locks: protect most of the file_lock handling with i_lock
[linux-2.6-block.git] / fs / locks.c
CommitLineData
1da177e4
LT
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.
395cf969 63 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
1da177e4
LT
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>
9f3acc31 119#include <linux/fdtable.h>
1da177e4
LT
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>
1da177e4
LT
125#include <linux/syscalls.h>
126#include <linux/time.h>
4fb3a538 127#include <linux/rcupdate.h>
ab1f1611 128#include <linux/pid_namespace.h>
1da177e4 129
1da177e4
LT
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
ab83fa4b
BF
136static bool lease_breaking(struct file_lock *fl)
137{
778fc546
BF
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;
ab83fa4b
BF
148}
149
1da177e4
LT
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
1c8c601a
JL
156/*
157 * The global file_lock_list is only used for displaying /proc/locks. Protected
158 * by the file_lock_lock.
159 */
26bcbf96 160static LIST_HEAD(file_lock_list);
88974691 161
1c8c601a
JL
162/*
163 * The blocked_list is used to find POSIX lock loops for deadlock detection.
164 * Protected by file_lock_lock.
165 */
1da177e4 166static LIST_HEAD(blocked_list);
88974691 167
1c8c601a
JL
168/*
169 * This lock protects the blocked_list, and the file_lock_list. Generally, if
170 * you're accessing one of those lists, you want to be holding this lock.
171 *
172 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
173 * pointer for file_lock structures that are acting as lock requests (in
174 * contrast to those that are acting as records of acquired locks).
175 *
176 * Note that when we acquire this lock in order to change the above fields,
177 * we often hold the i_lock as well. In certain cases, when reading the fields
178 * protected by this lock, we can skip acquiring it iff we already hold the
179 * i_lock.
180 *
181 * In particular, adding an entry to the fl_block list requires that you hold
182 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
183 * an entry from the list however only requires the file_lock_lock.
184 */
72f98e72 185static DEFINE_SPINLOCK(file_lock_lock);
1da177e4 186
e18b890b 187static struct kmem_cache *filelock_cache __read_mostly;
1da177e4 188
ee19cc40 189static void locks_init_lock_heads(struct file_lock *fl)
a51cb91d 190{
ee19cc40
MS
191 INIT_LIST_HEAD(&fl->fl_link);
192 INIT_LIST_HEAD(&fl->fl_block);
193 init_waitqueue_head(&fl->fl_wait);
a51cb91d
MS
194}
195
1da177e4 196/* Allocate an empty lock structure. */
c5b1f0d9 197struct file_lock *locks_alloc_lock(void)
1da177e4 198{
ee19cc40 199 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
a51cb91d
MS
200
201 if (fl)
ee19cc40 202 locks_init_lock_heads(fl);
a51cb91d
MS
203
204 return fl;
1da177e4 205}
c5b1f0d9 206EXPORT_SYMBOL_GPL(locks_alloc_lock);
1da177e4 207
a9e61e25 208void locks_release_private(struct file_lock *fl)
47831f35
TM
209{
210 if (fl->fl_ops) {
211 if (fl->fl_ops->fl_release_private)
212 fl->fl_ops->fl_release_private(fl);
213 fl->fl_ops = NULL;
214 }
068535f1 215 fl->fl_lmops = NULL;
47831f35
TM
216
217}
a9e61e25 218EXPORT_SYMBOL_GPL(locks_release_private);
47831f35 219
1da177e4 220/* Free a lock which is not in use. */
05fa3135 221void locks_free_lock(struct file_lock *fl)
1da177e4 222{
5ce29646
MS
223 BUG_ON(waitqueue_active(&fl->fl_wait));
224 BUG_ON(!list_empty(&fl->fl_block));
225 BUG_ON(!list_empty(&fl->fl_link));
1da177e4 226
47831f35 227 locks_release_private(fl);
1da177e4
LT
228 kmem_cache_free(filelock_cache, fl);
229}
05fa3135 230EXPORT_SYMBOL(locks_free_lock);
1da177e4
LT
231
232void locks_init_lock(struct file_lock *fl)
233{
ee19cc40
MS
234 memset(fl, 0, sizeof(struct file_lock));
235 locks_init_lock_heads(fl);
1da177e4
LT
236}
237
238EXPORT_SYMBOL(locks_init_lock);
239
47831f35
TM
240static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
241{
242 if (fl->fl_ops) {
243 if (fl->fl_ops->fl_copy_lock)
244 fl->fl_ops->fl_copy_lock(new, fl);
245 new->fl_ops = fl->fl_ops;
246 }
bb8430a2 247 if (fl->fl_lmops)
47831f35 248 new->fl_lmops = fl->fl_lmops;
47831f35
TM
249}
250
1da177e4
LT
251/*
252 * Initialize a new lock from an existing file_lock structure.
253 */
1a747ee0 254void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
1da177e4
LT
255{
256 new->fl_owner = fl->fl_owner;
257 new->fl_pid = fl->fl_pid;
0996905f 258 new->fl_file = NULL;
1da177e4
LT
259 new->fl_flags = fl->fl_flags;
260 new->fl_type = fl->fl_type;
261 new->fl_start = fl->fl_start;
262 new->fl_end = fl->fl_end;
0996905f
TM
263 new->fl_ops = NULL;
264 new->fl_lmops = NULL;
265}
3dd7b71c 266EXPORT_SYMBOL(__locks_copy_lock);
0996905f
TM
267
268void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
269{
270 locks_release_private(new);
271
272 __locks_copy_lock(new, fl);
273 new->fl_file = fl->fl_file;
1da177e4
LT
274 new->fl_ops = fl->fl_ops;
275 new->fl_lmops = fl->fl_lmops;
47831f35
TM
276
277 locks_copy_private(new, fl);
1da177e4
LT
278}
279
280EXPORT_SYMBOL(locks_copy_lock);
281
282static inline int flock_translate_cmd(int cmd) {
283 if (cmd & LOCK_MAND)
284 return cmd & (LOCK_MAND | LOCK_RW);
285 switch (cmd) {
286 case LOCK_SH:
287 return F_RDLCK;
288 case LOCK_EX:
289 return F_WRLCK;
290 case LOCK_UN:
291 return F_UNLCK;
292 }
293 return -EINVAL;
294}
295
296/* Fill in a file_lock structure with an appropriate FLOCK lock. */
297static int flock_make_lock(struct file *filp, struct file_lock **lock,
298 unsigned int cmd)
299{
300 struct file_lock *fl;
301 int type = flock_translate_cmd(cmd);
302 if (type < 0)
303 return type;
304
305 fl = locks_alloc_lock();
306 if (fl == NULL)
307 return -ENOMEM;
308
309 fl->fl_file = filp;
310 fl->fl_pid = current->tgid;
311 fl->fl_flags = FL_FLOCK;
312 fl->fl_type = type;
313 fl->fl_end = OFFSET_MAX;
314
315 *lock = fl;
316 return 0;
317}
318
0ec4f431 319static int assign_type(struct file_lock *fl, long type)
1da177e4
LT
320{
321 switch (type) {
322 case F_RDLCK:
323 case F_WRLCK:
324 case F_UNLCK:
325 fl->fl_type = type;
326 break;
327 default:
328 return -EINVAL;
329 }
330 return 0;
331}
332
333/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
334 * style lock.
335 */
336static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
337 struct flock *l)
338{
339 off_t start, end;
340
341 switch (l->l_whence) {
f5579f8c 342 case SEEK_SET:
1da177e4
LT
343 start = 0;
344 break;
f5579f8c 345 case SEEK_CUR:
1da177e4
LT
346 start = filp->f_pos;
347 break;
f5579f8c 348 case SEEK_END:
496ad9aa 349 start = i_size_read(file_inode(filp));
1da177e4
LT
350 break;
351 default:
352 return -EINVAL;
353 }
354
355 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
356 POSIX-2001 defines it. */
357 start += l->l_start;
4c780a46
TM
358 if (start < 0)
359 return -EINVAL;
360 fl->fl_end = OFFSET_MAX;
361 if (l->l_len > 0) {
362 end = start + l->l_len - 1;
363 fl->fl_end = end;
364 } else if (l->l_len < 0) {
1da177e4 365 end = start - 1;
4c780a46 366 fl->fl_end = end;
1da177e4 367 start += l->l_len;
4c780a46
TM
368 if (start < 0)
369 return -EINVAL;
1da177e4 370 }
1da177e4 371 fl->fl_start = start; /* we record the absolute position */
4c780a46
TM
372 if (fl->fl_end < fl->fl_start)
373 return -EOVERFLOW;
1da177e4
LT
374
375 fl->fl_owner = current->files;
376 fl->fl_pid = current->tgid;
377 fl->fl_file = filp;
378 fl->fl_flags = FL_POSIX;
379 fl->fl_ops = NULL;
380 fl->fl_lmops = NULL;
381
382 return assign_type(fl, l->l_type);
383}
384
385#if BITS_PER_LONG == 32
386static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
387 struct flock64 *l)
388{
389 loff_t start;
390
391 switch (l->l_whence) {
f5579f8c 392 case SEEK_SET:
1da177e4
LT
393 start = 0;
394 break;
f5579f8c 395 case SEEK_CUR:
1da177e4
LT
396 start = filp->f_pos;
397 break;
f5579f8c 398 case SEEK_END:
496ad9aa 399 start = i_size_read(file_inode(filp));
1da177e4
LT
400 break;
401 default:
402 return -EINVAL;
403 }
404
4c780a46
TM
405 start += l->l_start;
406 if (start < 0)
1da177e4 407 return -EINVAL;
4c780a46
TM
408 fl->fl_end = OFFSET_MAX;
409 if (l->l_len > 0) {
410 fl->fl_end = start + l->l_len - 1;
411 } else if (l->l_len < 0) {
412 fl->fl_end = start - 1;
413 start += l->l_len;
414 if (start < 0)
415 return -EINVAL;
416 }
1da177e4 417 fl->fl_start = start; /* we record the absolute position */
4c780a46
TM
418 if (fl->fl_end < fl->fl_start)
419 return -EOVERFLOW;
1da177e4
LT
420
421 fl->fl_owner = current->files;
422 fl->fl_pid = current->tgid;
423 fl->fl_file = filp;
424 fl->fl_flags = FL_POSIX;
425 fl->fl_ops = NULL;
426 fl->fl_lmops = NULL;
427
f32cb532 428 return assign_type(fl, l->l_type);
1da177e4
LT
429}
430#endif
431
432/* default lease lock manager operations */
433static void lease_break_callback(struct file_lock *fl)
434{
435 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
436}
437
7b021967 438static const struct lock_manager_operations lease_manager_ops = {
8fb47a4f 439 .lm_break = lease_break_callback,
8fb47a4f 440 .lm_change = lease_modify,
1da177e4
LT
441};
442
443/*
444 * Initialize a lease, use the default lock manager operations
445 */
0ec4f431 446static int lease_init(struct file *filp, long type, struct file_lock *fl)
1da177e4 447 {
75dff55a
TM
448 if (assign_type(fl, type) != 0)
449 return -EINVAL;
450
1da177e4
LT
451 fl->fl_owner = current->files;
452 fl->fl_pid = current->tgid;
453
454 fl->fl_file = filp;
455 fl->fl_flags = FL_LEASE;
1da177e4
LT
456 fl->fl_start = 0;
457 fl->fl_end = OFFSET_MAX;
458 fl->fl_ops = NULL;
459 fl->fl_lmops = &lease_manager_ops;
460 return 0;
461}
462
463/* Allocate a file_lock initialised to this type of lease */
0ec4f431 464static struct file_lock *lease_alloc(struct file *filp, long type)
1da177e4
LT
465{
466 struct file_lock *fl = locks_alloc_lock();
75dff55a 467 int error = -ENOMEM;
1da177e4
LT
468
469 if (fl == NULL)
e32b8ee2 470 return ERR_PTR(error);
1da177e4
LT
471
472 error = lease_init(filp, type, fl);
75dff55a
TM
473 if (error) {
474 locks_free_lock(fl);
e32b8ee2 475 return ERR_PTR(error);
75dff55a 476 }
e32b8ee2 477 return fl;
1da177e4
LT
478}
479
480/* Check if two locks overlap each other.
