Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/locks.c
4 *
5 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
6 * Doug Evans (dje@spiff.uucp), August 07, 1992
7 *
8 * Deadlock detection added.
9 * FIXME: one thing isn't handled yet:
10 * - mandatory locks (requires lots of changes elsewhere)
11 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
12 *
13 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
14 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
15 *
16 * Converted file_lock_table to a linked list from an array, which eliminates
17 * the limits on how many active file locks are open.
18 * Chad Page (pageone@netcom.com), November 27, 1994
19 *
20 * Removed dependency on file descriptors. dup()'ed file descriptors now
21 * get the same locks as the original file descriptors, and a close() on
22 * any file descriptor removes ALL the locks on the file for the current
23 * process. Since locks still depend on the process id, locks are inherited
24 * after an exec() but not after a fork(). This agrees with POSIX, and both
25 * BSD and SVR4 practice.
26 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
27 *
28 * Scrapped free list which is redundant now that we allocate locks
29 * dynamically with kmalloc()/kfree().
30 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
31 *
32 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
33 *
34 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
35 * fcntl() system call. They have the semantics described above.
36 *
37 * FL_FLOCK locks are created with calls to flock(), through the flock()
38 * system call, which is new. Old C libraries implement flock() via fcntl()
39 * and will continue to use the old, broken implementation.
40 *
41 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
42 * with a file pointer (filp). As a result they can be shared by a parent
43 * process and its children after a fork(). They are removed when the last
44 * file descriptor referring to the file pointer is closed (unless explicitly
45 * unlocked).
46 *
47 * FL_FLOCK locks never deadlock, an existing lock is always removed before
48 * upgrading from shared to exclusive (or vice versa). When this happens
49 * any processes blocked by the current lock are woken up and allowed to
50 * run before the new lock is applied.
51 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
52 *
53 * Removed some race conditions in flock_lock_file(), marked other possible
54 * races. Just grep for FIXME to see them.
55 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
56 *
57 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
58 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
59 * once we've checked for blocking and deadlocking.
60 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
61 *
62 * Initial implementation of mandatory locks. SunOS turned out to be
63 * a rotten model, so I implemented the "obvious" semantics.
64 * See 'Documentation/filesystems/mandatory-locking.rst' for details.
65 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
66 *
67 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
68 * check if a file has mandatory locks, used by mmap(), open() and creat() to
69 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
70 * Manual, Section 2.
71 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
72 *
73 * Tidied up block list handling. Added '/proc/locks' interface.
74 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
75 *
76 * Fixed deadlock condition for pathological code that mixes calls to
77 * flock() and fcntl().
78 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
79 *
80 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
81 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
82 * guarantee sensible behaviour in the case where file system modules might
83 * be compiled with different options than the kernel itself.
84 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
85 *
86 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
87 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
88 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
89 *
90 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
91 * locks. Changed process synchronisation to avoid dereferencing locks that
92 * have already been freed.
93 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
94 *
95 * Made the block list a circular list to minimise searching in the list.
96 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
97 *
98 * Made mandatory locking a mount option. Default is not to allow mandatory
99 * locking.
100 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
101 *
102 * Some adaptations for NFS support.
103 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
104 *
105 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
106 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
107 *
108 * Use slab allocator instead of kmalloc/kfree.
109 * Use generic list implementation from <linux/list.h>.
110 * Sped up posix_locks_deadlock by only considering blocked locks.
111 * Matthew Wilcox <willy@debian.org>, March, 2000.
112 *
113 * Leases and LOCK_MAND
114 * Matthew Wilcox <willy@debian.org>, June, 2000.
115 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
116 *
117 * Locking conflicts and dependencies:
118 * If multiple threads attempt to lock the same byte (or flock the same file)
119 * only one can be granted the lock, and other must wait their turn.
120 * The first lock has been "applied" or "granted", the others are "waiting"
121 * and are "blocked" by the "applied" lock..
122 *
123 * Waiting and applied locks are all kept in trees whose properties are:
124 *
125 * - the root of a tree may be an applied or waiting lock.
126 * - every other node in the tree is a waiting lock that
127 * conflicts with every ancestor of that node.
128 *
129 * Every such tree begins life as a waiting singleton which obviously
130 * satisfies the above properties.
131 *
132 * The only ways we modify trees preserve these properties:
133 *
134 * 1. We may add a new leaf node, but only after first verifying that it
135 * conflicts with all of its ancestors.
136 * 2. We may remove the root of a tree, creating a new singleton
137 * tree from the root and N new trees rooted in the immediate
138 * children.
139 * 3. If the root of a tree is not currently an applied lock, we may
140 * apply it (if possible).
141 * 4. We may upgrade the root of the tree (either extend its range,
142 * or upgrade its entire range from read to write).
143 *
144 * When an applied lock is modified in a way that reduces or downgrades any
145 * part of its range, we remove all its children (2 above). This particularly
146 * happens when a lock is unlocked.
147 *
148 * For each of those child trees we "wake up" the thread which is
149 * waiting for the lock so it can continue handling as follows: if the
150 * root of the tree applies, we do so (3). If it doesn't, it must
151 * conflict with some applied lock. We remove (wake up) all of its children
152 * (2), and add it is a new leaf to the tree rooted in the applied
153 * lock (1). We then repeat the process recursively with those
154 * children.
155 *
156 */
157
158#include <linux/capability.h>
159#include <linux/file.h>
160#include <linux/fdtable.h>
161#include <linux/fs.h>
162#include <linux/init.h>
163#include <linux/security.h>
164#include <linux/slab.h>
165#include <linux/syscalls.h>
166#include <linux/time.h>
167#include <linux/rcupdate.h>
168#include <linux/pid_namespace.h>
169#include <linux/hashtable.h>
170#include <linux/percpu.h>
171
172#define CREATE_TRACE_POINTS
173#include <trace/events/filelock.h>
174
175#include <linux/uaccess.h>
176
177#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
178#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
179#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
180#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
181#define IS_REMOTELCK(fl) (fl->fl_pid <= 0)
182
183static bool lease_breaking(struct file_lock *fl)
184{
185 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
186}
187
188static int target_leasetype(struct file_lock *fl)
189{
190 if (fl->fl_flags & FL_UNLOCK_PENDING)
191 return F_UNLCK;
192 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
193 return F_RDLCK;
194 return fl->fl_type;
195}
196
197int leases_enable = 1;
198int lease_break_time = 45;
199
200/*
201 * The global file_lock_list is only used for displaying /proc/locks, so we
202 * keep a list on each CPU, with each list protected by its own spinlock.
203 * Global serialization is done using file_rwsem.
204 *
205 * Note that alterations to the list also require that the relevant flc_lock is
206 * held.
207 */
208struct file_lock_list_struct {
209 spinlock_t lock;
210 struct hlist_head hlist;
211};
212static DEFINE_PER_CPU(struct file_lock_list_struct, file_lock_list);
213DEFINE_STATIC_PERCPU_RWSEM(file_rwsem);
214
215
216/*
217 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
218 * It is protected by blocked_lock_lock.
219 *
220 * We hash locks by lockowner in order to optimize searching for the lock a
221 * particular lockowner is waiting on.
222 *
223 * FIXME: make this value scale via some heuristic? We generally will want more
224 * buckets when we have more lockowners holding locks, but that's a little
225 * difficult to determine without knowing what the workload will look like.
226 */
227#define BLOCKED_HASH_BITS 7
228static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
229
230/*
231 * This lock protects the blocked_hash. Generally, if you're accessing it, you
232 * want to be holding this lock.
233 *
234 * In addition, it also protects the fl->fl_blocked_requests list, and the
235 * fl->fl_blocker pointer for file_lock structures that are acting as lock
236 * requests (in contrast to those that are acting as records of acquired locks).
237 *
238 * Note that when we acquire this lock in order to change the above fields,
239 * we often hold the flc_lock as well. In certain cases, when reading the fields
240 * protected by this lock, we can skip acquiring it iff we already hold the
241 * flc_lock.
242 */
243static DEFINE_SPINLOCK(blocked_lock_lock);
244
245static struct kmem_cache *flctx_cache __read_mostly;
246static struct kmem_cache *filelock_cache __read_mostly;
247
248static struct file_lock_context *
249locks_get_lock_context(struct inode *inode, int type)
250{
251 struct file_lock_context *ctx;
252
253 /* paired with cmpxchg() below */
254 ctx = smp_load_acquire(&inode->i_flctx);
255 if (likely(ctx) || type == F_UNLCK)
256 goto out;
257
258 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
259 if (!ctx)
260 goto out;
261
262 spin_lock_init(&ctx->flc_lock);
263 INIT_LIST_HEAD(&ctx->flc_flock);
264 INIT_LIST_HEAD(&ctx->flc_posix);
265 INIT_LIST_HEAD(&ctx->flc_lease);
266
267 /*
268 * Assign the pointer if it's not already assigned. If it is, then
269 * free the context we just allocated.
270 */
271 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
272 kmem_cache_free(flctx_cache, ctx);
273 ctx = smp_load_acquire(&inode->i_flctx);
274 }
275out:
276 trace_locks_get_lock_context(inode, type, ctx);
277 return ctx;
278}
279
280static void
281locks_dump_ctx_list(struct list_head *list, char *list_type)
282{
283 struct file_lock *fl;
284
285 list_for_each_entry(fl, list, fl_list) {
286 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);
287 }
288}
289
290static void
291locks_check_ctx_lists(struct inode *inode)
292{
293 struct file_lock_context *ctx = inode->i_flctx;
294
295 if (unlikely(!list_empty(&ctx->flc_flock) ||
296 !list_empty(&ctx->flc_posix) ||
297 !list_empty(&ctx->flc_lease))) {
298 pr_warn("Leaked locks on dev=0x%x:0x%x ino=0x%lx:\n",
299 MAJOR(inode->i_sb->s_dev), MINOR(inode->i_sb->s_dev),
300 inode->i_ino);
301 locks_dump_ctx_list(&ctx->flc_flock, "FLOCK");
302 locks_dump_ctx_list(&ctx->flc_posix, "POSIX");
303 locks_dump_ctx_list(&ctx->flc_lease, "LEASE");
304 }
305}
306
307static void
308locks_check_ctx_file_list(struct file *filp, struct list_head *list,
309 char *list_type)
310{
311 struct file_lock *fl;
312 struct inode *inode = locks_inode(filp);
313
314 list_for_each_entry(fl, list, fl_list)
315 if (fl->fl_file == filp)
316 pr_warn("Leaked %s lock on dev=0x%x:0x%x ino=0x%lx "
317 " fl_owner=%p fl_flags=0x%x fl_type=0x%x fl_pid=%u\n",
318 list_type, MAJOR(inode->i_sb->s_dev),
319 MINOR(inode->i_sb->s_dev), inode->i_ino,
320 fl->fl_owner, fl->fl_flags, fl->fl_type, fl->fl_pid);
321}
322
323void
324locks_free_lock_context(struct inode *inode)
325{
326 struct file_lock_context *ctx = inode->i_flctx;
327
328 if (unlikely(ctx)) {
329 locks_check_ctx_lists(inode);
330 kmem_cache_free(flctx_cache, ctx);
331 }
332}
333
334static void locks_init_lock_heads(struct file_lock *fl)
335{
336 INIT_HLIST_NODE(&fl->fl_link);
337 INIT_LIST_HEAD(&fl->fl_list);
338 INIT_LIST_HEAD(&fl->fl_blocked_requests);
339 INIT_LIST_HEAD(&fl->fl_blocked_member);
340 init_waitqueue_head(&fl->fl_wait);
341}
342
343/* Allocate an empty lock structure. */
344struct file_lock *locks_alloc_lock(void)
345{
346 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
347
348 if (fl)
349 locks_init_lock_heads(fl);
350
351 return fl;
352}
353EXPORT_SYMBOL_GPL(locks_alloc_lock);
354
355void locks_release_private(struct file_lock *fl)
356{
357 BUG_ON(waitqueue_active(&fl->fl_wait));
358 BUG_ON(!list_empty(&fl->fl_list));
359 BUG_ON(!list_empty(&fl->fl_blocked_requests));
360 BUG_ON(!list_empty(&fl->fl_blocked_member));
361 BUG_ON(!hlist_unhashed(&fl->fl_link));
362
363 if (fl->fl_ops) {
364 if (fl->fl_ops->fl_release_private)
365 fl->fl_ops->fl_release_private(fl);
366 fl->fl_ops = NULL;
367 }
368
369 if (fl->fl_lmops) {
370 if (fl->fl_lmops->lm_put_owner) {
371 fl->fl_lmops->lm_put_owner(fl->fl_owner);
372 fl->fl_owner = NULL;
373 }
374 fl->fl_lmops = NULL;
375 }
376}
377EXPORT_SYMBOL_GPL(locks_release_private);
378
379/* Free a lock which is not in use. */
380void locks_free_lock(struct file_lock *fl)
381{
382 locks_release_private(fl);
383 kmem_cache_free(filelock_cache, fl);
384}
385EXPORT_SYMBOL(locks_free_lock);
386
387static void
388locks_dispose_list(struct list_head *dispose)
389{
390 struct file_lock *fl;
391
392 while (!list_empty(dispose)) {
393 fl = list_first_entry(dispose, struct file_lock, fl_list);
394 list_del_init(&fl->fl_list);
395 locks_free_lock(fl);
396 }
397}
398
399void locks_init_lock(struct file_lock *fl)
400{
401 memset(fl, 0, sizeof(struct file_lock));
402 locks_init_lock_heads(fl);
403}
404EXPORT_SYMBOL(locks_init_lock);
405
406/*
407 * Initialize a new lock from an existing file_lock structure.
