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#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);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/locks.c
4 *
5 * We implement four types of file locks: BSD locks, posix locks, open
6 * file description locks, and leases. For details about BSD locks,
7 * see the flock(2) man page; for details about the other three, see
8 * fcntl(2).
9 *
10 *
11 * Locking conflicts and dependencies:
12 * If multiple threads attempt to lock the same byte (or flock the same file)
13 * only one can be granted the lock, and other must wait their turn.
14 * The first lock has been "applied" or "granted", the others are "waiting"
15 * and are "blocked" by the "applied" lock..
16 *
17 * Waiting and applied locks are all kept in trees whose properties are:
18 *
19 * - the root of a tree may be an applied or waiting lock.
20 * - every other node in the tree is a waiting lock that
21 * conflicts with every ancestor of that node.
22 *
23 * Every such tree begins life as a waiting singleton which obviously
24 * satisfies the above properties.
25 *
26 * The only ways we modify trees preserve these properties:
27 *
28 * 1. We may add a new leaf node, but only after first verifying that it
29 * conflicts with all of its ancestors.
30 * 2. We may remove the root of a tree, creating a new singleton
31 * tree from the root and N new trees rooted in the immediate
32 * children.
33 * 3. If the root of a tree is not currently an applied lock, we may
34 * apply it (if possible).
35 * 4. We may upgrade the root of the tree (either extend its range,
36 * or upgrade its entire range from read to write).
37 *
38 * When an applied lock is modified in a way that reduces or downgrades any
39 * part of its range, we remove all its children (2 above). This particularly
40 * happens when a lock is unlocked.
41 *
42 * For each of those child trees we "wake up" the thread which is
43 * waiting for the lock so it can continue handling as follows: if the
44 * root of the tree applies, we do so (3). If it doesn't, it must
45 * conflict with some applied lock. We remove (wake up) all of its children
46 * (2), and add it is a new leaf to the tree rooted in the applied
47 * lock (1). We then repeat the process recursively with those
48 * children.
49 *
50 */
51
52#include <linux/capability.h>
53#include <linux/file.h>
54#include <linux/fdtable.h>
55#include <linux/fs.h>
56#include <linux/init.h>
57#include <linux/security.h>
58#include <linux/slab.h>
59#include <linux/syscalls.h>
60#include <linux/time.h>
61#include <linux/rcupdate.h>
62#include <linux/pid_namespace.h>
63#include <linux/hashtable.h>
64#include <linux/percpu.h>
65#include <linux/sysctl.h>
66
67#define CREATE_TRACE_POINTS
68#include <trace/events/filelock.h>
69
70#include <linux/uaccess.h>
71
72#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
73#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
74#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
75#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
76#define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
77
78static bool lease_breaking(struct file_lock *fl)
79{
80 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
81}
82
83static int target_leasetype(struct file_lock *fl)
84{
85 if (fl->fl_flags & FL_UNLOCK_PENDING)
86 return F_UNLCK;
87 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
88 return F_RDLCK;
89 return fl->fl_type;
90}
91
92static int leases_enable = 1;
93static int lease_break_time = 45;
94
95#ifdef CONFIG_SYSCTL
96static struct ctl_table locks_sysctls[] = {
97 {
98 .procname = "leases-enable",
99 .data = &leases_enable,
100 .maxlen = sizeof(int),
101 .mode = 0644,
102 .proc_handler = proc_dointvec,
103 },
104#ifdef CONFIG_MMU
105 {
106 .procname = "lease-break-time",
107 .data = &lease_break_time,
108 .maxlen = sizeof(int),
109 .mode = 0644,
110 .proc_handler = proc_dointvec,
111 },
112#endif /* CONFIG_MMU */
113 {}
114};
115
116static int __init init_fs_locks_sysctls(void)
117{
118 register_sysctl_init("fs", locks_sysctls);
119 return 0;
120}
121early_initcall(init_fs_locks_sysctls);
122#endif /* CONFIG_SYSCTL */
123
124/*
125 * The global file_lock_list is only used for displaying /proc/locks, so we
126 * keep a list on each CPU, with each list protected by its own spinlock.
127 * Global serialization is done using file_rwsem.
128 *
129 * Note that alterations to the list also require that the relevant flc_lock is
130 * held.
131 */
132struct file_lock_list_struct {
133 spinlock_t lock;
134 struct hlist_head hlist;
135};
136static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
137DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
138
139
140/*
141 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
142 * It is protected by blocked_lock_lock.
143 *
144 * We hash locks by lockowner in order to optimize searching for the lock a
145 * particular lockowner is waiting on.
146 *
147 * FIXME: make this value scale via some heuristic? We generally will want more
148 * buckets when we have more lockowners holding locks, but that's a little
149 * difficult to determine without knowing what the workload will look like.
150 */
151#define BLOCKED_HASH_BITS 7
152static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
153
154/*
155 * This lock protects the blocked_hash. Generally, if you're accessing it, you
156 * want to be holding this lock.
157 *
158 * In addition, it also protects the fl->fl_blocked_requests list, and the
159 * fl->fl_blocker pointer for file_lock structures that are acting as lock
160 * requests (in contrast to those that are acting as records of acquired locks).
161 *
162 * Note that when we acquire this lock in order to change the above fields,
163 * we often hold the flc_lock as well. In certain cases, when reading the fields
164 * protected by this lock, we can skip acquiring it iff we already hold the
165 * flc_lock.
166 */
167static DEFINE_SPINLOCK(blocked_lock_lock);
168
169static struct kmem_cache *flctx_cache __read_mostly;
170static struct kmem_cache *filelock_cache __read_mostly;
171
172static struct file_lock_context *
173locks_get_lock_context(struct inode *inode, int type)
174{
175 struct file_lock_context *ctx;
176
177 /* paired with cmpxchg() below */
178 ctx = locks_inode_context(inode);
179 if (likely(ctx) || type == F_UNLCK)
180 goto out;
181
182 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
183 if (!ctx)
184 goto out;
185
186 spin_lock_init(&ctx->flc_lock);
187 INIT_LIST_HEAD(&ctx->flc_flock);
188 INIT_LIST_HEAD(&ctx->flc_posix);
189 INIT_LIST_HEAD(&ctx->flc_lease);
190
191 /*
192 * Assign the pointer if it's not already assigned. If it is, then
193 * free the context we just allocated.
194 */
195 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
196 kmem_cache_free(flctx_cache, ctx);
197 ctx = locks_inode_context(inode);
198 }
199out:
200 trace_locks_get_lock_context(inode, type, ctx);
201 return ctx;
202}
203
204static void
205locks_dump_ctx_list(struct list_head *list, char *list_type)
206{
207 struct file_lock *fl;
208
209 list_for_each_entry(fl, list, fl_list) {
210 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);
211 }
212}
213
214static void
215locks_check_ctx_lists(struct inode *inode)
216{
217 struct file_lock_context *ctx = inode->i_flctx;
218
219 if (unlikely(!list_empty(&ctx->flc_flock) ||
220 !list_empty(&ctx->flc_posix) ||
221 !list_empty(&ctx->flc_lease))) {
222 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
223 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
224 inode->i_ino);
225 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
226 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
227 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
228 }
229}
230
231static void
232locks_check_ctx_file_list(struct file *filp, struct list_head *list,
233 char *list_type)
234{
235 struct file_lock *fl;
236 struct inode *inode = locks_inode(filp);
237
238 list_for_each_entry(fl, list, fl_list)
239 if (fl->fl_file == filp)
240 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
241 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
242 list_type, MAJOR(inode->i_sb->s_dev),
243 MINOR(inode->i_sb->s_dev), inode->i_ino,
244 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
245}
246
247void
248locks_free_lock_context(struct inode *inode)
249{
250 struct file_lock_context *ctx = locks_inode_context(inode);
251
252 if (unlikely(ctx)) {
253 locks_check_ctx_lists(inode);
254 kmem_cache_free(flctx_cache, ctx);
255 }
256}
257
258static void locks_init_lock_heads(struct file_lock *fl)
259{
260 INIT_HLIST_NODE(&fl->fl_link);
261 INIT_LIST_HEAD(&fl->fl_list);
262 INIT_LIST_HEAD(&fl->fl_blocked_requests);
263 INIT_LIST_HEAD(&fl->fl_blocked_member);
264 init_waitqueue_head(&fl->fl_wait);
265}
266
267/* Allocate an empty lock structure. */
268struct file_lock *locks_alloc_lock(void)
269{
270 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
271
272 if (fl)
273 locks_init_lock_heads(fl);
274
275 return fl;
276}
277EXPORT_SYMBOL_GPL(locks_alloc_lock);
278
279void locks_release_private(struct file_lock *fl)
280{
281 BUG_ON(waitqueue_active(&fl->fl_wait));
282 BUG_ON(!list_empty(&fl->fl_list));
283 BUG_ON(!list_empty(&fl->fl_blocked_requests));
284 BUG_ON(!list_empty(&fl->fl_blocked_member));
285 BUG_ON(!hlist_unhashed(&fl->fl_link));
286
287 if (fl->fl_ops) {
288 if (fl->fl_ops->fl_release_private)
289 fl->fl_ops->fl_release_private(fl);
290 fl->fl_ops = NULL;
291 }
292
293 if (fl->fl_lmops) {
294 if (fl->fl_lmops->lm_put_owner) {
295 fl->fl_lmops->lm_put_owner(fl->fl_owner);
296 fl->fl_owner = NULL;
297 }
298 fl->fl_lmops = NULL;
299 }
300}
301EXPORT_SYMBOL_GPL(locks_release_private);
302
303/**
304 * locks_owner_has_blockers - Check for blocking lock requests
305 * @flctx: file lock context
306 * @owner: lock owner
307 *
308 * Return values:
309 * %true: @owner has at least one blocker
310 * %false: @owner has no blockers
311 */
312bool locks_owner_has_blockers(struct file_lock_context *flctx,
313 fl_owner_t owner)
314{
315 struct file_lock *fl;
316
317 spin_lock(&flctx->flc_lock);
318 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
319 if (fl->fl_owner != owner)
320 continue;
321 if (!list_empty(&fl->fl_blocked_requests)) {
322 spin_unlock(&flctx->flc_lock);
323 return true;
324 }
325 }
326 spin_unlock(&flctx->flc_lock);
327 return false;
328}
329EXPORT_SYMBOL_GPL(locks_owner_has_blockers);
330
331/* Free a lock which is not in use. */
332void locks_free_lock(struct file_lock *fl)
333{
334 locks_release_private(fl);
335 kmem_cache_free(filelock_cache, fl);
336}
337EXPORT_SYMBOL(locks_free_lock);
338
339static void
340locks_dispose_list(struct list_head *dispose)
341{
342 struct file_lock *fl;
343
344 while (!list_empty(dispose)) {
345 fl = list_first_entry(dispose, struct file_lock, fl_list);
346 list_del_init(&fl->fl_list);
347 locks_free_lock(fl);
348 }
349}
350
351void locks_init_lock(struct file_lock *fl)
352{
353 memset(fl, 0, sizeof(struct file_lock));
354 locks_init_lock_heads(fl);
355}
356EXPORT_SYMBOL(locks_init_lock);
357
358/*
359 * Initialize a new lock from an existing file_lock structure.
