "FLOCK MSNFS ");
2653 } else {
2654 seq_puts(f, "FLOCK ADVISORY ");
2655 }
2656 } else if (IS_LEASE(fl)) {
2657 if (fl->fl_flags & FL_DELEG)
2658 seq_puts(f, "DELEG ");
2659 else
2660 seq_puts(f, "LEASE ");
2661
2662 if (lease_breaking(fl))
2663 seq_puts(f, "BREAKING ");
2664 else if (fl->fl_file)
2665 seq_puts(f, "ACTIVE ");
2666 else
2667 seq_puts(f, "BREAKER ");
2668 } else {
2669 seq_puts(f, "UNKNOWN UNKNOWN ");
2670 }
2671 if (fl->fl_type & LOCK_MAND) {
2672 seq_printf(f, "%s ",
2673 (fl->fl_type & LOCK_READ)
2674 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2675 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2676 } else {
2677 seq_printf(f, "%s ",
2678 (lease_breaking(fl))
2679 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2680 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2681 }
2682 if (inode) {
2683 /* userspace relies on this representation of dev_t */
2684 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2685 MAJOR(inode->i_sb->s_dev),
2686 MINOR(inode->i_sb->s_dev), inode->i_ino);
2687 } else {
2688 seq_printf(f, "%d <none>:0 ", fl_pid);
2689 }
2690 if (IS_POSIX(fl)) {
2691 if (fl->fl_end == OFFSET_MAX)
2692 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2693 else
2694 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2695 } else {
2696 seq_puts(f, "0 EOF\n");
2697 }
2698}
2699
2700static int locks_show(struct seq_file *f, void *v)
2701{
2702 struct locks_iterator *iter = f->private;
2703 struct file_lock *fl, *bfl;
2704 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2705
2706 fl = hlist_entry(v, struct file_lock, fl_link);
2707
2708 if (fl->fl_nspid && !pid_nr_ns(fl->fl_nspid, proc_pidns))
2709 return 0;
2710
2711 lock_get_status(f, fl, iter->li_pos, "");
2712
2713 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2714 lock_get_status(f, bfl, iter->li_pos, " ->");
2715
2716 return 0;
2717}
2718
2719static void __show_fd_locks(struct seq_file *f,
2720 struct list_head *head, int *id,
2721 struct file *filp, struct files_struct *files)
2722{
2723 struct file_lock *fl;
2724
2725 list_for_each_entry(fl, head, fl_list) {
2726
2727 if (filp != fl->fl_file)
2728 continue;
2729 if (fl->fl_owner != files &&
2730 fl->fl_owner != filp)
2731 continue;
2732
2733 (*id)++;
2734 seq_puts(f, "lock:\t");
2735 lock_get_status(f, fl, *id, "");
2736 }
2737}
2738
2739void show_fd_locks(struct seq_file *f,
2740 struct file *filp, struct files_struct *files)
2741{
2742 struct inode *inode = locks_inode(filp);
2743 struct file_lock_context *ctx;
2744 int id = 0;
2745
2746 ctx = smp_load_acquire(&inode->i_flctx);
2747 if (!ctx)
2748 return;
2749
2750 spin_lock(&ctx->flc_lock);
2751 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2752 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2753 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2754 spin_unlock(&ctx->flc_lock);
2755}
2756
2757static void *locks_start(struct seq_file *f, loff_t *pos)
2758 __acquires(&blocked_lock_lock)
2759{
2760 struct locks_iterator *iter = f->private;
2761
2762 iter->li_pos = *pos + 1;
2763 percpu_down_write(&file_rwsem);
2764 spin_lock(&blocked_lock_lock);
2765 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2766}
2767
2768static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2769{
2770 struct locks_iterator *iter = f->private;
2771
2772 ++iter->li_pos;
2773 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2774}
2775
2776static void locks_stop(struct seq_file *f, void *v)
2777 __releases(&blocked_lock_lock)
2778{
2779 spin_unlock(&blocked_lock_lock);
2780 percpu_up_write(&file_rwsem);
2781}
2782
2783static const struct seq_operations locks_seq_operations = {
2784 .start = locks_start,
2785 .next = locks_next,
2786 .stop = locks_stop,
2787 .show = locks_show,
2788};
2789
2790static int locks_open(struct inode *inode, struct file *filp)
2791{
2792 return seq_open_private(filp, &locks_seq_operations,
2793 sizeof(struct locks_iterator));
2794}
2795
2796static const struct file_operations proc_locks_operations = {
2797 .open = locks_open,
2798 .read = seq_read,
2799 .llseek = seq_lseek,
2800 .release = seq_release_private,
2801};
2802
2803static int __init proc_locks_init(void)
2804{
2805 proc_create("locks", 0, NULL, &proc_locks_operations);
2806 return 0;
2807}
2808fs_initcall(proc_locks_init);
2809#endif
2810
2811static int __init filelock_init(void)
2812{
2813 int i;
2814
2815 flctx_cache = kmem_cache_create("file_lock_ctx",
2816 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2817
2818 filelock_cache = kmem_cache_create("file_lock_cache",
2819 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2820
2821
2822 for_each_possible_cpu(i) {
2823 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2824
2825 spin_lock_init(&fll->lock);
2826 INIT_HLIST_HEAD(&fll->hlist);
2827 }
2828
2829 return 0;
2830}
2831
2832core_initcall(filelock_init);
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
131#define CREATE_TRACE_POINTS
132#include <trace/events/filelock.h>
133
134#include <linux/uaccess.h>
135
136#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
137#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
138#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
139#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
140#define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
141
142static inline bool is_remote_lock(struct file *filp)
143{
144 return likely(!(filp->f_path.dentry->d_sb->s_flags & SB_NOREMOTELOCK));
145}
146
147static bool lease_breaking(struct file_lock *fl)
148{
149 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
150}
151
152static int target_leasetype(struct file_lock *fl)
153{
154 if (fl->fl_flags & FL_UNLOCK_PENDING)
155 return F_UNLCK;
156 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
157 return F_RDLCK;
158 return fl->fl_type;
159}
160
161int leases_enable = 1;
162int lease_break_time = 45;
163
164/*
165 * The global file_lock_list is only used for displaying /proc/locks, so we
166 * keep a list on each CPU, with each list protected by its own spinlock.
167 * Global serialization is done using file_rwsem.
168 *
169 * Note that alterations to the list also require that the relevant flc_lock is
170 * held.
171 */
172struct file_lock_list_struct {
173 spinlock_t lock;
174 struct hlist_head hlist;
175};
176static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
177DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
178
179/*
180 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
181 * It is protected by blocked_lock_lock.
182 *
183 * We hash locks by lockowner in order to optimize searching for the lock a
184 * particular lockowner is waiting on.
185 *
186 * FIXME: make this value scale via some heuristic? We generally will want more
187 * buckets when we have more lockowners holding locks, but that's a little
188 * difficult to determine without knowing what the workload will look like.
189 */
190#define BLOCKED_HASH_BITS 7
191static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
192
193/*
194 * This lock protects the blocked_hash. Generally, if you're accessing it, you
195 * want to be holding this lock.
196 *
197 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
198 * pointer for file_lock structures that are acting as lock requests (in
199 * contrast to those that are acting as records of acquired locks).
200 *
201 * Note that when we acquire this lock in order to change the above fields,
202 * we often hold the flc_lock as well. In certain cases, when reading the fields
203 * protected by this lock, we can skip acquiring it iff we already hold the
204 * flc_lock.
205 *
206 * In particular, adding an entry to the fl_block list requires that you hold
207 * both the flc_lock and the blocked_lock_lock (acquired in that order).
208 * Deleting an entry from the list however only requires the file_lock_lock.
209 */
210static DEFINE_SPINLOCK(blocked_lock_lock);
211
212static struct kmem_cache *flctx_cache __read_mostly;
213static struct kmem_cache *filelock_cache __read_mostly;
214
215static struct file_lock_context *
216locks_get_lock_context(struct inode *inode, int type)
217{
218 struct file_lock_context *ctx;
219
220 /* paired with cmpxchg() below */
221 ctx = smp_load_acquire(&inode->i_flctx);
222 if (likely(ctx) || type == F_UNLCK)
223 goto out;
224
225 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
226 if (!ctx)
227 goto out;
228
229 spin_lock_init(&ctx->flc_lock);
230 INIT_LIST_HEAD(&ctx->flc_flock);
231 INIT_LIST_HEAD(&ctx->flc_posix);
232 INIT_LIST_HEAD(&ctx->flc_lease);
233
234 /*
235 * Assign the pointer if it's not already assigned. If it is, then
236 * free the context we just allocated.
237 */
238 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
239 kmem_cache_free(flctx_cache, ctx);
240 ctx = smp_load_acquire(&inode->i_flctx);
241 }
242out:
243 trace_locks_get_lock_context(inode, type, ctx);
244 return ctx;
245}
246
247static void
248locks_dump_ctx_list(struct list_head *list, char *list_type)
249{
250 struct file_lock *fl;
251
252 list_for_each_entry(fl, list, fl_list) {
253 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);
254 }
255}
256
257static void
258locks_check_ctx_lists(struct inode *inode)
259{
260 struct file_lock_context *ctx = inode->i_flctx;
261
262 if (unlikely(!list_empty(&ctx->flc_flock) ||
263 !list_empty(&ctx->flc_posix) ||
264 !list_empty(&ctx->flc_lease))) {
265 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
266 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
267 inode->i_ino);
268 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
269 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
270 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
271 }
272}
273
274static void
275locks_check_ctx_file_list(struct file *filp, struct list_head *list,
276 char *list_type)
277{
278 struct file_lock *fl;
279 struct inode *inode = locks_inode(filp);
280
281 list_for_each_entry(fl, list, fl_list)
282 if (fl->fl_file == filp)
283 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
284 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
285 list_type, MAJOR(inode->i_sb->s_dev),
286 MINOR(inode->i_sb->s_dev), inode->i_ino,
287 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
288}
289
290void
291locks_free_lock_context(struct inode *inode)
292{
293 struct file_lock_context *ctx = inode->i_flctx;
294
295 if (unlikely(ctx)) {
296 locks_check_ctx_lists(inode);
297 kmem_cache_free(flctx_cache, ctx);
298 }
299}
300
301static void locks_init_lock_heads(struct file_lock *fl)
302{
303 INIT_HLIST_NODE(&fl->fl_link);
304 INIT_LIST_HEAD(&fl->fl_list);
305 INIT_LIST_HEAD(&fl->fl_block);
306 init_waitqueue_head(&fl->fl_wait);
307}
308
309/* Allocate an empty lock structure. */
310struct file_lock *locks_alloc_lock(void)
311{
312 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
313
314 if (fl)
315 locks_init_lock_heads(fl);
316
317 return fl;
318}
319EXPORT_SYMBOL_GPL(locks_alloc_lock);
320
321void locks_release_private(struct file_lock *fl)
322{
323 if (fl->fl_ops) {
324 if (fl->fl_ops->fl_release_private)
325 fl->fl_ops->fl_release_private(fl);
326 fl->fl_ops = NULL;
327 }
328
329 if (fl->fl_lmops) {
330 if (fl->fl_lmops->lm_put_owner) {
331 fl->fl_lmops->lm_put_owner(fl->fl_owner);
332 fl->fl_owner = NULL;
333 }
334 fl->fl_lmops = NULL;
335 }
336}
337EXPORT_SYMBOL_GPL(locks_release_private);
338
339/* Free a lock which is not in use. */
340void locks_free_lock(struct file_lock *fl)
341{
342 BUG_ON(waitqueue_active(&fl->fl_wait));
343 BUG_ON(!list_empty(&fl->fl_list));
344 BUG_ON(!list_empty(&fl->fl_block));
345 BUG_ON(!hlist_unhashed(&fl->fl_link));
346
347 locks_release_private(fl);
348 kmem_cache_free(filelock_cache, fl);
349}
350EXPORT_SYMBOL(locks_free_lock);
351
352static void
353locks_dispose_list(struct list_head *dispose)
354{
355 struct file_lock *fl;
356
357 while (!list_empty(dispose)) {
358 fl = list_first_entry(dispose, struct file_lock, fl_list);
359 list_del_init(&fl->fl_list);
360 locks_free_lock(fl);
361 }
362}
363
364void locks_init_lock(struct file_lock *fl)
365{
366 memset(fl, 0, sizeof(struct file_lock));
367 locks_init_lock_heads(fl);
368}
369
370EXPORT_SYMBOL(locks_init_lock);
371
372/*
373 * Initialize a new lock from an existing file_lock structure.