408 */
409void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
410{
411 new->fl_owner = fl->fl_owner;
412 new->fl_pid = fl->fl_pid;
413 new->fl_file = NULL;
414 new->fl_flags = fl->fl_flags;
415 new->fl_type = fl->fl_type;
416 new->fl_start = fl->fl_start;
417 new->fl_end = fl->fl_end;
418 new->fl_lmops = fl->fl_lmops;
419 new->fl_ops = NULL;
420
421 if (fl->fl_lmops) {
422 if (fl->fl_lmops->lm_get_owner)
423 fl->fl_lmops->lm_get_owner(fl->fl_owner);
424 }
425}
426EXPORT_SYMBOL(locks_copy_conflock);
427
428void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
429{
430 /* "new" must be a freshly-initialized lock */
431 WARN_ON_ONCE(new->fl_ops);
432
433 locks_copy_conflock(new, fl);
434
435 new->fl_file = fl->fl_file;
436 new->fl_ops = fl->fl_ops;
437
438 if (fl->fl_ops) {
439 if (fl->fl_ops->fl_copy_lock)
440 fl->fl_ops->fl_copy_lock(new, fl);
441 }
442}
443EXPORT_SYMBOL(locks_copy_lock);
444
445static void locks_move_blocks(struct file_lock *new, struct file_lock *fl)
446{
447 struct file_lock *f;
448
449 /*
450 * As ctx->flc_lock is held, new requests cannot be added to
451 * ->fl_blocked_requests, so we don't need a lock to check if it
452 * is empty.
453 */
454 if (list_empty(&fl->fl_blocked_requests))
455 return;
456 spin_lock(&blocked_lock_lock);
457 list_splice_init(&fl->fl_blocked_requests, &new->fl_blocked_requests);
458 list_for_each_entry(f, &new->fl_blocked_requests, fl_blocked_member)
459 f->fl_blocker = new;
460 spin_unlock(&blocked_lock_lock);
461}
462
463static inline int flock_translate_cmd(int cmd) {
464 if (cmd & LOCK_MAND)
465 return cmd & (LOCK_MAND | LOCK_RW);
466 switch (cmd) {
467 case LOCK_SH:
468 return F_RDLCK;
469 case LOCK_EX:
470 return F_WRLCK;
471 case LOCK_UN:
472 return F_UNLCK;
473 }
474 return -EINVAL;
475}
476
477/* Fill in a file_lock structure with an appropriate FLOCK lock. */
478static struct file_lock *
479flock_make_lock(struct file *filp, unsigned int cmd, struct file_lock *fl)
480{
481 int type = flock_translate_cmd(cmd);
482
483 if (type < 0)
484 return ERR_PTR(type);
485
486 if (fl == NULL) {
487 fl = locks_alloc_lock();
488 if (fl == NULL)
489 return ERR_PTR(-ENOMEM);
490 } else {
491 locks_init_lock(fl);
492 }
493
494 fl->fl_file = filp;
495 fl->fl_owner = filp;
496 fl->fl_pid = current->tgid;
497 fl->fl_flags = FL_FLOCK;
498 fl->fl_type = type;
499 fl->fl_end = OFFSET_MAX;
500
501 return fl;
502}
503
504static int assign_type(struct file_lock *fl, long type)
505{
506 switch (type) {
507 case F_RDLCK:
508 case F_WRLCK:
509 case F_UNLCK:
510 fl->fl_type = type;
511 break;
512 default:
513 return -EINVAL;
514 }
515 return 0;
516}
517
518static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
519 struct flock64 *l)
520{
521 switch (l->l_whence) {
522 case SEEK_SET:
523 fl->fl_start = 0;
524 break;
525 case SEEK_CUR:
526 fl->fl_start = filp->f_pos;
527 break;
528 case SEEK_END:
529 fl->fl_start = i_size_read(file_inode(filp));
530 break;
531 default:
532 return -EINVAL;
533 }
534 if (l->l_start > OFFSET_MAX - fl->fl_start)
535 return -EOVERFLOW;
536 fl->fl_start += l->l_start;
537 if (fl->fl_start < 0)
538 return -EINVAL;
539
540 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
541 POSIX-2001 defines it. */
542 if (l->l_len > 0) {
543 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
544 return -EOVERFLOW;
545 fl->fl_end = fl->fl_start + l->l_len - 1;
546
547 } else if (l->l_len < 0) {
548 if (fl->fl_start + l->l_len < 0)
549 return -EINVAL;
550 fl->fl_end = fl->fl_start - 1;
551 fl->fl_start += l->l_len;
552 } else
553 fl->fl_end = OFFSET_MAX;
554
555 fl->fl_owner = current->files;
556 fl->fl_pid = current->tgid;
557 fl->fl_file = filp;
558 fl->fl_flags = FL_POSIX;
559 fl->fl_ops = NULL;
560 fl->fl_lmops = NULL;
561
562 return assign_type(fl, l->l_type);
563}
564
565/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
566 * style lock.
567 */
568static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
569 struct flock *l)
570{
571 struct flock64 ll = {
572 .l_type = l->l_type,
573 .l_whence = l->l_whence,
574 .l_start = l->l_start,
575 .l_len = l->l_len,
576 };
577
578 return flock64_to_posix_lock(filp, fl, &ll);
579}
580
581/* default lease lock manager operations */
582static bool
583lease_break_callback(struct file_lock *fl)
584{
585 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
586 return false;
587}
588
589static void
590lease_setup(struct file_lock *fl, void **priv)
591{
592 struct file *filp = fl->fl_file;
593 struct fasync_struct *fa = *priv;
594
595 /*
596 * fasync_insert_entry() returns the old entry if any. If there was no
597 * old entry, then it used "priv" and inserted it into the fasync list.
598 * Clear the pointer to indicate that it shouldn't be freed.
599 */
600 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
601 *priv = NULL;
602
603 __f_setown(filp, task_pid(current), PIDTYPE_TGID, 0);
604}
605
606static const struct lock_manager_operations lease_manager_ops = {
607 .lm_break = lease_break_callback,
608 .lm_change = lease_modify,
609 .lm_setup = lease_setup,
610};
611
612/*
613 * Initialize a lease, use the default lock manager operations
614 */
615static int lease_init(struct file *filp, long type, struct file_lock *fl)
616{
617 if (assign_type(fl, type) != 0)
618 return -EINVAL;
619
620 fl->fl_owner = filp;
621 fl->fl_pid = current->tgid;
622
623 fl->fl_file = filp;
624 fl->fl_flags = FL_LEASE;
625 fl->fl_start = 0;
626 fl->fl_end = OFFSET_MAX;
627 fl->fl_ops = NULL;
628 fl->fl_lmops = &lease_manager_ops;
629 return 0;
630}
631
632/* Allocate a file_lock initialised to this type of lease */
633static struct file_lock *lease_alloc(struct file *filp, long type)
634{
635 struct file_lock *fl = locks_alloc_lock();
636 int error = -ENOMEM;
637
638 if (fl == NULL)
639 return ERR_PTR(error);
640
641 error = lease_init(filp, type, fl);
642 if (error) {
643 locks_free_lock(fl);
644 return ERR_PTR(error);
645 }
646 return fl;
647}
648
649/* Check if two locks overlap each other.
650 */
651static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
652{
653 return ((fl1->fl_end >= fl2->fl_start) &&
654 (fl2->fl_end >= fl1->fl_start));
655}
656
657/*
658 * Check whether two locks have the same owner.
659 */
660static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
661{
662 return fl1->fl_owner == fl2->fl_owner;
663}
664
665/* Must be called with the flc_lock held! */
666static void locks_insert_global_locks(struct file_lock *fl)
667{
668 struct file_lock_list_struct *fll = this_cpu_ptr(&file_lock_list);
669
670 percpu_rwsem_assert_held(&file_rwsem);
671
672 spin_lock(&fll->lock);
673 fl->fl_link_cpu = smp_processor_id();
674 hlist_add_head(&fl->fl_link, &fll->hlist);
675 spin_unlock(&fll->lock);
676}
677
678/* Must be called with the flc_lock held! */
679static void locks_delete_global_locks(struct file_lock *fl)
680{
681 struct file_lock_list_struct *fll;
682
683 percpu_rwsem_assert_held(&file_rwsem);
684
685 /*
686 * Avoid taking lock if already unhashed. This is safe since this check
687 * is done while holding the flc_lock, and new insertions into the list
688 * also require that it be held.
689 */
690 if (hlist_unhashed(&fl->fl_link))
691 return;
692
693 fll = per_cpu_ptr(&file_lock_list, fl->fl_link_cpu);
694 spin_lock(&fll->lock);
695 hlist_del_init(&fl->fl_link);
696 spin_unlock(&fll->lock);
697}
698
699static unsigned long
700posix_owner_key(struct file_lock *fl)
701{
702 return (unsigned long)fl->fl_owner;
703}
704
705static void locks_insert_global_blocked(struct file_lock *waiter)
706{
707 lockdep_assert_held(&blocked_lock_lock);
708
709 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
710}
711
712static void locks_delete_global_blocked(struct file_lock *waiter)
713{
714 lockdep_assert_held(&blocked_lock_lock);
715
716 hash_del(&waiter->fl_link);
717}
718
719/* Remove waiter from blocker's block list.
720 * When blocker ends up pointing to itself then the list is empty.
721 *
722 * Must be called with blocked_lock_lock held.
723 */
724static void __locks_delete_block(struct file_lock *waiter)
725{
726 locks_delete_global_blocked(waiter);
727 list_del_init(&waiter->fl_blocked_member);
728}
729
730static void __locks_wake_up_blocks(struct file_lock *blocker)
731{
732 while (!list_empty(&blocker->fl_blocked_requests)) {
733 struct file_lock *waiter;
734
735 waiter = list_first_entry(&blocker->fl_blocked_requests,
736 struct file_lock, fl_blocked_member);
737 __locks_delete_block(waiter);
738 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
739 waiter->fl_lmops->lm_notify(waiter);
740 else
741 wake_up(&waiter->fl_wait);
742
743 /*
744 * The setting of fl_blocker to NULL marks the "done"
745 * point in deleting a block. Paired with acquire at the top
746 * of locks_delete_block().
747 */
748 smp_store_release(&waiter->fl_blocker, NULL);
749 }
750}
751
752/**
753 * locks_delete_lock - stop waiting for a file lock
754 * @waiter: the lock which was waiting
755 *
756 * lockd/nfsd need to disconnect the lock while working on it.
757 */
758int locks_delete_block(struct file_lock *waiter)
759{
760 int status = -ENOENT;
761
762 /*
763 * If fl_blocker is NULL, it won't be set again as this thread "owns"
764 * the lock and is the only one that might try to claim the lock.
765 *
766 * We use acquire/release to manage fl_blocker so that we can
767 * optimize away taking the blocked_lock_lock in many cases.