481 */
482static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
483{
484 return ((fl1->fl_end >= fl2->fl_start) &&
485 (fl2->fl_end >= fl1->fl_start));
486}
487
488/*
489 * Check whether two locks have the same owner.
490 */
33443c42 491static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
1da177e4 492{
8fb47a4f 493 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
1da177e4 494 return fl2->fl_lmops == fl1->fl_lmops &&
8fb47a4f 495 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
1da177e4
LT
496 return fl1->fl_owner == fl2->fl_owner;
497}
498
88974691
JL
499static inline void
500locks_insert_global_locks(struct file_lock *fl)
501{
1c8c601a 502 spin_lock(&file_lock_lock);
88974691 503 list_add_tail(&fl->fl_link, &file_lock_list);
1c8c601a 504 spin_unlock(&file_lock_lock);
88974691
JL
505}
506
507static inline void
508locks_delete_global_locks(struct file_lock *fl)
509{
1c8c601a 510 spin_lock(&file_lock_lock);
88974691 511 list_del_init(&fl->fl_link);
1c8c601a 512 spin_unlock(&file_lock_lock);
88974691
JL
513}
514
515static inline void
516locks_insert_global_blocked(struct file_lock *waiter)
517{
518 list_add(&waiter->fl_link, &blocked_list);
519}
520
521static inline void
522locks_delete_global_blocked(struct file_lock *waiter)
523{
524 list_del_init(&waiter->fl_link);
525}
526
1da177e4
LT
527/* Remove waiter from blocker's block list.
528 * When blocker ends up pointing to itself then the list is empty.
1c8c601a
JL
529 *
530 * Must be called with file_lock_lock held.
1da177e4 531 */
33443c42 532static void __locks_delete_block(struct file_lock *waiter)
1da177e4 533{
88974691 534 locks_delete_global_blocked(waiter);
1da177e4 535 list_del_init(&waiter->fl_block);
1da177e4
LT
536 waiter->fl_next = NULL;
537}
538
1a9e64a7 539static void locks_delete_block(struct file_lock *waiter)
1da177e4 540{
1c8c601a 541 spin_lock(&file_lock_lock);
1da177e4 542 __locks_delete_block(waiter);
1c8c601a 543 spin_unlock(&file_lock_lock);
1da177e4
LT
544}
545
546/* Insert waiter into blocker's block list.
547 * We use a circular list so that processes can be easily woken up in
548 * the order they blocked. The documentation doesn't require this but
549 * it seems like the reasonable thing to do.
1c8c601a
JL
550 *
551 * Must be called with file_lock_lock held!
1da177e4 552 */
1c8c601a
JL
553static void __locks_insert_block(struct file_lock *blocker,
554 struct file_lock *waiter)
1da177e4 555{
6dc0fe8f 556 BUG_ON(!list_empty(&waiter->fl_block));
1da177e4 557 waiter->fl_next = blocker;
88974691 558 list_add_tail(&waiter->fl_block, &blocker->fl_block);
1da177e4 559 if (IS_POSIX(blocker))
1c8c601a
JL
560 locks_insert_global_blocked(waiter);
561}
562
563/* Must be called with i_lock held. */
564static void locks_insert_block(struct file_lock *blocker,
565 struct file_lock *waiter)
566{
567 spin_lock(&file_lock_lock);
568 __locks_insert_block(blocker, waiter);
569 spin_unlock(&file_lock_lock);
1da177e4
LT
570}
571
1cb36012
JL
572/*
573 * Wake up processes blocked waiting for blocker.
574 *
1c8c601a 575 * Must be called with the inode->i_lock held!
1da177e4
LT
576 */
577static void locks_wake_up_blocks(struct file_lock *blocker)
578{
1c8c601a 579 spin_lock(&file_lock_lock);
1da177e4 580 while (!list_empty(&blocker->fl_block)) {
f0c1cd0e
PE
581 struct file_lock *waiter;
582
583 waiter = list_first_entry(&blocker->fl_block,
1da177e4
LT
584 struct file_lock, fl_block);
585 __locks_delete_block(waiter);
8fb47a4f
BF
586 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
587 waiter->fl_lmops->lm_notify(waiter);
1da177e4
LT
588 else
589 wake_up(&waiter->fl_wait);
590 }
1c8c601a 591 spin_unlock(&file_lock_lock);
1da177e4
LT
592}
593
594/* Insert file lock fl into an inode's lock list at the position indicated
595 * by pos. At the same time add the lock to the global file lock list.
1c8c601a
JL
596 *
597 * Must be called with the i_lock held!
1da177e4
LT
598 */
599static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
600{
ab1f1611
VG
601 fl->fl_nspid = get_pid(task_tgid(current));
602
1da177e4
LT
603 /* insert into file's list */
604 fl->fl_next = *pos;
605 *pos = fl;
88974691
JL
606
607 locks_insert_global_locks(fl);
1da177e4
LT
608}
609
610/*
611 * Delete a lock and then free it.
612 * Wake up processes that are blocked waiting for this lock,
613 * notify the FS that the lock has been cleared and
614 * finally free the lock.
1c8c601a
JL
615 *
616 * Must be called with the i_lock held!
1da177e4
LT
617 */
618static void locks_delete_lock(struct file_lock **thisfl_p)
619{
620 struct file_lock *fl = *thisfl_p;
621
88974691
JL
622 locks_delete_global_locks(fl);
623
1da177e4
LT
624 *thisfl_p = fl->fl_next;
625 fl->fl_next = NULL;
1da177e4 626
ab1f1611
VG
627 if (fl->fl_nspid) {
628 put_pid(fl->fl_nspid);
629 fl->fl_nspid = NULL;
630 }
631
1da177e4
LT
632 locks_wake_up_blocks(fl);
633 locks_free_lock(fl);
634}
635
636/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
637 * checks for shared/exclusive status of overlapping locks.
638 */
639static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
640{
641 if (sys_fl->fl_type == F_WRLCK)
642 return 1;
643 if (caller_fl->fl_type == F_WRLCK)
644 return 1;
645 return 0;
646}
647
648/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
649 * checking before calling the locks_conflict().
650 */
651static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
652{
653 /* POSIX locks owned by the same process do not conflict with
654 * each other.
655 */
656 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
657 return (0);
658
659 /* Check whether they overlap */
660 if (!locks_overlap(caller_fl, sys_fl))
661 return 0;
662
663 return (locks_conflict(caller_fl, sys_fl));
664}
665
666/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
667 * checking before calling the locks_conflict().
668 */
669static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
670{
671 /* FLOCK locks referring to the same filp do not conflict with
672 * each other.
673 */
674 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
675 return (0);
676 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
677 return 0;
678
679 return (locks_conflict(caller_fl, sys_fl));
680}
681
6d34ac19 682void
9d6a8c5c 683posix_test_lock(struct file *filp, struct file_lock *fl)
1da177e4
LT
684{
685 struct file_lock *cfl;
1c8c601a 686 struct inode *inode = file_inode(filp);
1da177e4 687
1c8c601a 688 spin_lock(&inode->i_lock);
496ad9aa 689 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
1da177e4
LT
690 if (!IS_POSIX(cfl))
691 continue;
b842e240 692 if (posix_locks_conflict(fl, cfl))
1da177e4
LT
693 break;
694 }
ab1f1611 695 if (cfl) {
9d6a8c5c 696 __locks_copy_lock(fl, cfl);
ab1f1611 697 if (cfl->fl_nspid)
6c5f3e7b 698 fl->fl_pid = pid_vnr(cfl->fl_nspid);
ab1f1611 699 } else
129a84de 700 fl->fl_type = F_UNLCK;
1c8c601a 701 spin_unlock(&inode->i_lock);
6d34ac19 702 return;
1da177e4 703}
1da177e4
LT
704EXPORT_SYMBOL(posix_test_lock);
705
b533184f
BF
706/*
707 * Deadlock detection:
708 *
709 * We attempt to detect deadlocks that are due purely to posix file
710 * locks.
1da177e4 711 *
b533184f
BF
712 * We assume that a task can be waiting for at most one lock at a time.
713 * So for any acquired lock, the process holding that lock may be
714 * waiting on at most one other lock. That lock in turns may be held by
715 * someone waiting for at most one other lock. Given a requested lock
716 * caller_fl which is about to wait for a conflicting lock block_fl, we
717 * follow this chain of waiters to ensure we are not about to create a
718 * cycle.