374 */
375void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
376{
377 new->fl_owner = fl->fl_owner;
378 new->fl_pid = fl->fl_pid;
379 new->fl_file = NULL;
380 new->fl_flags = fl->fl_flags;
381 new->fl_type = fl->fl_type;
382 new->fl_start = fl->fl_start;
383 new->fl_end = fl->fl_end;
384 new->fl_lmops = fl->fl_lmops;
385 new->fl_ops = NULL;
386
387 if (fl->fl_lmops) {
388 if (fl->fl_lmops->lm_get_owner)
389 fl->fl_lmops->lm_get_owner(fl->fl_owner);
390 }
391}
392EXPORT_SYMBOL(locks_copy_conflock);
393
394void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
395{
396 /* "new" must be a freshly-initialized lock */
397 WARN_ON_ONCE(new->fl_ops);
398
399 locks_copy_conflock(new, fl);
400
401 new->fl_file = fl->fl_file;
402 new->fl_ops = fl->fl_ops;
403
404 if (fl->fl_ops) {
405 if (fl->fl_ops->fl_copy_lock)
406 fl->fl_ops->fl_copy_lock(new, fl);
407 }
408}
409
410EXPORT_SYMBOL(locks_copy_lock);
411
412static inline int flock_translate_cmd(int cmd) {
413 if (cmd & LOCK_MAND)
414 return cmd & (LOCK_MAND | LOCK_RW);
415 switch (cmd) {
416 case LOCK_SH:
417 return F_RDLCK;
418 case LOCK_EX:
419 return F_WRLCK;
420 case LOCK_UN:
421 return F_UNLCK;
422 }
423 return -EINVAL;
424}
425
426/* Fill in a file_lock structure with an appropriate FLOCK lock. */
427static struct file_lock *
428flock_make_lock(struct file *filp, unsigned int cmd)
429{
430 struct file_lock *fl;
431 int type = flock_translate_cmd(cmd);
432
433 if (type < 0)
434 return ERR_PTR(type);
435
436 fl = locks_alloc_lock();
437 if (fl == NULL)
438 return ERR_PTR(-ENOMEM);
439
440 fl->fl_file = filp;
441 fl->fl_owner = filp;
442 fl->fl_pid = current->tgid;
443 fl->fl_flags = FL_FLOCK;
444 fl->fl_type = type;
445 fl->fl_end = OFFSET_MAX;
446
447 return fl;
448}
449
450static int assign_type(struct file_lock *fl, long type)
451{
452 switch (type) {
453 case F_RDLCK:
454 case F_WRLCK:
455 case F_UNLCK:
456 fl->fl_type = type;
457 break;
458 default:
459 return -EINVAL;
460 }
461 return 0;
462}
463
464static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
465 struct flock64 *l)
466{
467 switch (l->l_whence) {
468 case SEEK_SET:
469 fl->fl_start = 0;
470 break;
471 case SEEK_CUR:
472 fl->fl_start = filp->f_pos;
473 break;
474 case SEEK_END:
475 fl->fl_start = i_size_read(file_inode(filp));
476 break;
477 default:
478 return -EINVAL;
479 }
480 if (l->l_start > OFFSET_MAX - fl->fl_start)
481 return -EOVERFLOW;
482 fl->fl_start += l->l_start;
483 if (fl->fl_start < 0)
484 return -EINVAL;
485
486 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
487 POSIX-2001 defines it. */
488 if (l->l_len > 0) {
489 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
490 return -EOVERFLOW;
491 fl->fl_end = fl->fl_start + l->l_len - 1;
492
493 } else if (l->l_len < 0) {
494 if (fl->fl_start + l->l_len < 0)
495 return -EINVAL;
496 fl->fl_end = fl->fl_start - 1;
497 fl->fl_start += l->l_len;
498 } else
499 fl->fl_end = OFFSET_MAX;
500
501 fl->fl_owner = current->files;
502 fl->fl_pid = current->tgid;
503 fl->fl_file = filp;
504 fl->fl_flags = FL_POSIX;
505 fl->fl_ops = NULL;
506 fl->fl_lmops = NULL;
507
508 return assign_type(fl, l->l_type);
509}
510
511/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
512 * style lock.
513 */
514static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
515 struct flock *l)
516{
517 struct flock64 ll = {
518 .l_type = l->l_type,
519 .l_whence = l->l_whence,
520 .l_start = l->l_start,
521 .l_len = l->l_len,
522 };
523
524 return flock64_to_posix_lock(filp, fl, &ll);
525}
526
527/* default lease lock manager operations */
528static bool
529lease_break_callback(struct file_lock *fl)
530{
531 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
532 return false;
533}
534
535static void
536lease_setup(struct file_lock *fl, void **priv)
537{
538 struct file *filp = fl->fl_file;
539 struct fasync_struct *fa = *priv;
540
541 /*
542 * fasync_insert_entry() returns the old entry if any. If there was no
543 * old entry, then it used "priv" and inserted it into the fasync list.
544 * Clear the pointer to indicate that it shouldn't be freed.
545 */
546 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
547 *priv = NULL;
548
549 __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
550}
551
552static const struct lock_manager_operations lease_manager_ops = {
553 .lm_break = lease_break_callback,
554 .lm_change = lease_modify,
555 .lm_setup = lease_setup,
556};
557
558/*
559 * Initialize a lease, use the default lock manager operations
560 */
561static int lease_init(struct file *filp, long type, struct file_lock *fl)
562{
563 if (assign_type(fl, type) != 0)
564 return -EINVAL;
565
566 fl->fl_owner = filp;
567 fl->fl_pid = current->tgid;
568
569 fl->fl_file = filp;
570 fl->fl_flags = FL_LEASE;
571 fl->fl_start = 0;
572 fl->fl_end = OFFSET_MAX;
573 fl->fl_ops = NULL;
574 fl->fl_lmops = &lease_manager_ops;
575 return 0;
576}
577
578/* Allocate a file_lock initialised to this type of lease */
579static struct file_lock *lease_alloc(struct file *filp, long type)
580{
581 struct file_lock *fl = locks_alloc_lock();
582 int error = -ENOMEM;
583
584 if (fl == NULL)
585 return ERR_PTR(error);
586
587 error = lease_init(filp, type, fl);
588 if (error) {
589 locks_free_lock(fl);
590 return ERR_PTR(error);
591 }
592 return fl;
593}
594
595/* Check if two locks overlap each other.
596 */
597static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
598{
599 return ((fl1->fl_end >= fl2->fl_start) &&
600 (fl2->fl_end >= fl1->fl_start));
601}
602
603/*
604 * Check whether two locks have the same owner.
605 */
606static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
607{
608 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
609 return fl2->fl_lmops == fl1->fl_lmops &&
610 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
611 return fl1->fl_owner == fl2->fl_owner;
612}
613
614/* Must be called with the flc_lock held! */
615static void locks_insert_global_locks(struct file_lock *fl)
616{
617 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
618
619 percpu_rwsem_assert_held(&file_rwsem);
620
621 spin_lock(&fll->lock);
622 fl->fl_link_cpu = smp_processor_id();
623 hlist_add_head(&fl->fl_link, &fll->hlist);
624 spin_unlock(&fll->lock);
625}
626
627/* Must be called with the flc_lock held! */
628static void locks_delete_global_locks(struct file_lock *fl)
629{
630 struct file_lock_list_struct *fll;
631
632 percpu_rwsem_assert_held(&file_rwsem);
633
634 /*
635 * Avoid taking lock if already unhashed. This is safe since this check
636 * is done while holding the flc_lock, and new insertions into the list
637 * also require that it be held.
638 */
639 if (hlist_unhashed(&fl->fl_link))
640 return;
641
642 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
643 spin_lock(&fll->lock);
644 hlist_del_init(&fl->fl_link);
645 spin_unlock(&fll->lock);
646}
647
648static unsigned long
649posix_owner_key(struct file_lock *fl)
650{
651 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
652 return fl->fl_lmops->lm_owner_key(fl);
653 return (unsigned long)fl->fl_owner;
654}
655
656static void locks_insert_global_blocked(struct file_lock *waiter)
657{
658 lockdep_assert_held(&blocked_lock_lock);
659
660 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
661}
662
663static void locks_delete_global_blocked(struct file_lock *waiter)
664{
665 lockdep_assert_held(&blocked_lock_lock);
666
667 hash_del(&waiter->fl_link);
668}
669
670/* Remove waiter from blocker's block list.
671 * When blocker ends up pointing to itself then the list is empty.
672 *
673 * Must be called with blocked_lock_lock held.
674 */
675static void __locks_delete_block(struct file_lock *waiter)
676{
677 locks_delete_global_blocked(waiter);
678 list_del_init(&waiter->fl_block);
679 waiter->fl_next = NULL;
680}
681
682static void locks_delete_block(struct file_lock *waiter)
683{
684 spin_lock(&blocked_lock_lock);
685 __locks_delete_block(waiter);
686 spin_unlock(&blocked_lock_lock);
687}
688
689/* Insert waiter into blocker's block list.
690 * We use a circular list so that processes can be easily woken up in
691 * the order they blocked. The documentation doesn't require this but
692 * it seems like the reasonable thing to do.
693 *
694 * Must be called with both the flc_lock and blocked_lock_lock held. The
695 * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
696 * that the flc_lock is also held on insertions we can avoid taking the
697 * blocked_lock_lock in some cases when we see that the fl_block list is empty.