768 *
769 * The smp_load_acquire guarantees two things:
770 *
771 * 1/ that fl_blocked_requests can be tested locklessly. If something
772 * was recently added to that list it must have been in a locked region
773 * *before* the locked region when fl_blocker was set to NULL.
774 *
775 * 2/ that no other thread is accessing 'waiter', so it is safe to free
776 * it. __locks_wake_up_blocks is careful not to touch waiter after
777 * fl_blocker is released.
778 *
779 * If a lockless check of fl_blocker shows it to be NULL, we know that
780 * no new locks can be inserted into its fl_blocked_requests list, and
781 * can avoid doing anything further if the list is empty.
782 */
783 if (!smp_load_acquire(&waiter->fl_blocker) &&
784 list_empty(&waiter->fl_blocked_requests))
785 return status;
786
787 spin_lock(&blocked_lock_lock);
788 if (waiter->fl_blocker)
789 status = 0;
790 __locks_wake_up_blocks(waiter);
791 __locks_delete_block(waiter);
792
793 /*
794 * The setting of fl_blocker to NULL marks the "done" point in deleting
795 * a block. Paired with acquire at the top of this function.
796 */
797 smp_store_release(&waiter->fl_blocker, NULL);
798 spin_unlock(&blocked_lock_lock);
799 return status;
800}
801EXPORT_SYMBOL(locks_delete_block);
802
803/* Insert waiter into blocker's block list.
804 * We use a circular list so that processes can be easily woken up in
805 * the order they blocked. The documentation doesn't require this but
806 * it seems like the reasonable thing to do.
807 *
808 * Must be called with both the flc_lock and blocked_lock_lock held. The
809 * fl_blocked_requests list itself is protected by the blocked_lock_lock,
810 * but by ensuring that the flc_lock is also held on insertions we can avoid
811 * taking the blocked_lock_lock in some cases when we see that the
812 * fl_blocked_requests list is empty.
813 *
814 * Rather than just adding to the list, we check for conflicts with any existing
815 * waiters, and add beneath any waiter that blocks the new waiter.
816 * Thus wakeups don't happen until needed.
817 */
818static void __locks_insert_block(struct file_lock *blocker,
819 struct file_lock *waiter,
820 bool conflict(struct file_lock *,
821 struct file_lock *))
822{
823 struct file_lock *fl;
824 BUG_ON(!list_empty(&waiter->fl_blocked_member));
825
826new_blocker:
827 list_for_each_entry(fl, &blocker->fl_blocked_requests, fl_blocked_member)
828 if (conflict(fl, waiter)) {
829 blocker = fl;
830 goto new_blocker;
831 }
832 waiter->fl_blocker = blocker;
833 list_add_tail(&waiter->fl_blocked_member, &blocker->fl_blocked_requests);
834 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
835 locks_insert_global_blocked(waiter);
836
837 /* The requests in waiter->fl_blocked are known to conflict with
838 * waiter, but might not conflict with blocker, or the requests
839 * and lock which block it. So they all need to be woken.
840 */
841 __locks_wake_up_blocks(waiter);
842}
843
844/* Must be called with flc_lock held. */
845static void locks_insert_block(struct file_lock *blocker,
846 struct file_lock *waiter,
847 bool conflict(struct file_lock *,
848 struct file_lock *))
849{
850 spin_lock(&blocked_lock_lock);
851 __locks_insert_block(blocker, waiter, conflict);
852 spin_unlock(&blocked_lock_lock);
853}
854
855/*
856 * Wake up processes blocked waiting for blocker.
857 *
858 * Must be called with the inode->flc_lock held!
859 */
860static void locks_wake_up_blocks(struct file_lock *blocker)
861{
862 /*
863 * Avoid taking global lock if list is empty. This is safe since new
864 * blocked requests are only added to the list under the flc_lock, and
865 * the flc_lock is always held here. Note that removal from the
866 * fl_blocked_requests list does not require the flc_lock, so we must
867 * recheck list_empty() after acquiring the blocked_lock_lock.
868 */
869 if (list_empty(&blocker->fl_blocked_requests))
870 return;
871
872 spin_lock(&blocked_lock_lock);
873 __locks_wake_up_blocks(blocker);
874 spin_unlock(&blocked_lock_lock);
875}
876
877static void
878locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
879{
880 list_add_tail(&fl->fl_list, before);
881 locks_insert_global_locks(fl);
882}
883
884static void
885locks_unlink_lock_ctx(struct file_lock *fl)
886{
887 locks_delete_global_locks(fl);
888 list_del_init(&fl->fl_list);
889 locks_wake_up_blocks(fl);
890}
891
892static void
893locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
894{
895 locks_unlink_lock_ctx(fl);
896 if (dispose)
897 list_add(&fl->fl_list, dispose);
898 else
899 locks_free_lock(fl);
900}
901
902/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
903 * checks for shared/exclusive status of overlapping locks.
904 */
905static bool locks_conflict(struct file_lock *caller_fl,
906 struct file_lock *sys_fl)
907{
908 if (sys_fl->fl_type == F_WRLCK)
909 return true;
910 if (caller_fl->fl_type == F_WRLCK)
911 return true;
912 return false;
913}
914
915/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
916 * checking before calling the locks_conflict().
917 */
918static bool posix_locks_conflict(struct file_lock *caller_fl,
919 struct file_lock *sys_fl)
920{
921 /* POSIX locks owned by the same process do not conflict with
922 * each other.
923 */
924 if (posix_same_owner(caller_fl, sys_fl))
925 return false;
926
927 /* Check whether they overlap */
928 if (!locks_overlap(caller_fl, sys_fl))
929 return false;
930
931 return locks_conflict(caller_fl, sys_fl);
932}
933
934/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
935 * checking before calling the locks_conflict().
936 */
937static bool flock_locks_conflict(struct file_lock *caller_fl,
938 struct file_lock *sys_fl)
939{
940 /* FLOCK locks referring to the same filp do not conflict with
941 * each other.
942 */
943 if (caller_fl->fl_file == sys_fl->fl_file)
944 return false;
945 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
946 return false;
947
948 return locks_conflict(caller_fl, sys_fl);
949}
950
951void
952posix_test_lock(struct file *filp, struct file_lock *fl)
953{
954 struct file_lock *cfl;
955 struct file_lock_context *ctx;
956 struct inode *inode = locks_inode(filp);
957
958 ctx = smp_load_acquire(&inode->i_flctx);
959 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
960 fl->fl_type = F_UNLCK;
961 return;
962 }
963
964 spin_lock(&ctx->flc_lock);
965 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
966 if (posix_locks_conflict(fl, cfl)) {
967 locks_copy_conflock(fl, cfl);
968 goto out;
969 }
970 }
971 fl->fl_type = F_UNLCK;
972out:
973 spin_unlock(&ctx->flc_lock);
974 return;
975}
976EXPORT_SYMBOL(posix_test_lock);
977
978/*
979 * Deadlock detection:
980 *
981 * We attempt to detect deadlocks that are due purely to posix file
982 * locks.
983 *
984 * We assume that a task can be waiting for at most one lock at a time.
985 * So for any acquired lock, the process holding that lock may be
986 * waiting on at most one other lock. That lock in turns may be held by
987 * someone waiting for at most one other lock. Given a requested lock
988 * caller_fl which is about to wait for a conflicting lock block_fl, we
989 * follow this chain of waiters to ensure we are not about to create a
990 * cycle.
991 *
992 * Since we do this before we ever put a process to sleep on a lock, we
993 * are ensured that there is never a cycle; that is what guarantees that
994 * the while() loop in posix_locks_deadlock() eventually completes.
995 *
996 * Note: the above assumption may not be true when handling lock
997 * requests from a broken NFS client. It may also fail in the presence
998 * of tasks (such as posix threads) sharing the same open file table.
999 * To handle those cases, we just bail out after a few iterations.
1000 *
1001 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
1002 * Because the owner is not even nominally tied to a thread of
1003 * execution, the deadlock detection below can't reasonably work well. Just
1004 * skip it for those.
1005 *
1006 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
1007 * locks that just checks for the case where two tasks are attempting to
1008 * upgrade from read to write locks on the same inode.
1009 */
1010
1011#define MAX_DEADLK_ITERATIONS 10
1012
1013/* Find a lock that the owner of the given block_fl is blocking on. */
1014static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
1015{
1016 struct file_lock *fl;
1017
1018 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
1019 if (posix_same_owner(fl, block_fl)) {
1020 while (fl->fl_blocker)
1021 fl = fl->fl_blocker;
1022 return fl;
1023 }
1024 }
1025 return NULL;
1026}
1027
1028/* Must be called with the blocked_lock_lock held! */
1029static int posix_locks_deadlock(struct file_lock *caller_fl,
1030 struct file_lock *block_fl)
1031{
1032 int i = 0;
1033
1034 lockdep_assert_held(&blocked_lock_lock);
1035
1036 /*
1037 * This deadlock detector can't reasonably detect deadlocks with
1038 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
1039 */
1040 if (IS_OFDLCK(caller_fl))
1041 return 0;
1042
1043 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
1044 if (i++ > MAX_DEADLK_ITERATIONS)
1045 return 0;
1046 if (posix_same_owner(caller_fl, block_fl))
1047 return 1;
1048 }
1049 return 0;
1050}
1051
1052/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
1053 * after any leases, but before any posix locks.
1054 *
1055 * Note that if called with an FL_EXISTS argument, the caller may determine
1056 * whether or not a lock was successfully freed by testing the return
1057 * value for -ENOENT.
1058 */
1059static int flock_lock_inode(struct inode *inode, struct file_lock *request)
1060{
1061 struct file_lock *new_fl = NULL;
1062 struct file_lock *fl;
1063 struct file_lock_context *ctx;
1064 int error = 0;
1065 bool found = false;
1066 LIST_HEAD(dispose);
1067
1068 ctx = locks_get_lock_context(inode, request->fl_type);
1069 if (!ctx) {
1070 if (request->fl_type != F_UNLCK)
1071 return -ENOMEM;
1072 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
1073 }
1074
1075 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
1076 new_fl = locks_alloc_lock();
1077 if (!new_fl)
1078 return -ENOMEM;
1079 }
1080
1081 percpu_down_read(&file_rwsem);
1082 spin_lock(&ctx->flc_lock);
1083 if (request->fl_flags & FL_ACCESS)
1084 goto find_conflict;
1085
1086 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1087 if (request->fl_file != fl->fl_file)
1088 continue;
1089 if (request->fl_type == fl->fl_type)
1090 goto out;
1091 found = true;
1092 locks_delete_lock_ctx(fl, &dispose);
1093 break;
1094 }
1095
1096 if (request->fl_type == F_UNLCK) {
1097 if ((request->fl_flags & FL_EXISTS) && !found)
1098 error = -ENOENT;
1099 goto out;
1100 }
1101
1102find_conflict:
1103 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
1104 if (!flock_locks_conflict(request, fl))
1105 continue;
1106 error = -EAGAIN;
1107 if (!(request->fl_flags & FL_SLEEP))
1108 goto out;
1109 error = FILE_LOCK_DEFERRED;
1110 locks_insert_block(fl, request, flock_locks_conflict);
1111 goto out;
1112 }
1113 if (request->fl_flags & FL_ACCESS)
1114 goto out;
1115 locks_copy_lock(new_fl, request);
1116 locks_move_blocks(new_fl, request);
1117 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
1118 new_fl = NULL;
1119 error = 0;
1120
1121out:
1122 spin_unlock(&ctx->flc_lock);
1123 percpu_up_read(&file_rwsem);
1124 if (new_fl)
1125 locks_free_lock(new_fl);
1126 locks_dispose_list(&dispose);
1127 trace_flock_lock_inode(inode, request, error);
1128 return error;
1129}
1130
1131static int posix_lock_inode(struct inode *inode, struct file_lock *request,
1132 struct file_lock *conflock)
1133{
1134 struct file_lock *fl, *tmp;
1135 struct file_lock *new_fl = NULL;
1136 struct file_lock *new_fl2 = NULL;
1137 struct file_lock *left = NULL;
1138 struct file_lock *right = NULL;
1139 struct file_lock_context *ctx;
1140 int error;
1141 bool added = false;
1142 LIST_HEAD(dispose);
1143
1144 ctx = locks_get_lock_context(inode, request->fl_type);
1145 if (!ctx)
1146 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
1147
1148 /*
1149 * We may need two file_lock structures for this operation,
1150 * so we get them in advance to avoid races.
