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