698 */
699static void __locks_insert_block(struct file_lock *blocker,
700 struct file_lock *waiter)
701{
702 BUG_ON(!list_empty(&waiter->fl_block));
703 waiter->fl_next = blocker;
704 list_add_tail(&waiter->fl_block, &blocker->fl_block);
705 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
706 locks_insert_global_blocked(waiter);
707}
708
709/* Must be called with flc_lock held. */
710static void locks_insert_block(struct file_lock *blocker,
711 struct file_lock *waiter)
712{
713 spin_lock(&blocked_lock_lock);
714 __locks_insert_block(blocker, waiter);
715 spin_unlock(&blocked_lock_lock);
716}
717
718/*
719 * Wake up processes blocked waiting for blocker.
720 *
721 * Must be called with the inode->flc_lock held!
722 */
723static void locks_wake_up_blocks(struct file_lock *blocker)
724{
725 /*
726 * Avoid taking global lock if list is empty. This is safe since new
727 * blocked requests are only added to the list under the flc_lock, and
728 * the flc_lock is always held here. Note that removal from the fl_block
729 * list does not require the flc_lock, so we must recheck list_empty()
730 * after acquiring the blocked_lock_lock.
731 */
732 if (list_empty(&blocker->fl_block))
733 return;
734
735 spin_lock(&blocked_lock_lock);
736 while (!list_empty(&blocker->fl_block)) {
737 struct file_lock *waiter;
738
739 waiter = list_first_entry(&blocker->fl_block,
740 struct file_lock, fl_block);
741 __locks_delete_block(waiter);
742 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
743 waiter->fl_lmops->lm_notify(waiter);
744 else
745 wake_up(&waiter->fl_wait);
746 }
747 spin_unlock(&blocked_lock_lock);
748}
749
750static void
751locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
752{
753 list_add_tail(&fl->fl_list, before);
754 locks_insert_global_locks(fl);
755}
756
757static void
758locks_unlink_lock_ctx(struct file_lock *fl)
759{
760 locks_delete_global_locks(fl);
761 list_del_init(&fl->fl_list);
762 locks_wake_up_blocks(fl);
763}
764
765static void
766locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
767{
768 locks_unlink_lock_ctx(fl);
769 if (dispose)
770 list_add(&fl->fl_list, dispose);
771 else
772 locks_free_lock(fl);
773}
774
775/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
776 * checks for shared/exclusive status of overlapping locks.
777 */
778static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
779{
780 if (sys_fl->fl_type == F_WRLCK)
781 return 1;
782 if (caller_fl->fl_type == F_WRLCK)
783 return 1;
784 return 0;
785}
786
787/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
788 * checking before calling the locks_conflict().
789 */
790static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
791{
792 /* POSIX locks owned by the same process do not conflict with
793 * each other.
794 */
795 if (posix_same_owner(caller_fl, sys_fl))
796 return (0);
797
798 /* Check whether they overlap */
799 if (!locks_overlap(caller_fl, sys_fl))
800 return 0;
801
802 return (locks_conflict(caller_fl, sys_fl));
803}
804
805/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
806 * checking before calling the locks_conflict().
807 */
808static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
809{
810 /* FLOCK locks referring to the same filp do not conflict with
811 * each other.
812 */
813 if (caller_fl->fl_file == sys_fl->fl_file)
814 return (0);
815 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
816 return 0;
817
818 return (locks_conflict(caller_fl, sys_fl));
819}
820
821void
822posix_test_lock(struct file *filp, struct file_lock *fl)
823{
824 struct file_lock *cfl;
825 struct file_lock_context *ctx;
826 struct inode *inode = locks_inode(filp);
827
828 ctx = smp_load_acquire(&inode->i_flctx);
829 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
830 fl->fl_type = F_UNLCK;
831 return;
832 }
833
834 spin_lock(&ctx->flc_lock);
835 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
836 if (posix_locks_conflict(fl, cfl)) {
837 locks_copy_conflock(fl, cfl);
838 goto out;
839 }
840 }
841 fl->fl_type = F_UNLCK;
842out:
843 spin_unlock(&ctx->flc_lock);
844 return;
845}
846EXPORT_SYMBOL(posix_test_lock);
847
848/*
849 * Deadlock detection:
850 *
851 * We attempt to detect deadlocks that are due purely to posix file
852 * locks.
853 *
854 * We assume that a task can be waiting for at most one lock at a time.
855 * So for any acquired lock, the process holding that lock may be
856 * waiting on at most one other lock. That lock in turns may be held by
857 * someone waiting for at most one other lock. Given a requested lock
858 * caller_fl which is about to wait for a conflicting lock block_fl, we
859 * follow this chain of waiters to ensure we are not about to create a
860 * cycle.
861 *
862 * Since we do this before we ever put a process to sleep on a lock, we
863 * are ensured that there is never a cycle; that is what guarantees that
864 * the while() loop in posix_locks_deadlock() eventually completes.
865 *
866 * Note: the above assumption may not be true when handling lock
867 * requests from a broken NFS client. It may also fail in the presence
868 * of tasks (such as posix threads) sharing the same open file table.
869 * To handle those cases, we just bail out after a few iterations.
870 *
871 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
872 * Because the owner is not even nominally tied to a thread of
873 * execution, the deadlock detection below can't reasonably work well. Just
874 * skip it for those.
875 *
876 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
877 * locks that just checks for the case where two tasks are attempting to
878 * upgrade from read to write locks on the same inode.
879 */
880
881#define MAX_DEADLK_ITERATIONS 10
882
883/* Find a lock that the owner of the given block_fl is blocking on. */
884static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
885{
886 struct file_lock *fl;
887
888 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
889 if (posix_same_owner(fl, block_fl))
890 return fl->fl_next;
891 }
892 return NULL;
893}
894
895/* Must be called with the blocked_lock_lock held! */
896static int posix_locks_deadlock(struct file_lock *caller_fl,
897 struct file_lock *block_fl)
898{
899 int i = 0;
900
901 lockdep_assert_held(&blocked_lock_lock);
902
903 /*
904 * This deadlock detector can't reasonably detect deadlocks with
905 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
906 */
907 if (IS_OFDLCK(caller_fl))
908 return 0;
909
910 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
911 if (i++ > MAX_DEADLK_ITERATIONS)
912 return 0;
913 if (posix_same_owner(caller_fl, block_fl))
914 return 1;
915 }
916 return 0;
917}
918
919/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
920 * after any leases, but before any posix locks.
921 *
922 * Note that if called with an FL_EXISTS argument, the caller may determine
923 * whether or not a lock was successfully freed by testing the return
924 * value for -ENOENT.
925 */
926static int flock_lock_inode(struct inode *inode, struct file_lock *request)
927{
928 struct file_lock *new_fl = NULL;
929 struct file_lock *fl;
930 struct file_lock_context *ctx;
931 int error = 0;
932 bool found = false;
933 LIST_HEAD(dispose);
934
935 ctx = locks_get_lock_context(inode, request->fl_type);
936 if (!ctx) {
937 if (request->fl_type != F_UNLCK)
938 return -ENOMEM;
939 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
940 }
941
942 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
943 new_fl = locks_alloc_lock();
944 if (!new_fl)
945 return -ENOMEM;
946 }
947
948 percpu_down_read_preempt_disable(&file_rwsem);
949 spin_lock(&ctx->flc_lock);
950 if (request->fl_flags & FL_ACCESS)
951 goto find_conflict;
952
953 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
954 if (request->fl_file != fl->fl_file)
955 continue;
956 if (request->fl_type == fl->fl_type)
957 goto out;
958 found = true;
959 locks_delete_lock_ctx(fl, &dispose);
960 break;
961 }
962
963 if (request->fl_type == F_UNLCK) {
964 if ((request->fl_flags & FL_EXISTS) && !found)
965 error = -ENOENT;
966 goto out;
967 }
968
969find_conflict:
970 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
971 if (!flock_locks_conflict(request, fl))
972 continue;
973 error = -EAGAIN;
974 if (!(request->fl_flags & FL_SLEEP))
975 goto out;
976 error = FILE_LOCK_DEFERRED;
977 locks_insert_block(fl, request);
978 goto out;
979 }
980 if (request->fl_flags & FL_ACCESS)
981 goto out;
982 locks_copy_lock(new_fl, request);
983 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
984 new_fl = NULL;
985 error = 0;
986
987out:
988 spin_unlock(&ctx->flc_lock);
989 percpu_up_read_preempt_enable(&file_rwsem);
990 if (new_fl)
991 locks_free_lock(new_fl);
992 locks_dispose_list(&dispose);
993 return error;
994}
995
996static int posix_lock_inode(struct inode *inode, struct file_lock *request,
997 struct file_lock *conflock)
998{
999 struct file_lock *fl, *tmp;
1000 struct file_lock *new_fl = NULL;
1001 struct file_lock *new_fl2 = NULL;
1002 struct file_lock *left = NULL;
1003 struct file_lock *right = NULL;
1004 struct file_lock_context *ctx;
1005 int error;
1006 bool added = false;
1007 LIST_HEAD(dispose);
1008
1009 ctx = locks_get_lock_context(inode, request->fl_type);
1010 if (!ctx)
1011 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1012
1013 /*
1014 * We may need two file_lock structures for this operation,
1015 * so we get them in advance to avoid races.
1016 *
1017 * In some cases we can be sure, that no new locks will be needed
1018 */
1019 if (!(request->fl_flags & FL_ACCESS) &&
1020 (request->fl_type != F_UNLCK ||
1021 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1022 new_fl = locks_alloc_lock();
1023 new_fl2 = locks_alloc_lock();
1024 }
1025
1026 percpu_down_read_preempt_disable(&file_rwsem);
1027 spin_lock(&ctx->flc_lock);
1028 /*
1029 * New lock request. Walk all POSIX locks and look for conflicts. If
1030 * there are any, either return error or put the request on the
1031 * blocker's list of waiters and the global blocked_hash.
1032 */
1033 if (request->fl_type != F_UNLCK) {
1034 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1035 if (!posix_locks_conflict(request, fl))
1036 continue;
1037 if (conflock)
1038 locks_copy_conflock(conflock, fl);
1039 error = -EAGAIN;
1040 if (!(request->fl_flags & FL_SLEEP))
1041 goto out;
1042 /*
1043 * Deadlock detection and insertion into the blocked
1044 * locks list must be done while holding the same lock!