1da177e4 719 *
b533184f
BF
720 * Since we do this before we ever put a process to sleep on a lock, we
721 * are ensured that there is never a cycle; that is what guarantees that
722 * the while() loop in posix_locks_deadlock() eventually completes.
97855b49 723 *
b533184f
BF
724 * Note: the above assumption may not be true when handling lock
725 * requests from a broken NFS client. It may also fail in the presence
726 * of tasks (such as posix threads) sharing the same open file table.
727 *
728 * To handle those cases, we just bail out after a few iterations.
1da177e4 729 */
97855b49
BF
730
731#define MAX_DEADLK_ITERATIONS 10
732
b533184f
BF
733/* Find a lock that the owner of the given block_fl is blocking on. */
734static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
735{
736 struct file_lock *fl;
737
738 list_for_each_entry(fl, &blocked_list, fl_link) {
739 if (posix_same_owner(fl, block_fl))
740 return fl->fl_next;
741 }
742 return NULL;
743}
744
1c8c601a 745/* Must be called with the file_lock_lock held! */
b0904e14 746static int posix_locks_deadlock(struct file_lock *caller_fl,
1da177e4
LT
747 struct file_lock *block_fl)
748{
97855b49 749 int i = 0;
1da177e4 750
b533184f
BF
751 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
752 if (i++ > MAX_DEADLK_ITERATIONS)
753 return 0;
754 if (posix_same_owner(caller_fl, block_fl))
755 return 1;
1da177e4
LT
756 }
757 return 0;
758}
759
1da177e4 760/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
02888f41 761 * after any leases, but before any posix locks.
f475ae95
TM
762 *
763 * Note that if called with an FL_EXISTS argument, the caller may determine
764 * whether or not a lock was successfully freed by testing the return
765 * value for -ENOENT.
1da177e4 766 */
993dfa87 767static int flock_lock_file(struct file *filp, struct file_lock *request)
1da177e4 768{
993dfa87 769 struct file_lock *new_fl = NULL;
1da177e4 770 struct file_lock **before;
496ad9aa 771 struct inode * inode = file_inode(filp);
1da177e4
LT
772 int error = 0;
773 int found = 0;
774
b89f4321 775 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
84d535ad 776 new_fl = locks_alloc_lock();
b89f4321
AB
777 if (!new_fl)
778 return -ENOMEM;
84d535ad
PE
779 }
780
1c8c601a 781 spin_lock(&inode->i_lock);
b89f4321
AB
782 if (request->fl_flags & FL_ACCESS)
783 goto find_conflict;
784
1da177e4
LT
785 for_each_lock(inode, before) {
786 struct file_lock *fl = *before;
787 if (IS_POSIX(fl))
788 break;
789 if (IS_LEASE(fl))
790 continue;
791 if (filp != fl->fl_file)
792 continue;
993dfa87 793 if (request->fl_type == fl->fl_type)
1da177e4
LT
794 goto out;
795 found = 1;
796 locks_delete_lock(before);
797 break;
798 }
1da177e4 799
f475ae95
TM
800 if (request->fl_type == F_UNLCK) {
801 if ((request->fl_flags & FL_EXISTS) && !found)
802 error = -ENOENT;
993dfa87 803 goto out;
f475ae95 804 }
1da177e4
LT
805
806 /*
807 * If a higher-priority process was blocked on the old file lock,
808 * give it the opportunity to lock the file.
809 */
b89f4321 810 if (found) {
1c8c601a 811 spin_unlock(&inode->i_lock);
def01bc5 812 cond_resched();
1c8c601a 813 spin_lock(&inode->i_lock);
b89f4321 814 }
1da177e4 815
f07f18dd 816find_conflict:
1da177e4
LT
817 for_each_lock(inode, before) {
818 struct file_lock *fl = *before;
819 if (IS_POSIX(fl))
820 break;
821 if (IS_LEASE(fl))
822 continue;
993dfa87 823 if (!flock_locks_conflict(request, fl))
1da177e4
LT
824 continue;
825 error = -EAGAIN;
bde74e4b
MS
826 if (!(request->fl_flags & FL_SLEEP))
827 goto out;
828 error = FILE_LOCK_DEFERRED;
829 locks_insert_block(fl, request);
1da177e4
LT
830 goto out;
831 }
f07f18dd
TM
832 if (request->fl_flags & FL_ACCESS)
833 goto out;
993dfa87 834 locks_copy_lock(new_fl, request);
0e2f6db8 835 locks_insert_lock(before, new_fl);
993dfa87 836 new_fl = NULL;
9cedc194 837 error = 0;
1da177e4
LT
838
839out:
1c8c601a 840 spin_unlock(&inode->i_lock);
993dfa87
TM
841 if (new_fl)
842 locks_free_lock(new_fl);
1da177e4
LT
843 return error;
844}
845
150b3934 846static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
1da177e4
LT
847{
848 struct file_lock *fl;
39005d02
MS
849 struct file_lock *new_fl = NULL;
850 struct file_lock *new_fl2 = NULL;
1da177e4
LT
851 struct file_lock *left = NULL;
852 struct file_lock *right = NULL;
853 struct file_lock **before;
b9746ef8
JL
854 int error;
855 bool added = false;
1da177e4
LT
856
857 /*
858 * We may need two file_lock structures for this operation,
859 * so we get them in advance to avoid races.
39005d02
MS
860 *
861 * In some cases we can be sure, that no new locks will be needed
1da177e4 862 */
39005d02
MS
863 if (!(request->fl_flags & FL_ACCESS) &&
864 (request->fl_type != F_UNLCK ||
865 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
866 new_fl = locks_alloc_lock();
867 new_fl2 = locks_alloc_lock();
868 }
1da177e4 869
1c8c601a 870 spin_lock(&inode->i_lock);
1cb36012
JL
871 /*
872 * New lock request. Walk all POSIX locks and look for conflicts. If
873 * there are any, either return error or put the request on the
874 * blocker's list of waiters and the global blocked_list.
875 */
1da177e4
LT
876 if (request->fl_type != F_UNLCK) {
877 for_each_lock(inode, before) {
526985b9 878 fl = *before;
1da177e4
LT
879 if (!IS_POSIX(fl))
880 continue;
881 if (!posix_locks_conflict(request, fl))
882 continue;
5842add2 883 if (conflock)
1a747ee0 884 __locks_copy_lock(conflock, fl);
1da177e4
LT
885 error = -EAGAIN;
886 if (!(request->fl_flags & FL_SLEEP))
887 goto out;
1c8c601a
JL
888 /*
889 * Deadlock detection and insertion into the blocked
890 * locks list must be done while holding the same lock!
891 */
1da177e4 892 error = -EDEADLK;
1c8c601a
JL
893 spin_lock(&file_lock_lock);
894 if (likely(!posix_locks_deadlock(request, fl))) {
895 error = FILE_LOCK_DEFERRED;
896 __locks_insert_block(fl, request);
897 }
898 spin_unlock(&file_lock_lock);
1da177e4
LT
899 goto out;
900 }
901 }
902
903 /* If we're just looking for a conflict, we're done. */
904 error = 0;
905 if (request->fl_flags & FL_ACCESS)
906 goto out;
907
1da177e4 908 /*
1da177e4
LT
909 * Find the first old lock with the same owner as the new lock.
910 */
911
912 before = &inode->i_flock;
913
914 /* First skip locks owned by other processes. */
915 while ((fl = *before) && (!IS_POSIX(fl) ||
916 !posix_same_owner(request, fl))) {
917 before = &fl->fl_next;
918 }
919
1cb36012 920 /* Process locks with this owner. */
1da177e4
LT
921 while ((fl = *before) && posix_same_owner(request, fl)) {
922 /* Detect adjacent or overlapping regions (if same lock type)
923 */
924 if (request->fl_type == fl->fl_type) {
449231d6
OK
925 /* In all comparisons of start vs end, use
926 * "start - 1" rather than "end + 1". If end
927 * is OFFSET_MAX, end + 1 will become negative.
928 */
1da177e4
LT
929 if (fl->fl_end < request->fl_start - 1)
930 goto next_lock;
931 /* If the next lock in the list has entirely bigger
932 * addresses than the new one, insert the lock here.
933 */
449231d6 934 if (fl->fl_start - 1 > request->fl_end)
1da177e4
LT
935 break;
936
937 /* If we come here, the new and old lock are of the
938 * same type and adjacent or overlapping. Make one
939 * lock yielding from the lower start address of both
940 * locks to the higher end address.
941 */
942 if (fl->fl_start > request->fl_start)
943 fl->fl_start = request->fl_start;
944 else
945 request->fl_start = fl->fl_start;
946 if (fl->fl_end < request->fl_end)
947 fl->fl_end = request->fl_end;
948 else
949 request->fl_end = fl->fl_end;
950 if (added) {
951 locks_delete_lock(before);
952 continue;
953 }
954 request = fl;
b9746ef8 955 added = true;
1da177e4
LT
956 }
957 else {
958 /* Processing for different lock types is a bit
959 * more complex.
960 */
961 if (fl->fl_end < request->fl_start)
962 goto next_lock;
963 if (fl->fl_start > request->fl_end)
964 break;
965 if (request->fl_type == F_UNLCK)
b9746ef8 966 added = true;
1da177e4
LT
967 if (fl->fl_start < request->fl_start)
968 left = fl;
969 /* If the next lock in the list has a higher end
970 * address than the new one, insert the new one here.
971 */
972 if (fl->fl_end > request->fl_end) {
973 right = fl;
974 break;
975 }
976 if (fl->fl_start >= request->fl_start) {
977 /* The new lock completely replaces an old
978 * one (This may happen several times).
979 */
980 if (added) {
981 locks_delete_lock(before);
982 continue;
983 }
984 /* Replace the old lock with the new one.
985 * Wake up anybody waiting for the old one,
986 * as the change in lock type might satisfy
987 * their needs.
988 */
989 locks_wake_up_blocks(fl);
990 fl->fl_start = request->fl_start;
991 fl->fl_end = request->fl_end;
992 fl->fl_type = request->fl_type;
47831f35
TM
993 locks_release_private(fl);
994 locks_copy_private(fl, request);
1da177e4 995 request = fl;
b9746ef8 996 added = true;
1da177e4
LT
997 }
998 }
999 /* Go on to next lock.