360 */
361void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
362{
363 new->fl_owner = fl->fl_owner;
364 new->fl_pid = fl->fl_pid;
365 new->fl_file = NULL;
366 new->fl_flags = fl->fl_flags;
367 new->fl_type = fl->fl_type;
368 new->fl_start = fl->fl_start;
369 new->fl_end = fl->fl_end;
370 new->fl_lmops = fl->fl_lmops;
371 new->fl_ops = NULL;
372
373 if (fl->fl_lmops) {
374 if (fl->fl_lmops->lm_get_owner)
375 fl->fl_lmops->lm_get_owner(fl->fl_owner);
376 }
377}
378EXPORT_SYMBOL(locks_copy_conflock);
379
380void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
381{
382 /* "new" must be a freshly-initialized lock */
383 WARN_ON_ONCE(new->fl_ops);
384
385 locks_copy_conflock(new, fl);
386
387 new->fl_file = fl->fl_file;
388 new->fl_ops = fl->fl_ops;
389
390 if (fl->fl_ops) {
391 if (fl->fl_ops->fl_copy_lock)
392 fl->fl_ops->fl_copy_lock(new, fl);
393 }
394}
395EXPORT_SYMBOL(locks_copy_lock);
396
397static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
398{
399 struct file_lock *f;
400
401 /*
402 * As ctx->flc_lock is held, new requests cannot be added to
403 * ->fl_blocked_requests, so we don't need a lock to check if it
404 * is empty.
405 */
406 if (list_empty(&fl->fl_blocked_requests))
407 return;
408 spin_lock(&blocked_lock_lock);
409 list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
410 list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
411 f->fl_blocker = new;
412 spin_unlock(&blocked_lock_lock);
413}
414
415static inline int flock_translate_cmd(int cmd) {
416 switch (cmd) {
417 case LOCK_SH:
418 return F_RDLCK;
419 case LOCK_EX:
420 return F_WRLCK;
421 case LOCK_UN:
422 return F_UNLCK;
423 }
424 return -EINVAL;
425}
426
427/* Fill in a file_lock structure with an appropriate FLOCK lock. */
428static void flock_make_lock(struct file *filp, struct file_lock *fl, int type)
429{
430 locks_init_lock(fl);
431
432 fl->fl_file = filp;
433 fl->fl_owner = filp;
434 fl->fl_pid = current->tgid;
435 fl->fl_flags = FL_FLOCK;
436 fl->fl_type = type;
437 fl->fl_end = OFFSET_MAX;
438}
439
440static int assign_type(struct file_lock *fl, long type)
441{
442 switch (type) {
443 case F_RDLCK:
444 case F_WRLCK:
445 case F_UNLCK:
446 fl->fl_type = type;
447 break;
448 default:
449 return -EINVAL;
450 }
451 return 0;
452}
453
454static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
455 struct flock64 *l)
456{
457 switch (l->l_whence) {
458 case SEEK_SET:
459 fl->fl_start = 0;
460 break;
461 case SEEK_CUR:
462 fl->fl_start = filp->f_pos;
463 break;
464 case SEEK_END:
465 fl->fl_start = i_size_read(file_inode(filp));
466 break;
467 default:
468 return -EINVAL;
469 }
470 if (l->l_start > OFFSET_MAX - fl->fl_start)
471 return -EOVERFLOW;
472 fl->fl_start += l->l_start;
473 if (fl->fl_start < 0)
474 return -EINVAL;
475
476 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
477 POSIX-2001 defines it. */
478 if (l->l_len > 0) {
479 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
480 return -EOVERFLOW;
481 fl->fl_end = fl->fl_start + (l->l_len - 1);
482
483 } else if (l->l_len < 0) {
484 if (fl->fl_start + l->l_len < 0)
485 return -EINVAL;
486 fl->fl_end = fl->fl_start - 1;
487 fl->fl_start += l->l_len;
488 } else
489 fl->fl_end = OFFSET_MAX;
490
491 fl->fl_owner = current->files;
492 fl->fl_pid = current->tgid;
493 fl->fl_file = filp;
494 fl->fl_flags = FL_POSIX;
495 fl->fl_ops = NULL;
496 fl->fl_lmops = NULL;
497
498 return assign_type(fl, l->l_type);
499}
500
501/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
502 * style lock.
503 */
504static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
505 struct flock *l)
506{
507 struct flock64 ll = {
508 .l_type = l->l_type,
509 .l_whence = l->l_whence,
510 .l_start = l->l_start,
511 .l_len = l->l_len,
512 };
513
514 return flock64_to_posix_lock(filp, fl, &ll);
515}
516
517/* default lease lock manager operations */
518static bool
519lease_break_callback(struct file_lock *fl)
520{
521 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
522 return false;
523}
524
525static void
526lease_setup(struct file_lock *fl, void **priv)
527{
528 struct file *filp = fl->fl_file;
529 struct fasync_struct *fa = *priv;
530
531 /*
532 * fasync_insert_entry() returns the old entry if any. If there was no
533 * old entry, then it used "priv" and inserted it into the fasync list.
534 * Clear the pointer to indicate that it shouldn't be freed.
535 */
536 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
537 *priv = NULL;
538
539 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
540}
541
542static const struct lock_manager_operations lease_manager_ops = {
543 .lm_break = lease_break_callback,
544 .lm_change = lease_modify,
545 .lm_setup = lease_setup,
546};
547
548/*
549 * Initialize a lease, use the default lock manager operations
550 */
551static int lease_init(struct file *filp, long type, struct file_lock *fl)
552{
553 if (assign_type(fl, type) != 0)
554 return -EINVAL;
555
556 fl->fl_owner = filp;
557 fl->fl_pid = current->tgid;
558
559 fl->fl_file = filp;
560 fl->fl_flags = FL_LEASE;
561 fl->fl_start = 0;
562 fl->fl_end = OFFSET_MAX;
563 fl->fl_ops = NULL;
564 fl->fl_lmops = &lease_manager_ops;
565 return 0;
566}
567
568/* Allocate a file_lock initialised to this type of lease */
569static struct file_lock *lease_alloc(struct file *filp, long type)
570{
571 struct file_lock *fl = locks_alloc_lock();
572 int error = -ENOMEM;
573
574 if (fl == NULL)
575 return ERR_PTR(error);
576
577 error = lease_init(filp, type, fl);
578 if (error) {
579 locks_free_lock(fl);
580 return ERR_PTR(error);
581 }
582 return fl;
583}
584
585/* Check if two locks overlap each other.
586 */
587static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
588{
589 return ((fl1->fl_end >= fl2->fl_start) &&
590 (fl2->fl_end >= fl1->fl_start));
591}
592
593/*
594 * Check whether two locks have the same owner.
595 */
596static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
597{
598 return fl1->fl_owner == fl2->fl_owner;
599}
600
601/* Must be called with the flc_lock held! */
602static void locks_insert_global_locks(struct file_lock *fl)
603{
604 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
605
606 percpu_rwsem_assert_held(&file_rwsem);
607
608 spin_lock(&fll->lock);
609 fl->fl_link_cpu = smp_processor_id();
610 hlist_add_head(&fl->fl_link, &fll->hlist);
611 spin_unlock(&fll->lock);
612}
613
614/* Must be called with the flc_lock held! */
615static void locks_delete_global_locks(struct file_lock *fl)
616{
617 struct file_lock_list_struct *fll;
618
619 percpu_rwsem_assert_held(&file_rwsem);
620
621 /*
622 * Avoid taking lock if already unhashed. This is safe since this check
623 * is done while holding the flc_lock, and new insertions into the list
624 * also require that it be held.
625 */
626 if (hlist_unhashed(&fl->fl_link))
627 return;
628
629 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
630 spin_lock(&fll->lock);
631 hlist_del_init(&fl->fl_link);
632 spin_unlock(&fll->lock);
633}
634
635static unsigned long
636posix_owner_key(struct file_lock *fl)
637{
638 return (unsigned long)fl->fl_owner;
639}
640
641static void locks_insert_global_blocked(struct file_lock *waiter)
642{
643 lockdep_assert_held(&blocked_lock_lock);
644
645 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
646}
647
648static void locks_delete_global_blocked(struct file_lock *waiter)
649{
650 lockdep_assert_held(&blocked_lock_lock);
651
652 hash_del(&waiter->fl_link);
653}
654
655/* Remove waiter from blocker's block list.
656 * When blocker ends up pointing to itself then the list is empty.
657 *
658 * Must be called with blocked_lock_lock held.
659 */
660static void __locks_delete_block(struct file_lock *waiter)
661{
662 locks_delete_global_blocked(waiter);
663 list_del_init(&waiter->fl_blocked_member);
664}
665
666static void __locks_wake_up_blocks(struct file_lock *blocker)
667{
668 while (!list_empty(&blocker->fl_blocked_requests)) {
669 struct file_lock *waiter;
670
671 waiter = list_first_entry(&blocker->fl_blocked_requests,
672 struct file_lock, fl_blocked_member);
673 __locks_delete_block(waiter);
674 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
675 waiter->fl_lmops->lm_notify(waiter);
676 else
677 wake_up(&waiter->fl_wait);
678
679 /*
680 * The setting of fl_blocker to NULL marks the "done"
681 * point in deleting a block. Paired with acquire at the top
682 * of locks_delete_block().
683 */
684 smp_store_release(&waiter->fl_blocker, NULL);
685 }
686}
687
688/**
689 * locks_delete_block - stop waiting for a file lock
690 * @waiter: the lock which was waiting
691 *
692 * lockd/nfsd need to disconnect the lock while working on it.
693 */
694int locks_delete_block(struct file_lock *waiter)
695{
696 int status = -ENOENT;
697
698 /*
699 * If fl_blocker is NULL, it won't be set again as this thread "owns"
700 * the lock and is the only one that might try to claim the lock.
701 *
702 * We use acquire/release to manage fl_blocker so that we can
703 * optimize away taking the blocked_lock_lock in many cases.
704 *
705 * The smp_load_acquire guarantees two things:
706 *
707 * 1/ that fl_blocked_requests can be tested locklessly. If something
708 * was recently added to that list it must have been in a locked region
709 * *before* the locked region when fl_blocker was set to NULL.
710 *
711 * 2/ that no other thread is accessing 'waiter', so it is safe to free
712 * it. __locks_wake_up_blocks is careful not to touch waiter after
713 * fl_blocker is released.
714 *
715 * If a lockless check of fl_blocker shows it to be NULL, we know that
716 * no new locks can be inserted into its fl_blocked_requests list, and
717 * can avoid doing anything further if the list is empty.
718 */
719 if (!smp_load_acquire(&waiter->fl_blocker) &&
720 list_empty(&waiter->fl_blocked_requests))
721 return status;
722
723 spin_lock(&blocked_lock_lock);
724 if (waiter->fl_blocker)
725 status = 0;
726 __locks_wake_up_blocks(waiter);
727 __locks_delete_block(waiter);
728
729 /*
730 * The setting of fl_blocker to NULL marks the "done" point in deleting
731 * a block. Paired with acquire at the top of this function.
732 */
733 smp_store_release(&waiter->fl_blocker, NULL);
734 spin_unlock(&blocked_lock_lock);
735 return status;
736}
737EXPORT_SYMBOL(locks_delete_block);
738
739/* Insert waiter into blocker's block list.
740 * We use a circular list so that processes can be easily woken up in
741 * the order they blocked. The documentation doesn't require this but
742 * it seems like the reasonable thing to do.
743 *
744 * Must be called with both the flc_lock and blocked_lock_lock held. The
745 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
746 * but by ensuring that the flc_lock is also held on insertions we can avoid
747 * taking the blocked_lock_lock in some cases when we see that the
748 * fl_blocked_requests list is empty.