1045 */
1046 error = -EDEADLK;
1047 spin_lock(&blocked_lock_lock);
1048 if (likely(!posix_locks_deadlock(request, fl))) {
1049 error = FILE_LOCK_DEFERRED;
1050 __locks_insert_block(fl, request);
1051 }
1052 spin_unlock(&blocked_lock_lock);
1053 goto out;
1054 }
1055 }
1056
1057 /* If we're just looking for a conflict, we're done. */
1058 error = 0;
1059 if (request->fl_flags & FL_ACCESS)
1060 goto out;
1061
1062 /* Find the first old lock with the same owner as the new lock */
1063 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1064 if (posix_same_owner(request, fl))
1065 break;
1066 }
1067
1068 /* Process locks with this owner. */
1069 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1070 if (!posix_same_owner(request, fl))
1071 break;
1072
1073 /* Detect adjacent or overlapping regions (if same lock type) */
1074 if (request->fl_type == fl->fl_type) {
1075 /* In all comparisons of start vs end, use
1076 * "start - 1" rather than "end + 1". If end
1077 * is OFFSET_MAX, end + 1 will become negative.
1078 */
1079 if (fl->fl_end < request->fl_start - 1)
1080 continue;
1081 /* If the next lock in the list has entirely bigger
1082 * addresses than the new one, insert the lock here.
1083 */
1084 if (fl->fl_start - 1 > request->fl_end)
1085 break;
1086
1087 /* If we come here, the new and old lock are of the
1088 * same type and adjacent or overlapping. Make one
1089 * lock yielding from the lower start address of both
1090 * locks to the higher end address.
1091 */
1092 if (fl->fl_start > request->fl_start)
1093 fl->fl_start = request->fl_start;
1094 else
1095 request->fl_start = fl->fl_start;
1096 if (fl->fl_end < request->fl_end)
1097 fl->fl_end = request->fl_end;
1098 else
1099 request->fl_end = fl->fl_end;
1100 if (added) {
1101 locks_delete_lock_ctx(fl, &dispose);
1102 continue;
1103 }
1104 request = fl;
1105 added = true;
1106 } else {
1107 /* Processing for different lock types is a bit
1108 * more complex.
1109 */
1110 if (fl->fl_end < request->fl_start)
1111 continue;
1112 if (fl->fl_start > request->fl_end)
1113 break;
1114 if (request->fl_type == F_UNLCK)
1115 added = true;
1116 if (fl->fl_start < request->fl_start)
1117 left = fl;
1118 /* If the next lock in the list has a higher end
1119 * address than the new one, insert the new one here.
1120 */
1121 if (fl->fl_end > request->fl_end) {
1122 right = fl;
1123 break;
1124 }
1125 if (fl->fl_start >= request->fl_start) {
1126 /* The new lock completely replaces an old
1127 * one (This may happen several times).
1128 */
1129 if (added) {
1130 locks_delete_lock_ctx(fl, &dispose);
1131 continue;
1132 }
1133 /*
1134 * Replace the old lock with new_fl, and
1135 * remove the old one. It's safe to do the
1136 * insert here since we know that we won't be
1137 * using new_fl later, and that the lock is
1138 * just replacing an existing lock.
1139 */
1140 error = -ENOLCK;
1141 if (!new_fl)
1142 goto out;
1143 locks_copy_lock(new_fl, request);
1144 request = new_fl;
1145 new_fl = NULL;
1146 locks_insert_lock_ctx(request, &fl->fl_list);
1147 locks_delete_lock_ctx(fl, &dispose);
1148 added = true;
1149 }
1150 }
1151 }
1152
1153 /*
1154 * The above code only modifies existing locks in case of merging or
1155 * replacing. If new lock(s) need to be inserted all modifications are
1156 * done below this, so it's safe yet to bail out.
1157 */
1158 error = -ENOLCK; /* "no luck" */
1159 if (right && left == right && !new_fl2)
1160 goto out;
1161
1162 error = 0;
1163 if (!added) {
1164 if (request->fl_type == F_UNLCK) {
1165 if (request->fl_flags & FL_EXISTS)
1166 error = -ENOENT;
1167 goto out;
1168 }
1169
1170 if (!new_fl) {
1171 error = -ENOLCK;
1172 goto out;
1173 }
1174 locks_copy_lock(new_fl, request);
1175 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1176 fl = new_fl;
1177 new_fl = NULL;
1178 }
1179 if (right) {
1180 if (left == right) {
1181 /* The new lock breaks the old one in two pieces,
1182 * so we have to use the second new lock.
1183 */
1184 left = new_fl2;
1185 new_fl2 = NULL;
1186 locks_copy_lock(left, right);
1187 locks_insert_lock_ctx(left, &fl->fl_list);
1188 }
1189 right->fl_start = request->fl_end + 1;
1190 locks_wake_up_blocks(right);
1191 }
1192 if (left) {
1193 left->fl_end = request->fl_start - 1;
1194 locks_wake_up_blocks(left);
1195 }
1196 out:
1197 spin_unlock(&ctx->flc_lock);
1198 percpu_up_read_preempt_enable(&file_rwsem);
1199 /*
1200 * Free any unused locks.
1201 */
1202 if (new_fl)
1203 locks_free_lock(new_fl);
1204 if (new_fl2)
1205 locks_free_lock(new_fl2);
1206 locks_dispose_list(&dispose);
1207 trace_posix_lock_inode(inode, request, error);
1208
1209 return error;
1210}
1211
1212/**
1213 * posix_lock_file - Apply a POSIX-style lock to a file
1214 * @filp: The file to apply the lock to
1215 * @fl: The lock to be applied
1216 * @conflock: Place to return a copy of the conflicting lock, if found.
1217 *
1218 * Add a POSIX style lock to a file.
1219 * We merge adjacent & overlapping locks whenever possible.
1220 * POSIX locks are sorted by owner task, then by starting address
1221 *
1222 * Note that if called with an FL_EXISTS argument, the caller may determine
1223 * whether or not a lock was successfully freed by testing the return
1224 * value for -ENOENT.
1225 */
1226int posix_lock_file(struct file *filp, struct file_lock *fl,
1227 struct file_lock *conflock)
1228{
1229 return posix_lock_inode(locks_inode(filp), fl, conflock);
1230}
1231EXPORT_SYMBOL(posix_lock_file);
1232
1233/**
1234 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1235 * @inode: inode of file to which lock request should be applied
1236 * @fl: The lock to be applied
1237 *
1238 * Apply a POSIX style lock request to an inode.
1239 */
1240static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1241{
1242 int error;
1243 might_sleep ();
1244 for (;;) {
1245 error = posix_lock_inode(inode, fl, NULL);
1246 if (error != FILE_LOCK_DEFERRED)
1247 break;
1248 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1249 if (!error)
1250 continue;
1251
1252 locks_delete_block(fl);
1253 break;
1254 }
1255 return error;
1256}
1257
1258#ifdef CONFIG_MANDATORY_FILE_LOCKING
1259/**
1260 * locks_mandatory_locked - Check for an active lock
1261 * @file: the file to check
1262 *
1263 * Searches the inode's list of locks to find any POSIX locks which conflict.
1264 * This function is called from locks_verify_locked() only.
1265 */
1266int locks_mandatory_locked(struct file *file)
1267{
1268 int ret;
1269 struct inode *inode = locks_inode(file);
1270 struct file_lock_context *ctx;
1271 struct file_lock *fl;
1272
1273 ctx = smp_load_acquire(&inode->i_flctx);
1274 if (!ctx || list_empty_careful(&ctx->flc_posix))
1275 return 0;
1276
1277 /*
1278 * Search the lock list for this inode for any POSIX locks.
1279 */
1280 spin_lock(&ctx->flc_lock);
1281 ret = 0;
1282 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1283 if (fl->fl_owner != current->files &&
1284 fl->fl_owner != file) {
1285 ret = -EAGAIN;
1286 break;
1287 }
1288 }
1289 spin_unlock(&ctx->flc_lock);
1290 return ret;
1291}
1292
1293/**
1294 * locks_mandatory_area - Check for a conflicting lock
1295 * @inode: the file to check
1296 * @filp: how the file was opened (if it was)
1297 * @start: first byte in the file to check
1298 * @end: lastbyte in the file to check
1299 * @type: %F_WRLCK for a write lock, else %F_RDLCK
1300 *
1301 * Searches the inode's list of locks to find any POSIX locks which conflict.
1302 */
1303int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1304 loff_t end, unsigned char type)
1305{
1306 struct file_lock fl;
1307 int error;
1308 bool sleep = false;
1309
1310 locks_init_lock(&fl);
1311 fl.fl_pid = current->tgid;
1312 fl.fl_file = filp;
1313 fl.fl_flags = FL_POSIX | FL_ACCESS;
1314 if (filp && !(filp->f_flags & O_NONBLOCK))
1315 sleep = true;
1316 fl.fl_type = type;
1317 fl.fl_start = start;
1318 fl.fl_end = end;
1319
1320 for (;;) {
1321 if (filp) {
1322 fl.fl_owner = filp;
1323 fl.fl_flags &= ~FL_SLEEP;
1324 error = posix_lock_inode(inode, &fl, NULL);
1325 if (!error)
1326 break;
1327 }
1328
1329 if (sleep)
1330 fl.fl_flags |= FL_SLEEP;
1331 fl.fl_owner = current->files;
1332 error = posix_lock_inode(inode, &fl, NULL);
1333 if (error != FILE_LOCK_DEFERRED)
1334 break;
1335 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1336 if (!error) {
1337 /*
1338 * If we've been sleeping someone might have
1339 * changed the permissions behind our back.
1340 */
1341 if (__mandatory_lock(inode))
1342 continue;
1343 }
1344
1345 locks_delete_block(&fl);
1346 break;
1347 }
1348
1349 return error;
1350}
1351
1352EXPORT_SYMBOL(locks_mandatory_area);
1353#endif /* CONFIG_MANDATORY_FILE_LOCKING */
1354
1355static void lease_clear_pending(struct file_lock *fl, int arg)
1356{
1357 switch (arg) {
1358 case F_UNLCK:
1359 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1360 /* fall through: */
1361 case F_RDLCK:
1362 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1363 }
1364}
1365
1366/* We already had a lease on this file; just change its type */
1367int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1368{
1369 int error = assign_type(fl, arg);
1370
1371 if (error)
1372 return error;
1373 lease_clear_pending(fl, arg);
1374 locks_wake_up_blocks(fl);
1375 if (arg == F_UNLCK) {
1376 struct file *filp = fl->fl_file;
1377
1378 f_delown(filp);
1379 filp->f_owner.signum = 0;
1380 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1381 if (fl->fl_fasync != NULL) {
1382 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1383 fl->fl_fasync = NULL;
1384 }
1385 locks_delete_lock_ctx(fl, dispose);
1386 }
1387 return 0;
1388}
1389EXPORT_SYMBOL(lease_modify);
1390
1391static bool past_time(unsigned long then)
1392{
1393 if (!then)
1394 /* 0 is a special value meaning "this never expires": */
1395 return false;
1396 return time_after(jiffies, then);
1397}
1398
1399static void time_out_leases(struct inode *inode, struct list_head *dispose)
1400{
1401 struct file_lock_context *ctx = inode->i_flctx;
1402 struct file_lock *fl, *tmp;
1403
1404 lockdep_assert_held(&ctx->flc_lock);
1405
1406 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1407 trace_time_out_leases(inode, fl);
1408 if (past_time(fl->fl_downgrade_time))
1409 lease_modify(fl, F_RDLCK, dispose);
1410 if (past_time(fl->fl_break_time))
1411 lease_modify(fl, F_UNLCK, dispose);
1412 }
1413}
1414
1415static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1416{
1417 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1418 return false;
1419 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1420 return false;
1421 return locks_conflict(breaker, lease);
1422}
1423
1424static bool
1425any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1426{
1427 struct file_lock_context *ctx = inode->i_flctx;
1428 struct file_lock *fl;
1429
1430 lockdep_assert_held(&ctx->flc_lock);
1431
1432 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1433 if (leases_conflict(fl, breaker))
1434 return true;
1435 }
1436 return false;
1437}
1438
1439/**
1440 * __break_lease - revoke all outstanding leases on file
1441 * @inode: the inode of the file to return
1442 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1443 * break all leases
1444 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1445 * only delegations
1446 *
1447 * break_lease (inlined for speed) has checked there already is at least
1448 * some kind of lock (maybe a lease) on this file. Leases are broken on
1449 * a call to open() or truncate(). This function can sleep unless you
1450 * specified %O_NONBLOCK to your open().