1000 */
1001 next_lock:
1002 before = &fl->fl_next;
1003 }
1004
0d9a490a 1005 /*
1cb36012
JL
1006 * The above code only modifies existing locks in case of merging or
1007 * replacing. If new lock(s) need to be inserted all modifications are
1008 * done below this, so it's safe yet to bail out.
0d9a490a
MS
1009 */
1010 error = -ENOLCK; /* "no luck" */
1011 if (right && left == right && !new_fl2)
1012 goto out;
1013
1da177e4
LT
1014 error = 0;
1015 if (!added) {
f475ae95
TM
1016 if (request->fl_type == F_UNLCK) {
1017 if (request->fl_flags & FL_EXISTS)
1018 error = -ENOENT;
1da177e4 1019 goto out;
f475ae95 1020 }
0d9a490a
MS
1021
1022 if (!new_fl) {
1023 error = -ENOLCK;
1024 goto out;
1025 }
1da177e4
LT
1026 locks_copy_lock(new_fl, request);
1027 locks_insert_lock(before, new_fl);
1028 new_fl = NULL;
1029 }
1030 if (right) {
1031 if (left == right) {
1032 /* The new lock breaks the old one in two pieces,
1033 * so we have to use the second new lock.
1034 */
1035 left = new_fl2;
1036 new_fl2 = NULL;
1037 locks_copy_lock(left, right);
1038 locks_insert_lock(before, left);
1039 }
1040 right->fl_start = request->fl_end + 1;
1041 locks_wake_up_blocks(right);
1042 }
1043 if (left) {
1044 left->fl_end = request->fl_start - 1;
1045 locks_wake_up_blocks(left);
1046 }
1047 out:
1c8c601a 1048 spin_unlock(&inode->i_lock);
1da177e4
LT
1049 /*
1050 * Free any unused locks.
1051 */
1052 if (new_fl)
1053 locks_free_lock(new_fl);
1054 if (new_fl2)
1055 locks_free_lock(new_fl2);
1056 return error;
1057}
1058
1059/**
1060 * posix_lock_file - Apply a POSIX-style lock to a file
1061 * @filp: The file to apply the lock to
1062 * @fl: The lock to be applied
150b3934 1063 * @conflock: Place to return a copy of the conflicting lock, if found.
1da177e4
LT
1064 *
1065 * Add a POSIX style lock to a file.
1066 * We merge adjacent & overlapping locks whenever possible.
1067 * POSIX locks are sorted by owner task, then by starting address
f475ae95
TM
1068 *
1069 * Note that if called with an FL_EXISTS argument, the caller may determine
1070 * whether or not a lock was successfully freed by testing the return
1071 * value for -ENOENT.
1da177e4 1072 */
150b3934 1073int posix_lock_file(struct file *filp, struct file_lock *fl,
5842add2
AA
1074 struct file_lock *conflock)
1075{
496ad9aa 1076 return __posix_lock_file(file_inode(filp), fl, conflock);
1da177e4 1077}
150b3934 1078EXPORT_SYMBOL(posix_lock_file);
1da177e4
LT
1079
1080/**
1081 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1082 * @filp: The file to apply the lock to
1083 * @fl: The lock to be applied
1084 *
1085 * Add a POSIX style lock to a file.
1086 * We merge adjacent & overlapping locks whenever possible.
1087 * POSIX locks are sorted by owner task, then by starting address
1088 */
1089int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1090{
1091 int error;
1092 might_sleep ();
1093 for (;;) {
150b3934 1094 error = posix_lock_file(filp, fl, NULL);
bde74e4b 1095 if (error != FILE_LOCK_DEFERRED)
1da177e4
LT
1096 break;
1097 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1098 if (!error)
1099 continue;
1100
1101 locks_delete_block(fl);
1102 break;
1103 }
1104 return error;
1105}
1106EXPORT_SYMBOL(posix_lock_file_wait);
1107
1108/**
1109 * locks_mandatory_locked - Check for an active lock
1110 * @inode: the file to check
1111 *
1112 * Searches the inode's list of locks to find any POSIX locks which conflict.
1113 * This function is called from locks_verify_locked() only.
1114 */
1115int locks_mandatory_locked(struct inode *inode)
1116{
1117 fl_owner_t owner = current->files;
1118 struct file_lock *fl;
1119
1120 /*
1121 * Search the lock list for this inode for any POSIX locks.
1122 */
1c8c601a 1123 spin_lock(&inode->i_lock);
1da177e4
LT
1124 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1125 if (!IS_POSIX(fl))
1126 continue;
1127 if (fl->fl_owner != owner)
1128 break;
1129 }
1c8c601a 1130 spin_unlock(&inode->i_lock);
1da177e4
LT
1131 return fl ? -EAGAIN : 0;
1132}
1133
1134/**
1135 * locks_mandatory_area - Check for a conflicting lock
1136 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1137 * for shared
1138 * @inode: the file to check
1139 * @filp: how the file was opened (if it was)
1140 * @offset: start of area to check
1141 * @count: length of area to check
1142 *
1143 * Searches the inode's list of locks to find any POSIX locks which conflict.
1144 * This function is called from rw_verify_area() and
1145 * locks_verify_truncate().
1146 */
1147int locks_mandatory_area(int read_write, struct inode *inode,
1148 struct file *filp, loff_t offset,
1149 size_t count)
1150{
1151 struct file_lock fl;
1152 int error;
1153
1154 locks_init_lock(&fl);
1155 fl.fl_owner = current->files;
1156 fl.fl_pid = current->tgid;
1157 fl.fl_file = filp;
1158 fl.fl_flags = FL_POSIX | FL_ACCESS;
1159 if (filp && !(filp->f_flags & O_NONBLOCK))
1160 fl.fl_flags |= FL_SLEEP;
1161 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1162 fl.fl_start = offset;
1163 fl.fl_end = offset + count - 1;
1164
1165 for (;;) {
150b3934 1166 error = __posix_lock_file(inode, &fl, NULL);
bde74e4b 1167 if (error != FILE_LOCK_DEFERRED)
1da177e4
LT
1168 break;
1169 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1170 if (!error) {
1171 /*
1172 * If we've been sleeping someone might have
1173 * changed the permissions behind our back.
1174 */
a16877ca 1175 if (__mandatory_lock(inode))
1da177e4
LT
1176 continue;
1177 }
1178
1179 locks_delete_block(&fl);
1180 break;
1181 }
1182
1183 return error;
1184}
1185
1186EXPORT_SYMBOL(locks_mandatory_area);
1187
778fc546
BF
1188static void lease_clear_pending(struct file_lock *fl, int arg)
1189{
1190 switch (arg) {
1191 case F_UNLCK:
1192 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1193 /* fall through: */
1194 case F_RDLCK:
1195 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1196 }
1197}
1198
1da177e4
LT
1199/* We already had a lease on this file; just change its type */
1200int lease_modify(struct file_lock **before, int arg)
1201{
1202 struct file_lock *fl = *before;
1203 int error = assign_type(fl, arg);
1204
1205 if (error)
1206 return error;
778fc546 1207 lease_clear_pending(fl, arg);
1da177e4 1208 locks_wake_up_blocks(fl);
3b6e2723
FB
1209 if (arg == F_UNLCK) {
1210 struct file *filp = fl->fl_file;
1211
1212 f_delown(filp);
1213 filp->f_owner.signum = 0;
96d6d59c
BF
1214 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1215 if (fl->fl_fasync != NULL) {
1216 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1217 fl->fl_fasync = NULL;
1218 }
1da177e4 1219 locks_delete_lock(before);
3b6e2723 1220 }
1da177e4
LT
1221 return 0;
1222}
1223
1224EXPORT_SYMBOL(lease_modify);
1225
778fc546
BF
1226static bool past_time(unsigned long then)
1227{
1228 if (!then)
1229 /* 0 is a special value meaning "this never expires": */
1230 return false;
1231 return time_after(jiffies, then);
1232}
1233
1da177e4
LT
1234static void time_out_leases(struct inode *inode)
1235{
1236 struct file_lock **before;
1237 struct file_lock *fl;
1238
1239 before = &inode->i_flock;
ab83fa4b 1240 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
778fc546
BF
1241 if (past_time(fl->fl_downgrade_time))
1242 lease_modify(before, F_RDLCK);
1243 if (past_time(fl->fl_break_time))
1244 lease_modify(before, F_UNLCK);
1da177e4
LT
1245 if (fl == *before) /* lease_modify may have freed fl */
1246 before = &fl->fl_next;
1247 }
1248}
1249
1250/**
1251 * __break_lease - revoke all outstanding leases on file
1252 * @inode: the inode of the file to return
1253 * @mode: the open mode (read or write)
1254 *
87250dd2 1255 * break_lease (inlined for speed) has checked there already is at least
1256 * some kind of lock (maybe a lease) on this file. Leases are broken on
1257 * a call to open() or truncate(). This function can sleep unless you
1da177e4
LT
1258 * specified %O_NONBLOCK to your open().