749 *
750 * Rather than just adding to the list, we check for conflicts with any existing
751 * waiters, and add beneath any waiter that blocks the new waiter.
752 * Thus wakeups don't happen until needed.
753 */
754static void __locks_insert_block(struct file_lock *blocker,
755 struct file_lock *waiter,
756 bool conflict(struct file_lock *,
757 struct file_lock *))
758{
759 struct file_lock *fl;
760 BUG_ON(!list_empty(&waiter->fl_blocked_member));
761
762new_blocker:
763 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
764 if (conflict(fl, waiter)) {
765 blocker = fl;
766 goto new_blocker;
767 }
768 waiter->fl_blocker = blocker;
769 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
770 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
771 locks_insert_global_blocked(waiter);
772
773 /* The requests in waiter->fl_blocked are known to conflict with
774 * waiter, but might not conflict with blocker, or the requests
775 * and lock which block it. So they all need to be woken.
776 */
777 __locks_wake_up_blocks(waiter);
778}
779
780/* Must be called with flc_lock held. */
781static void locks_insert_block(struct file_lock *blocker,
782 struct file_lock *waiter,
783 bool conflict(struct file_lock *,
784 struct file_lock *))
785{
786 spin_lock(&blocked_lock_lock);
787 __locks_insert_block(blocker, waiter, conflict);
788 spin_unlock(&blocked_lock_lock);
789}
790
791/*
792 * Wake up processes blocked waiting for blocker.
793 *
794 * Must be called with the inode->flc_lock held!
795 */
796static void locks_wake_up_blocks(struct file_lock *blocker)
797{
798 /*
799 * Avoid taking global lock if list is empty. This is safe since new
800 * blocked requests are only added to the list under the flc_lock, and
801 * the flc_lock is always held here. Note that removal from the
802 * fl_blocked_requests list does not require the flc_lock, so we must
803 * recheck list_empty() after acquiring the blocked_lock_lock.
804 */
805 if (list_empty(&blocker->fl_blocked_requests))
806 return;
807
808 spin_lock(&blocked_lock_lock);
809 __locks_wake_up_blocks(blocker);
810 spin_unlock(&blocked_lock_lock);
811}
812
813static void
814locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
815{
816 list_add_tail(&fl->fl_list, before);
817 locks_insert_global_locks(fl);
818}
819
820static void
821locks_unlink_lock_ctx(struct file_lock *fl)
822{
823 locks_delete_global_locks(fl);
824 list_del_init(&fl->fl_list);
825 locks_wake_up_blocks(fl);
826}
827
828static void
829locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
830{
831 locks_unlink_lock_ctx(fl);
832 if (dispose)
833 list_add(&fl->fl_list, dispose);
834 else
835 locks_free_lock(fl);
836}
837
838/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
839 * checks for shared/exclusive status of overlapping locks.
840 */
841static bool locks_conflict(struct file_lock *caller_fl,
842 struct file_lock *sys_fl)
843{
844 if (sys_fl->fl_type == F_WRLCK)
845 return true;
846 if (caller_fl->fl_type == F_WRLCK)
847 return true;
848 return false;
849}
850
851/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
852 * checking before calling the locks_conflict().
853 */
854static bool posix_locks_conflict(struct file_lock *caller_fl,
855 struct file_lock *sys_fl)
856{
857 /* POSIX locks owned by the same process do not conflict with
858 * each other.
859 */
860 if (posix_same_owner(caller_fl, sys_fl))
861 return false;
862
863 /* Check whether they overlap */
864 if (!locks_overlap(caller_fl, sys_fl))
865 return false;
866
867 return locks_conflict(caller_fl, sys_fl);
868}
869
870/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
871 * checking before calling the locks_conflict().
872 */
873static bool flock_locks_conflict(struct file_lock *caller_fl,
874 struct file_lock *sys_fl)
875{
876 /* FLOCK locks referring to the same filp do not conflict with
877 * each other.
878 */
879 if (caller_fl->fl_file == sys_fl->fl_file)
880 return false;
881
882 return locks_conflict(caller_fl, sys_fl);
883}
884
885void
886posix_test_lock(struct file *filp, struct file_lock *fl)
887{
888 struct file_lock *cfl;
889 struct file_lock_context *ctx;
890 struct inode *inode = locks_inode(filp);
891 void *owner;
892 void (*func)(void);
893
894 ctx = locks_inode_context(inode);
895 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
896 fl->fl_type = F_UNLCK;
897 return;
898 }
899
900retry:
901 spin_lock(&ctx->flc_lock);
902 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
903 if (!posix_locks_conflict(fl, cfl))
904 continue;
905 if (cfl->fl_lmops && cfl->fl_lmops->lm_lock_expirable
906 && (*cfl->fl_lmops->lm_lock_expirable)(cfl)) {
907 owner = cfl->fl_lmops->lm_mod_owner;
908 func = cfl->fl_lmops->lm_expire_lock;
909 __module_get(owner);
910 spin_unlock(&ctx->flc_lock);
911 (*func)();
912 module_put(owner);
913 goto retry;
914 }
915 locks_copy_conflock(fl, cfl);
916 goto out;
917 }
918 fl->fl_type = F_UNLCK;
919out:
920 spin_unlock(&ctx->flc_lock);
921 return;
922}
923EXPORT_SYMBOL(posix_test_lock);
924
925/*
926 * Deadlock detection:
927 *
928 * We attempt to detect deadlocks that are due purely to posix file
929 * locks.
930 *
931 * We assume that a task can be waiting for at most one lock at a time.
932 * So for any acquired lock, the process holding that lock may be
933 * waiting on at most one other lock. That lock in turns may be held by
934 * someone waiting for at most one other lock. Given a requested lock
935 * caller_fl which is about to wait for a conflicting lock block_fl, we
936 * follow this chain of waiters to ensure we are not about to create a
937 * cycle.
938 *
939 * Since we do this before we ever put a process to sleep on a lock, we
940 * are ensured that there is never a cycle; that is what guarantees that
941 * the while() loop in posix_locks_deadlock() eventually completes.
942 *
943 * Note: the above assumption may not be true when handling lock
944 * requests from a broken NFS client. It may also fail in the presence
945 * of tasks (such as posix threads) sharing the same open file table.
946 * To handle those cases, we just bail out after a few iterations.
947 *
948 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
949 * Because the owner is not even nominally tied to a thread of
950 * execution, the deadlock detection below can't reasonably work well. Just
951 * skip it for those.
952 *
953 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
954 * locks that just checks for the case where two tasks are attempting to
955 * upgrade from read to write locks on the same inode.
956 */
957
958#define MAX_DEADLK_ITERATIONS 10
959
960/* Find a lock that the owner of the given block_fl is blocking on. */
961static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
962{
963 struct file_lock *fl;
964
965 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
966 if (posix_same_owner(fl, block_fl)) {
967 while (fl->fl_blocker)
968 fl = fl->fl_blocker;
969 return fl;
970 }
971 }
972 return NULL;
973}
974
975/* Must be called with the blocked_lock_lock held! */
976static int posix_locks_deadlock(struct file_lock *caller_fl,
977 struct file_lock *block_fl)
978{
979 int i = 0;
980
981 lockdep_assert_held(&blocked_lock_lock);
982
983 /*
984 * This deadlock detector can't reasonably detect deadlocks with
985 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
986 */
987 if (IS_OFDLCK(caller_fl))
988 return 0;
989
990 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
991 if (i++ > MAX_DEADLK_ITERATIONS)
992 return 0;
993 if (posix_same_owner(caller_fl, block_fl))
994 return 1;
995 }
996 return 0;
997}
998
999/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1000 * after any leases, but before any posix locks.
1001 *
1002 * Note that if called with an FL_EXISTS argument, the caller may determine
1003 * whether or not a lock was successfully freed by testing the return
1004 * value for -ENOENT.
1005 */
1006static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1007{
1008 struct file_lock *new_fl = NULL;
1009 struct file_lock *fl;
1010 struct file_lock_context *ctx;
1011 int error = 0;
1012 bool found = false;
1013 LIST_HEAD(dispose);
1014
1015 ctx = locks_get_lock_context(inode, request->fl_type);
1016 if (!ctx) {
1017 if (request->fl_type != F_UNLCK)
1018 return -ENOMEM;
1019 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1020 }
1021
1022 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1023 new_fl = locks_alloc_lock();
1024 if (!new_fl)
1025 return -ENOMEM;
1026 }
1027
1028 percpu_down_read(&file_rwsem);
1029 spin_lock(&ctx->flc_lock);
1030 if (request->fl_flags & FL_ACCESS)
1031 goto find_conflict;
1032
1033 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1034 if (request->fl_file != fl->fl_file)
1035 continue;
1036 if (request->fl_type == fl->fl_type)
1037 goto out;
1038 found = true;
1039 locks_delete_lock_ctx(fl, &dispose);
1040 break;
1041 }
1042
1043 if (request->fl_type == F_UNLCK) {
1044 if ((request->fl_flags & FL_EXISTS) && !found)
1045 error = -ENOENT;
1046 goto out;
1047 }
1048
1049find_conflict:
1050 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1051 if (!flock_locks_conflict(request, fl))
1052 continue;
1053 error = -EAGAIN;
1054 if (!(request->fl_flags & FL_SLEEP))
1055 goto out;
1056 error = FILE_LOCK_DEFERRED;
1057 locks_insert_block(fl, request, flock_locks_conflict);
1058 goto out;
1059 }
1060 if (request->fl_flags & FL_ACCESS)
1061 goto out;
1062 locks_copy_lock(new_fl, request);
1063 locks_move_blocks(new_fl, request);
1064 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1065 new_fl = NULL;
1066 error = 0;
1067
1068out:
1069 spin_unlock(&ctx->flc_lock);
1070 percpu_up_read(&file_rwsem);
1071 if (new_fl)
1072 locks_free_lock(new_fl);
1073 locks_dispose_list(&dispose);
1074 trace_flock_lock_inode(inode, request, error);
1075 return error;
1076}
1077
1078static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1079 struct file_lock *conflock)
1080{
1081 struct file_lock *fl, *tmp;
1082 struct file_lock *new_fl = NULL;
1083 struct file_lock *new_fl2 = NULL;
1084 struct file_lock *left = NULL;
1085 struct file_lock *right = NULL;
1086 struct file_lock_context *ctx;
1087 int error;
1088 bool added = false;
1089 LIST_HEAD(dispose);
1090 void *owner;
1091 void (*func)(void);
1092
1093 ctx = locks_get_lock_context(inode, request->fl_type);
1094 if (!ctx)
1095 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1096
1097 /*
1098 * We may need two file_lock structures for this operation,
1099 * so we get them in advance to avoid races.
1100 *
1101 * In some cases we can be sure, that no new locks will be needed
1102 */
1103 if (!(request->fl_flags & FL_ACCESS) &&
1104 (request->fl_type != F_UNLCK ||
1105 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1106 new_fl = locks_alloc_lock();
1107 new_fl2 = locks_alloc_lock();
1108 }
1109
1110retry:
1111 percpu_down_read(&file_rwsem);
1112 spin_lock(&ctx->flc_lock);
1113 /*
1114 * New lock request. Walk all POSIX locks and look for conflicts. If
1115 * there are any, either return error or put the request on the
1116 * blocker's list of waiters and the global blocked_hash.