1151 *
1152 * In some cases we can be sure, that no new locks will be needed
1153 */
1154 if (!(request->fl_flags & FL_ACCESS) &&
1155 (request->fl_type != F_UNLCK ||
1156 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
1157 new_fl = locks_alloc_lock();
1158 new_fl2 = locks_alloc_lock();
1159 }
1160
1161 percpu_down_read(&file_rwsem);
1162 spin_lock(&ctx->flc_lock);
1163 /*
1164 * New lock request. Walk all POSIX locks and look for conflicts. If
1165 * there are any, either return error or put the request on the
1166 * blocker's list of waiters and the global blocked_hash.
1167 */
1168 if (request->fl_type != F_UNLCK) {
1169 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1170 if (!posix_locks_conflict(request, fl))
1171 continue;
1172 if (conflock)
1173 locks_copy_conflock(conflock, fl);
1174 error = -EAGAIN;
1175 if (!(request->fl_flags & FL_SLEEP))
1176 goto out;
1177 /*
1178 * Deadlock detection and insertion into the blocked
1179 * locks list must be done while holding the same lock!
1180 */
1181 error = -EDEADLK;
1182 spin_lock(&blocked_lock_lock);
1183 /*
1184 * Ensure that we don't find any locks blocked on this
1185 * request during deadlock detection.
1186 */
1187 __locks_wake_up_blocks(request);
1188 if (likely(!posix_locks_deadlock(request, fl))) {
1189 error = FILE_LOCK_DEFERRED;
1190 __locks_insert_block(fl, request,
1191 posix_locks_conflict);
1192 }
1193 spin_unlock(&blocked_lock_lock);
1194 goto out;
1195 }
1196 }
1197
1198 /* If we're just looking for a conflict, we're done. */
1199 error = 0;
1200 if (request->fl_flags & FL_ACCESS)
1201 goto out;
1202
1203 /* Find the first old lock with the same owner as the new lock */
1204 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1205 if (posix_same_owner(request, fl))
1206 break;
1207 }
1208
1209 /* Process locks with this owner. */
1210 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1211 if (!posix_same_owner(request, fl))
1212 break;
1213
1214 /* Detect adjacent or overlapping regions (if same lock type) */
1215 if (request->fl_type == fl->fl_type) {
1216 /* In all comparisons of start vs end, use
1217 * "start - 1" rather than "end + 1". If end
1218 * is OFFSET_MAX, end + 1 will become negative.
1219 */
1220 if (fl->fl_end < request->fl_start - 1)
1221 continue;
1222 /* If the next lock in the list has entirely bigger
1223 * addresses than the new one, insert the lock here.
1224 */
1225 if (fl->fl_start - 1 > request->fl_end)
1226 break;
1227
1228 /* If we come here, the new and old lock are of the
1229 * same type and adjacent or overlapping. Make one
1230 * lock yielding from the lower start address of both
1231 * locks to the higher end address.
1232 */
1233 if (fl->fl_start > request->fl_start)
1234 fl->fl_start = request->fl_start;
1235 else
1236 request->fl_start = fl->fl_start;
1237 if (fl->fl_end < request->fl_end)
1238 fl->fl_end = request->fl_end;
1239 else
1240 request->fl_end = fl->fl_end;
1241 if (added) {
1242 locks_delete_lock_ctx(fl, &dispose);
1243 continue;
1244 }
1245 request = fl;
1246 added = true;
1247 } else {
1248 /* Processing for different lock types is a bit
1249 * more complex.
1250 */
1251 if (fl->fl_end < request->fl_start)
1252 continue;
1253 if (fl->fl_start > request->fl_end)
1254 break;
1255 if (request->fl_type == F_UNLCK)
1256 added = true;
1257 if (fl->fl_start < request->fl_start)
1258 left = fl;
1259 /* If the next lock in the list has a higher end
1260 * address than the new one, insert the new one here.
1261 */
1262 if (fl->fl_end > request->fl_end) {
1263 right = fl;
1264 break;
1265 }
1266 if (fl->fl_start >= request->fl_start) {
1267 /* The new lock completely replaces an old
1268 * one (This may happen several times).
1269 */
1270 if (added) {
1271 locks_delete_lock_ctx(fl, &dispose);
1272 continue;
1273 }
1274 /*
1275 * Replace the old lock with new_fl, and
1276 * remove the old one. It's safe to do the
1277 * insert here since we know that we won't be
1278 * using new_fl later, and that the lock is
1279 * just replacing an existing lock.
1280 */
1281 error = -ENOLCK;
1282 if (!new_fl)
1283 goto out;
1284 locks_copy_lock(new_fl, request);
1285 locks_move_blocks(new_fl, request);
1286 request = new_fl;
1287 new_fl = NULL;
1288 locks_insert_lock_ctx(request, &fl->fl_list);
1289 locks_delete_lock_ctx(fl, &dispose);
1290 added = true;
1291 }
1292 }
1293 }
1294
1295 /*
1296 * The above code only modifies existing locks in case of merging or
1297 * replacing. If new lock(s) need to be inserted all modifications are
1298 * done below this, so it's safe yet to bail out.
1299 */
1300 error = -ENOLCK; /* "no luck" */
1301 if (right && left == right && !new_fl2)
1302 goto out;
1303
1304 error = 0;
1305 if (!added) {
1306 if (request->fl_type == F_UNLCK) {
1307 if (request->fl_flags & FL_EXISTS)
1308 error = -ENOENT;
1309 goto out;
1310 }
1311
1312 if (!new_fl) {
1313 error = -ENOLCK;
1314 goto out;
1315 }
1316 locks_copy_lock(new_fl, request);
1317 locks_move_blocks(new_fl, request);
1318 locks_insert_lock_ctx(new_fl, &fl->fl_list);
1319 fl = new_fl;
1320 new_fl = NULL;
1321 }
1322 if (right) {
1323 if (left == right) {
1324 /* The new lock breaks the old one in two pieces,
1325 * so we have to use the second new lock.
1326 */
1327 left = new_fl2;
1328 new_fl2 = NULL;
1329 locks_copy_lock(left, right);
1330 locks_insert_lock_ctx(left, &fl->fl_list);
1331 }
1332 right->fl_start = request->fl_end + 1;
1333 locks_wake_up_blocks(right);
1334 }
1335 if (left) {
1336 left->fl_end = request->fl_start - 1;
1337 locks_wake_up_blocks(left);
1338 }
1339 out:
1340 spin_unlock(&ctx->flc_lock);
1341 percpu_up_read(&file_rwsem);
1342 /*
1343 * Free any unused locks.
1344 */
1345 if (new_fl)
1346 locks_free_lock(new_fl);
1347 if (new_fl2)
1348 locks_free_lock(new_fl2);
1349 locks_dispose_list(&dispose);
1350 trace_posix_lock_inode(inode, request, error);
1351
1352 return error;
1353}
1354
1355/**
1356 * posix_lock_file - Apply a POSIX-style lock to a file
1357 * @filp: The file to apply the lock to
1358 * @fl: The lock to be applied
1359 * @conflock: Place to return a copy of the conflicting lock, if found.
1360 *
1361 * Add a POSIX style lock to a file.
1362 * We merge adjacent & overlapping locks whenever possible.
1363 * POSIX locks are sorted by owner task, then by starting address
1364 *
1365 * Note that if called with an FL_EXISTS argument, the caller may determine
1366 * whether or not a lock was successfully freed by testing the return
1367 * value for -ENOENT.
1368 */
1369int posix_lock_file(struct file *filp, struct file_lock *fl,
1370 struct file_lock *conflock)
1371{
1372 return posix_lock_inode(locks_inode(filp), fl, conflock);
1373}
1374EXPORT_SYMBOL(posix_lock_file);
1375
1376/**
1377 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1378 * @inode: inode of file to which lock request should be applied
1379 * @fl: The lock to be applied
1380 *
1381 * Apply a POSIX style lock request to an inode.
1382 */
1383static int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1384{
1385 int error;
1386 might_sleep ();
1387 for (;;) {
1388 error = posix_lock_inode(inode, fl, NULL);
1389 if (error != FILE_LOCK_DEFERRED)
1390 break;
1391 error = wait_event_interruptible(fl->fl_wait,
1392 list_empty(&fl->fl_blocked_member));
1393 if (error)
1394 break;
1395 }
1396 locks_delete_block(fl);
1397 return error;
1398}
1399
1400#ifdef CONFIG_MANDATORY_FILE_LOCKING
1401/**
1402 * locks_mandatory_locked - Check for an active lock
1403 * @file: the file to check
1404 *
1405 * Searches the inode's list of locks to find any POSIX locks which conflict.
1406 * This function is called from locks_verify_locked() only.
1407 */
1408int locks_mandatory_locked(struct file *file)
1409{
1410 int ret;
1411 struct inode *inode = locks_inode(file);
1412 struct file_lock_context *ctx;
1413 struct file_lock *fl;
1414
1415 ctx = smp_load_acquire(&inode->i_flctx);
1416 if (!ctx || list_empty_careful(&ctx->flc_posix))
1417 return 0;
1418
1419 /*
1420 * Search the lock list for this inode for any POSIX locks.
1421 */
1422 spin_lock(&ctx->flc_lock);
1423 ret = 0;
1424 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1425 if (fl->fl_owner != current->files &&
1426 fl->fl_owner != file) {
1427 ret = -EAGAIN;
1428 break;
1429 }
1430 }
1431 spin_unlock(&ctx->flc_lock);
1432 return ret;
1433}
1434
1435/**
1436 * locks_mandatory_area - Check for a conflicting lock
1437 * @inode: the file to check
1438 * @filp: how the file was opened (if it was)
1439 * @start: first byte in the file to check
1440 * @end: lastbyte in the file to check
1441 * @type: %F_WRLCK for a write lock, else %F_RDLCK
1442 *
1443 * Searches the inode's list of locks to find any POSIX locks which conflict.
1444 */
1445int locks_mandatory_area(struct inode *inode, struct file *filp, loff_t start,
1446 loff_t end, unsigned char type)
1447{
1448 struct file_lock fl;
1449 int error;
1450 bool sleep = false;
1451
1452 locks_init_lock(&fl);
1453 fl.fl_pid = current->tgid;
1454 fl.fl_file = filp;
1455 fl.fl_flags = FL_POSIX | FL_ACCESS;
1456 if (filp && !(filp->f_flags & O_NONBLOCK))
1457 sleep = true;
1458 fl.fl_type = type;
1459 fl.fl_start = start;
1460 fl.fl_end = end;
1461
1462 for (;;) {
1463 if (filp) {
1464 fl.fl_owner = filp;
1465 fl.fl_flags &= ~FL_SLEEP;
1466 error = posix_lock_inode(inode, &fl, NULL);
1467 if (!error)
1468 break;
1469 }
1470
1471 if (sleep)
1472 fl.fl_flags |= FL_SLEEP;
1473 fl.fl_owner = current->files;
1474 error = posix_lock_inode(inode, &fl, NULL);
1475 if (error != FILE_LOCK_DEFERRED)
1476 break;
1477 error = wait_event_interruptible(fl.fl_wait,
1478 list_empty(&fl.fl_blocked_member));
1479 if (!error) {
1480 /*
1481 * If we've been sleeping someone might have
1482 * changed the permissions behind our back.