1259 */
1260int __break_lease(struct inode *inode, unsigned int mode)
1261{
778fc546 1262 int error = 0;
1da177e4
LT
1263 struct file_lock *new_fl, *flock;
1264 struct file_lock *fl;
1da177e4
LT
1265 unsigned long break_time;
1266 int i_have_this_lease = 0;
8737c930 1267 int want_write = (mode & O_ACCMODE) != O_RDONLY;
1da177e4 1268
8737c930 1269 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
6d4b9e38
LT
1270 if (IS_ERR(new_fl))
1271 return PTR_ERR(new_fl);
1da177e4 1272
1c8c601a 1273 spin_lock(&inode->i_lock);
1da177e4
LT
1274
1275 time_out_leases(inode);
1276
1277 flock = inode->i_flock;
1278 if ((flock == NULL) || !IS_LEASE(flock))
1279 goto out;
1280
778fc546
BF
1281 if (!locks_conflict(flock, new_fl))
1282 goto out;
1283
1da177e4
LT
1284 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1285 if (fl->fl_owner == current->files)
1286 i_have_this_lease = 1;
1287
1da177e4
LT
1288 break_time = 0;
1289 if (lease_break_time > 0) {
1290 break_time = jiffies + lease_break_time * HZ;
1291 if (break_time == 0)
1292 break_time++; /* so that 0 means no break time */
1293 }
1294
1295 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
778fc546
BF
1296 if (want_write) {
1297 if (fl->fl_flags & FL_UNLOCK_PENDING)
1298 continue;
1299 fl->fl_flags |= FL_UNLOCK_PENDING;
1da177e4 1300 fl->fl_break_time = break_time;
778fc546
BF
1301 } else {
1302 if (lease_breaking(flock))
1303 continue;
1304 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1305 fl->fl_downgrade_time = break_time;
1da177e4 1306 }
778fc546 1307 fl->fl_lmops->lm_break(fl);
1da177e4
LT
1308 }
1309
1310 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1311 error = -EWOULDBLOCK;
1312 goto out;
1313 }
1314
1315restart:
1316 break_time = flock->fl_break_time;
1317 if (break_time != 0) {
1318 break_time -= jiffies;
1319 if (break_time == 0)
1320 break_time++;
1321 }
4321e01e 1322 locks_insert_block(flock, new_fl);
1c8c601a 1323 spin_unlock(&inode->i_lock);
4321e01e
MW
1324 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1325 !new_fl->fl_next, break_time);
1c8c601a
JL
1326 spin_lock(&inode->i_lock);
1327 locks_delete_block(new_fl);
1da177e4
LT
1328 if (error >= 0) {
1329 if (error == 0)
1330 time_out_leases(inode);
778fc546
BF
1331 /*
1332 * Wait for the next conflicting lease that has not been
1333 * broken yet
1334 */
1da177e4
LT
1335 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1336 flock = flock->fl_next) {
778fc546 1337 if (locks_conflict(new_fl, flock))
1da177e4
LT
1338 goto restart;
1339 }
1340 error = 0;
1341 }
1342
1343out:
1c8c601a 1344 spin_unlock(&inode->i_lock);
6d4b9e38 1345 locks_free_lock(new_fl);
1da177e4
LT
1346 return error;
1347}
1348
1349EXPORT_SYMBOL(__break_lease);
1350
1351/**
a6b91919 1352 * lease_get_mtime - get the last modified time of an inode
1da177e4
LT
1353 * @inode: the inode
1354 * @time: pointer to a timespec which will contain the last modified time
1355 *
1356 * This is to force NFS clients to flush their caches for files with
1357 * exclusive leases. The justification is that if someone has an
a6b91919 1358 * exclusive lease, then they could be modifying it.
1da177e4
LT
1359 */
1360void lease_get_mtime(struct inode *inode, struct timespec *time)
1361{
1362 struct file_lock *flock = inode->i_flock;
0ee5c6d6 1363 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
1da177e4
LT
1364 *time = current_fs_time(inode->i_sb);
1365 else
1366 *time = inode->i_mtime;
1367}
1368
1369EXPORT_SYMBOL(lease_get_mtime);
1370
1371/**
1372 * fcntl_getlease - Enquire what lease is currently active
1373 * @filp: the file
1374 *
1375 * The value returned by this function will be one of
1376 * (if no lease break is pending):
1377 *
1378 * %F_RDLCK to indicate a shared lease is held.
1379 *
1380 * %F_WRLCK to indicate an exclusive lease is held.
1381 *
1382 * %F_UNLCK to indicate no lease is held.
1383 *
1384 * (if a lease break is pending):
1385 *
1386 * %F_RDLCK to indicate an exclusive lease needs to be
1387 * changed to a shared lease (or removed).
1388 *
1389 * %F_UNLCK to indicate the lease needs to be removed.
1390 *
1391 * XXX: sfr & willy disagree over whether F_INPROGRESS
1392 * should be returned to userspace.
1393 */
1394int fcntl_getlease(struct file *filp)
1395{
1396 struct file_lock *fl;
1c8c601a 1397 struct inode *inode = file_inode(filp);
1da177e4
LT
1398 int type = F_UNLCK;
1399
1c8c601a 1400 spin_lock(&inode->i_lock);
496ad9aa
AV
1401 time_out_leases(file_inode(filp));
1402 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
1da177e4
LT
1403 fl = fl->fl_next) {
1404 if (fl->fl_file == filp) {
778fc546 1405 type = target_leasetype(fl);
1da177e4
LT
1406 break;
1407 }
1408 }
1c8c601a 1409 spin_unlock(&inode->i_lock);
1da177e4
LT
1410 return type;
1411}
1412
d4f22d19 1413static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
1da177e4 1414{
7eaae282 1415 struct file_lock *fl, **before, **my_before = NULL, *lease;
0f7fc9e4 1416 struct dentry *dentry = filp->f_path.dentry;
1da177e4 1417 struct inode *inode = dentry->d_inode;
c1f24ef4 1418 int error;
1da177e4 1419
096657b6
BF
1420 lease = *flp;
1421
8335ebd9
BF
1422 error = -EAGAIN;
1423 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
096657b6 1424 goto out;
8335ebd9
BF
1425 if ((arg == F_WRLCK)
1426 && ((dentry->d_count > 1)
1427 || (atomic_read(&inode->i_count) > 1)))
096657b6 1428 goto out;
6d5e8b05 1429
1da177e4
LT
1430 /*
1431 * At this point, we know that if there is an exclusive
1432 * lease on this file, then we hold it on this filp
1433 * (otherwise our open of this file would have blocked).
1434 * And if we are trying to acquire an exclusive lease,
1435 * then the file is not open by anyone (including us)
1436 * except for this filp.
1437 */
c1f24ef4 1438 error = -EAGAIN;
1da177e4
LT
1439 for (before = &inode->i_flock;
1440 ((fl = *before) != NULL) && IS_LEASE(fl);
1441 before = &fl->fl_next) {
c1f24ef4 1442 if (fl->fl_file == filp) {
1da177e4 1443 my_before = before;
c1f24ef4
BF
1444 continue;
1445 }
1446 /*
1447 * No exclusive leases if someone else has a lease on
1448 * this file:
1449 */
1450 if (arg == F_WRLCK)
1451 goto out;
1452 /*
1453 * Modifying our existing lease is OK, but no getting a
1454 * new lease if someone else is opening for write:
1455 */
1456 if (fl->fl_flags & FL_UNLOCK_PENDING)
1457 goto out;
1da177e4
LT
1458 }
1459
1da177e4 1460 if (my_before != NULL) {
8fb47a4f 1461 error = lease->fl_lmops->lm_change(my_before, arg);
51ee4b84
CH
1462 if (!error)
1463 *flp = *my_before;
1da177e4
LT
1464 goto out;
1465 }
1466
1da177e4
LT
1467 error = -EINVAL;
1468 if (!leases_enable)
1469 goto out;
1470
c5b1f0d9 1471 locks_insert_lock(before, lease);
85c59580 1472 return 0;
1da177e4 1473
1da177e4
LT
1474out:
1475 return error;
1476}
8335ebd9 1477
d4f22d19 1478static int generic_delete_lease(struct file *filp, struct file_lock **flp)
8335ebd9
BF
1479{
1480 struct file_lock *fl, **before;
1481 struct dentry *dentry = filp->f_path.dentry;
1482 struct inode *inode = dentry->d_inode;
1483
1484 for (before = &inode->i_flock;
1485 ((fl = *before) != NULL) && IS_LEASE(fl);
1486 before = &fl->fl_next) {
1487 if (fl->fl_file != filp)
1488 continue;
1489 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1490 }
1491 return -EAGAIN;
1492}
1493
1494/**
1495 * generic_setlease - sets a lease on an open file
1496 * @filp: file pointer
1497 * @arg: type of lease to obtain
1498 * @flp: input - file_lock to use, output - file_lock inserted
1499 *
1500 * The (input) flp->fl_lmops->lm_break function is required
1501 * by break_lease().
1502 *
1c8c601a 1503 * Called with inode->i_lock held.
8335ebd9
BF
1504 */
1505int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1506{
1507 struct dentry *dentry = filp->f_path.dentry;
1508 struct inode *inode = dentry->d_inode;
1509 int error;
1510
8e96e3b7 1511 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
8335ebd9
BF
1512 return -EACCES;
1513 if (!S_ISREG(inode->i_mode))
1514 return -EINVAL;
1515 error = security_file_lock(filp, arg);
1516 if (error)
1517 return error;
1518
1519 time_out_leases(inode);
1520
1521 BUG_ON(!(*flp)->fl_lmops->lm_break);
1522
1523 switch (arg) {
1524 case F_UNLCK:
1525 return generic_delete_lease(filp, flp);
1526 case F_RDLCK:
1527 case F_WRLCK:
1528 return generic_add_lease(filp, arg, flp);
1529 default:
8d657eb3 1530 return -EINVAL;
8335ebd9
BF
1531 }
1532}
0af1a450 1533EXPORT_SYMBOL(generic_setlease);
1da177e4 1534
b89f4321
AB
1535static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1536{
1537 if (filp->f_op && filp->f_op->setlease)
1538 return filp->f_op->setlease(filp, arg, lease);
1539 else
1540 return generic_setlease(filp, arg, lease);
1541}
1542
1543/**
a9933cea 1544 * vfs_setlease - sets a lease on an open file
1da177e4
LT
1545 * @filp: file pointer
1546 * @arg: type of lease to obtain
1547 * @lease: file_lock to use
1548 *
1549 * Call this to establish a lease on the file.
8fb47a4f 1550 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
f9ffed26
BF
1551 * break_lease will oops!
1552 *
1553 * This will call the filesystem's setlease file method, if
1554 * defined. Note that there is no getlease method; instead, the
1555 * filesystem setlease method should call back to setlease() to
1556 * add a lease to the inode's lease list, where fcntl_getlease() can
1557 * find it. Since fcntl_getlease() only reports whether the current
1558 * task holds a lease, a cluster filesystem need only do this for
1559 * leases held by processes on this node.
1560 *
1561 * There is also no break_lease method; filesystems that
c9404c9c 1562 * handle their own leases should break leases themselves from the
f9ffed26
BF
1563 * filesystem's open, create, and (on truncate) setattr methods.
1564 *
1565 * Warning: the only current setlease methods exist only to disable
1566 * leases in certain cases. More vfs changes may be required to
1567 * allow a full filesystem lease implementation.