1451 */
1452int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1453{
1454 int error = 0;
1455 struct file_lock_context *ctx;
1456 struct file_lock *new_fl, *fl, *tmp;
1457 unsigned long break_time;
1458 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1459 LIST_HEAD(dispose);
1460
1461 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1462 if (IS_ERR(new_fl))
1463 return PTR_ERR(new_fl);
1464 new_fl->fl_flags = type;
1465
1466 /* typically we will check that ctx is non-NULL before calling */
1467 ctx = smp_load_acquire(&inode->i_flctx);
1468 if (!ctx) {
1469 WARN_ON_ONCE(1);
1470 return error;
1471 }
1472
1473 percpu_down_read_preempt_disable(&file_rwsem);
1474 spin_lock(&ctx->flc_lock);
1475
1476 time_out_leases(inode, &dispose);
1477
1478 if (!any_leases_conflict(inode, new_fl))
1479 goto out;
1480
1481 break_time = 0;
1482 if (lease_break_time > 0) {
1483 break_time = jiffies + lease_break_time * HZ;
1484 if (break_time == 0)
1485 break_time++; /* so that 0 means no break time */
1486 }
1487
1488 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1489 if (!leases_conflict(fl, new_fl))
1490 continue;
1491 if (want_write) {
1492 if (fl->fl_flags & FL_UNLOCK_PENDING)
1493 continue;
1494 fl->fl_flags |= FL_UNLOCK_PENDING;
1495 fl->fl_break_time = break_time;
1496 } else {
1497 if (lease_breaking(fl))
1498 continue;
1499 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1500 fl->fl_downgrade_time = break_time;
1501 }
1502 if (fl->fl_lmops->lm_break(fl))
1503 locks_delete_lock_ctx(fl, &dispose);
1504 }
1505
1506 if (list_empty(&ctx->flc_lease))
1507 goto out;
1508
1509 if (mode & O_NONBLOCK) {
1510 trace_break_lease_noblock(inode, new_fl);
1511 error = -EWOULDBLOCK;
1512 goto out;
1513 }
1514
1515restart:
1516 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1517 break_time = fl->fl_break_time;
1518 if (break_time != 0)
1519 break_time -= jiffies;
1520 if (break_time == 0)
1521 break_time++;
1522 locks_insert_block(fl, new_fl);
1523 trace_break_lease_block(inode, new_fl);
1524 spin_unlock(&ctx->flc_lock);
1525 percpu_up_read_preempt_enable(&file_rwsem);
1526
1527 locks_dispose_list(&dispose);
1528 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1529 !new_fl->fl_next, break_time);
1530
1531 percpu_down_read_preempt_disable(&file_rwsem);
1532 spin_lock(&ctx->flc_lock);
1533 trace_break_lease_unblock(inode, new_fl);
1534 locks_delete_block(new_fl);
1535 if (error >= 0) {
1536 /*
1537 * Wait for the next conflicting lease that has not been
1538 * broken yet
1539 */
1540 if (error == 0)
1541 time_out_leases(inode, &dispose);
1542 if (any_leases_conflict(inode, new_fl))
1543 goto restart;
1544 error = 0;
1545 }
1546out:
1547 spin_unlock(&ctx->flc_lock);
1548 percpu_up_read_preempt_enable(&file_rwsem);
1549 locks_dispose_list(&dispose);
1550 locks_free_lock(new_fl);
1551 return error;
1552}
1553
1554EXPORT_SYMBOL(__break_lease);
1555
1556/**
1557 * lease_get_mtime - update modified time of an inode with exclusive lease
1558 * @inode: the inode
1559 * @time: pointer to a timespec which contains the last modified time
1560 *
1561 * This is to force NFS clients to flush their caches for files with
1562 * exclusive leases. The justification is that if someone has an
1563 * exclusive lease, then they could be modifying it.
1564 */
1565void lease_get_mtime(struct inode *inode, struct timespec *time)
1566{
1567 bool has_lease = false;
1568 struct file_lock_context *ctx;
1569 struct file_lock *fl;
1570
1571 ctx = smp_load_acquire(&inode->i_flctx);
1572 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1573 spin_lock(&ctx->flc_lock);
1574 fl = list_first_entry_or_null(&ctx->flc_lease,
1575 struct file_lock, fl_list);
1576 if (fl && (fl->fl_type == F_WRLCK))
1577 has_lease = true;
1578 spin_unlock(&ctx->flc_lock);
1579 }
1580
1581 if (has_lease)
1582 *time = current_time(inode);
1583}
1584
1585EXPORT_SYMBOL(lease_get_mtime);
1586
1587/**
1588 * fcntl_getlease - Enquire what lease is currently active
1589 * @filp: the file
1590 *
1591 * The value returned by this function will be one of
1592 * (if no lease break is pending):
1593 *
1594 * %F_RDLCK to indicate a shared lease is held.
1595 *
1596 * %F_WRLCK to indicate an exclusive lease is held.
1597 *
1598 * %F_UNLCK to indicate no lease is held.
1599 *
1600 * (if a lease break is pending):
1601 *
1602 * %F_RDLCK to indicate an exclusive lease needs to be
1603 * changed to a shared lease (or removed).
1604 *
1605 * %F_UNLCK to indicate the lease needs to be removed.
1606 *
1607 * XXX: sfr & willy disagree over whether F_INPROGRESS
1608 * should be returned to userspace.
1609 */
1610int fcntl_getlease(struct file *filp)
1611{
1612 struct file_lock *fl;
1613 struct inode *inode = locks_inode(filp);
1614 struct file_lock_context *ctx;
1615 int type = F_UNLCK;
1616 LIST_HEAD(dispose);
1617
1618 ctx = smp_load_acquire(&inode->i_flctx);
1619 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1620 percpu_down_read_preempt_disable(&file_rwsem);
1621 spin_lock(&ctx->flc_lock);
1622 time_out_leases(inode, &dispose);
1623 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1624 if (fl->fl_file != filp)
1625 continue;
1626 type = target_leasetype(fl);
1627 break;
1628 }
1629 spin_unlock(&ctx->flc_lock);
1630 percpu_up_read_preempt_enable(&file_rwsem);
1631
1632 locks_dispose_list(&dispose);
1633 }
1634 return type;
1635}
1636
1637/**
1638 * check_conflicting_open - see if the given dentry points to a file that has
1639 * an existing open that would conflict with the
1640 * desired lease.
1641 * @dentry: dentry to check
1642 * @arg: type of lease that we're trying to acquire
1643 * @flags: current lock flags
1644 *
1645 * Check to see if there's an existing open fd on this file that would
1646 * conflict with the lease we're trying to set.
1647 */
1648static int
1649check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
1650{
1651 int ret = 0;
1652 struct inode *inode = dentry->d_inode;
1653
1654 if (flags & FL_LAYOUT)
1655 return 0;
1656
1657 if ((arg == F_RDLCK) &&
1658 (atomic_read(&d_real_inode(dentry)->i_writecount) > 0))
1659 return -EAGAIN;
1660
1661 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1662 (atomic_read(&inode->i_count) > 1)))
1663 ret = -EAGAIN;
1664
1665 return ret;
1666}
1667
1668static int
1669generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1670{
1671 struct file_lock *fl, *my_fl = NULL, *lease;
1672 struct dentry *dentry = filp->f_path.dentry;
1673 struct inode *inode = dentry->d_inode;
1674 struct file_lock_context *ctx;
1675 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1676 int error;
1677 LIST_HEAD(dispose);
1678
1679 lease = *flp;
1680 trace_generic_add_lease(inode, lease);
1681
1682 /* Note that arg is never F_UNLCK here */
1683 ctx = locks_get_lock_context(inode, arg);
1684 if (!ctx)
1685 return -ENOMEM;
1686
1687 /*
1688 * In the delegation case we need mutual exclusion with
1689 * a number of operations that take the i_mutex. We trylock
1690 * because delegations are an optional optimization, and if
1691 * there's some chance of a conflict--we'd rather not
1692 * bother, maybe that's a sign this just isn't a good file to
1693 * hand out a delegation on.
1694 */
1695 if (is_deleg && !inode_trylock(inode))
1696 return -EAGAIN;
1697
1698 if (is_deleg && arg == F_WRLCK) {
1699 /* Write delegations are not currently supported: */
1700 inode_unlock(inode);
1701 WARN_ON_ONCE(1);
1702 return -EINVAL;
1703 }
1704
1705 percpu_down_read_preempt_disable(&file_rwsem);
1706 spin_lock(&ctx->flc_lock);
1707 time_out_leases(inode, &dispose);
1708 error = check_conflicting_open(dentry, arg, lease->fl_flags);
1709 if (error)
1710 goto out;
1711
1712 /*
1713 * At this point, we know that if there is an exclusive
1714 * lease on this file, then we hold it on this filp
1715 * (otherwise our open of this file would have blocked).
1716 * And if we are trying to acquire an exclusive lease,
1717 * then the file is not open by anyone (including us)
1718 * except for this filp.
1719 */
1720 error = -EAGAIN;
1721 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1722 if (fl->fl_file == filp &&
1723 fl->fl_owner == lease->fl_owner) {
1724 my_fl = fl;
1725 continue;
1726 }
1727
1728 /*
1729 * No exclusive leases if someone else has a lease on
1730 * this file:
1731 */
1732 if (arg == F_WRLCK)
1733 goto out;
1734 /*
1735 * Modifying our existing lease is OK, but no getting a
1736 * new lease if someone else is opening for write:
1737 */
1738 if (fl->fl_flags & FL_UNLOCK_PENDING)
1739 goto out;
1740 }
1741
1742 if (my_fl != NULL) {
1743 lease = my_fl;
1744 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1745 if (error)
1746 goto out;
1747 goto out_setup;
1748 }
1749
1750 error = -EINVAL;
1751 if (!leases_enable)
1752 goto out;
1753
1754 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1755 /*
1756 * The check in break_lease() is lockless. It's possible for another
1757 * open to race in after we did the earlier check for a conflicting
1758 * open but before the lease was inserted. Check again for a
1759 * conflicting open and cancel the lease if there is one.