1117 */
1118 if (request->fl_type != F_UNLCK) {
1119 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1120 if (!posix_locks_conflict(request, fl))
1121 continue;
1122 if (fl->fl_lmops && fl->fl_lmops->lm_lock_expirable
1123 && (*fl->fl_lmops->lm_lock_expirable)(fl)) {
1124 owner = fl->fl_lmops->lm_mod_owner;
1125 func = fl->fl_lmops->lm_expire_lock;
1126 __module_get(owner);
1127 spin_unlock(&ctx->flc_lock);
1128 percpu_up_read(&file_rwsem);
1129 (*func)();
1130 module_put(owner);
1131 goto retry;
1132 }
1133 if (conflock)
1134 locks_copy_conflock(conflock, fl);
1135 error = -EAGAIN;
1136 if (!(request->fl_flags & FL_SLEEP))
1137 goto out;
1138 /*
1139 * Deadlock detection and insertion into the blocked
1140 * locks list must be done while holding the same lock!
1141 */
1142 error = -EDEADLK;
1143 spin_lock(&blocked_lock_lock);
1144 /*
1145 * Ensure that we don't find any locks blocked on this
1146 * request during deadlock detection.
1147 */
1148 __locks_wake_up_blocks(request);
1149 if (likely(!posix_locks_deadlock(request, fl))) {
1150 error = FILE_LOCK_DEFERRED;
1151 __locks_insert_block(fl, request,
1152 posix_locks_conflict);
1153 }
1154 spin_unlock(&blocked_lock_lock);
1155 goto out;
1156 }
1157 }
1158
1159 /* If we're just looking for a conflict, we're done. */
1160 error = 0;
1161 if (request->fl_flags & FL_ACCESS)
1162 goto out;
1163
1164 /* Find the first old lock with the same owner as the new lock */
1165 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1166 if (posix_same_owner(request, fl))
1167 break;
1168 }
1169
1170 /* Process locks with this owner. */
1171 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1172 if (!posix_same_owner(request, fl))
1173 break;
1174
1175 /* Detect adjacent or overlapping regions (if same lock type) */
1176 if (request->fl_type == fl->fl_type) {
1177 /* In all comparisons of start vs end, use
1178 * "start - 1" rather than "end + 1". If end
1179 * is OFFSET_MAX, end + 1 will become negative.
1180 */
1181 if (fl->fl_end < request->fl_start - 1)
1182 continue;
1183 /* If the next lock in the list has entirely bigger
1184 * addresses than the new one, insert the lock here.
1185 */
1186 if (fl->fl_start - 1 > request->fl_end)
1187 break;
1188
1189 /* If we come here, the new and old lock are of the
1190 * same type and adjacent or overlapping. Make one
1191 * lock yielding from the lower start address of both
1192 * locks to the higher end address.
1193 */
1194 if (fl->fl_start > request->fl_start)
1195 fl->fl_start = request->fl_start;
1196 else
1197 request->fl_start = fl->fl_start;
1198 if (fl->fl_end < request->fl_end)
1199 fl->fl_end = request->fl_end;
1200 else
1201 request->fl_end = fl->fl_end;
1202 if (added) {
1203 locks_delete_lock_ctx(fl, &dispose);
1204 continue;
1205 }
1206 request = fl;
1207 added = true;
1208 } else {
1209 /* Processing for different lock types is a bit
1210 * more complex.
1211 */
1212 if (fl->fl_end < request->fl_start)
1213 continue;
1214 if (fl->fl_start > request->fl_end)
1215 break;
1216 if (request->fl_type == F_UNLCK)
1217 added = true;
1218 if (fl->fl_start < request->fl_start)
1219 left = fl;
1220 /* If the next lock in the list has a higher end
1221 * address than the new one, insert the new one here.
1222 */
1223 if (fl->fl_end > request->fl_end) {
1224 right = fl;
1225 break;
1226 }
1227 if (fl->fl_start >= request->fl_start) {
1228 /* The new lock completely replaces an old
1229 * one (This may happen several times).
1230 */
1231 if (added) {
1232 locks_delete_lock_ctx(fl, &dispose);
1233 continue;
1234 }
1235 /*
1236 * Replace the old lock with new_fl, and
1237 * remove the old one. It's safe to do the
1238 * insert here since we know that we won't be
1239 * using new_fl later, and that the lock is
1240 * just replacing an existing lock.
1241 */
1242 error = -ENOLCK;
1243 if (!new_fl)
1244 goto out;
1245 locks_copy_lock(new_fl, request);
1246 locks_move_blocks(new_fl, request);
1247 request = new_fl;
1248 new_fl = NULL;
1249 locks_insert_lock_ctx(request, &fl->fl_list);
1250 locks_delete_lock_ctx(fl, &dispose);
1251 added = true;
1252 }
1253 }
1254 }
1255
1256 /*
1257 * The above code only modifies existing locks in case of merging or
1258 * replacing. If new lock(s) need to be inserted all modifications are
1259 * done below this, so it's safe yet to bail out.
1260 */
1261 error = -ENOLCK; /* "no luck" */
1262 if (right && left == right && !new_fl2)
1263 goto out;
1264
1265 error = 0;
1266 if (!added) {
1267 if (request->fl_type == F_UNLCK) {
1268 if (request->fl_flags & FL_EXISTS)
1269 error = -ENOENT;
1270 goto out;
1271 }
1272
1273 if (!new_fl) {
1274 error = -ENOLCK;
1275 goto out;
1276 }
1277 locks_copy_lock(new_fl, request);
1278 locks_move_blocks(new_fl, request);
1279 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1280 fl = new_fl;
1281 new_fl = NULL;
1282 }
1283 if (right) {
1284 if (left == right) {
1285 /* The new lock breaks the old one in two pieces,
1286 * so we have to use the second new lock.
1287 */
1288 left = new_fl2;
1289 new_fl2 = NULL;
1290 locks_copy_lock(left, right);
1291 locks_insert_lock_ctx(left, &fl->fl_list);
1292 }
1293 right->fl_start = request->fl_end + 1;
1294 locks_wake_up_blocks(right);
1295 }
1296 if (left) {
1297 left->fl_end = request->fl_start - 1;
1298 locks_wake_up_blocks(left);
1299 }
1300 out:
1301 spin_unlock(&ctx->flc_lock);
1302 percpu_up_read(&file_rwsem);
1303 /*
1304 * Free any unused locks.
1305 */
1306 if (new_fl)
1307 locks_free_lock(new_fl);
1308 if (new_fl2)
1309 locks_free_lock(new_fl2);
1310 locks_dispose_list(&dispose);
1311 trace_posix_lock_inode(inode, request, error);
1312
1313 return error;
1314}
1315
1316/**
1317 * posix_lock_file - Apply a POSIX-style lock to a file
1318 * @filp: The file to apply the lock to
1319 * @fl: The lock to be applied
1320 * @conflock: Place to return a copy of the conflicting lock, if found.
1321 *
1322 * Add a POSIX style lock to a file.
1323 * We merge adjacent & overlapping locks whenever possible.
1324 * POSIX locks are sorted by owner task, then by starting address
1325 *
1326 * Note that if called with an FL_EXISTS argument, the caller may determine
1327 * whether or not a lock was successfully freed by testing the return
1328 * value for -ENOENT.
1329 */
1330int posix_lock_file(struct file *filp, struct file_lock *fl,
1331 struct file_lock *conflock)
1332{
1333 return posix_lock_inode(locks_inode(filp), fl, conflock);
1334}
1335EXPORT_SYMBOL(posix_lock_file);
1336
1337/**
1338 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1339 * @inode: inode of file to which lock request should be applied
1340 * @fl: The lock to be applied
1341 *
1342 * Apply a POSIX style lock request to an inode.
1343 */
1344static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1345{
1346 int error;
1347 might_sleep ();
1348 for (;;) {
1349 error = posix_lock_inode(inode, fl, NULL);
1350 if (error != FILE_LOCK_DEFERRED)
1351 break;
1352 error = wait_event_interruptible(fl->fl_wait,
1353 list_empty(&fl->fl_blocked_member));
1354 if (error)
1355 break;
1356 }
1357 locks_delete_block(fl);
1358 return error;
1359}
1360
1361static void lease_clear_pending(struct file_lock *fl, int arg)
1362{
1363 switch (arg) {
1364 case F_UNLCK:
1365 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1366 fallthrough;
1367 case F_RDLCK:
1368 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1369 }
1370}
1371
1372/* We already had a lease on this file; just change its type */
1373int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1374{
1375 int error = assign_type(fl, arg);
1376
1377 if (error)
1378 return error;
1379 lease_clear_pending(fl, arg);
1380 locks_wake_up_blocks(fl);
1381 if (arg == F_UNLCK) {
1382 struct file *filp = fl->fl_file;
1383
1384 f_delown(filp);
1385 filp->f_owner.signum = 0;
1386 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1387 if (fl->fl_fasync != NULL) {
1388 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1389 fl->fl_fasync = NULL;
1390 }
1391 locks_delete_lock_ctx(fl, dispose);
1392 }
1393 return 0;
1394}
1395EXPORT_SYMBOL(lease_modify);
1396
1397static bool past_time(unsigned long then)
1398{
1399 if (!then)
1400 /* 0 is a special value meaning "this never expires": */
1401 return false;
1402 return time_after(jiffies, then);
1403}
1404
1405static void time_out_leases(struct inode *inode, struct list_head *dispose)
1406{
1407 struct file_lock_context *ctx = inode->i_flctx;
1408 struct file_lock *fl, *tmp;
1409
1410 lockdep_assert_held(&ctx->flc_lock);
1411
1412 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1413 trace_time_out_leases(inode, fl);
1414 if (past_time(fl->fl_downgrade_time))
1415 lease_modify(fl, F_RDLCK, dispose);
1416 if (past_time(fl->fl_break_time))
1417 lease_modify(fl, F_UNLCK, dispose);
1418 }
1419}
1420
1421static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1422{
1423 bool rc;
1424
1425 if (lease->fl_lmops->lm_breaker_owns_lease
1426 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1427 return false;
1428 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1429 rc = false;
1430 goto trace;
1431 }
1432 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1433 rc = false;
1434 goto trace;
1435 }
1436
1437 rc = locks_conflict(breaker, lease);
1438trace:
1439 trace_leases_conflict(rc, lease, breaker);
1440 return rc;
1441}
1442
1443static bool
1444any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1445{
1446 struct file_lock_context *ctx = inode->i_flctx;
1447 struct file_lock *fl;
1448
1449 lockdep_assert_held(&ctx->flc_lock);
1450
1451 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1452 if (leases_conflict(fl, breaker))
1453 return true;
1454 }
1455 return false;
1456}
1457
1458/**
1459 * __break_lease - revoke all outstanding leases on file
1460 * @inode: the inode of the file to return
1461 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1462 * break all leases
1463 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1464 * only delegations
1465 *
1466 * break_lease (inlined for speed) has checked there already is at least
1467 * some kind of lock (maybe a lease) on this file. Leases are broken on
1468 * a call to open() or truncate(). This function can sleep unless you
1469 * specified %O_NONBLOCK to your open().