1483 */
1484 if (__mandatory_lock(inode))
1485 continue;
1486 }
1487
1488 break;
1489 }
1490 locks_delete_block(&fl);
1491
1492 return error;
1493}
1494EXPORT_SYMBOL(locks_mandatory_area);
1495#endif /* CONFIG_MANDATORY_FILE_LOCKING */
1496
1497static void lease_clear_pending(struct file_lock *fl, int arg)
1498{
1499 switch (arg) {
1500 case F_UNLCK:
1501 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1502 fallthrough;
1503 case F_RDLCK:
1504 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1505 }
1506}
1507
1508/* We already had a lease on this file; just change its type */
1509int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
1510{
1511 int error = assign_type(fl, arg);
1512
1513 if (error)
1514 return error;
1515 lease_clear_pending(fl, arg);
1516 locks_wake_up_blocks(fl);
1517 if (arg == F_UNLCK) {
1518 struct file *filp = fl->fl_file;
1519
1520 f_delown(filp);
1521 filp->f_owner.signum = 0;
1522 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1523 if (fl->fl_fasync != NULL) {
1524 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1525 fl->fl_fasync = NULL;
1526 }
1527 locks_delete_lock_ctx(fl, dispose);
1528 }
1529 return 0;
1530}
1531EXPORT_SYMBOL(lease_modify);
1532
1533static bool past_time(unsigned long then)
1534{
1535 if (!then)
1536 /* 0 is a special value meaning "this never expires": */
1537 return false;
1538 return time_after(jiffies, then);
1539}
1540
1541static void time_out_leases(struct inode *inode, struct list_head *dispose)
1542{
1543 struct file_lock_context *ctx = inode->i_flctx;
1544 struct file_lock *fl, *tmp;
1545
1546 lockdep_assert_held(&ctx->flc_lock);
1547
1548 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1549 trace_time_out_leases(inode, fl);
1550 if (past_time(fl->fl_downgrade_time))
1551 lease_modify(fl, F_RDLCK, dispose);
1552 if (past_time(fl->fl_break_time))
1553 lease_modify(fl, F_UNLCK, dispose);
1554 }
1555}
1556
1557static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1558{
1559 bool rc;
1560
1561 if (lease->fl_lmops->lm_breaker_owns_lease
1562 && lease->fl_lmops->lm_breaker_owns_lease(lease))
1563 return false;
1564 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT)) {
1565 rc = false;
1566 goto trace;
1567 }
1568 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE)) {
1569 rc = false;
1570 goto trace;
1571 }
1572
1573 rc = locks_conflict(breaker, lease);
1574trace:
1575 trace_leases_conflict(rc, lease, breaker);
1576 return rc;
1577}
1578
1579static bool
1580any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1581{
1582 struct file_lock_context *ctx = inode->i_flctx;
1583 struct file_lock *fl;
1584
1585 lockdep_assert_held(&ctx->flc_lock);
1586
1587 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1588 if (leases_conflict(fl, breaker))
1589 return true;
1590 }
1591 return false;
1592}
1593
1594/**
1595 * __break_lease - revoke all outstanding leases on file
1596 * @inode: the inode of the file to return
1597 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1598 * break all leases
1599 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1600 * only delegations
1601 *
1602 * break_lease (inlined for speed) has checked there already is at least
1603 * some kind of lock (maybe a lease) on this file. Leases are broken on
1604 * a call to open() or truncate(). This function can sleep unless you
1605 * specified %O_NONBLOCK to your open().
1606 */
1607int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
1608{
1609 int error = 0;
1610 struct file_lock_context *ctx;
1611 struct file_lock *new_fl, *fl, *tmp;
1612 unsigned long break_time;
1613 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1614 LIST_HEAD(dispose);
1615
1616 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
1617 if (IS_ERR(new_fl))
1618 return PTR_ERR(new_fl);
1619 new_fl->fl_flags = type;
1620
1621 /* typically we will check that ctx is non-NULL before calling */
1622 ctx = smp_load_acquire(&inode->i_flctx);
1623 if (!ctx) {
1624 WARN_ON_ONCE(1);
1625 goto free_lock;
1626 }
1627
1628 percpu_down_read(&file_rwsem);
1629 spin_lock(&ctx->flc_lock);
1630
1631 time_out_leases(inode, &dispose);
1632
1633 if (!any_leases_conflict(inode, new_fl))
1634 goto out;
1635
1636 break_time = 0;
1637 if (lease_break_time > 0) {
1638 break_time = jiffies + lease_break_time * HZ;
1639 if (break_time == 0)
1640 break_time++; /* so that 0 means no break time */
1641 }
1642
1643 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
1644 if (!leases_conflict(fl, new_fl))
1645 continue;
1646 if (want_write) {
1647 if (fl->fl_flags & FL_UNLOCK_PENDING)
1648 continue;
1649 fl->fl_flags |= FL_UNLOCK_PENDING;
1650 fl->fl_break_time = break_time;
1651 } else {
1652 if (lease_breaking(fl))
1653 continue;
1654 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1655 fl->fl_downgrade_time = break_time;
1656 }
1657 if (fl->fl_lmops->lm_break(fl))
1658 locks_delete_lock_ctx(fl, &dispose);
1659 }
1660
1661 if (list_empty(&ctx->flc_lease))
1662 goto out;
1663
1664 if (mode & O_NONBLOCK) {
1665 trace_break_lease_noblock(inode, new_fl);
1666 error = -EWOULDBLOCK;
1667 goto out;
1668 }
1669
1670restart:
1671 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1672 break_time = fl->fl_break_time;
1673 if (break_time != 0)
1674 break_time -= jiffies;
1675 if (break_time == 0)
1676 break_time++;
1677 locks_insert_block(fl, new_fl, leases_conflict);
1678 trace_break_lease_block(inode, new_fl);
1679 spin_unlock(&ctx->flc_lock);
1680 percpu_up_read(&file_rwsem);
1681
1682 locks_dispose_list(&dispose);
1683 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1684 list_empty(&new_fl->fl_blocked_member),
1685 break_time);
1686
1687 percpu_down_read(&file_rwsem);
1688 spin_lock(&ctx->flc_lock);
1689 trace_break_lease_unblock(inode, new_fl);
1690 locks_delete_block(new_fl);
1691 if (error >= 0) {
1692 /*
1693 * Wait for the next conflicting lease that has not been
1694 * broken yet
1695 */
1696 if (error == 0)
1697 time_out_leases(inode, &dispose);
1698 if (any_leases_conflict(inode, new_fl))
1699 goto restart;
1700 error = 0;
1701 }
1702out:
1703 spin_unlock(&ctx->flc_lock);
1704 percpu_up_read(&file_rwsem);
1705 locks_dispose_list(&dispose);
1706free_lock:
1707 locks_free_lock(new_fl);
1708 return error;
1709}
1710EXPORT_SYMBOL(__break_lease);
1711
1712/**
1713 * lease_get_mtime - update modified time of an inode with exclusive lease
1714 * @inode: the inode
1715 * @time: pointer to a timespec which contains the last modified time
1716 *
1717 * This is to force NFS clients to flush their caches for files with
1718 * exclusive leases. The justification is that if someone has an
1719 * exclusive lease, then they could be modifying it.
1720 */
1721void lease_get_mtime(struct inode *inode, struct timespec64 *time)
1722{
1723 bool has_lease = false;
1724 struct file_lock_context *ctx;
1725 struct file_lock *fl;
1726
1727 ctx = smp_load_acquire(&inode->i_flctx);
1728 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1729 spin_lock(&ctx->flc_lock);
1730 fl = list_first_entry_or_null(&ctx->flc_lease,
1731 struct file_lock, fl_list);
1732 if (fl && (fl->fl_type == F_WRLCK))
1733 has_lease = true;
1734 spin_unlock(&ctx->flc_lock);
1735 }
1736
1737 if (has_lease)
1738 *time = current_time(inode);
1739}
1740EXPORT_SYMBOL(lease_get_mtime);
1741
1742/**
1743 * fcntl_getlease - Enquire what lease is currently active
1744 * @filp: the file
1745 *
1746 * The value returned by this function will be one of
1747 * (if no lease break is pending):
1748 *
1749 * %F_RDLCK to indicate a shared lease is held.
1750 *
1751 * %F_WRLCK to indicate an exclusive lease is held.
1752 *
1753 * %F_UNLCK to indicate no lease is held.
1754 *
1755 * (if a lease break is pending):
1756 *
1757 * %F_RDLCK to indicate an exclusive lease needs to be
1758 * changed to a shared lease (or removed).
1759 *
1760 * %F_UNLCK to indicate the lease needs to be removed.
1761 *
1762 * XXX: sfr & willy disagree over whether F_INPROGRESS
1763 * should be returned to userspace.
1764 */
1765int fcntl_getlease(struct file *filp)
1766{
1767 struct file_lock *fl;
1768 struct inode *inode = locks_inode(filp);
1769 struct file_lock_context *ctx;
1770 int type = F_UNLCK;
1771 LIST_HEAD(dispose);
1772
1773 ctx = smp_load_acquire(&inode->i_flctx);
1774 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1775 percpu_down_read(&file_rwsem);
1776 spin_lock(&ctx->flc_lock);
1777 time_out_leases(inode, &dispose);
1778 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1779 if (fl->fl_file != filp)
1780 continue;
1781 type = target_leasetype(fl);
1782 break;
1783 }
1784 spin_unlock(&ctx->flc_lock);
1785 percpu_up_read(&file_rwsem);
1786
1787 locks_dispose_list(&dispose);
1788 }
1789 return type;
1790}
1791
1792/**
1793 * check_conflicting_open - see if the given file points to an inode that has
1794 * an existing open that would conflict with the
1795 * desired lease.
1796 * @filp: file to check
1797 * @arg: type of lease that we're trying to acquire
1798 * @flags: current lock flags
1799 *
1800 * Check to see if there's an existing open fd on this file that would
1801 * conflict with the lease we're trying to set.
1802 */
1803static int
1804check_conflicting_open(struct file *filp, const long arg, int flags)
1805{
1806 struct inode *inode = locks_inode(filp);
1807 int self_wcount = 0, self_rcount = 0;
1808
1809 if (flags & FL_LAYOUT)
1810 return 0;
1811 if (flags & FL_DELEG)
1812 /* We leave these checks to the caller. */
1813 return 0;
1814
1815 if (arg == F_RDLCK)
1816 return inode_is_open_for_write(inode) ? -EAGAIN : 0;
1817 else if (arg != F_WRLCK)
1818 return 0;
1819
1820 /*
1821 * Make sure that only read/write count is from lease requestor.
1822 * Note that this will result in denying write leases when i_writecount
1823 * is negative, which is what we want. (We shouldn't grant write leases
1824 * on files open for execution.)
1825 */
1826 if (filp->f_mode & FMODE_WRITE)
1827 self_wcount = 1;
1828 else if (filp->f_mode & FMODE_READ)
1829 self_rcount = 1;
1830
1831 if (atomic_read(&inode->i_writecount) != self_wcount ||
1832 atomic_read(&inode->i_readcount) != self_rcount)
1833 return -EAGAIN;
1834
1835 return 0;
1836}
1837
1838static int
1839generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
1840{
1841 struct file_lock *fl, *my_fl = NULL, *lease;
1842 struct inode *inode = locks_inode(filp);
1843 struct file_lock_context *ctx;
1844 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
1845 int error;
1846 LIST_HEAD(dispose);
1847
1848 lease = *flp;
1849 trace_generic_add_lease(inode, lease);
1850
1851 /* Note that arg is never F_UNLCK here */
1852 ctx = locks_get_lock_context(inode, arg);
1853 if (!ctx)
1854 return -ENOMEM;
1855
1856 /*
1857 * In the delegation case we need mutual exclusion with
1858 * a number of operations that take the i_mutex. We trylock
1859 * because delegations are an optional optimization, and if
1860 * there's some chance of a conflict--we'd rather not
1861 * bother, maybe that's a sign this just isn't a good file to
1862 * hand out a delegation on.
1863 */
1864 if (is_deleg && !inode_trylock(inode))
1865 return -EAGAIN;
1866
1867 if (is_deleg && arg == F_WRLCK) {
1868 /* Write delegations are not currently supported: */
1869 inode_unlock(inode);
1870 WARN_ON_ONCE(1);
1871 return -EINVAL;
1872 }
1873
1874 percpu_down_read(&file_rwsem);
1875 spin_lock(&ctx->flc_lock);
1876 time_out_leases(inode, &dispose);
1877 error = check_conflicting_open(filp, arg, lease->fl_flags);
1878 if (error)
1879 goto out;
1880
1881 /*
1882 * At this point, we know that if there is an exclusive
1883 * lease on this file, then we hold it on this filp
1884 * (otherwise our open of this file would have blocked).
1885 * And if we are trying to acquire an exclusive lease,
1886 * then the file is not open by anyone (including us)
1887 * except for this filp.