1760 *
1761 * We also add a barrier here to ensure that the insertion of the lock
1762 * precedes these checks.
1763 */
1764 smp_mb();
1765 error = check_conflicting_open(dentry, arg, lease->fl_flags);
1766 if (error) {
1767 locks_unlink_lock_ctx(lease);
1768 goto out;
1769 }
1770
1771out_setup:
1772 if (lease->fl_lmops->lm_setup)
1773 lease->fl_lmops->lm_setup(lease, priv);
1774out:
1775 spin_unlock(&ctx->flc_lock);
1776 percpu_up_read_preempt_enable(&file_rwsem);
1777 locks_dispose_list(&dispose);
1778 if (is_deleg)
1779 inode_unlock(inode);
1780 if (!error && !my_fl)
1781 *flp = NULL;
1782 return error;
1783}
1784
1785static int generic_delete_lease(struct file *filp, void *owner)
1786{
1787 int error = -EAGAIN;
1788 struct file_lock *fl, *victim = NULL;
1789 struct inode *inode = locks_inode(filp);
1790 struct file_lock_context *ctx;
1791 LIST_HEAD(dispose);
1792
1793 ctx = smp_load_acquire(&inode->i_flctx);
1794 if (!ctx) {
1795 trace_generic_delete_lease(inode, NULL);
1796 return error;
1797 }
1798
1799 percpu_down_read_preempt_disable(&file_rwsem);
1800 spin_lock(&ctx->flc_lock);
1801 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1802 if (fl->fl_file == filp &&
1803 fl->fl_owner == owner) {
1804 victim = fl;
1805 break;
1806 }
1807 }
1808 trace_generic_delete_lease(inode, victim);
1809 if (victim)
1810 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1811 spin_unlock(&ctx->flc_lock);
1812 percpu_up_read_preempt_enable(&file_rwsem);
1813 locks_dispose_list(&dispose);
1814 return error;
1815}
1816
1817/**
1818 * generic_setlease - sets a lease on an open file
1819 * @filp: file pointer
1820 * @arg: type of lease to obtain
1821 * @flp: input - file_lock to use, output - file_lock inserted
1822 * @priv: private data for lm_setup (may be NULL if lm_setup
1823 * doesn't require it)
1824 *
1825 * The (input) flp->fl_lmops->lm_break function is required
1826 * by break_lease().
1827 */
1828int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1829 void **priv)
1830{
1831 struct inode *inode = locks_inode(filp);
1832 int error;
1833
1834 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1835 return -EACCES;
1836 if (!S_ISREG(inode->i_mode))
1837 return -EINVAL;
1838 error = security_file_lock(filp, arg);
1839 if (error)
1840 return error;
1841
1842 switch (arg) {
1843 case F_UNLCK:
1844 return generic_delete_lease(filp, *priv);
1845 case F_RDLCK:
1846 case F_WRLCK:
1847 if (!(*flp)->fl_lmops->lm_break) {
1848 WARN_ON_ONCE(1);
1849 return -ENOLCK;
1850 }
1851
1852 return generic_add_lease(filp, arg, flp, priv);
1853 default:
1854 return -EINVAL;
1855 }
1856}
1857EXPORT_SYMBOL(generic_setlease);
1858
1859/**
1860 * vfs_setlease - sets a lease on an open file
1861 * @filp: file pointer
1862 * @arg: type of lease to obtain
1863 * @lease: file_lock to use when adding a lease
1864 * @priv: private info for lm_setup when adding a lease (may be
1865 * NULL if lm_setup doesn't require it)
1866 *
1867 * Call this to establish a lease on the file. The "lease" argument is not
1868 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1869 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1870 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1871 * stack trace).
1872 *
1873 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1874 * may be NULL if the lm_setup operation doesn't require it.
1875 */
1876int
1877vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1878{
1879 if (filp->f_op->setlease && is_remote_lock(filp))
1880 return filp->f_op->setlease(filp, arg, lease, priv);
1881 else
1882 return generic_setlease(filp, arg, lease, priv);
1883}
1884EXPORT_SYMBOL_GPL(vfs_setlease);
1885
1886static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1887{
1888 struct file_lock *fl;
1889 struct fasync_struct *new;
1890 int error;
1891
1892 fl = lease_alloc(filp, arg);
1893 if (IS_ERR(fl))
1894 return PTR_ERR(fl);
1895
1896 new = fasync_alloc();
1897 if (!new) {
1898 locks_free_lock(fl);
1899 return -ENOMEM;
1900 }
1901 new->fa_fd = fd;
1902
1903 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1904 if (fl)
1905 locks_free_lock(fl);
1906 if (new)
1907 fasync_free(new);
1908 return error;
1909}
1910
1911/**
1912 * fcntl_setlease - sets a lease on an open file
1913 * @fd: open file descriptor
1914 * @filp: file pointer
1915 * @arg: type of lease to obtain
1916 *
1917 * Call this fcntl to establish a lease on the file.
1918 * Note that you also need to call %F_SETSIG to
1919 * receive a signal when the lease is broken.
1920 */
1921int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1922{
1923 if (arg == F_UNLCK)
1924 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
1925 return do_fcntl_add_lease(fd, filp, arg);
1926}
1927
1928/**
1929 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
1930 * @inode: inode of the file to apply to
1931 * @fl: The lock to be applied
1932 *
1933 * Apply a FLOCK style lock request to an inode.
1934 */
1935static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1936{
1937 int error;
1938 might_sleep();
1939 for (;;) {
1940 error = flock_lock_inode(inode, fl);
1941 if (error != FILE_LOCK_DEFERRED)
1942 break;
1943 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1944 if (!error)
1945 continue;
1946
1947 locks_delete_block(fl);
1948 break;
1949 }
1950 return error;
1951}
1952
1953/**
1954 * locks_lock_inode_wait - Apply a lock to an inode
1955 * @inode: inode of the file to apply to
1956 * @fl: The lock to be applied
1957 *
1958 * Apply a POSIX or FLOCK style lock request to an inode.
1959 */
1960int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1961{
1962 int res = 0;
1963 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
1964 case FL_POSIX:
1965 res = posix_lock_inode_wait(inode, fl);
1966 break;
1967 case FL_FLOCK:
1968 res = flock_lock_inode_wait(inode, fl);
1969 break;
1970 default:
1971 BUG();
1972 }
1973 return res;
1974}
1975EXPORT_SYMBOL(locks_lock_inode_wait);
1976
1977/**
1978 * sys_flock: - flock() system call.
1979 * @fd: the file descriptor to lock.
1980 * @cmd: the type of lock to apply.
1981 *
1982 * Apply a %FL_FLOCK style lock to an open file descriptor.
1983 * The @cmd can be one of:
1984 *
1985 * - %LOCK_SH -- a shared lock.
1986 * - %LOCK_EX -- an exclusive lock.
1987 * - %LOCK_UN -- remove an existing lock.
1988 * - %LOCK_MAND -- a 'mandatory' flock.
1989 * This exists to emulate Windows Share Modes.
1990 *
1991 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1992 * processes read and write access respectively.
1993 */
1994SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1995{
1996 struct fd f = fdget(fd);
1997 struct file_lock *lock;
1998 int can_sleep, unlock;
1999 int error;
2000
2001 error = -EBADF;
2002 if (!f.file)
2003 goto out;
2004
2005 can_sleep = !(cmd & LOCK_NB);
2006 cmd &= ~LOCK_NB;
2007 unlock = (cmd == LOCK_UN);
2008
2009 if (!unlock && !(cmd & LOCK_MAND) &&
2010 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2011 goto out_putf;
2012
2013 lock = flock_make_lock(f.file, cmd);
2014 if (IS_ERR(lock)) {
2015 error = PTR_ERR(lock);
2016 goto out_putf;
2017 }
2018
2019 if (can_sleep)
2020 lock->fl_flags |= FL_SLEEP;
2021
2022 error = security_file_lock(f.file, lock->fl_type);
2023 if (error)
2024 goto out_free;
2025
2026 if (f.file->f_op->flock && is_remote_lock(f.file))
2027 error = f.file->f_op->flock(f.file,
2028 (can_sleep) ? F_SETLKW : F_SETLK,
2029 lock);
2030 else
2031 error = locks_lock_file_wait(f.file, lock);
2032
2033 out_free:
2034 locks_free_lock(lock);
2035
2036 out_putf:
2037 fdput(f);
2038 out:
2039 return error;
2040}
2041
2042/**
2043 * vfs_test_lock - test file byte range lock
2044 * @filp: The file to test lock for
2045 * @fl: The lock to test; also used to hold result
2046 *
2047 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2048 * setting conf->fl_type to something other than F_UNLCK.
2049 */
2050int vfs_test_lock(struct file *filp, struct file_lock *fl)
2051{
2052 if (filp->f_op->lock && is_remote_lock(filp))
2053 return filp->f_op->lock(filp, F_GETLK, fl);
2054 posix_test_lock(filp, fl);
2055 return 0;
2056}
2057EXPORT_SYMBOL_GPL(vfs_test_lock);
2058
2059/**
2060 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2061 * @fl: The file_lock who's fl_pid should be translated
2062 * @ns: The namespace into which the pid should be translated
2063 *
2064 * Used to tranlate a fl_pid into a namespace virtual pid number
2065 */
2066static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2067{
2068 pid_t vnr;
2069 struct pid *pid;
2070
2071 if (IS_OFDLCK(fl))
2072 return -1;
2073 if (IS_REMOTELCK(fl))
2074 return fl->fl_pid;
2075
2076 rcu_read_lock();
2077 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2078 vnr = pid_nr_ns(pid, ns);
2079 rcu_read_unlock();
2080 return vnr;
2081}
2082
2083static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2084{
2085 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2086#if BITS_PER_LONG == 32
2087 /*
2088 * Make sure we can represent the posix lock via
2089 * legacy 32bit flock.
2090 */
2091 if (fl->fl_start > OFFT_OFFSET_MAX)
2092 return -EOVERFLOW;
2093 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2094 return -EOVERFLOW;
2095#endif
2096 flock->l_start = fl->fl_start;
2097 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2098 fl->fl_end - fl->fl_start + 1;
2099 flock->l_whence = 0;
2100 flock->l_type = fl->fl_type;
2101 return 0;
2102}
2103
2104#if BITS_PER_LONG == 32
2105static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2106{
2107 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2108 flock->l_start = fl->fl_start;
2109 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2110 fl->fl_end - fl->fl_start + 1;
2111 flock->l_whence = 0;
2112 flock->l_type = fl->fl_type;
2113}
2114#endif
2115
2116/* Report the first existing lock that would conflict with l.
2117 * This implements the F_GETLK command of fcntl().