1da177e4
LT
1568 */
1569
a9933cea 1570int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1da177e4 1571{
1c8c601a 1572 struct inode *inode = file_inode(filp);
1da177e4
LT
1573 int error;
1574
1c8c601a 1575 spin_lock(&inode->i_lock);
b89f4321 1576 error = __vfs_setlease(filp, arg, lease);
1c8c601a 1577 spin_unlock(&inode->i_lock);
1da177e4
LT
1578
1579 return error;
1580}
a9933cea 1581EXPORT_SYMBOL_GPL(vfs_setlease);
1da177e4 1582
0ceaf6c7
BF
1583static int do_fcntl_delete_lease(struct file *filp)
1584{
1585 struct file_lock fl, *flp = &fl;
1586
1587 lease_init(filp, F_UNLCK, flp);
1588
1589 return vfs_setlease(filp, F_UNLCK, &flp);
1590}
1591
1592static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
1da177e4 1593{
3df057ac 1594 struct file_lock *fl, *ret;
1c8c601a 1595 struct inode *inode = file_inode(filp);
f7347ce4 1596 struct fasync_struct *new;
1da177e4
LT
1597 int error;
1598
c5b1f0d9
AB
1599 fl = lease_alloc(filp, arg);
1600 if (IS_ERR(fl))
1601 return PTR_ERR(fl);
1da177e4 1602
f7347ce4
LT
1603 new = fasync_alloc();
1604 if (!new) {
1605 locks_free_lock(fl);
1606 return -ENOMEM;
1607 }
3df057ac 1608 ret = fl;
1c8c601a 1609 spin_lock(&inode->i_lock);
8896b93f 1610 error = __vfs_setlease(filp, arg, &ret);
51ee4b84 1611 if (error) {
1c8c601a 1612 spin_unlock(&inode->i_lock);
51ee4b84
CH
1613 locks_free_lock(fl);
1614 goto out_free_fasync;
1615 }
3df057ac
BF
1616 if (ret != fl)
1617 locks_free_lock(fl);
1da177e4 1618
f7347ce4
LT
1619 /*
1620 * fasync_insert_entry() returns the old entry if any.
1621 * If there was no old entry, then it used 'new' and
1622 * inserted it into the fasync list. Clear new so that
1623 * we don't release it here.
1624 */
3df057ac 1625 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
f7347ce4
LT
1626 new = NULL;
1627
8896b93f 1628 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
1c8c601a 1629 spin_unlock(&inode->i_lock);
51ee4b84
CH
1630
1631out_free_fasync:
f7347ce4
LT
1632 if (new)
1633 fasync_free(new);
1da177e4
LT
1634 return error;
1635}
1636
0ceaf6c7
BF
1637/**
1638 * fcntl_setlease - sets a lease on an open file
1639 * @fd: open file descriptor
1640 * @filp: file pointer
1641 * @arg: type of lease to obtain
1642 *
1643 * Call this fcntl to establish a lease on the file.
1644 * Note that you also need to call %F_SETSIG to
1645 * receive a signal when the lease is broken.
1646 */
1647int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1648{
1649 if (arg == F_UNLCK)
1650 return do_fcntl_delete_lease(filp);
1651 return do_fcntl_add_lease(fd, filp, arg);
1652}
1653
1da177e4
LT
1654/**
1655 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1656 * @filp: The file to apply the lock to
1657 * @fl: The lock to be applied
1658 *
1659 * Add a FLOCK style lock to a file.
1660 */
1661int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1662{
1663 int error;
1664 might_sleep();
1665 for (;;) {
1666 error = flock_lock_file(filp, fl);
bde74e4b 1667 if (error != FILE_LOCK_DEFERRED)
1da177e4
LT
1668 break;
1669 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1670 if (!error)
1671 continue;
1672
1673 locks_delete_block(fl);
1674 break;
1675 }
1676 return error;
1677}
1678
1679EXPORT_SYMBOL(flock_lock_file_wait);
1680
1681/**
1682 * sys_flock: - flock() system call.
1683 * @fd: the file descriptor to lock.
1684 * @cmd: the type of lock to apply.
1685 *
1686 * Apply a %FL_FLOCK style lock to an open file descriptor.
1687 * The @cmd can be one of
1688 *
1689 * %LOCK_SH -- a shared lock.
1690 *
1691 * %LOCK_EX -- an exclusive lock.
1692 *
1693 * %LOCK_UN -- remove an existing lock.
1694 *
1695 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1696 *
1697 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1698 * processes read and write access respectively.
1699 */
002c8976 1700SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
1da177e4 1701{
2903ff01 1702 struct fd f = fdget(fd);
1da177e4
LT
1703 struct file_lock *lock;
1704 int can_sleep, unlock;
1705 int error;
1706
1707 error = -EBADF;
2903ff01 1708 if (!f.file)
1da177e4
LT
1709 goto out;
1710
1711 can_sleep = !(cmd & LOCK_NB);
1712 cmd &= ~LOCK_NB;
1713 unlock = (cmd == LOCK_UN);
1714
aeb5d727 1715 if (!unlock && !(cmd & LOCK_MAND) &&
2903ff01 1716 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
1da177e4
LT
1717 goto out_putf;
1718
2903ff01 1719 error = flock_make_lock(f.file, &lock, cmd);
1da177e4
LT
1720 if (error)
1721 goto out_putf;
1722 if (can_sleep)
1723 lock->fl_flags |= FL_SLEEP;
1724
2903ff01 1725 error = security_file_lock(f.file, lock->fl_type);
1da177e4
LT
1726 if (error)
1727 goto out_free;
1728
2903ff01
AV
1729 if (f.file->f_op && f.file->f_op->flock)
1730 error = f.file->f_op->flock(f.file,
1da177e4
LT
1731 (can_sleep) ? F_SETLKW : F_SETLK,
1732 lock);
1733 else
2903ff01 1734 error = flock_lock_file_wait(f.file, lock);
1da177e4
LT
1735
1736 out_free:
993dfa87 1737 locks_free_lock(lock);
1da177e4
LT
1738
1739 out_putf:
2903ff01 1740 fdput(f);
1da177e4
LT
1741 out:
1742 return error;
1743}
1744
3ee17abd
BF
1745/**
1746 * vfs_test_lock - test file byte range lock
1747 * @filp: The file to test lock for
6924c554 1748 * @fl: The lock to test; also used to hold result
3ee17abd
BF
1749 *
1750 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1751 * setting conf->fl_type to something other than F_UNLCK.
1752 */
1753int vfs_test_lock(struct file *filp, struct file_lock *fl)
1754{
1755 if (filp->f_op && filp->f_op->lock)
1756 return filp->f_op->lock(filp, F_GETLK, fl);
1757 posix_test_lock(filp, fl);
1758 return 0;
1759}
1760EXPORT_SYMBOL_GPL(vfs_test_lock);
1761
c2fa1b8a
BF
1762static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1763{
1764 flock->l_pid = fl->fl_pid;
1765#if BITS_PER_LONG == 32
1766 /*
1767 * Make sure we can represent the posix lock via
1768 * legacy 32bit flock.
1769 */
1770 if (fl->fl_start > OFFT_OFFSET_MAX)
1771 return -EOVERFLOW;
1772 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1773 return -EOVERFLOW;
1774#endif
1775 flock->l_start = fl->fl_start;
1776 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1777 fl->fl_end - fl->fl_start + 1;
1778 flock->l_whence = 0;
129a84de 1779 flock->l_type = fl->fl_type;
c2fa1b8a
BF
1780 return 0;
1781}
1782
1783#if BITS_PER_LONG == 32
1784static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1785{
1786 flock->l_pid = fl->fl_pid;
1787 flock->l_start = fl->fl_start;
1788 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1789 fl->fl_end - fl->fl_start + 1;
1790 flock->l_whence = 0;
1791 flock->l_type = fl->fl_type;
1792}
1793#endif
1794
1da177e4
LT
1795/* Report the first existing lock that would conflict with l.
1796 * This implements the F_GETLK command of fcntl().
1797 */
1798int fcntl_getlk(struct file *filp, struct flock __user *l)
1799{
9d6a8c5c 1800 struct file_lock file_lock;
1da177e4
LT
1801 struct flock flock;
1802 int error;
1803
1804 error = -EFAULT;
1805 if (copy_from_user(&flock, l, sizeof(flock)))
1806 goto out;
1807 error = -EINVAL;
1808 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1809 goto out;
1810
1811 error = flock_to_posix_lock(filp, &file_lock, &flock);
1812 if (error)
1813 goto out;
1814
3ee17abd
BF
1815 error = vfs_test_lock(filp, &file_lock);
1816 if (error)
1817 goto out;
1da177e4 1818
9d6a8c5c
ME
1819 flock.l_type = file_lock.fl_type;
1820 if (file_lock.fl_type != F_UNLCK) {
1821 error = posix_lock_to_flock(&flock, &file_lock);
c2fa1b8a 1822 if (error)
1da177e4 1823 goto out;
1da177e4
LT
1824 }
1825 error = -EFAULT;
1826 if (!copy_to_user(l, &flock, sizeof(flock)))
1827 error = 0;
1828out:
1829 return error;
1830}
1831
7723ec97
ME
1832/**
1833 * vfs_lock_file - file byte range lock
1834 * @filp: The file to apply the lock to
1835 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1836 * @fl: The lock to be applied
150b3934
ME
1837 * @conf: Place to return a copy of the conflicting lock, if found.
1838 *
1839 * A caller that doesn't care about the conflicting lock may pass NULL
1840 * as the final argument.
1841 *
1842 * If the filesystem defines a private ->lock() method, then @conf will
1843 * be left unchanged; so a caller that cares should initialize it to
1844 * some acceptable default.
2beb6614
ME
1845 *
1846 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1847 * locks, the ->lock() interface may return asynchronously, before the lock has
1848 * been granted or denied by the underlying filesystem, if (and only if)
8fb47a4f 1849 * lm_grant is set. Callers expecting ->lock() to return asynchronously
2beb6614
ME
1850 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1851 * the request is for a blocking lock. When ->lock() does return asynchronously,
8fb47a4f 1852 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
2beb6614
ME
1853 * request completes.
1854 * If the request is for non-blocking lock the file system should return
bde74e4b
MS
1855 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1856 * with the result. If the request timed out the callback routine will return a
2beb6614
ME
1857 * nonzero return code and the file system should release the lock. The file
1858 * system is also responsible to keep a corresponding posix lock when it
1859 * grants a lock so the VFS can find out which locks are locally held and do
1860 * the correct lock cleanup when required.