1888 */
1889 error = -EAGAIN;
1890 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1891 if (fl->fl_file == filp &&
1892 fl->fl_owner == lease->fl_owner) {
1893 my_fl = fl;
1894 continue;
1895 }
1896
1897 /*
1898 * No exclusive leases if someone else has a lease on
1899 * this file:
1900 */
1901 if (arg == F_WRLCK)
1902 goto out;
1903 /*
1904 * Modifying our existing lease is OK, but no getting a
1905 * new lease if someone else is opening for write:
1906 */
1907 if (fl->fl_flags & FL_UNLOCK_PENDING)
1908 goto out;
1909 }
1910
1911 if (my_fl != NULL) {
1912 lease = my_fl;
1913 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
1914 if (error)
1915 goto out;
1916 goto out_setup;
1917 }
1918
1919 error = -EINVAL;
1920 if (!leases_enable)
1921 goto out;
1922
1923 locks_insert_lock_ctx(lease, &ctx->flc_lease);
1924 /*
1925 * The check in break_lease() is lockless. It's possible for another
1926 * open to race in after we did the earlier check for a conflicting
1927 * open but before the lease was inserted. Check again for a
1928 * conflicting open and cancel the lease if there is one.
1929 *
1930 * We also add a barrier here to ensure that the insertion of the lock
1931 * precedes these checks.
1932 */
1933 smp_mb();
1934 error = check_conflicting_open(filp, arg, lease->fl_flags);
1935 if (error) {
1936 locks_unlink_lock_ctx(lease);
1937 goto out;
1938 }
1939
1940out_setup:
1941 if (lease->fl_lmops->lm_setup)
1942 lease->fl_lmops->lm_setup(lease, priv);
1943out:
1944 spin_unlock(&ctx->flc_lock);
1945 percpu_up_read(&file_rwsem);
1946 locks_dispose_list(&dispose);
1947 if (is_deleg)
1948 inode_unlock(inode);
1949 if (!error && !my_fl)
1950 *flp = NULL;
1951 return error;
1952}
1953
1954static int generic_delete_lease(struct file *filp, void *owner)
1955{
1956 int error = -EAGAIN;
1957 struct file_lock *fl, *victim = NULL;
1958 struct inode *inode = locks_inode(filp);
1959 struct file_lock_context *ctx;
1960 LIST_HEAD(dispose);
1961
1962 ctx = smp_load_acquire(&inode->i_flctx);
1963 if (!ctx) {
1964 trace_generic_delete_lease(inode, NULL);
1965 return error;
1966 }
1967
1968 percpu_down_read(&file_rwsem);
1969 spin_lock(&ctx->flc_lock);
1970 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1971 if (fl->fl_file == filp &&
1972 fl->fl_owner == owner) {
1973 victim = fl;
1974 break;
1975 }
1976 }
1977 trace_generic_delete_lease(inode, victim);
1978 if (victim)
1979 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
1980 spin_unlock(&ctx->flc_lock);
1981 percpu_up_read(&file_rwsem);
1982 locks_dispose_list(&dispose);
1983 return error;
1984}
1985
1986/**
1987 * generic_setlease - sets a lease on an open file
1988 * @filp: file pointer
1989 * @arg: type of lease to obtain
1990 * @flp: input - file_lock to use, output - file_lock inserted
1991 * @priv: private data for lm_setup (may be NULL if lm_setup
1992 * doesn't require it)
1993 *
1994 * The (input) flp->fl_lmops->lm_break function is required
1995 * by break_lease().
1996 */
1997int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1998 void **priv)
1999{
2000 struct inode *inode = locks_inode(filp);
2001 int error;
2002
2003 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
2004 return -EACCES;
2005 if (!S_ISREG(inode->i_mode))
2006 return -EINVAL;
2007 error = security_file_lock(filp, arg);
2008 if (error)
2009 return error;
2010
2011 switch (arg) {
2012 case F_UNLCK:
2013 return generic_delete_lease(filp, *priv);
2014 case F_RDLCK:
2015 case F_WRLCK:
2016 if (!(*flp)->fl_lmops->lm_break) {
2017 WARN_ON_ONCE(1);
2018 return -ENOLCK;
2019 }
2020
2021 return generic_add_lease(filp, arg, flp, priv);
2022 default:
2023 return -EINVAL;
2024 }
2025}
2026EXPORT_SYMBOL(generic_setlease);
2027
2028#if IS_ENABLED(CONFIG_SRCU)
2029/*
2030 * Kernel subsystems can register to be notified on any attempt to set
2031 * a new lease with the lease_notifier_chain. This is used by (e.g.) nfsd
2032 * to close files that it may have cached when there is an attempt to set a
2033 * conflicting lease.
2034 */
2035static struct srcu_notifier_head lease_notifier_chain;
2036
2037static inline void
2038lease_notifier_chain_init(void)
2039{
2040 srcu_init_notifier_head(&lease_notifier_chain);
2041}
2042
2043static inline void
2044setlease_notifier(long arg, struct file_lock *lease)
2045{
2046 if (arg != F_UNLCK)
2047 srcu_notifier_call_chain(&lease_notifier_chain, arg, lease);
2048}
2049
2050int lease_register_notifier(struct notifier_block *nb)
2051{
2052 return srcu_notifier_chain_register(&lease_notifier_chain, nb);
2053}
2054EXPORT_SYMBOL_GPL(lease_register_notifier);
2055
2056void lease_unregister_notifier(struct notifier_block *nb)
2057{
2058 srcu_notifier_chain_unregister(&lease_notifier_chain, nb);
2059}
2060EXPORT_SYMBOL_GPL(lease_unregister_notifier);
2061
2062#else /* !IS_ENABLED(CONFIG_SRCU) */
2063static inline void
2064lease_notifier_chain_init(void)
2065{
2066}
2067
2068static inline void
2069setlease_notifier(long arg, struct file_lock *lease)
2070{
2071}
2072
2073int lease_register_notifier(struct notifier_block *nb)
2074{
2075 return 0;
2076}
2077EXPORT_SYMBOL_GPL(lease_register_notifier);
2078
2079void lease_unregister_notifier(struct notifier_block *nb)
2080{
2081}
2082EXPORT_SYMBOL_GPL(lease_unregister_notifier);
2083
2084#endif /* IS_ENABLED(CONFIG_SRCU) */
2085
2086/**
2087 * vfs_setlease - sets a lease on an open file
2088 * @filp: file pointer
2089 * @arg: type of lease to obtain
2090 * @lease: file_lock to use when adding a lease
2091 * @priv: private info for lm_setup when adding a lease (may be
2092 * NULL if lm_setup doesn't require it)
2093 *
2094 * Call this to establish a lease on the file. The "lease" argument is not
2095 * used for F_UNLCK requests and may be NULL. For commands that set or alter
2096 * an existing lease, the ``(*lease)->fl_lmops->lm_break`` operation must be
2097 * set; if not, this function will return -ENOLCK (and generate a scary-looking
2098 * stack trace).
2099 *
2100 * The "priv" pointer is passed directly to the lm_setup function as-is. It
2101 * may be NULL if the lm_setup operation doesn't require it.
2102 */
2103int
2104vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
2105{
2106 if (lease)
2107 setlease_notifier(arg, *lease);
2108 if (filp->f_op->setlease)
2109 return filp->f_op->setlease(filp, arg, lease, priv);
2110 else
2111 return generic_setlease(filp, arg, lease, priv);
2112}
2113EXPORT_SYMBOL_GPL(vfs_setlease);
2114
2115static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
2116{
2117 struct file_lock *fl;
2118 struct fasync_struct *new;
2119 int error;
2120
2121 fl = lease_alloc(filp, arg);
2122 if (IS_ERR(fl))
2123 return PTR_ERR(fl);
2124
2125 new = fasync_alloc();
2126 if (!new) {
2127 locks_free_lock(fl);
2128 return -ENOMEM;
2129 }
2130 new->fa_fd = fd;
2131
2132 error = vfs_setlease(filp, arg, &fl, (void **)&new);
2133 if (fl)
2134 locks_free_lock(fl);
2135 if (new)
2136 fasync_free(new);
2137 return error;
2138}
2139
2140/**
2141 * fcntl_setlease - sets a lease on an open file
2142 * @fd: open file descriptor
2143 * @filp: file pointer
2144 * @arg: type of lease to obtain
2145 *
2146 * Call this fcntl to establish a lease on the file.
2147 * Note that you also need to call %F_SETSIG to
2148 * receive a signal when the lease is broken.
2149 */
2150int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
2151{
2152 if (arg == F_UNLCK)
2153 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
2154 return do_fcntl_add_lease(fd, filp, arg);
2155}
2156
2157/**
2158 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
2159 * @inode: inode of the file to apply to
2160 * @fl: The lock to be applied
2161 *
2162 * Apply a FLOCK style lock request to an inode.
2163 */
2164static int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2165{
2166 int error;
2167 might_sleep();
2168 for (;;) {
2169 error = flock_lock_inode(inode, fl);
2170 if (error != FILE_LOCK_DEFERRED)
2171 break;
2172 error = wait_event_interruptible(fl->fl_wait,
2173 list_empty(&fl->fl_blocked_member));
2174 if (error)
2175 break;
2176 }
2177 locks_delete_block(fl);
2178 return error;
2179}
2180
2181/**
2182 * locks_lock_inode_wait - Apply a lock to an inode
2183 * @inode: inode of the file to apply to
2184 * @fl: The lock to be applied
2185 *
2186 * Apply a POSIX or FLOCK style lock request to an inode.
2187 */
2188int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
2189{
2190 int res = 0;
2191 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
2192 case FL_POSIX:
2193 res = posix_lock_inode_wait(inode, fl);
2194 break;
2195 case FL_FLOCK:
2196 res = flock_lock_inode_wait(inode, fl);
2197 break;
2198 default:
2199 BUG();
2200 }
2201 return res;
2202}
2203EXPORT_SYMBOL(locks_lock_inode_wait);
2204
2205/**
2206 * sys_flock: - flock() system call.
2207 * @fd: the file descriptor to lock.
2208 * @cmd: the type of lock to apply.
2209 *
2210 * Apply a %FL_FLOCK style lock to an open file descriptor.
2211 * The @cmd can be one of:
2212 *
2213 * - %LOCK_SH -- a shared lock.
2214 * - %LOCK_EX -- an exclusive lock.
2215 * - %LOCK_UN -- remove an existing lock.
2216 * - %LOCK_MAND -- a 'mandatory' flock.
2217 * This exists to emulate Windows Share Modes.
2218 *
2219 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
2220 * processes read and write access respectively.
2221 */
2222SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
2223{
2224 struct fd f = fdget(fd);
2225 struct file_lock *lock;
2226 int can_sleep, unlock;
2227 int error;
2228
2229 error = -EBADF;
2230 if (!f.file)
2231 goto out;
2232
2233 can_sleep = !(cmd & LOCK_NB);
2234 cmd &= ~LOCK_NB;
2235 unlock = (cmd == LOCK_UN);
2236
2237 if (!unlock && !(cmd & LOCK_MAND) &&
2238 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
2239 goto out_putf;
2240
2241 lock = flock_make_lock(f.file, cmd, NULL);
2242 if (IS_ERR(lock)) {
2243 error = PTR_ERR(lock);
2244 goto out_putf;
2245 }
2246
2247 if (can_sleep)
2248 lock->fl_flags |= FL_SLEEP;
2249
2250 error = security_file_lock(f.file, lock->fl_type);
2251 if (error)
2252 goto out_free;
2253
2254 if (f.file->f_op->flock)
2255 error = f.file->f_op->flock(f.file,
2256 (can_sleep) ? F_SETLKW : F_SETLK,
2257 lock);
2258 else
2259 error = locks_lock_file_wait(f.file, lock);
2260
2261 out_free:
2262 locks_free_lock(lock);
2263
2264 out_putf:
2265 fdput(f);
2266 out:
2267 return error;
2268}
2269
2270/**
2271 * vfs_test_lock - test file byte range lock
2272 * @filp: The file to test lock for
2273 * @fl: The lock to test; also used to hold result
2274 *
2275 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
2276 * setting conf->fl_type to something other than F_UNLCK.