2118 */
2119int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2120{
2121 struct file_lock *fl;
2122 int error;
2123
2124 fl = locks_alloc_lock();
2125 if (fl == NULL)
2126 return -ENOMEM;
2127 error = -EINVAL;
2128 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2129 goto out;
2130
2131 error = flock_to_posix_lock(filp, fl, flock);
2132 if (error)
2133 goto out;
2134
2135 if (cmd == F_OFD_GETLK) {
2136 error = -EINVAL;
2137 if (flock->l_pid != 0)
2138 goto out;
2139
2140 cmd = F_GETLK;
2141 fl->fl_flags |= FL_OFDLCK;
2142 fl->fl_owner = filp;
2143 }
2144
2145 error = vfs_test_lock(filp, fl);
2146 if (error)
2147 goto out;
2148
2149 flock->l_type = fl->fl_type;
2150 if (fl->fl_type != F_UNLCK) {
2151 error = posix_lock_to_flock(flock, fl);
2152 if (error)
2153 goto out;
2154 }
2155out:
2156 locks_free_lock(fl);
2157 return error;
2158}
2159
2160/**
2161 * vfs_lock_file - file byte range lock
2162 * @filp: The file to apply the lock to
2163 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2164 * @fl: The lock to be applied
2165 * @conf: Place to return a copy of the conflicting lock, if found.
2166 *
2167 * A caller that doesn't care about the conflicting lock may pass NULL
2168 * as the final argument.
2169 *
2170 * If the filesystem defines a private ->lock() method, then @conf will
2171 * be left unchanged; so a caller that cares should initialize it to
2172 * some acceptable default.
2173 *
2174 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2175 * locks, the ->lock() interface may return asynchronously, before the lock has
2176 * been granted or denied by the underlying filesystem, if (and only if)
2177 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2178 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2179 * the request is for a blocking lock. When ->lock() does return asynchronously,
2180 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2181 * request completes.
2182 * If the request is for non-blocking lock the file system should return
2183 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2184 * with the result. If the request timed out the callback routine will return a
2185 * nonzero return code and the file system should release the lock. The file
2186 * system is also responsible to keep a corresponding posix lock when it
2187 * grants a lock so the VFS can find out which locks are locally held and do
2188 * the correct lock cleanup when required.
2189 * The underlying filesystem must not drop the kernel lock or call
2190 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2191 * return code.
2192 */
2193int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2194{
2195 if (filp->f_op->lock && is_remote_lock(filp))
2196 return filp->f_op->lock(filp, cmd, fl);
2197 else
2198 return posix_lock_file(filp, fl, conf);
2199}
2200EXPORT_SYMBOL_GPL(vfs_lock_file);
2201
2202static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2203 struct file_lock *fl)
2204{
2205 int error;
2206
2207 error = security_file_lock(filp, fl->fl_type);
2208 if (error)
2209 return error;
2210
2211 for (;;) {
2212 error = vfs_lock_file(filp, cmd, fl, NULL);
2213 if (error != FILE_LOCK_DEFERRED)
2214 break;
2215 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2216 if (!error)
2217 continue;
2218
2219 locks_delete_block(fl);
2220 break;
2221 }
2222
2223 return error;
2224}
2225
2226/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2227static int
2228check_fmode_for_setlk(struct file_lock *fl)
2229{
2230 switch (fl->fl_type) {
2231 case F_RDLCK:
2232 if (!(fl->fl_file->f_mode & FMODE_READ))
2233 return -EBADF;
2234 break;
2235 case F_WRLCK:
2236 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2237 return -EBADF;
2238 }
2239 return 0;
2240}
2241
2242/* Apply the lock described by l to an open file descriptor.
2243 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2244 */
2245int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2246 struct flock *flock)
2247{
2248 struct file_lock *file_lock = locks_alloc_lock();
2249 struct inode *inode = locks_inode(filp);
2250 struct file *f;
2251 int error;
2252
2253 if (file_lock == NULL)
2254 return -ENOLCK;
2255
2256 /* Don't allow mandatory locks on files that may be memory mapped
2257 * and shared.
2258 */
2259 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2260 error = -EAGAIN;
2261 goto out;
2262 }
2263
2264 error = flock_to_posix_lock(filp, file_lock, flock);
2265 if (error)
2266 goto out;
2267
2268 error = check_fmode_for_setlk(file_lock);
2269 if (error)
2270 goto out;
2271
2272 /*
2273 * If the cmd is requesting file-private locks, then set the
2274 * FL_OFDLCK flag and override the owner.
2275 */
2276 switch (cmd) {
2277 case F_OFD_SETLK:
2278 error = -EINVAL;
2279 if (flock->l_pid != 0)
2280 goto out;
2281
2282 cmd = F_SETLK;
2283 file_lock->fl_flags |= FL_OFDLCK;
2284 file_lock->fl_owner = filp;
2285 break;
2286 case F_OFD_SETLKW:
2287 error = -EINVAL;
2288 if (flock->l_pid != 0)
2289 goto out;
2290
2291 cmd = F_SETLKW;
2292 file_lock->fl_flags |= FL_OFDLCK;
2293 file_lock->fl_owner = filp;
2294 /* Fallthrough */
2295 case F_SETLKW:
2296 file_lock->fl_flags |= FL_SLEEP;
2297 }
2298
2299 error = do_lock_file_wait(filp, cmd, file_lock);
2300
2301 /*
2302 * Attempt to detect a close/fcntl race and recover by releasing the
2303 * lock that was just acquired. There is no need to do that when we're
2304 * unlocking though, or for OFD locks.
2305 */
2306 if (!error && file_lock->fl_type != F_UNLCK &&
2307 !(file_lock->fl_flags & FL_OFDLCK)) {
2308 /*
2309 * We need that spin_lock here - it prevents reordering between
2310 * update of i_flctx->flc_posix and check for it done in
2311 * close(). rcu_read_lock() wouldn't do.
2312 */
2313 spin_lock(¤t->files->file_lock);
2314 f = fcheck(fd);
2315 spin_unlock(¤t->files->file_lock);
2316 if (f != filp) {
2317 file_lock->fl_type = F_UNLCK;
2318 error = do_lock_file_wait(filp, cmd, file_lock);
2319 WARN_ON_ONCE(error);
2320 error = -EBADF;
2321 }
2322 }
2323out:
2324 trace_fcntl_setlk(inode, file_lock, error);
2325 locks_free_lock(file_lock);
2326 return error;
2327}
2328
2329#if BITS_PER_LONG == 32
2330/* Report the first existing lock that would conflict with l.
2331 * This implements the F_GETLK command of fcntl().
2332 */
2333int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2334{
2335 struct file_lock *fl;
2336 int error;
2337
2338 fl = locks_alloc_lock();
2339 if (fl == NULL)
2340 return -ENOMEM;
2341
2342 error = -EINVAL;
2343 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2344 goto out;
2345
2346 error = flock64_to_posix_lock(filp, fl, flock);
2347 if (error)
2348 goto out;
2349
2350 if (cmd == F_OFD_GETLK) {
2351 error = -EINVAL;
2352 if (flock->l_pid != 0)
2353 goto out;
2354
2355 cmd = F_GETLK64;
2356 fl->fl_flags |= FL_OFDLCK;
2357 fl->fl_owner = filp;
2358 }
2359
2360 error = vfs_test_lock(filp, fl);
2361 if (error)
2362 goto out;
2363
2364 flock->l_type = fl->fl_type;
2365 if (fl->fl_type != F_UNLCK)
2366 posix_lock_to_flock64(flock, fl);
2367
2368out:
2369 locks_free_lock(fl);
2370 return error;
2371}
2372
2373/* Apply the lock described by l to an open file descriptor.
2374 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2375 */
2376int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2377 struct flock64 *flock)
2378{
2379 struct file_lock *file_lock = locks_alloc_lock();
2380 struct inode *inode = locks_inode(filp);
2381 struct file *f;
2382 int error;
2383
2384 if (file_lock == NULL)
2385 return -ENOLCK;
2386
2387 /* Don't allow mandatory locks on files that may be memory mapped
2388 * and shared.
2389 */
2390 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2391 error = -EAGAIN;
2392 goto out;
2393 }
2394
2395 error = flock64_to_posix_lock(filp, file_lock, flock);
2396 if (error)
2397 goto out;
2398
2399 error = check_fmode_for_setlk(file_lock);
2400 if (error)
2401 goto out;
2402
2403 /*
2404 * If the cmd is requesting file-private locks, then set the
2405 * FL_OFDLCK flag and override the owner.
2406 */
2407 switch (cmd) {
2408 case F_OFD_SETLK:
2409 error = -EINVAL;
2410 if (flock->l_pid != 0)
2411 goto out;
2412
2413 cmd = F_SETLK64;
2414 file_lock->fl_flags |= FL_OFDLCK;
2415 file_lock->fl_owner = filp;
2416 break;
2417 case F_OFD_SETLKW:
2418 error = -EINVAL;
2419 if (flock->l_pid != 0)
2420 goto out;
2421
2422 cmd = F_SETLKW64;
2423 file_lock->fl_flags |= FL_OFDLCK;
2424 file_lock->fl_owner = filp;
2425 /* Fallthrough */
2426 case F_SETLKW64:
2427 file_lock->fl_flags |= FL_SLEEP;
2428 }
2429
2430 error = do_lock_file_wait(filp, cmd, file_lock);
2431
2432 /*
2433 * Attempt to detect a close/fcntl race and recover by releasing the
2434 * lock that was just acquired. There is no need to do that when we're
2435 * unlocking though, or for OFD locks.
2436 */
2437 if (!error && file_lock->fl_type != F_UNLCK &&
2438 !(file_lock->fl_flags & FL_OFDLCK)) {
2439 /*
2440 * We need that spin_lock here - it prevents reordering between
2441 * update of i_flctx->flc_posix and check for it done in
2442 * close(). rcu_read_lock() wouldn't do.
2443 */
2444 spin_lock(¤t->files->file_lock);
2445 f = fcheck(fd);
2446 spin_unlock(¤t->files->file_lock);
2447 if (f != filp) {
2448 file_lock->fl_type = F_UNLCK;
2449 error = do_lock_file_wait(filp, cmd, file_lock);
2450 WARN_ON_ONCE(error);
2451 error = -EBADF;
2452 }
2453 }
2454out:
2455 locks_free_lock(file_lock);
2456 return error;
2457}
2458#endif /* BITS_PER_LONG == 32 */
2459
2460/*
2461 * This function is called when the file is being removed
2462 * from the task's fd array. POSIX locks belonging to this task
2463 * are deleted at this time.
2464 */
2465void locks_remove_posix(struct file *filp, fl_owner_t owner)
2466{
2467 int error;
2468 struct inode *inode = locks_inode(filp);
2469 struct file_lock lock;
2470 struct file_lock_context *ctx;
2471
2472 /*
2473 * If there are no locks held on this file, we don't need to call
2474 * posix_lock_file(). Another process could be setting a lock on this
2475 * file at the same time, but we wouldn't remove that lock anyway.