1470 */
1471int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1472{
1473 int error = 0;
1474 struct file_lock_context *ctx;
1475 struct file_lock *new_fl, *fl, *tmp;
1476 unsigned long break_time;
1477 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1478 LIST_HEAD(dispose);
1479
1480 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1481 if (IS_ERR(new_fl))
1482 return PTR_ERR(new_fl);
1483 new_fl->fl_flags = type;
1484
1485 /* typically we will check that ctx is non-NULL before calling */
1486 ctx = locks_inode_context(inode);
1487 if (!ctx) {
1488 WARN_ON_ONCE(1);
1489 goto free_lock;
1490 }
1491
1492 percpu_down_read(&file_rwsem);
1493 spin_lock(&ctx->flc_lock);
1494
1495 time_out_leases(inode, &dispose);
1496
1497 if (!any_leases_conflict(inode, new_fl))
1498 goto out;
1499
1500 break_time = 0;
1501 if (lease_break_time > 0) {
1502 break_time = jiffies + lease_break_time * HZ;
1503 if (break_time == 0)
1504 break_time++; /* so that 0 means no break time */
1505 }
1506
1507 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1508 if (!leases_conflict(fl, new_fl))
1509 continue;
1510 if (want_write) {
1511 if (fl->fl_flags & FL_UNLOCK_PENDING)
1512 continue;
1513 fl->fl_flags |= FL_UNLOCK_PENDING;
1514 fl->fl_break_time = break_time;
1515 } else {
1516 if (lease_breaking(fl))
1517 continue;
1518 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1519 fl->fl_downgrade_time = break_time;
1520 }
1521 if (fl->fl_lmops->lm_break(fl))
1522 locks_delete_lock_ctx(fl, &dispose);
1523 }
1524
1525 if (list_empty(&ctx->flc_lease))
1526 goto out;
1527
1528 if (mode & O_NONBLOCK) {
1529 trace_break_lease_noblock(inode, new_fl);
1530 error = -EWOULDBLOCK;
1531 goto out;
1532 }
1533
1534restart:
1535 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1536 break_time = fl->fl_break_time;
1537 if (break_time != 0)
1538 break_time -= jiffies;
1539 if (break_time == 0)
1540 break_time++;
1541 locks_insert_block(fl, new_fl, leases_conflict);
1542 trace_break_lease_block(inode, new_fl);
1543 spin_unlock(&ctx->flc_lock);
1544 percpu_up_read(&file_rwsem);
1545
1546 locks_dispose_list(&dispose);
1547 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1548 list_empty(&new_fl->fl_blocked_member),
1549 break_time);
1550
1551 percpu_down_read(&file_rwsem);
1552 spin_lock(&ctx->flc_lock);
1553 trace_break_lease_unblock(inode, new_fl);
1554 locks_delete_block(new_fl);
1555 if (error >= 0) {
1556 /*
1557 * Wait for the next conflicting lease that has not been
1558 * broken yet
1559 */
1560 if (error == 0)
1561 time_out_leases(inode, &dispose);
1562 if (any_leases_conflict(inode, new_fl))
1563 goto restart;
1564 error = 0;
1565 }
1566out:
1567 spin_unlock(&ctx->flc_lock);
1568 percpu_up_read(&file_rwsem);
1569 locks_dispose_list(&dispose);
1570free_lock:
1571 locks_free_lock(new_fl);
1572 return error;
1573}
1574EXPORT_SYMBOL(__break_lease);
1575
1576/**
1577 * lease_get_mtime - update modified time of an inode with exclusive lease
1578 * @inode: the inode
1579 * @time: pointer to a timespec which contains the last modified time
1580 *
1581 * This is to force NFS clients to flush their caches for files with
1582 * exclusive leases. The justification is that if someone has an
1583 * exclusive lease, then they could be modifying it.
1584 */
1585void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1586{
1587 bool has_lease = false;
1588 struct file_lock_context *ctx;
1589 struct file_lock *fl;
1590
1591 ctx = locks_inode_context(inode);
1592 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1593 spin_lock(&ctx->flc_lock);
1594 fl = list_first_entry_or_null(&ctx->flc_lease,
1595 struct file_lock, fl_list);
1596 if (fl && (fl->fl_type == F_WRLCK))
1597 has_lease = true;
1598 spin_unlock(&ctx->flc_lock);
1599 }
1600
1601 if (has_lease)
1602 *time = current_time(inode);
1603}
1604EXPORT_SYMBOL(lease_get_mtime);
1605
1606/**
1607 * fcntl_getlease - Enquire what lease is currently active
1608 * @filp: the file
1609 *
1610 * The value returned by this function will be one of
1611 * (if no lease break is pending):
1612 *
1613 * %F_RDLCK to indicate a shared lease is held.
1614 *
1615 * %F_WRLCK to indicate an exclusive lease is held.
1616 *
1617 * %F_UNLCK to indicate no lease is held.
1618 *
1619 * (if a lease break is pending):
1620 *
1621 * %F_RDLCK to indicate an exclusive lease needs to be
1622 * changed to a shared lease (or removed).
1623 *
1624 * %F_UNLCK to indicate the lease needs to be removed.
1625 *
1626 * XXX: sfr & willy disagree over whether F_INPROGRESS
1627 * should be returned to userspace.
1628 */
1629int fcntl_getlease(struct file *filp)
1630{
1631 struct file_lock *fl;
1632 struct inode *inode = locks_inode(filp);
1633 struct file_lock_context *ctx;
1634 int type = F_UNLCK;
1635 LIST_HEAD(dispose);
1636
1637 ctx = locks_inode_context(inode);
1638 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1639 percpu_down_read(&file_rwsem);
1640 spin_lock(&ctx->flc_lock);
1641 time_out_leases(inode, &dispose);
1642 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1643 if (fl->fl_file != filp)
1644 continue;
1645 type = target_leasetype(fl);
1646 break;
1647 }
1648 spin_unlock(&ctx->flc_lock);
1649 percpu_up_read(&file_rwsem);
1650
1651 locks_dispose_list(&dispose);
1652 }
1653 return type;
1654}
1655
1656/**
1657 * check_conflicting_open - see if the given file points to an inode that has
1658 * an existing open that would conflict with the
1659 * desired lease.
1660 * @filp: file to check
1661 * @arg: type of lease that we're trying to acquire
1662 * @flags: current lock flags
1663 *
1664 * Check to see if there's an existing open fd on this file that would
1665 * conflict with the lease we're trying to set.
1666 */
1667static int
1668check_conflicting_open(struct file *filp, const long arg, int flags)
1669{
1670 struct inode *inode = locks_inode(filp);
1671 int self_wcount = 0, self_rcount = 0;
1672
1673 if (flags & FL_LAYOUT)
1674 return 0;
1675 if (flags & FL_DELEG)
1676 /* We leave these checks to the caller */
1677 return 0;
1678
1679 if (arg == F_RDLCK)
1680 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1681 else if (arg != F_WRLCK)
1682 return 0;
1683
1684 /*
1685 * Make sure that only read/write count is from lease requestor.
1686 * Note that this will result in denying write leases when i_writecount
1687 * is negative, which is what we want. (We shouldn't grant write leases
1688 * on files open for execution.)
1689 */
1690 if (filp->f_mode & FMODE_WRITE)
1691 self_wcount = 1;
1692 else if (filp->f_mode & FMODE_READ)
1693 self_rcount = 1;
1694
1695 if (atomic_read(&inode->i_writecount) != self_wcount ||
1696 atomic_read(&inode->i_readcount) != self_rcount)
1697 return -EAGAIN;
1698
1699 return 0;
1700}
1701
1702static int
1703generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1704{
1705 struct file_lock *fl, *my_fl = NULL, *lease;
1706 struct inode *inode = locks_inode(filp);
1707 struct file_lock_context *ctx;
1708 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1709 int error;
1710 LIST_HEAD(dispose);
1711
1712 lease = *flp;
1713 trace_generic_add_lease(inode, lease);
1714
1715 /* Note that arg is never F_UNLCK here */
1716 ctx = locks_get_lock_context(inode, arg);
1717 if (!ctx)
1718 return -ENOMEM;
1719
1720 /*
1721 * In the delegation case we need mutual exclusion with
1722 * a number of operations that take the i_mutex. We trylock
1723 * because delegations are an optional optimization, and if
1724 * there's some chance of a conflict--we'd rather not
1725 * bother, maybe that's a sign this just isn't a good file to
1726 * hand out a delegation on.
1727 */
1728 if (is_deleg && !inode_trylock(inode))
1729 return -EAGAIN;
1730
1731 if (is_deleg && arg == F_WRLCK) {
1732 /* Write delegations are not currently supported: */
1733 inode_unlock(inode);
1734 WARN_ON_ONCE(1);
1735 return -EINVAL;
1736 }
1737
1738 percpu_down_read(&file_rwsem);
1739 spin_lock(&ctx->flc_lock);
1740 time_out_leases(inode, &dispose);
1741 error = check_conflicting_open(filp, arg, lease->fl_flags);
1742 if (error)
1743 goto out;
1744
1745 /*
1746 * At this point, we know that if there is an exclusive
1747 * lease on this file, then we hold it on this filp
1748 * (otherwise our open of this file would have blocked).
1749 * And if we are trying to acquire an exclusive lease,
1750 * then the file is not open by anyone (including us)
1751 * except for this filp.
1752 */
1753 error = -EAGAIN;
1754 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1755 if (fl->fl_file == filp &&
1756 fl->fl_owner == lease->fl_owner) {
1757 my_fl = fl;
1758 continue;
1759 }
1760
1761 /*
1762 * No exclusive leases if someone else has a lease on
1763 * this file:
1764 */
1765 if (arg == F_WRLCK)
1766 goto out;
1767 /*
1768 * Modifying our existing lease is OK, but no getting a
1769 * new lease if someone else is opening for write:
1770 */
1771 if (fl->fl_flags & FL_UNLOCK_PENDING)
1772 goto out;
1773 }
1774
1775 if (my_fl != NULL) {
1776 lease = my_fl;
1777 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1778 if (error)
1779 goto out;
1780 goto out_setup;
1781 }
1782
1783 error = -EINVAL;
1784 if (!leases_enable)
1785 goto out;
1786
1787 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1788 /*
1789 * The check in break_lease() is lockless. It's possible for another
1790 * open to race in after we did the earlier check for a conflicting
1791 * open but before the lease was inserted. Check again for a
1792 * conflicting open and cancel the lease if there is one.
1793 *
1794 * We also add a barrier here to ensure that the insertion of the lock
1795 * precedes these checks.
1796 */
1797 smp_mb();
1798 error = check_conflicting_open(filp, arg, lease->fl_flags);
1799 if (error) {
1800 locks_unlink_lock_ctx(lease);
1801 goto out;
1802 }
1803
1804out_setup:
1805 if (lease->fl_lmops->lm_setup)
1806 lease->fl_lmops->lm_setup(lease, priv);
1807out:
1808 spin_unlock(&ctx->flc_lock);
1809 percpu_up_read(&file_rwsem);
1810 locks_dispose_list(&dispose);
1811 if (is_deleg)
1812 inode_unlock(inode);
1813 if (!error && !my_fl)
1814 *flp = NULL;
1815 return error;
1816}
1817
1818static int generic_delete_lease(struct file *filp, void *owner)
1819{
1820 int error = -EAGAIN;
1821 struct file_lock *fl, *victim = NULL;
1822 struct inode *inode = locks_inode(filp);
1823 struct file_lock_context *ctx;
1824 LIST_HEAD(dispose);
1825
1826 ctx = locks_inode_context(inode);
1827 if (!ctx) {
1828 trace_generic_delete_lease(inode, NULL);
1829 return error;
1830 }
1831
1832 percpu_down_read(&file_rwsem);
1833 spin_lock(&ctx->flc_lock);
1834 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1835 if (fl->fl_file == filp &&
1836 fl->fl_owner == owner) {
1837 victim = fl;
1838 break;
1839 }
1840 }
1841 trace_generic_delete_lease(inode, victim);
1842 if (victim)
1843 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1844 spin_unlock(&ctx->flc_lock);
1845 percpu_up_read(&file_rwsem);
1846 locks_dispose_list(&dispose);
1847 return error;
1848}
1849
1850/**
1851 * generic_setlease - sets a lease on an open file
1852 * @filp: file pointer
1853 * @arg: type of lease to obtain
1854 * @flp: input - file_lock to use, output - file_lock inserted
1855 * @priv: private data for lm_setup (may be NULL if lm_setup
1856 * doesn't require it)
1857 *
1858 * The (input) flp->fl_lmops->lm_break function is required
1859 * by break_lease().