1861 * The underlying filesystem must not drop the kernel lock or call
8fb47a4f 1862 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
2beb6614 1863 * return code.
7723ec97 1864 */
150b3934 1865int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
7723ec97
ME
1866{
1867 if (filp->f_op && filp->f_op->lock)
1868 return filp->f_op->lock(filp, cmd, fl);
1869 else
150b3934 1870 return posix_lock_file(filp, fl, conf);
7723ec97
ME
1871}
1872EXPORT_SYMBOL_GPL(vfs_lock_file);
1873
b648a6de
MS
1874static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1875 struct file_lock *fl)
1876{
1877 int error;
1878
1879 error = security_file_lock(filp, fl->fl_type);
1880 if (error)
1881 return error;
1882
764c76b3
MS
1883 for (;;) {
1884 error = vfs_lock_file(filp, cmd, fl, NULL);
1885 if (error != FILE_LOCK_DEFERRED)
b648a6de 1886 break;
764c76b3
MS
1887 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1888 if (!error)
1889 continue;
1890
1891 locks_delete_block(fl);
1892 break;
b648a6de
MS
1893 }
1894
1895 return error;
1896}
1897
1da177e4
LT
1898/* Apply the lock described by l to an open file descriptor.
1899 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1900 */
c293621b
PS
1901int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1902 struct flock __user *l)
1da177e4
LT
1903{
1904 struct file_lock *file_lock = locks_alloc_lock();
1905 struct flock flock;
1906 struct inode *inode;
0b2bac2f 1907 struct file *f;
1da177e4
LT
1908 int error;
1909
1910 if (file_lock == NULL)
1911 return -ENOLCK;
1912
1913 /*
1914 * This might block, so we do it before checking the inode.
1915 */
1916 error = -EFAULT;
1917 if (copy_from_user(&flock, l, sizeof(flock)))
1918 goto out;
1919
496ad9aa 1920 inode = file_inode(filp);
1da177e4
LT
1921
1922 /* Don't allow mandatory locks on files that may be memory mapped
1923 * and shared.
1924 */
a16877ca 1925 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1da177e4
LT
1926 error = -EAGAIN;
1927 goto out;
1928 }
1929
c293621b 1930again:
1da177e4
LT
1931 error = flock_to_posix_lock(filp, file_lock, &flock);
1932 if (error)
1933 goto out;
1934 if (cmd == F_SETLKW) {
1935 file_lock->fl_flags |= FL_SLEEP;
1936 }
1937
1938 error = -EBADF;
1939 switch (flock.l_type) {
1940 case F_RDLCK:
1941 if (!(filp->f_mode & FMODE_READ))
1942 goto out;
1943 break;
1944 case F_WRLCK:
1945 if (!(filp->f_mode & FMODE_WRITE))
1946 goto out;
1947 break;
1948 case F_UNLCK:
1949 break;
1950 default:
1951 error = -EINVAL;
1952 goto out;
1953 }
1954
b648a6de 1955 error = do_lock_file_wait(filp, cmd, file_lock);
1da177e4 1956
c293621b
PS
1957 /*
1958 * Attempt to detect a close/fcntl race and recover by
1959 * releasing the lock that was just acquired.
1960 */
0b2bac2f
AV
1961 /*
1962 * we need that spin_lock here - it prevents reordering between
1963 * update of inode->i_flock and check for it done in close().
1964 * rcu_read_lock() wouldn't do.
1965 */
1966 spin_lock(&current->files->file_lock);
1967 f = fcheck(fd);
1968 spin_unlock(&current->files->file_lock);
1969 if (!error && f != filp && flock.l_type != F_UNLCK) {
c293621b
PS
1970 flock.l_type = F_UNLCK;
1971 goto again;
1da177e4
LT
1972 }
1973
c293621b 1974out:
1da177e4
LT
1975 locks_free_lock(file_lock);
1976 return error;
1977}
1978
1979#if BITS_PER_LONG == 32
1980/* Report the first existing lock that would conflict with l.
1981 * This implements the F_GETLK command of fcntl().
1982 */
1983int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1984{
9d6a8c5c 1985 struct file_lock file_lock;
1da177e4
LT
1986 struct flock64 flock;
1987 int error;
1988
1989 error = -EFAULT;
1990 if (copy_from_user(&flock, l, sizeof(flock)))
1991 goto out;
1992 error = -EINVAL;
1993 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1994 goto out;
1995
1996 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1997 if (error)
1998 goto out;
1999
3ee17abd
BF
2000 error = vfs_test_lock(filp, &file_lock);
2001 if (error)
2002 goto out;
2003
9d6a8c5c
ME
2004 flock.l_type = file_lock.fl_type;
2005 if (file_lock.fl_type != F_UNLCK)
2006 posix_lock_to_flock64(&flock, &file_lock);
2007
1da177e4
LT
2008 error = -EFAULT;
2009 if (!copy_to_user(l, &flock, sizeof(flock)))
2010 error = 0;
2011
2012out:
2013 return error;
2014}
2015
2016/* Apply the lock described by l to an open file descriptor.
2017 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2018 */
c293621b
PS
2019int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2020 struct flock64 __user *l)
1da177e4
LT
2021{
2022 struct file_lock *file_lock = locks_alloc_lock();
2023 struct flock64 flock;
2024 struct inode *inode;
0b2bac2f 2025 struct file *f;
1da177e4
LT
2026 int error;
2027
2028 if (file_lock == NULL)
2029 return -ENOLCK;
2030
2031 /*
2032 * This might block, so we do it before checking the inode.
2033 */
2034 error = -EFAULT;
2035 if (copy_from_user(&flock, l, sizeof(flock)))
2036 goto out;
2037
496ad9aa 2038 inode = file_inode(filp);
1da177e4
LT
2039
2040 /* Don't allow mandatory locks on files that may be memory mapped
2041 * and shared.
2042 */
a16877ca 2043 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
1da177e4
LT
2044 error = -EAGAIN;
2045 goto out;
2046 }
2047
c293621b 2048again:
1da177e4
LT
2049 error = flock64_to_posix_lock(filp, file_lock, &flock);
2050 if (error)
2051 goto out;
2052 if (cmd == F_SETLKW64) {
2053 file_lock->fl_flags |= FL_SLEEP;
2054 }
2055
2056 error = -EBADF;
2057 switch (flock.l_type) {
2058 case F_RDLCK:
2059 if (!(filp->f_mode & FMODE_READ))
2060 goto out;
2061 break;
2062 case F_WRLCK:
2063 if (!(filp->f_mode & FMODE_WRITE))
2064 goto out;
2065 break;
2066 case F_UNLCK:
2067 break;
2068 default:
2069 error = -EINVAL;
2070 goto out;
2071 }
2072
b648a6de 2073 error = do_lock_file_wait(filp, cmd, file_lock);
1da177e4 2074
c293621b
PS
2075 /*
2076 * Attempt to detect a close/fcntl race and recover by
2077 * releasing the lock that was just acquired.
2078 */
0b2bac2f
AV
2079 spin_lock(&current->files->file_lock);
2080 f = fcheck(fd);
2081 spin_unlock(&current->files->file_lock);
2082 if (!error && f != filp && flock.l_type != F_UNLCK) {
c293621b
PS
2083 flock.l_type = F_UNLCK;
2084 goto again;
1da177e4
LT
2085 }
2086
2087out:
2088 locks_free_lock(file_lock);
2089 return error;
2090}
2091#endif /* BITS_PER_LONG == 32 */
2092
2093/*
2094 * This function is called when the file is being removed
2095 * from the task's fd array. POSIX locks belonging to this task
2096 * are deleted at this time.
2097 */
2098void locks_remove_posix(struct file *filp, fl_owner_t owner)
2099{
ff7b86b8 2100 struct file_lock lock;
1da177e4
LT
2101
2102 /*
2103 * If there are no locks held on this file, we don't need to call
2104 * posix_lock_file(). Another process could be setting a lock on this
2105 * file at the same time, but we wouldn't remove that lock anyway.
2106 */
496ad9aa 2107 if (!file_inode(filp)->i_flock)
1da177e4
LT
2108 return;
2109
2110 lock.fl_type = F_UNLCK;
75e1fcc0 2111 lock.fl_flags = FL_POSIX | FL_CLOSE;
1da177e4
LT
2112 lock.fl_start = 0;
2113 lock.fl_end = OFFSET_MAX;
2114 lock.fl_owner = owner;
2115 lock.fl_pid = current->tgid;
2116 lock.fl_file = filp;
2117 lock.fl_ops = NULL;
2118 lock.fl_lmops = NULL;
2119
150b3934 2120 vfs_lock_file(filp, F_SETLK, &lock, NULL);
1da177e4 2121
1da177e4
LT
2122 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2123 lock.fl_ops->fl_release_private(&lock);
2124}
2125
2126EXPORT_SYMBOL(locks_remove_posix);
2127
2128/*
2129 * This function is called on the last close of an open file.
2130 */
2131void locks_remove_flock(struct file *filp)
2132{
496ad9aa 2133 struct inode * inode = file_inode(filp);
1da177e4
LT
2134 struct file_lock *fl;
2135 struct file_lock **before;
2136
2137 if (!inode->i_flock)
2138 return;
2139
2140 if (filp->f_op && filp->f_op->flock) {
2141 struct file_lock fl = {
2142 .fl_pid = current->tgid,
2143 .fl_file = filp,
2144 .fl_flags = FL_FLOCK,
2145 .fl_type = F_UNLCK,
2146 .fl_end = OFFSET_MAX,
2147 };
2148 filp->f_op->flock(filp, F_SETLKW, &fl);
80fec4c6
TM
2149 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2150 fl.fl_ops->fl_release_private(&fl);
1da177e4
LT
2151 }
2152
1c8c601a 2153 spin_lock(&inode->i_lock);
1da177e4
LT
2154 before = &inode->i_flock;
2155
2156 while ((fl = *before) != NULL) {
2157 if (fl->fl_file == filp) {
c293621b 2158 if (IS_FLOCK(fl)) {
1da177e4
LT
2159 locks_delete_lock(before);
2160 continue;
2161 }
2162 if (IS_LEASE(fl)) {
2163 lease_modify(before, F_UNLCK);
2164 continue;
2165 }
2166 /* What? */
2167 BUG();
2168 }
2169 before = &fl->fl_next;
2170 }
1c8c601a 2171 spin_unlock(&inode->i_lock);
1da177e4
LT
2172}
2173
1da177e4
LT
2174/**
2175 * posix_unblock_lock - stop waiting for a file lock
1da177e4
LT
2176 * @waiter: the lock which was waiting
2177 *
2178 * lockd needs to block waiting for locks.