2277 */
2278int vfs_test_lock(struct file *filp, struct file_lock *fl)
2279{
2280 if (filp->f_op->lock)
2281 return filp->f_op->lock(filp, F_GETLK, fl);
2282 posix_test_lock(filp, fl);
2283 return 0;
2284}
2285EXPORT_SYMBOL_GPL(vfs_test_lock);
2286
2287/**
2288 * locks_translate_pid - translate a file_lock's fl_pid number into a namespace
2289 * @fl: The file_lock who's fl_pid should be translated
2290 * @ns: The namespace into which the pid should be translated
2291 *
2292 * Used to tranlate a fl_pid into a namespace virtual pid number
2293 */
2294static pid_t locks_translate_pid(struct file_lock *fl, struct pid_namespace *ns)
2295{
2296 pid_t vnr;
2297 struct pid *pid;
2298
2299 if (IS_OFDLCK(fl))
2300 return -1;
2301 if (IS_REMOTELCK(fl))
2302 return fl->fl_pid;
2303 /*
2304 * If the flock owner process is dead and its pid has been already
2305 * freed, the translation below won't work, but we still want to show
2306 * flock owner pid number in init pidns.
2307 */
2308 if (ns == &init_pid_ns)
2309 return (pid_t)fl->fl_pid;
2310
2311 rcu_read_lock();
2312 pid = find_pid_ns(fl->fl_pid, &init_pid_ns);
2313 vnr = pid_nr_ns(pid, ns);
2314 rcu_read_unlock();
2315 return vnr;
2316}
2317
2318static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
2319{
2320 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2321#if BITS_PER_LONG == 32
2322 /*
2323 * Make sure we can represent the posix lock via
2324 * legacy 32bit flock.
2325 */
2326 if (fl->fl_start > OFFT_OFFSET_MAX)
2327 return -EOVERFLOW;
2328 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2329 return -EOVERFLOW;
2330#endif
2331 flock->l_start = fl->fl_start;
2332 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2333 fl->fl_end - fl->fl_start + 1;
2334 flock->l_whence = 0;
2335 flock->l_type = fl->fl_type;
2336 return 0;
2337}
2338
2339#if BITS_PER_LONG == 32
2340static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2341{
2342 flock->l_pid = locks_translate_pid(fl, task_active_pid_ns(current));
2343 flock->l_start = fl->fl_start;
2344 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2345 fl->fl_end - fl->fl_start + 1;
2346 flock->l_whence = 0;
2347 flock->l_type = fl->fl_type;
2348}
2349#endif
2350
2351/* Report the first existing lock that would conflict with l.
2352 * This implements the F_GETLK command of fcntl().
2353 */
2354int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock *flock)
2355{
2356 struct file_lock *fl;
2357 int error;
2358
2359 fl = locks_alloc_lock();
2360 if (fl == NULL)
2361 return -ENOMEM;
2362 error = -EINVAL;
2363 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2364 goto out;
2365
2366 error = flock_to_posix_lock(filp, fl, flock);
2367 if (error)
2368 goto out;
2369
2370 if (cmd == F_OFD_GETLK) {
2371 error = -EINVAL;
2372 if (flock->l_pid != 0)
2373 goto out;
2374
2375 cmd = F_GETLK;
2376 fl->fl_flags |= FL_OFDLCK;
2377 fl->fl_owner = filp;
2378 }
2379
2380 error = vfs_test_lock(filp, fl);
2381 if (error)
2382 goto out;
2383
2384 flock->l_type = fl->fl_type;
2385 if (fl->fl_type != F_UNLCK) {
2386 error = posix_lock_to_flock(flock, fl);
2387 if (error)
2388 goto out;
2389 }
2390out:
2391 locks_free_lock(fl);
2392 return error;
2393}
2394
2395/**
2396 * vfs_lock_file - file byte range lock
2397 * @filp: The file to apply the lock to
2398 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2399 * @fl: The lock to be applied
2400 * @conf: Place to return a copy of the conflicting lock, if found.
2401 *
2402 * A caller that doesn't care about the conflicting lock may pass NULL
2403 * as the final argument.
2404 *
2405 * If the filesystem defines a private ->lock() method, then @conf will
2406 * be left unchanged; so a caller that cares should initialize it to
2407 * some acceptable default.
2408 *
2409 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2410 * locks, the ->lock() interface may return asynchronously, before the lock has
2411 * been granted or denied by the underlying filesystem, if (and only if)
2412 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2413 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2414 * the request is for a blocking lock. When ->lock() does return asynchronously,
2415 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2416 * request completes.
2417 * If the request is for non-blocking lock the file system should return
2418 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2419 * with the result. If the request timed out the callback routine will return a
2420 * nonzero return code and the file system should release the lock. The file
2421 * system is also responsible to keep a corresponding posix lock when it
2422 * grants a lock so the VFS can find out which locks are locally held and do
2423 * the correct lock cleanup when required.
2424 * The underlying filesystem must not drop the kernel lock or call
2425 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2426 * return code.
2427 */
2428int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
2429{
2430 if (filp->f_op->lock)
2431 return filp->f_op->lock(filp, cmd, fl);
2432 else
2433 return posix_lock_file(filp, fl, conf);
2434}
2435EXPORT_SYMBOL_GPL(vfs_lock_file);
2436
2437static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2438 struct file_lock *fl)
2439{
2440 int error;
2441
2442 error = security_file_lock(filp, fl->fl_type);
2443 if (error)
2444 return error;
2445
2446 for (;;) {
2447 error = vfs_lock_file(filp, cmd, fl, NULL);
2448 if (error != FILE_LOCK_DEFERRED)
2449 break;
2450 error = wait_event_interruptible(fl->fl_wait,
2451 list_empty(&fl->fl_blocked_member));
2452 if (error)
2453 break;
2454 }
2455 locks_delete_block(fl);
2456
2457 return error;
2458}
2459
2460/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
2461static int
2462check_fmode_for_setlk(struct file_lock *fl)
2463{
2464 switch (fl->fl_type) {
2465 case F_RDLCK:
2466 if (!(fl->fl_file->f_mode & FMODE_READ))
2467 return -EBADF;
2468 break;
2469 case F_WRLCK:
2470 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2471 return -EBADF;
2472 }
2473 return 0;
2474}
2475
2476/* Apply the lock described by l to an open file descriptor.
2477 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2478 */
2479int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2480 struct flock *flock)
2481{
2482 struct file_lock *file_lock = locks_alloc_lock();
2483 struct inode *inode = locks_inode(filp);
2484 struct file *f;
2485 int error;
2486
2487 if (file_lock == NULL)
2488 return -ENOLCK;
2489
2490 /* Don't allow mandatory locks on files that may be memory mapped
2491 * and shared.
2492 */
2493 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2494 error = -EAGAIN;
2495 goto out;
2496 }
2497
2498 error = flock_to_posix_lock(filp, file_lock, flock);
2499 if (error)
2500 goto out;
2501
2502 error = check_fmode_for_setlk(file_lock);
2503 if (error)
2504 goto out;
2505
2506 /*
2507 * If the cmd is requesting file-private locks, then set the
2508 * FL_OFDLCK flag and override the owner.
2509 */
2510 switch (cmd) {
2511 case F_OFD_SETLK:
2512 error = -EINVAL;
2513 if (flock->l_pid != 0)
2514 goto out;
2515
2516 cmd = F_SETLK;
2517 file_lock->fl_flags |= FL_OFDLCK;
2518 file_lock->fl_owner = filp;
2519 break;
2520 case F_OFD_SETLKW:
2521 error = -EINVAL;
2522 if (flock->l_pid != 0)
2523 goto out;
2524
2525 cmd = F_SETLKW;
2526 file_lock->fl_flags |= FL_OFDLCK;
2527 file_lock->fl_owner = filp;
2528 fallthrough;
2529 case F_SETLKW:
2530 file_lock->fl_flags |= FL_SLEEP;
2531 }
2532
2533 error = do_lock_file_wait(filp, cmd, file_lock);
2534
2535 /*
2536 * Attempt to detect a close/fcntl race and recover by releasing the
2537 * lock that was just acquired. There is no need to do that when we're
2538 * unlocking though, or for OFD locks.
2539 */
2540 if (!error && file_lock->fl_type != F_UNLCK &&
2541 !(file_lock->fl_flags & FL_OFDLCK)) {
2542 /*
2543 * We need that spin_lock here - it prevents reordering between
2544 * update of i_flctx->flc_posix and check for it done in
2545 * close(). rcu_read_lock() wouldn't do.
2546 */
2547 spin_lock(¤t->files->file_lock);
2548 f = fcheck(fd);
2549 spin_unlock(¤t->files->file_lock);
2550 if (f != filp) {
2551 file_lock->fl_type = F_UNLCK;
2552 error = do_lock_file_wait(filp, cmd, file_lock);
2553 WARN_ON_ONCE(error);
2554 error = -EBADF;
2555 }
2556 }
2557out:
2558 trace_fcntl_setlk(inode, file_lock, error);
2559 locks_free_lock(file_lock);
2560 return error;
2561}
2562
2563#if BITS_PER_LONG == 32
2564/* Report the first existing lock that would conflict with l.
2565 * This implements the F_GETLK command of fcntl().
2566 */
2567int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 *flock)
2568{
2569 struct file_lock *fl;
2570 int error;
2571
2572 fl = locks_alloc_lock();
2573 if (fl == NULL)
2574 return -ENOMEM;
2575
2576 error = -EINVAL;
2577 if (flock->l_type != F_RDLCK && flock->l_type != F_WRLCK)
2578 goto out;
2579
2580 error = flock64_to_posix_lock(filp, fl, flock);
2581 if (error)
2582 goto out;
2583
2584 if (cmd == F_OFD_GETLK) {
2585 error = -EINVAL;
2586 if (flock->l_pid != 0)
2587 goto out;
2588
2589 cmd = F_GETLK64;
2590 fl->fl_flags |= FL_OFDLCK;
2591 fl->fl_owner = filp;
2592 }
2593
2594 error = vfs_test_lock(filp, fl);
2595 if (error)
2596 goto out;
2597
2598 flock->l_type = fl->fl_type;
2599 if (fl->fl_type != F_UNLCK)
2600 posix_lock_to_flock64(flock, fl);
2601
2602out:
2603 locks_free_lock(fl);
2604 return error;
2605}
2606
2607/* Apply the lock described by l to an open file descriptor.
2608 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2609 */
2610int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2611 struct flock64 *flock)
2612{
2613 struct file_lock *file_lock = locks_alloc_lock();
2614 struct inode *inode = locks_inode(filp);
2615 struct file *f;
2616 int error;
2617
2618 if (file_lock == NULL)
2619 return -ENOLCK;
2620
2621 /* Don't allow mandatory locks on files that may be memory mapped
2622 * and shared.
2623 */
2624 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
2625 error = -EAGAIN;
2626 goto out;
2627 }
2628
2629 error = flock64_to_posix_lock(filp, file_lock, flock);
2630 if (error)
2631 goto out;
2632
2633 error = check_fmode_for_setlk(file_lock);
2634 if (error)
2635 goto out;
2636
2637 /*
2638 * If the cmd is requesting file-private locks, then set the
2639 * FL_OFDLCK flag and override the owner.
2640 */
2641 switch (cmd) {
2642 case F_OFD_SETLK:
2643 error = -EINVAL;
2644 if (flock->l_pid != 0)
2645 goto out;
2646
2647 cmd = F_SETLK64;
2648 file_lock->fl_flags |= FL_OFDLCK;
2649 file_lock->fl_owner = filp;
2650 break;
2651 case F_OFD_SETLKW:
2652 error = -EINVAL;
2653 if (flock->l_pid != 0)
2654 goto out;
2655
2656 cmd = F_SETLKW64;
2657 file_lock->fl_flags |= FL_OFDLCK;
2658 file_lock->fl_owner = filp;
2659 fallthrough;
2660 case F_SETLKW64:
2661 file_lock->fl_flags |= FL_SLEEP;
2662 }
2663
2664 error = do_lock_file_wait(filp, cmd, file_lock);
2665
2666 /*
2667 * Attempt to detect a close/fcntl race and recover by releasing the
2668 * lock that was just acquired. There is no need to do that when we're
2669 * unlocking though, or for OFD locks.