1860 */
1861int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1862 void **priv)
1863{
1864 struct inode *inode = locks_inode(filp);
1865 int error;
1866
1867 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
1868 return -EACCES;
1869 if (!S_ISREG(inode->i_mode))
1870 return -EINVAL;
1871 error = security_file_lock(filp, arg);
1872 if (error)
1873 return error;
1874
1875 switch (arg) {
1876 case F_UNLCK:
1877 return generic_delete_lease(filp, *priv);
1878 case F_RDLCK:
1879 case F_WRLCK:
1880 if (!(*flp)->fl_lmops->lm_break) {
1881 WARN_ON_ONCE(1);
1882 return -ENOLCK;
1883 }
1884
1885 return generic_add_lease(filp, arg, flp, priv);
1886 default:
1887 return -EINVAL;
1888 }
1889}
1890EXPORT_SYMBOL(generic_setlease);
1891
1892#if IS_ENABLED(CONFIG_SRCU)
1893/*
1894 * Kernel subsystems can register to be notified on any attempt to set
1895 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
1896 * to close files that it may have cached when there is an attempt to set a
1897 * conflicting lease.
1898 */
1899static struct srcu_notifier_head lease_notifier_chain;
1900
1901static inline void
1902lease_notifier_chain_init(void)
1903{
1904 srcu_init_notifier_head(&lease_notifier_chain);
1905}
1906
1907static inline void
1908setlease_notifier(long arg, struct file_lock *lease)
1909{
1910 if (arg != F_UNLCK)
1911 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
1912}
1913
1914int lease_register_notifier(struct notifier_block *nb)
1915{
1916 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
1917}
1918EXPORT_SYMBOL_GPL(lease_register_notifier);
1919
1920void lease_unregister_notifier(struct notifier_block *nb)
1921{
1922 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
1923}
1924EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1925
1926#else /* !IS_ENABLED(CONFIG_SRCU) */
1927static inline void
1928lease_notifier_chain_init(void)
1929{
1930}
1931
1932static inline void
1933setlease_notifier(long arg, struct file_lock *lease)
1934{
1935}
1936
1937int lease_register_notifier(struct notifier_block *nb)
1938{
1939 return 0;
1940}
1941EXPORT_SYMBOL_GPL(lease_register_notifier);
1942
1943void lease_unregister_notifier(struct notifier_block *nb)
1944{
1945}
1946EXPORT_SYMBOL_GPL(lease_unregister_notifier);
1947
1948#endif /* IS_ENABLED(CONFIG_SRCU) */
1949
1950/**
1951 * vfs_setlease - sets a lease on an open file
1952 * @filp: file pointer
1953 * @arg: type of lease to obtain
1954 * @lease: file_lock to use when adding a lease
1955 * @priv: private info for lm_setup when adding a lease (may be
1956 * NULL if lm_setup doesn't require it)
1957 *
1958 * Call this to establish a lease on the file. The "lease" argument is not
1959 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1960 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
1961 * set; if not, this function will return -ENOLCK (and generate a scary-looking
1962 * stack trace).
1963 *
1964 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1965 * may be NULL if the lm_setup operation doesn't require it.
1966 */
1967int
1968vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
1969{
1970 if (lease)
1971 setlease_notifier(arg, *lease);
1972 if (filp->f_op->setlease)
1973 return filp->f_op->setlease(filp, arg, lease, priv);
1974 else
1975 return generic_setlease(filp, arg, lease, priv);
1976}
1977EXPORT_SYMBOL_GPL(vfs_setlease);
1978
1979static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1980{
1981 struct file_lock *fl;
1982 struct fasync_struct *new;
1983 int error;
1984
1985 fl = lease_alloc(filp, arg);
1986 if (IS_ERR(fl))
1987 return PTR_ERR(fl);
1988
1989 new = fasync_alloc();
1990 if (!new) {
1991 locks_free_lock(fl);
1992 return -ENOMEM;
1993 }
1994 new->fa_fd = fd;
1995
1996 error = vfs_setlease(filp, arg, &fl, (void **)&new);
1997 if (fl)
1998 locks_free_lock(fl);
1999 if (new)
2000 fasync_free(new);
2001 return error;
2002}
2003
2004/**
2005 * fcntl_setlease - sets a lease on an open file
2006 * @fd: open file descriptor
2007 * @filp: file pointer
2008 * @arg: type of lease to obtain
2009 *
2010 * Call this fcntl to establish a lease on the file.
2011 * Note that you also need to call %F_SETSIG to
2012 * receive a signal when the lease is broken.
2013 */
2014int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2015{
2016 if (arg == F_UNLCK)
2017 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2018 return do_fcntl_add_lease(fd, filp, arg);
2019}
2020
2021/**
2022 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2023 * @inode: inode of the file to apply to
2024 * @fl: The lock to be applied
2025 *
2026 * Apply a FLOCK style lock request to an inode.
2027 */
2028static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2029{
2030 int error;
2031 might_sleep();
2032 for (;;) {
2033 error = flock_lock_inode(inode, fl);
2034 if (error != FILE_LOCK_DEFERRED)
2035 break;
2036 error = wait_event_interruptible(fl->fl_wait,
2037 list_empty(&fl->fl_blocked_member));
2038 if (error)
2039 break;
2040 }
2041 locks_delete_block(fl);
2042 return error;
2043}
2044
2045/**
2046 * locks_lock_inode_wait - Apply a lock to an inode
2047 * @inode: inode of the file to apply to
2048 * @fl: The lock to be applied
2049 *
2050 * Apply a POSIX or FLOCK style lock request to an inode.
2051 */
2052int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2053{
2054 int res = 0;
2055 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2056 case FL_POSIX:
2057 res = posix_lock_inode_wait(inode, fl);
2058 break;
2059 case FL_FLOCK:
2060 res = flock_lock_inode_wait(inode, fl);
2061 break;
2062 default:
2063 BUG();
2064 }
2065 return res;
2066}
2067EXPORT_SYMBOL(locks_lock_inode_wait);
2068
2069/**
2070 * sys_flock: - flock() system call.
2071 * @fd: the file descriptor to lock.
2072 * @cmd: the type of lock to apply.
2073 *
2074 * Apply a %FL_FLOCK style lock to an open file descriptor.
2075 * The @cmd can be one of:
2076 *
2077 * - %LOCK_SH -- a shared lock.
2078 * - %LOCK_EX -- an exclusive lock.
2079 * - %LOCK_UN -- remove an existing lock.
2080 * - %LOCK_MAND -- a 'mandatory' flock. (DEPRECATED)
2081 *
2082 * %LOCK_MAND support has been removed from the kernel.
2083 */
2084SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2085{
2086 int can_sleep, error, type;
2087 struct file_lock fl;
2088 struct fd f;
2089
2090 /*
2091 * LOCK_MAND locks were broken for a long time in that they never
2092 * conflicted with one another and didn't prevent any sort of open,
2093 * read or write activity.
2094 *
2095 * Just ignore these requests now, to preserve legacy behavior, but
2096 * throw a warning to let people know that they don't actually work.
2097 */
2098 if (cmd & LOCK_MAND) {
2099 pr_warn_once("%s(%d): Attempt to set a LOCK_MAND lock via flock(2). This support has been removed and the request ignored.\n", current->comm, current->pid);
2100 return 0;
2101 }
2102
2103 type = flock_translate_cmd(cmd & ~LOCK_NB);
2104 if (type < 0)
2105 return type;
2106
2107 error = -EBADF;
2108 f = fdget(fd);
2109 if (!f.file)
2110 return error;
2111
2112 if (type != F_UNLCK && !(f.file->f_mode & (FMODE_READ | FMODE_WRITE)))
2113 goto out_putf;
2114
2115 flock_make_lock(f.file, &fl, type);
2116
2117 error = security_file_lock(f.file, fl.fl_type);
2118 if (error)
2119 goto out_putf;
2120
2121 can_sleep = !(cmd & LOCK_NB);
2122 if (can_sleep)
2123 fl.fl_flags |= FL_SLEEP;
2124
2125 if (f.file->f_op->flock)
2126 error = f.file->f_op->flock(f.file,
2127 (can_sleep) ? F_SETLKW : F_SETLK,
2128 &fl);
2129 else
2130 error = locks_lock_file_wait(f.file, &fl);
2131
2132 locks_release_private(&fl);
2133 out_putf:
2134 fdput(f);
2135
2136 return error;
2137}
2138
2139/**
2140 * vfs_test_lock - test file byte range lock
2141 * @filp: The file to test lock for
2142 * @fl: The lock to test; also used to hold result
2143 *
2144 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2145 * setting conf->fl_type to something other than F_UNLCK.
2146 */
2147int vfs_test_lock(struct file *filp, struct file_lock *fl)
2148{
2149 WARN_ON_ONCE(filp != fl->fl_file);
2150 if (filp->f_op->lock)
2151 return filp->f_op->lock(filp, F_GETLK, fl);
2152 posix_test_lock(filp, fl);
2153 return 0;
2154}
2155EXPORT_SYMBOL_GPL(vfs_test_lock);
2156
2157/**
2158 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2159 * @fl: The file_lock who's fl_pid should be translated
2160 * @ns: The namespace into which the pid should be translated
2161 *
2162 * Used to tranlate a fl_pid into a namespace virtual pid number
2163 */
2164static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2165{
2166 pid_t vnr;
2167 struct pid *pid;
2168
2169 if (IS_OFDLCK(fl))
2170 return -1;
2171 if (IS_REMOTELCK(fl))
2172 return fl->fl_pid;
2173 /*
2174 * If the flock owner process is dead and its pid has been already
2175 * freed, the translation below won't work, but we still want to show
2176 * flock owner pid number in init pidns.
2177 */
2178 if (ns == &init_pid_ns)
2179 return (pid_t)fl->fl_pid;
2180
2181 rcu_read_lock();
2182 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2183 vnr = pid_nr_ns(pid, ns);
2184 rcu_read_unlock();
2185 return vnr;
2186}
2187
2188static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2189{
2190 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2191#if BITS_PER_LONG == 32
2192 /*
2193 * Make sure we can represent the posix lock via
2194 * legacy 32bit flock.
2195 */
2196 if (fl->fl_start > OFFT_OFFSET_MAX)
2197 return -EOVERFLOW;
2198 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2199 return -EOVERFLOW;
2200#endif
2201 flock->l_start = fl->fl_start;
2202 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2203 fl->fl_end - fl->fl_start + 1;
2204 flock->l_whence = 0;
2205 flock->l_type = fl->fl_type;
2206 return 0;
2207}
2208
2209#if BITS_PER_LONG == 32
2210static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2211{
2212 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2213 flock->l_start = fl->fl_start;
2214 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2215 fl->fl_end - fl->fl_start + 1;
2216 flock->l_whence = 0;
2217 flock->l_type = fl->fl_type;
2218}
2219#endif
2220
2221/* Report the first existing lock that would conflict with l.