2476 */
2477 ctx = smp_load_acquire(&inode->i_flctx);
2478 if (!ctx || list_empty(&ctx->flc_posix))
2479 return;
2480
2481 lock.fl_type = F_UNLCK;
2482 lock.fl_flags = FL_POSIX | FL_CLOSE;
2483 lock.fl_start = 0;
2484 lock.fl_end = OFFSET_MAX;
2485 lock.fl_owner = owner;
2486 lock.fl_pid = current->tgid;
2487 lock.fl_file = filp;
2488 lock.fl_ops = NULL;
2489 lock.fl_lmops = NULL;
2490
2491 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2492
2493 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2494 lock.fl_ops->fl_release_private(&lock);
2495 trace_locks_remove_posix(inode, &lock, error);
2496}
2497
2498EXPORT_SYMBOL(locks_remove_posix);
2499
2500/* The i_flctx must be valid when calling into here */
2501static void
2502locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2503{
2504 struct file_lock fl = {
2505 .fl_owner = filp,
2506 .fl_pid = current->tgid,
2507 .fl_file = filp,
2508 .fl_flags = FL_FLOCK | FL_CLOSE,
2509 .fl_type = F_UNLCK,
2510 .fl_end = OFFSET_MAX,
2511 };
2512 struct inode *inode = locks_inode(filp);
2513
2514 if (list_empty(&flctx->flc_flock))
2515 return;
2516
2517 if (filp->f_op->flock && is_remote_lock(filp))
2518 filp->f_op->flock(filp, F_SETLKW, &fl);
2519 else
2520 flock_lock_inode(inode, &fl);
2521
2522 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2523 fl.fl_ops->fl_release_private(&fl);
2524}
2525
2526/* The i_flctx must be valid when calling into here */
2527static void
2528locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2529{
2530 struct file_lock *fl, *tmp;
2531 LIST_HEAD(dispose);
2532
2533 if (list_empty(&ctx->flc_lease))
2534 return;
2535
2536 percpu_down_read_preempt_disable(&file_rwsem);
2537 spin_lock(&ctx->flc_lock);
2538 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2539 if (filp == fl->fl_file)
2540 lease_modify(fl, F_UNLCK, &dispose);
2541 spin_unlock(&ctx->flc_lock);
2542 percpu_up_read_preempt_enable(&file_rwsem);
2543
2544 locks_dispose_list(&dispose);
2545}
2546
2547/*
2548 * This function is called on the last close of an open file.
2549 */
2550void locks_remove_file(struct file *filp)
2551{
2552 struct file_lock_context *ctx;
2553
2554 ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2555 if (!ctx)
2556 return;
2557
2558 /* remove any OFD locks */
2559 locks_remove_posix(filp, filp);
2560
2561 /* remove flock locks */
2562 locks_remove_flock(filp, ctx);
2563
2564 /* remove any leases */
2565 locks_remove_lease(filp, ctx);
2566
2567 spin_lock(&ctx->flc_lock);
2568 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2569 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2570 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2571 spin_unlock(&ctx->flc_lock);
2572}
2573
2574/**
2575 * posix_unblock_lock - stop waiting for a file lock
2576 * @waiter: the lock which was waiting
2577 *
2578 * lockd needs to block waiting for locks.
2579 */
2580int
2581posix_unblock_lock(struct file_lock *waiter)
2582{
2583 int status = 0;
2584
2585 spin_lock(&blocked_lock_lock);
2586 if (waiter->fl_next)
2587 __locks_delete_block(waiter);
2588 else
2589 status = -ENOENT;
2590 spin_unlock(&blocked_lock_lock);
2591 return status;
2592}
2593EXPORT_SYMBOL(posix_unblock_lock);
2594
2595/**
2596 * vfs_cancel_lock - file byte range unblock lock
2597 * @filp: The file to apply the unblock to
2598 * @fl: The lock to be unblocked
2599 *
2600 * Used by lock managers to cancel blocked requests
2601 */
2602int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2603{
2604 if (filp->f_op->lock && is_remote_lock(filp))
2605 return filp->f_op->lock(filp, F_CANCELLK, fl);
2606 return 0;
2607}
2608
2609EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2610
2611#ifdef CONFIG_PROC_FS
2612#include <linux/proc_fs.h>
2613#include <linux/seq_file.h>
2614
2615struct locks_iterator {
2616 int li_cpu;
2617 loff_t li_pos;
2618};
2619
2620static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2621 loff_t id, char *pfx)
2622{
2623 struct inode *inode = NULL;
2624 unsigned int fl_pid;
2625 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2626
2627 fl_pid = locks_translate_pid(fl, proc_pidns);
2628 /*
2629 * If there isn't a fl_pid don't display who is waiting on
2630 * the lock if we are called from locks_show, or if we are
2631 * called from __show_fd_info - skip lock entirely
2632 */
2633 if (fl_pid == 0)
2634 return;
2635
2636 if (fl->fl_file != NULL)
2637 inode = locks_inode(fl->fl_file);
2638
2639 seq_printf(f, "%lld:%s ", id, pfx);
2640 if (IS_POSIX(fl)) {
2641 if (fl->fl_flags & FL_ACCESS)
2642 seq_puts(f, "ACCESS");
2643 else if (IS_OFDLCK(fl))
2644 seq_puts(f, "OFDLCK");
2645 else
2646 seq_puts(f, "POSIX ");
2647
2648 seq_printf(f, " %s ",
2649 (inode == NULL) ? "*NOINODE*" :
2650 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2651 } else if (IS_FLOCK(fl)) {
2652 if (fl->fl_type & LOCK_MAND) {
2653 seq_puts(f, "FLOCK MSNFS ");
2654 } else {
2655 seq_puts(f, "FLOCK ADVISORY ");
2656 }
2657 } else if (IS_LEASE(fl)) {
2658 if (fl->fl_flags & FL_DELEG)
2659 seq_puts(f, "DELEG ");
2660 else
2661 seq_puts(f, "LEASE ");
2662
2663 if (lease_breaking(fl))
2664 seq_puts(f, "BREAKING ");
2665 else if (fl->fl_file)
2666 seq_puts(f, "ACTIVE ");
2667 else
2668 seq_puts(f, "BREAKER ");
2669 } else {
2670 seq_puts(f, "UNKNOWN UNKNOWN ");
2671 }
2672 if (fl->fl_type & LOCK_MAND) {
2673 seq_printf(f, "%s ",
2674 (fl->fl_type & LOCK_READ)
2675 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2676 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2677 } else {
2678 seq_printf(f, "%s ",
2679 (lease_breaking(fl))
2680 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2681 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
2682 }
2683 if (inode) {
2684 /* userspace relies on this representation of dev_t */
2685 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
2686 MAJOR(inode->i_sb->s_dev),
2687 MINOR(inode->i_sb->s_dev), inode->i_ino);
2688 } else {
2689 seq_printf(f, "%d <none>:0 ", fl_pid);
2690 }
2691 if (IS_POSIX(fl)) {
2692 if (fl->fl_end == OFFSET_MAX)
2693 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2694 else
2695 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2696 } else {
2697 seq_puts(f, "0 EOF\n");
2698 }
2699}
2700
2701static int locks_show(struct seq_file *f, void *v)
2702{
2703 struct locks_iterator *iter = f->private;
2704 struct file_lock *fl, *bfl;
2705 struct pid_namespace *proc_pidns = file_inode(f->file)->i_sb->s_fs_info;
2706
2707 fl = hlist_entry(v, struct file_lock, fl_link);
2708
2709 if (locks_translate_pid(fl, proc_pidns) == 0)
2710 return 0;
2711
2712 lock_get_status(f, fl, iter->li_pos, "");
2713
2714 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2715 lock_get_status(f, bfl, iter->li_pos, " ->");
2716
2717 return 0;
2718}
2719
2720static void __show_fd_locks(struct seq_file *f,
2721 struct list_head *head, int *id,
2722 struct file *filp, struct files_struct *files)
2723{
2724 struct file_lock *fl;
2725
2726 list_for_each_entry(fl, head, fl_list) {
2727
2728 if (filp != fl->fl_file)
2729 continue;
2730 if (fl->fl_owner != files &&
2731 fl->fl_owner != filp)
2732 continue;
2733
2734 (*id)++;
2735 seq_puts(f, "lock:\t");
2736 lock_get_status(f, fl, *id, "");
2737 }
2738}
2739
2740void show_fd_locks(struct seq_file *f,
2741 struct file *filp, struct files_struct *files)
2742{
2743 struct inode *inode = locks_inode(filp);
2744 struct file_lock_context *ctx;
2745 int id = 0;
2746
2747 ctx = smp_load_acquire(&inode->i_flctx);
2748 if (!ctx)
2749 return;
2750
2751 spin_lock(&ctx->flc_lock);
2752 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2753 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2754 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2755 spin_unlock(&ctx->flc_lock);
2756}
2757
2758static void *locks_start(struct seq_file *f, loff_t *pos)
2759 __acquires(&blocked_lock_lock)
2760{
2761 struct locks_iterator *iter = f->private;
2762
2763 iter->li_pos = *pos + 1;
2764 percpu_down_write(&file_rwsem);
2765 spin_lock(&blocked_lock_lock);
2766 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2767}
2768
2769static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2770{
2771 struct locks_iterator *iter = f->private;
2772
2773 ++iter->li_pos;
2774 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2775}
2776
2777static void locks_stop(struct seq_file *f, void *v)
2778 __releases(&blocked_lock_lock)
2779{
2780 spin_unlock(&blocked_lock_lock);
2781 percpu_up_write(&file_rwsem);
2782}
2783
2784static const struct seq_operations locks_seq_operations = {
2785 .start = locks_start,
2786 .next = locks_next,
2787 .stop = locks_stop,
2788 .show = locks_show,
2789};
2790
2791static int locks_open(struct inode *inode, struct file *filp)
2792{
2793 return seq_open_private(filp, &locks_seq_operations,
2794 sizeof(struct locks_iterator));
2795}
2796
2797static const struct file_operations proc_locks_operations = {
2798 .open = locks_open,
2799 .read = seq_read,
2800 .llseek = seq_lseek,
2801 .release = seq_release_private,
2802};
2803
2804static int __init proc_locks_init(void)
2805{
2806 proc_create("locks", 0, NULL, &proc_locks_operations);
2807 return 0;
2808}
2809fs_initcall(proc_locks_init);
2810#endif
2811
2812static int __init filelock_init(void)
2813{
2814 int i;
2815
2816 flctx_cache = kmem_cache_create("file_lock_ctx",
2817 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2818
2819 filelock_cache = kmem_cache_create("file_lock_cache",
2820 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2821
2822
2823 for_each_possible_cpu(i) {
2824 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2825
2826 spin_lock_init(&fll->lock);
2827 INIT_HLIST_HEAD(&fll->hlist);
2828 }
2829
2830 return 0;
2831}
2832
2833core_initcall(filelock_init);