2670 */
2671 if (!error && file_lock->fl_type != F_UNLCK &&
2672 !(file_lock->fl_flags & FL_OFDLCK)) {
2673 /*
2674 * We need that spin_lock here - it prevents reordering between
2675 * update of i_flctx->flc_posix and check for it done in
2676 * close(). rcu_read_lock() wouldn't do.
2677 */
2678 spin_lock(¤t->files->file_lock);
2679 f = fcheck(fd);
2680 spin_unlock(¤t->files->file_lock);
2681 if (f != filp) {
2682 file_lock->fl_type = F_UNLCK;
2683 error = do_lock_file_wait(filp, cmd, file_lock);
2684 WARN_ON_ONCE(error);
2685 error = -EBADF;
2686 }
2687 }
2688out:
2689 locks_free_lock(file_lock);
2690 return error;
2691}
2692#endif /* BITS_PER_LONG == 32 */
2693
2694/*
2695 * This function is called when the file is being removed
2696 * from the task's fd array. POSIX locks belonging to this task
2697 * are deleted at this time.
2698 */
2699void locks_remove_posix(struct file *filp, fl_owner_t owner)
2700{
2701 int error;
2702 struct inode *inode = locks_inode(filp);
2703 struct file_lock lock;
2704 struct file_lock_context *ctx;
2705
2706 /*
2707 * If there are no locks held on this file, we don't need to call
2708 * posix_lock_file(). Another process could be setting a lock on this
2709 * file at the same time, but we wouldn't remove that lock anyway.
2710 */
2711 ctx = smp_load_acquire(&inode->i_flctx);
2712 if (!ctx || list_empty(&ctx->flc_posix))
2713 return;
2714
2715 locks_init_lock(&lock);
2716 lock.fl_type = F_UNLCK;
2717 lock.fl_flags = FL_POSIX | FL_CLOSE;
2718 lock.fl_start = 0;
2719 lock.fl_end = OFFSET_MAX;
2720 lock.fl_owner = owner;
2721 lock.fl_pid = current->tgid;
2722 lock.fl_file = filp;
2723 lock.fl_ops = NULL;
2724 lock.fl_lmops = NULL;
2725
2726 error = vfs_lock_file(filp, F_SETLK, &lock, NULL);
2727
2728 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2729 lock.fl_ops->fl_release_private(&lock);
2730 trace_locks_remove_posix(inode, &lock, error);
2731}
2732EXPORT_SYMBOL(locks_remove_posix);
2733
2734/* The i_flctx must be valid when calling into here */
2735static void
2736locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
2737{
2738 struct file_lock fl;
2739 struct inode *inode = locks_inode(filp);
2740
2741 if (list_empty(&flctx->flc_flock))
2742 return;
2743
2744 flock_make_lock(filp, LOCK_UN, &fl);
2745 fl.fl_flags |= FL_CLOSE;
2746
2747 if (filp->f_op->flock)
2748 filp->f_op->flock(filp, F_SETLKW, &fl);
2749 else
2750 flock_lock_inode(inode, &fl);
2751
2752 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2753 fl.fl_ops->fl_release_private(&fl);
2754}
2755
2756/* The i_flctx must be valid when calling into here */
2757static void
2758locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
2759{
2760 struct file_lock *fl, *tmp;
2761 LIST_HEAD(dispose);
2762
2763 if (list_empty(&ctx->flc_lease))
2764 return;
2765
2766 percpu_down_read(&file_rwsem);
2767 spin_lock(&ctx->flc_lock);
2768 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2769 if (filp == fl->fl_file)
2770 lease_modify(fl, F_UNLCK, &dispose);
2771 spin_unlock(&ctx->flc_lock);
2772 percpu_up_read(&file_rwsem);
2773
2774 locks_dispose_list(&dispose);
2775}
2776
2777/*
2778 * This function is called on the last close of an open file.
2779 */
2780void locks_remove_file(struct file *filp)
2781{
2782 struct file_lock_context *ctx;
2783
2784 ctx = smp_load_acquire(&locks_inode(filp)->i_flctx);
2785 if (!ctx)
2786 return;
2787
2788 /* remove any OFD locks */
2789 locks_remove_posix(filp, filp);
2790
2791 /* remove flock locks */
2792 locks_remove_flock(filp, ctx);
2793
2794 /* remove any leases */
2795 locks_remove_lease(filp, ctx);
2796
2797 spin_lock(&ctx->flc_lock);
2798 locks_check_ctx_file_list(filp, &ctx->flc_posix, "POSIX");
2799 locks_check_ctx_file_list(filp, &ctx->flc_flock, "FLOCK");
2800 locks_check_ctx_file_list(filp, &ctx->flc_lease, "LEASE");
2801 spin_unlock(&ctx->flc_lock);
2802}
2803
2804/**
2805 * vfs_cancel_lock - file byte range unblock lock
2806 * @filp: The file to apply the unblock to
2807 * @fl: The lock to be unblocked
2808 *
2809 * Used by lock managers to cancel blocked requests
2810 */
2811int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2812{
2813 if (filp->f_op->lock)
2814 return filp->f_op->lock(filp, F_CANCELLK, fl);
2815 return 0;
2816}
2817EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2818
2819#ifdef CONFIG_PROC_FS
2820#include <linux/proc_fs.h>
2821#include <linux/seq_file.h>
2822
2823struct locks_iterator {
2824 int li_cpu;
2825 loff_t li_pos;
2826};
2827
2828static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2829 loff_t id, char *pfx)
2830{
2831 struct inode *inode = NULL;
2832 unsigned int fl_pid;
2833 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2834
2835 fl_pid = locks_translate_pid(fl, proc_pidns);
2836 /*
2837 * If lock owner is dead (and pid is freed) or not visible in current
2838 * pidns, zero is shown as a pid value. Check lock info from
2839 * init_pid_ns to get saved lock pid value.
2840 */
2841
2842 if (fl->fl_file != NULL)
2843 inode = locks_inode(fl->fl_file);
2844
2845 seq_printf(f, "%lld:%s ", id, pfx);
2846 if (IS_POSIX(fl)) {
2847 if (fl->fl_flags & FL_ACCESS)
2848 seq_puts(f, "ACCESS");
2849 else if (IS_OFDLCK(fl))
2850 seq_puts(f, "OFDLCK");
2851 else
2852 seq_puts(f, "POSIX ");
2853
2854 seq_printf(f, " %s ",
2855 (inode == NULL) ? "*NOINODE*" :
2856 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
2857 } else if (IS_FLOCK(fl)) {
2858 if (fl->fl_type & LOCK_MAND) {
2859 seq_puts(f, "FLOCK MSNFS ");
2860 } else {
2861 seq_puts(f, "FLOCK ADVISORY ");
2862 }
2863 } else if (IS_LEASE(fl)) {
2864 if (fl->fl_flags & FL_DELEG)
2865 seq_puts(f, "DELEG ");
2866 else
2867 seq_puts(f, "LEASE ");
2868
2869 if (lease_breaking(fl))
2870 seq_puts(f, "BREAKING ");
2871 else if (fl->fl_file)
2872 seq_puts(f, "ACTIVE ");
2873 else
2874 seq_puts(f, "BREAKER ");
2875 } else {
2876 seq_puts(f, "UNKNOWN UNKNOWN ");
2877 }
2878 if (fl->fl_type & LOCK_MAND) {
2879 seq_printf(f, "%s ",
2880 (fl->fl_type & LOCK_READ)
2881 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2882 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2883 } else {
2884 int type = IS_LEASE(fl) ? target_leasetype(fl) : fl->fl_type;
2885
2886 seq_printf(f, "%s ", (type == F_WRLCK) ? "WRITE" :
2887 (type == F_RDLCK) ? "READ" : "UNLCK");
2888 }
2889 if (inode) {
2890 /* userspace relies on this representation of dev_t */
2891 seq_printf(f, "%d %02x:%02x:%lu ", fl_pid,
2892 MAJOR(inode->i_sb->s_dev),
2893 MINOR(inode->i_sb->s_dev), inode->i_ino);
2894 } else {
2895 seq_printf(f, "%d <none>:0 ", fl_pid);
2896 }
2897 if (IS_POSIX(fl)) {
2898 if (fl->fl_end == OFFSET_MAX)
2899 seq_printf(f, "%Ld EOF\n", fl->fl_start);
2900 else
2901 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
2902 } else {
2903 seq_puts(f, "0 EOF\n");
2904 }
2905}
2906
2907static int locks_show(struct seq_file *f, void *v)
2908{
2909 struct locks_iterator *iter = f->private;
2910 struct file_lock *fl, *bfl;
2911 struct pid_namespace *proc_pidns = proc_pid_ns(file_inode(f->file)->i_sb);
2912
2913 fl = hlist_entry(v, struct file_lock, fl_link);
2914
2915 if (locks_translate_pid(fl, proc_pidns) == 0)
2916 return 0;
2917
2918 lock_get_status(f, fl, iter->li_pos, "");
2919
2920 list_for_each_entry(bfl, &fl->fl_blocked_requests, fl_blocked_member)
2921 lock_get_status(f, bfl, iter->li_pos, " ->");
2922
2923 return 0;
2924}
2925
2926static void __show_fd_locks(struct seq_file *f,
2927 struct list_head *head, int *id,
2928 struct file *filp, struct files_struct *files)
2929{
2930 struct file_lock *fl;
2931
2932 list_for_each_entry(fl, head, fl_list) {
2933
2934 if (filp != fl->fl_file)
2935 continue;
2936 if (fl->fl_owner != files &&
2937 fl->fl_owner != filp)
2938 continue;
2939
2940 (*id)++;
2941 seq_puts(f, "lock:\t");
2942 lock_get_status(f, fl, *id, "");
2943 }
2944}
2945
2946void show_fd_locks(struct seq_file *f,
2947 struct file *filp, struct files_struct *files)
2948{
2949 struct inode *inode = locks_inode(filp);
2950 struct file_lock_context *ctx;
2951 int id = 0;
2952
2953 ctx = smp_load_acquire(&inode->i_flctx);
2954 if (!ctx)
2955 return;
2956
2957 spin_lock(&ctx->flc_lock);
2958 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2959 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2960 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2961 spin_unlock(&ctx->flc_lock);
2962}
2963
2964static void *locks_start(struct seq_file *f, loff_t *pos)
2965 __acquires(&blocked_lock_lock)
2966{
2967 struct locks_iterator *iter = f->private;
2968
2969 iter->li_pos = *pos + 1;
2970 percpu_down_write(&file_rwsem);
2971 spin_lock(&blocked_lock_lock);
2972 return seq_hlist_start_percpu(&file_lock_list.hlist, &iter->li_cpu, *pos);
2973}
2974
2975static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2976{
2977 struct locks_iterator *iter = f->private;
2978
2979 ++iter->li_pos;
2980 return seq_hlist_next_percpu(v, &file_lock_list.hlist, &iter->li_cpu, pos);
2981}
2982
2983static void locks_stop(struct seq_file *f, void *v)
2984 __releases(&blocked_lock_lock)
2985{
2986 spin_unlock(&blocked_lock_lock);
2987 percpu_up_write(&file_rwsem);
2988}
2989
2990static const struct seq_operations locks_seq_operations = {
2991 .start = locks_start,
2992 .next = locks_next,
2993 .stop = locks_stop,
2994 .show = locks_show,
2995};
2996
2997static int __init proc_locks_init(void)
2998{
2999 proc_create_seq_private("locks", 0, NULL, &locks_seq_operations,
3000 sizeof(struct locks_iterator), NULL);
3001 return 0;
3002}
3003fs_initcall(proc_locks_init);
3004#endif
3005
3006static int __init filelock_init(void)
3007{
3008 int i;
3009
3010 flctx_cache = kmem_cache_create("file_lock_ctx",
3011 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
3012
3013 filelock_cache = kmem_cache_create("file_lock_cache",
3014 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
3015
3016 for_each_possible_cpu(i) {
3017 struct file_lock_list_struct *fll = per_cpu_ptr(&file_lock_list, i);
3018
3019 spin_lock_init(&fll->lock);
3020 INIT_HLIST_HEAD(&fll->hlist);
3021 }
3022
3023 lease_notifier_chain_init();
3024 return 0;
3025}
3026core_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);