2222 * This implements the F_GETLK command of fcntl().
2223 */
2224int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2225{
2226 struct file_lock *fl;
2227 int error;
2228
2229 fl = locks_alloc_lock();
2230 if (fl == NULL)
2231 return -ENOMEM;
2232 error = -EINVAL;
2233 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2234 goto out;
2235
2236 error = flock_to_posix_lock(filp, fl, flock);
2237 if (error)
2238 goto out;
2239
2240 if (cmd == F_OFD_GETLK) {
2241 error = -EINVAL;
2242 if (flock->l_pid != 0)
2243 goto out;
2244
2245 fl->fl_flags |= FL_OFDLCK;
2246 fl->fl_owner = filp;
2247 }
2248
2249 error = vfs_test_lock(filp, fl);
2250 if (error)
2251 goto out;
2252
2253 flock->l_type = fl->fl_type;
2254 if (fl->fl_type != F_UNLCK) {
2255 error = posix_lock_to_flock(flock, fl);
2256 if (error)
2257 goto out;
2258 }
2259out:
2260 locks_free_lock(fl);
2261 return error;
2262}
2263
2264/**
2265 * vfs_lock_file - file byte range lock
2266 * @filp: The file to apply the lock to
2267 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2268 * @fl: The lock to be applied
2269 * @conf: Place to return a copy of the conflicting lock, if found.
2270 *
2271 * A caller that doesn't care about the conflicting lock may pass NULL
2272 * as the final argument.
2273 *
2274 * If the filesystem defines a private ->lock() method, then @conf will
2275 * be left unchanged; so a caller that cares should initialize it to
2276 * some acceptable default.
2277 *
2278 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2279 * locks, the ->lock() interface may return asynchronously, before the lock has
2280 * been granted or denied by the underlying filesystem, if (and only if)
2281 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2282 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2283 * the request is for a blocking lock. When ->lock() does return asynchronously,
2284 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2285 * request completes.
2286 * If the request is for non-blocking lock the file system should return
2287 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2288 * with the result. If the request timed out the callback routine will return a
2289 * nonzero return code and the file system should release the lock. The file
2290 * system is also responsible to keep a corresponding posix lock when it
2291 * grants a lock so the VFS can find out which locks are locally held and do
2292 * the correct lock cleanup when required.
2293 * The underlying filesystem must not drop the kernel lock or call
2294 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2295 * return code.
2296 */
2297int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2298{
2299 WARN_ON_ONCE(filp != fl->fl_file);
2300 if (filp->f_op->lock)
2301 return filp->f_op->lock(filp, cmd, fl);
2302 else
2303 return posix_lock_file(filp, fl, conf);
2304}
2305EXPORT_SYMBOL_GPL(vfs_lock_file);
2306
2307static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2308 struct file_lock *fl)
2309{
2310 int error;
2311
2312 error = security_file_lock(filp, fl->fl_type);
2313 if (error)
2314 return error;
2315
2316 for (;;) {
2317 error = vfs_lock_file(filp, cmd, fl, NULL);
2318 if (error != FILE_LOCK_DEFERRED)
2319 break;
2320 error = wait_event_interruptible(fl->fl_wait,
2321 list_empty(&fl->fl_blocked_member));
2322 if (error)
2323 break;
2324 }
2325 locks_delete_block(fl);
2326
2327 return error;
2328}
2329
2330/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2331static int
2332check_fmode_for_setlk(struct file_lock *fl)
2333{
2334 switch (fl->fl_type) {
2335 case F_RDLCK:
2336 if (!(fl->fl_file->f_mode & FMODE_READ))
2337 return -EBADF;
2338 break;
2339 case F_WRLCK:
2340 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2341 return -EBADF;
2342 }
2343 return 0;
2344}
2345
2346/* Apply the lock described by l to an open file descriptor.
2347 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2348 */
2349int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2350 struct flock *flock)
2351{
2352 struct file_lock *file_lock = locks_alloc_lock();
2353 struct inode *inode = locks_inode(filp);
2354 struct file *f;
2355 int error;
2356
2357 if (file_lock == NULL)
2358 return -ENOLCK;
2359
2360 error = flock_to_posix_lock(filp, file_lock, flock);
2361 if (error)
2362 goto out;
2363
2364 error = check_fmode_for_setlk(file_lock);
2365 if (error)
2366 goto out;
2367
2368 /*
2369 * If the cmd is requesting file-private locks, then set the
2370 * FL_OFDLCK flag and override the owner.
2371 */
2372 switch (cmd) {
2373 case F_OFD_SETLK:
2374 error = -EINVAL;
2375 if (flock->l_pid != 0)
2376 goto out;
2377
2378 cmd = F_SETLK;
2379 file_lock->fl_flags |= FL_OFDLCK;
2380 file_lock->fl_owner = filp;
2381 break;
2382 case F_OFD_SETLKW:
2383 error = -EINVAL;
2384 if (flock->l_pid != 0)
2385 goto out;
2386
2387 cmd = F_SETLKW;
2388 file_lock->fl_flags |= FL_OFDLCK;
2389 file_lock->fl_owner = filp;
2390 fallthrough;
2391 case F_SETLKW:
2392 file_lock->fl_flags |= FL_SLEEP;
2393 }
2394
2395 error = do_lock_file_wait(filp, cmd, file_lock);
2396
2397 /*
2398 * Attempt to detect a close/fcntl race and recover by releasing the
2399 * lock that was just acquired. There is no need to do that when we're
2400 * unlocking though, or for OFD locks.
2401 */
2402 if (!error && file_lock->fl_type != F_UNLCK &&
2403 !(file_lock->fl_flags & FL_OFDLCK)) {
2404 struct files_struct *files = current->files;
2405 /*
2406 * We need that spin_lock here - it prevents reordering between
2407 * update of i_flctx->flc_posix and check for it done in
2408 * close(). rcu_read_lock() wouldn't do.
2409 */
2410 spin_lock(&files->file_lock);
2411 f = files_lookup_fd_locked(files, fd);
2412 spin_unlock(&files->file_lock);
2413 if (f != filp) {
2414 file_lock->fl_type = F_UNLCK;
2415 error = do_lock_file_wait(filp, cmd, file_lock);
2416 WARN_ON_ONCE(error);
2417 error = -EBADF;
2418 }
2419 }
2420out:
2421 trace_fcntl_setlk(inode, file_lock, error);
2422 locks_free_lock(file_lock);
2423 return error;
2424}
2425
2426#if BITS_PER_LONG == 32
2427/* Report the first existing lock that would conflict with l.
2428 * This implements the F_GETLK command of fcntl().
2429 */
2430int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2431{
2432 struct file_lock *fl;
2433 int error;
2434
2435 fl = locks_alloc_lock();
2436 if (fl == NULL)
2437 return -ENOMEM;
2438
2439 error = -EINVAL;
2440 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2441 goto out;
2442
2443 error = flock64_to_posix_lock(filp, fl, flock);
2444 if (error)
2445 goto out;
2446
2447 if (cmd == F_OFD_GETLK) {
2448 error = -EINVAL;
2449 if (flock->l_pid != 0)
2450 goto out;
2451
2452 cmd = F_GETLK64;
2453 fl->fl_flags |= FL_OFDLCK;
2454 fl->fl_owner = filp;
2455 }
2456
2457 error = vfs_test_lock(filp, fl);
2458 if (error)
2459 goto out;
2460
2461 flock->l_type = fl->fl_type;
2462 if (fl->fl_type != F_UNLCK)
2463 posix_lock_to_flock64(flock, fl);
2464
2465out:
2466 locks_free_lock(fl);
2467 return error;
2468}
2469
2470/* Apply the lock described by l to an open file descriptor.
2471 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2472 */
2473int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2474 struct flock64 *flock)
2475{
2476 struct file_lock *file_lock = locks_alloc_lock();
2477 struct file *f;
2478 int error;
2479
2480 if (file_lock == NULL)
2481 return -ENOLCK;
2482
2483 error = flock64_to_posix_lock(filp, file_lock, flock);
2484 if (error)
2485 goto out;
2486
2487 error = check_fmode_for_setlk(file_lock);
2488 if (error)
2489 goto out;
2490
2491 /*
2492 * If the cmd is requesting file-private locks, then set the
2493 * FL_OFDLCK flag and override the owner.
2494 */
2495 switch (cmd) {
2496 case F_OFD_SETLK:
2497 error = -EINVAL;
2498 if (flock->l_pid != 0)
2499 goto out;
2500
2501 cmd = F_SETLK64;
2502 file_lock->fl_flags |= FL_OFDLCK;
2503 file_lock->fl_owner = filp;
2504 break;
2505 case F_OFD_SETLKW:
2506 error = -EINVAL;
2507 if (flock->l_pid != 0)
2508 goto out;
2509
2510 cmd = F_SETLKW64;
2511 file_lock->fl_flags |= FL_OFDLCK;
2512 file_lock->fl_owner = filp;
2513 fallthrough;
2514 case F_SETLKW64:
2515 file_lock->fl_flags |= FL_SLEEP;
2516 }
2517
2518 error = do_lock_file_wait(filp, cmd, file_lock);
2519
2520 /*
2521 * Attempt to detect a close/fcntl race and recover by releasing the
2522 * lock that was just acquired. There is no need to do that when we're
2523 * unlocking though, or for OFD locks.
2524 */
2525 if (!error && file_lock->fl_type != F_UNLCK &&
2526 !(file_lock->fl_flags & FL_OFDLCK)) {
2527 struct files_struct *files = current->files;
2528 /*
2529 * We need that spin_lock here - it prevents reordering between
2530 * update of i_flctx->flc_posix and check for it done in
2531 * close(). rcu_read_lock() wouldn't do.
2532 */
2533 spin_lock(&files->file_lock);
2534 f = files_lookup_fd_locked(files, fd);
2535 spin_unlock(&files->file_lock);
2536 if (f != filp) {
2537 file_lock->fl_type = F_UNLCK;
2538 error = do_lock_file_wait(filp, cmd, file_lock);
2539 WARN_ON_ONCE(error);
2540 error = -EBADF;
2541 }
2542 }
2543out:
2544 locks_free_lock(file_lock);
2545 return error;
2546}
2547#endif /* BITS_PER_LONG == 32 */
2548
2549/*
2550 * This function is called when the file is being removed
2551 * from the task's fd array. POSIX locks belonging to this task
2552 * are deleted at this time.
2553 */
2554void locks_remove_posix(struct file *filp, fl_owner_t owner)
2555{
2556 int error;
2557 struct inode *inode = locks_inode(filp);
2558 struct file_lock lock;
2559 struct file_lock_context *ctx;
2560
2561 /*
2562 * If there are no locks held on this file, we don't need to call
2563 * posix_lock_file(). Another process could be setting a lock on this
2564 * file at the same time, but we wouldn't remove that lock anyway.