2179 */
64a318ee 2180int
f891a29f 2181posix_unblock_lock(struct file_lock *waiter)
1da177e4 2182{
64a318ee
BF
2183 int status = 0;
2184
1c8c601a 2185 spin_lock(&file_lock_lock);
5996a298 2186 if (waiter->fl_next)
1da177e4 2187 __locks_delete_block(waiter);
64a318ee
BF
2188 else
2189 status = -ENOENT;
1c8c601a 2190 spin_unlock(&file_lock_lock);
64a318ee 2191 return status;
1da177e4 2192}
1da177e4
LT
2193EXPORT_SYMBOL(posix_unblock_lock);
2194
9b9d2ab4
ME
2195/**
2196 * vfs_cancel_lock - file byte range unblock lock
2197 * @filp: The file to apply the unblock to
2198 * @fl: The lock to be unblocked
2199 *
2200 * Used by lock managers to cancel blocked requests
2201 */
2202int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2203{
2204 if (filp->f_op && filp->f_op->lock)
2205 return filp->f_op->lock(filp, F_CANCELLK, fl);
2206 return 0;
2207}
2208
2209EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2210
7f8ada98 2211#ifdef CONFIG_PROC_FS
d8ba7a36 2212#include <linux/proc_fs.h>
7f8ada98
PE
2213#include <linux/seq_file.h>
2214
2215static void lock_get_status(struct seq_file *f, struct file_lock *fl,
99dc8292 2216 loff_t id, char *pfx)
1da177e4
LT
2217{
2218 struct inode *inode = NULL;
ab1f1611
VG
2219 unsigned int fl_pid;
2220
2221 if (fl->fl_nspid)
6c5f3e7b 2222 fl_pid = pid_vnr(fl->fl_nspid);
ab1f1611
VG
2223 else
2224 fl_pid = fl->fl_pid;
1da177e4
LT
2225
2226 if (fl->fl_file != NULL)
496ad9aa 2227 inode = file_inode(fl->fl_file);
1da177e4 2228
99dc8292 2229 seq_printf(f, "%lld:%s ", id, pfx);
1da177e4 2230 if (IS_POSIX(fl)) {
7f8ada98 2231 seq_printf(f, "%6s %s ",
1da177e4
LT
2232 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2233 (inode == NULL) ? "*NOINODE*" :
a16877ca 2234 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
1da177e4
LT
2235 } else if (IS_FLOCK(fl)) {
2236 if (fl->fl_type & LOCK_MAND) {
7f8ada98 2237 seq_printf(f, "FLOCK MSNFS ");
1da177e4 2238 } else {
7f8ada98 2239 seq_printf(f, "FLOCK ADVISORY ");
1da177e4
LT
2240 }
2241 } else if (IS_LEASE(fl)) {
7f8ada98 2242 seq_printf(f, "LEASE ");
ab83fa4b 2243 if (lease_breaking(fl))
7f8ada98 2244 seq_printf(f, "BREAKING ");
1da177e4 2245 else if (fl->fl_file)
7f8ada98 2246 seq_printf(f, "ACTIVE ");
1da177e4 2247 else
7f8ada98 2248 seq_printf(f, "BREAKER ");
1da177e4 2249 } else {
7f8ada98 2250 seq_printf(f, "UNKNOWN UNKNOWN ");
1da177e4
LT
2251 }
2252 if (fl->fl_type & LOCK_MAND) {
7f8ada98 2253 seq_printf(f, "%s ",
1da177e4
LT
2254 (fl->fl_type & LOCK_READ)
2255 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2256 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2257 } else {
7f8ada98 2258 seq_printf(f, "%s ",
ab83fa4b 2259 (lease_breaking(fl))
0ee5c6d6
JL
2260 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2261 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
1da177e4
LT
2262 }
2263 if (inode) {
2264#ifdef WE_CAN_BREAK_LSLK_NOW
ab1f1611 2265 seq_printf(f, "%d %s:%ld ", fl_pid,
1da177e4
LT
2266 inode->i_sb->s_id, inode->i_ino);
2267#else
2268 /* userspace relies on this representation of dev_t ;-( */
ab1f1611 2269 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
1da177e4
LT
2270 MAJOR(inode->i_sb->s_dev),
2271 MINOR(inode->i_sb->s_dev), inode->i_ino);
2272#endif
2273 } else {
ab1f1611 2274 seq_printf(f, "%d <none>:0 ", fl_pid);
1da177e4
LT
2275 }
2276 if (IS_POSIX(fl)) {
2277 if (fl->fl_end == OFFSET_MAX)
7f8ada98 2278 seq_printf(f, "%Ld EOF\n", fl->fl_start);
1da177e4 2279 else
7f8ada98 2280 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
1da177e4 2281 } else {
7f8ada98 2282 seq_printf(f, "0 EOF\n");
1da177e4
LT
2283 }
2284}
2285
7f8ada98 2286static int locks_show(struct seq_file *f, void *v)
1da177e4 2287{
7f8ada98 2288 struct file_lock *fl, *bfl;
1da177e4 2289
7f8ada98 2290 fl = list_entry(v, struct file_lock, fl_link);
1da177e4 2291
99dc8292 2292 lock_get_status(f, fl, *((loff_t *)f->private), "");
1da177e4 2293
7f8ada98 2294 list_for_each_entry(bfl, &fl->fl_block, fl_block)
99dc8292 2295 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
094f2825 2296
7f8ada98
PE
2297 return 0;
2298}
1da177e4 2299
7f8ada98
PE
2300static void *locks_start(struct seq_file *f, loff_t *pos)
2301{
99dc8292
JM
2302 loff_t *p = f->private;
2303
1c8c601a 2304 spin_lock(&file_lock_lock);
99dc8292 2305 *p = (*pos + 1);
7f8ada98
PE
2306 return seq_list_start(&file_lock_list, *pos);
2307}
1da177e4 2308
7f8ada98
PE
2309static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2310{
99dc8292
JM
2311 loff_t *p = f->private;
2312 ++*p;
7f8ada98
PE
2313 return seq_list_next(v, &file_lock_list, pos);
2314}
1da177e4 2315
7f8ada98
PE
2316static void locks_stop(struct seq_file *f, void *v)
2317{
1c8c601a 2318 spin_unlock(&file_lock_lock);
1da177e4
LT
2319}
2320
d8ba7a36 2321static const struct seq_operations locks_seq_operations = {
7f8ada98
PE
2322 .start = locks_start,
2323 .next = locks_next,
2324 .stop = locks_stop,
2325 .show = locks_show,
2326};
d8ba7a36
AD
2327
2328static int locks_open(struct inode *inode, struct file *filp)
2329{
99dc8292 2330 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
d8ba7a36
AD
2331}
2332
2333static const struct file_operations proc_locks_operations = {
2334 .open = locks_open,
2335 .read = seq_read,
2336 .llseek = seq_lseek,
99dc8292 2337 .release = seq_release_private,
d8ba7a36
AD
2338};
2339
2340static int __init proc_locks_init(void)
2341{
2342 proc_create("locks", 0, NULL, &proc_locks_operations);
2343 return 0;
2344}
2345module_init(proc_locks_init);
7f8ada98
PE
2346#endif
2347
1da177e4
LT
2348/**
2349 * lock_may_read - checks that the region is free of locks
2350 * @inode: the inode that is being read
2351 * @start: the first byte to read
2352 * @len: the number of bytes to read
2353 *
2354 * Emulates Windows locking requirements. Whole-file
2355 * mandatory locks (share modes) can prohibit a read and
2356 * byte-range POSIX locks can prohibit a read if they overlap.
2357 *
2358 * N.B. this function is only ever called
2359 * from knfsd and ownership of locks is never checked.
2360 */
2361int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2362{
2363 struct file_lock *fl;
2364 int result = 1;
1c8c601a
JL
2365
2366 spin_lock(&inode->i_lock);
1da177e4
LT
2367 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2368 if (IS_POSIX(fl)) {
2369 if (fl->fl_type == F_RDLCK)
2370 continue;
2371 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2372 continue;
2373 } else if (IS_FLOCK(fl)) {
2374 if (!(fl->fl_type & LOCK_MAND))
2375 continue;
2376 if (fl->fl_type & LOCK_READ)
2377 continue;
2378 } else
2379 continue;
2380 result = 0;
2381 break;
2382 }
1c8c601a 2383 spin_unlock(&inode->i_lock);
1da177e4
LT
2384 return result;
2385}
2386
2387EXPORT_SYMBOL(lock_may_read);
2388
2389/**
2390 * lock_may_write - checks that the region is free of locks
2391 * @inode: the inode that is being written
2392 * @start: the first byte to write
2393 * @len: the number of bytes to write
2394 *
2395 * Emulates Windows locking requirements. Whole-file
2396 * mandatory locks (share modes) can prohibit a write and
2397 * byte-range POSIX locks can prohibit a write if they overlap.
2398 *
2399 * N.B. this function is only ever called
2400 * from knfsd and ownership of locks is never checked.
2401 */
2402int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2403{
2404 struct file_lock *fl;
2405 int result = 1;
1c8c601a
JL
2406
2407 spin_lock(&inode->i_lock);
1da177e4
LT
2408 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2409 if (IS_POSIX(fl)) {
2410 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2411 continue;
2412 } else if (IS_FLOCK(fl)) {
2413 if (!(fl->fl_type & LOCK_MAND))
2414 continue;
2415 if (fl->fl_type & LOCK_WRITE)
2416 continue;
2417 } else
2418 continue;
2419 result = 0;
2420 break;
2421 }
1c8c601a 2422 spin_unlock(&inode->i_lock);
1da177e4
LT
2423 return result;
2424}
2425
2426EXPORT_SYMBOL(lock_may_write);
2427
1da177e4
LT
2428static int __init filelock_init(void)
2429{
2430 filelock_cache = kmem_cache_create("file_lock_cache",
ee19cc40
MS
2431 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2432
1da177e4
LT
2433 return 0;
2434}
2435
2436core_initcall(filelock_init);