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