2565 */
2566 ctx = locks_inode_context(inode);
2567 if (!ctx || list_empty(&ctx->flc_posix))
2568 return;
2569
2570 locks_init_lock(&lock);
2571 lock.fl_type = F_UNLCK;
2572 lock.fl_flags = FL_POSIX | FL_CLOSE;
2573 lock.fl_start = 0;
2574 lock.fl_end = OFFSET_MAX;
2575 lock.fl_owner = owner;
2576 lock.fl_pid = current->tgid;
2577 lock.fl_file = filp;
2578 lock.fl_ops = NULL;
2579 lock.fl_lmops = NULL;
2580
2581 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2582
2583 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2584 lock.fl_ops->fl_release_private(&lock);
2585 trace_locks_remove_posix(inode, &lock, error);
2586}
2587EXPORT_SYMBOL(locks_remove_posix);
2588
2589/* The i_flctx must be valid when calling into here */
2590static void
2591locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2592{
2593 struct file_lock fl;
2594 struct inode *inode = locks_inode(filp);
2595
2596 if (list_empty(&flctx->flc_flock))
2597 return;
2598
2599 flock_make_lock(filp, &fl, F_UNLCK);
2600 fl.fl_flags |= FL_CLOSE;
2601
2602 if (filp->f_op->flock)
2603 filp->f_op->flock(filp, F_SETLKW, &fl);
2604 else
2605 flock_lock_inode(inode, &fl);
2606
2607 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2608 fl.fl_ops->fl_release_private(&fl);
2609}
2610
2611/* The i_flctx must be valid when calling into here */
2612static void
2613locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2614{
2615 struct file_lock *fl, *tmp;
2616 LIST_HEAD(dispose);
2617
2618 if (list_empty(&ctx->flc_lease))
2619 return;
2620
2621 percpu_down_read(&file_rwsem);
2622 spin_lock(&ctx->flc_lock);
2623 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2624 if (filp == fl->fl_file)
2625 lease_modify(fl, F_UNLCK, &dispose);
2626 spin_unlock(&ctx->flc_lock);
2627 percpu_up_read(&file_rwsem);
2628
2629 locks_dispose_list(&dispose);
2630}
2631
2632/*
2633 * This function is called on the last close of an open file.
2634 */
2635void locks_remove_file(struct file *filp)
2636{
2637 struct file_lock_context *ctx;
2638
2639 ctx = locks_inode_context(locks_inode(filp));
2640 if (!ctx)
2641 return;
2642
2643 /* remove any OFD locks */
2644 locks_remove_posix(filp, filp);
2645
2646 /* remove flock locks */
2647 locks_remove_flock(filp, ctx);
2648
2649 /* remove any leases */
2650 locks_remove_lease(filp, ctx);
2651
2652 spin_lock(&ctx->flc_lock);
2653 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2654 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2655 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2656 spin_unlock(&ctx->flc_lock);
2657}
2658
2659/**
2660 * vfs_cancel_lock - file byte range unblock lock
2661 * @filp: The file to apply the unblock to
2662 * @fl: The lock to be unblocked
2663 *
2664 * Used by lock managers to cancel blocked requests
2665 */
2666int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2667{
2668 WARN_ON_ONCE(filp != fl->fl_file);
2669 if (filp->f_op->lock)
2670 return filp->f_op->lock(filp, F_CANCELLK, fl);
2671 return 0;
2672}
2673EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2674
2675/**
2676 * vfs_inode_has_locks - are any file locks held on @inode?
2677 * @inode: inode to check for locks
2678 *
2679 * Return true if there are any FL_POSIX or FL_FLOCK locks currently
2680 * set on @inode.
2681 */
2682bool vfs_inode_has_locks(struct inode *inode)
2683{
2684 struct file_lock_context *ctx;
2685 bool ret;
2686
2687 ctx = locks_inode_context(inode);
2688 if (!ctx)
2689 return false;
2690
2691 spin_lock(&ctx->flc_lock);
2692 ret = !list_empty(&ctx->flc_posix) || !list_empty(&ctx->flc_flock);
2693 spin_unlock(&ctx->flc_lock);
2694 return ret;
2695}
2696EXPORT_SYMBOL_GPL(vfs_inode_has_locks);
2697
2698#ifdef CONFIG_PROC_FS
2699#include <linux/proc_fs.h>
2700#include <linux/seq_file.h>
2701
2702struct locks_iterator {
2703 int li_cpu;
2704 loff_t li_pos;
2705};
2706
2707static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2708 loff_t id, char *pfx, int repeat)
2709{
2710 struct inode *inode = NULL;
2711 unsigned int fl_pid;
2712 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2713 int type;
2714
2715 fl_pid = locks_translate_pid(fl, proc_pidns);
2716 /*
2717 * If lock owner is dead (and pid is freed) or not visible in current
2718 * pidns, zero is shown as a pid value. Check lock info from
2719 * init_pid_ns to get saved lock pid value.
2720 */
2721
2722 if (fl->fl_file != NULL)
2723 inode = locks_inode(fl->fl_file);
2724
2725 seq_printf(f, "%lld: ", id);
2726
2727 if (repeat)
2728 seq_printf(f, "%*s", repeat - 1 + (int)strlen(pfx), pfx);
2729
2730 if (IS_POSIX(fl)) {
2731 if (fl->fl_flags & FL_ACCESS)
2732 seq_puts(f, "ACCESS");
2733 else if (IS_OFDLCK(fl))
2734 seq_puts(f, "OFDLCK");
2735 else
2736 seq_puts(f, "POSIX ");
2737
2738 seq_printf(f, " %s ",
2739 (inode == NULL) ? "*NOINODE*" : "ADVISORY ");
2740 } else if (IS_FLOCK(fl)) {
2741 seq_puts(f, "FLOCK ADVISORY ");
2742 } else if (IS_LEASE(fl)) {
2743 if (fl->fl_flags & FL_DELEG)
2744 seq_puts(f, "DELEG ");
2745 else
2746 seq_puts(f, "LEASE ");
2747
2748 if (lease_breaking(fl))
2749 seq_puts(f, "BREAKING ");
2750 else if (fl->fl_file)
2751 seq_puts(f, "ACTIVE ");
2752 else
2753 seq_puts(f, "BREAKER ");
2754 } else {
2755 seq_puts(f, "UNKNOWN UNKNOWN ");
2756 }
2757 type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2758
2759 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2760 (type == F_RDLCK) ? "READ" : "UNLCK");
2761 if (inode) {
2762 /* userspace relies on this representation of dev_t */
2763 seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2764 MAJOR(inode->i_sb->s_dev),
2765 MINOR(inode->i_sb->s_dev), inode->i_ino);
2766 } else {
2767 seq_printf(f, "%d <none>:0 ", fl_pid);
2768 }
2769 if (IS_POSIX(fl)) {
2770 if (fl->fl_end == OFFSET_MAX)
2771 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2772 else
2773 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2774 } else {
2775 seq_puts(f, "0 EOF\n");
2776 }
2777}
2778
2779static struct file_lock *get_next_blocked_member(struct file_lock *node)
2780{
2781 struct file_lock *tmp;
2782
2783 /* NULL node or root node */
2784 if (node == NULL || node->fl_blocker == NULL)
2785 return NULL;
2786
2787 /* Next member in the linked list could be itself */
2788 tmp = list_next_entry(node, fl_blocked_member);
2789 if (list_entry_is_head(tmp, &node->fl_blocker->fl_blocked_requests, fl_blocked_member)
2790 || tmp == node) {
2791 return NULL;
2792 }
2793
2794 return tmp;
2795}
2796
2797static int locks_show(struct seq_file *f, void *v)
2798{
2799 struct locks_iterator *iter = f->private;
2800 struct file_lock *cur, *tmp;
2801 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2802 int level = 0;
2803
2804 cur = hlist_entry(v, struct file_lock, fl_link);
2805
2806 if (locks_translate_pid(cur, proc_pidns) == 0)
2807 return 0;
2808
2809 /* View this crossed linked list as a binary tree, the first member of fl_blocked_requests
2810 * is the left child of current node, the next silibing in fl_blocked_member is the
2811 * right child, we can alse get the parent of current node from fl_blocker, so this
2812 * question becomes traversal of a binary tree
2813 */
2814 while (cur != NULL) {
2815 if (level)
2816 lock_get_status(f, cur, iter->li_pos, "-> ", level);
2817 else
2818 lock_get_status(f, cur, iter->li_pos, "", level);
2819
2820 if (!list_empty(&cur->fl_blocked_requests)) {
2821 /* Turn left */
2822 cur = list_first_entry_or_null(&cur->fl_blocked_requests,
2823 struct file_lock, fl_blocked_member);
2824 level++;
2825 } else {
2826 /* Turn right */
2827 tmp = get_next_blocked_member(cur);
2828 /* Fall back to parent node */
2829 while (tmp == NULL && cur->fl_blocker != NULL) {
2830 cur = cur->fl_blocker;
2831 level--;
2832 tmp = get_next_blocked_member(cur);
2833 }
2834 cur = tmp;
2835 }
2836 }
2837
2838 return 0;
2839}
2840
2841static void __show_fd_locks(struct seq_file *f,
2842 struct list_head *head, int *id,
2843 struct file *filp, struct files_struct *files)
2844{
2845 struct file_lock *fl;
2846
2847 list_for_each_entry(fl, head, fl_list) {
2848
2849 if (filp != fl->fl_file)
2850 continue;
2851 if (fl->fl_owner != files &&
2852 fl->fl_owner != filp)
2853 continue;
2854
2855 (*id)++;
2856 seq_puts(f, "lock:\t");
2857 lock_get_status(f, fl, *id, "", 0);
2858 }
2859}
2860
2861void show_fd_locks(struct seq_file *f,
2862 struct file *filp, struct files_struct *files)
2863{
2864 struct inode *inode = locks_inode(filp);
2865 struct file_lock_context *ctx;
2866 int id = 0;
2867
2868 ctx = locks_inode_context(inode);
2869 if (!ctx)
2870 return;
2871
2872 spin_lock(&ctx->flc_lock);
2873 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2874 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2875 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2876 spin_unlock(&ctx->flc_lock);
2877}
2878
2879static void *locks_start(struct seq_file *f, loff_t *pos)
2880 __acquires(&blocked_lock_lock)
2881{
2882 struct locks_iterator *iter = f->private;
2883
2884 iter->li_pos = *pos + 1;
2885 percpu_down_write(&file_rwsem);
2886 spin_lock(&blocked_lock_lock);
2887 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2888}
2889
2890static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2891{
2892 struct locks_iterator *iter = f->private;
2893
2894 ++iter->li_pos;
2895 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2896}
2897
2898static void locks_stop(struct seq_file *f, void *v)
2899 __releases(&blocked_lock_lock)
2900{
2901 spin_unlock(&blocked_lock_lock);
2902 percpu_up_write(&file_rwsem);
2903}
2904
2905static const struct seq_operations locks_seq_operations = {
2906 .start = locks_start,
2907 .next = locks_next,
2908 .stop = locks_stop,
2909 .show = locks_show,
2910};
2911
2912static int __init proc_locks_init(void)
2913{
2914 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
2915 sizeof(struct locks_iterator), NULL);
2916 return 0;
2917}
2918fs_initcall(proc_locks_init);
2919#endif
2920
2921static int __init filelock_init(void)
2922{
2923 int i;
2924
2925 flctx_cache = kmem_cache_create("file_lock_ctx",
2926 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2927
2928 filelock_cache = kmem_cache_create("file_lock_cache",
2929 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2930
2931 for_each_possible_cpu(i) {
2932 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
2933
2934 spin_lock_init(&fll->lock);
2935 INIT_HLIST_HEAD(&fll->hlist);
2936 }
2937
2938 lease_notifier_chain_init();
2939 return 0;
2940}
2941core_initcall(filelock_init);