blob: 4a8072736efa2716b46121bb8d6b994c7f573054 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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/mandatory.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/fs.h>
120#include <linux/init.h>
121#include <linux/module.h>
122#include <linux/security.h>
123#include <linux/slab.h>
124#include <linux/smp_lock.h>
125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129#include <asm/semaphore.h>
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
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
144
Christoph Lametere18b8902006-12-06 20:33:20 -0800145static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147/* Allocate an empty lock structure. */
148static struct file_lock *locks_alloc_lock(void)
149{
Christoph Lametere94b1762006-12-06 20:33:17 -0800150 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Trond Myklebust47831f32006-03-20 13:44:05 -0500153static void locks_release_private(struct file_lock *fl)
154{
155 if (fl->fl_ops) {
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
158 fl->fl_ops = NULL;
159 }
160 if (fl->fl_lmops) {
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
163 fl->fl_lmops = NULL;
164 }
165
166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800169static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Trond Myklebust47831f32006-03-20 13:44:05 -0500175 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(filelock_cache, fl);
177}
178
179void locks_init_lock(struct file_lock *fl)
180{
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
184 fl->fl_next = NULL;
185 fl->fl_fasync = NULL;
186 fl->fl_owner = NULL;
187 fl->fl_pid = 0;
188 fl->fl_file = NULL;
189 fl->fl_flags = 0;
190 fl->fl_type = 0;
191 fl->fl_start = fl->fl_end = 0;
192 fl->fl_ops = NULL;
193 fl->fl_lmops = NULL;
194}
195
196EXPORT_SYMBOL(locks_init_lock);
197
198/*
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
201 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800202static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 struct file_lock *lock = (struct file_lock *) foo;
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 locks_init_lock(lock);
207}
208
Trond Myklebust47831f32006-03-20 13:44:05 -0500209static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
210{
211 if (fl->fl_ops) {
212 if (fl->fl_ops->fl_copy_lock)
213 fl->fl_ops->fl_copy_lock(new, fl);
214 new->fl_ops = fl->fl_ops;
215 }
216 if (fl->fl_lmops) {
217 if (fl->fl_lmops->fl_copy_lock)
218 fl->fl_lmops->fl_copy_lock(new, fl);
219 new->fl_lmops = fl->fl_lmops;
220 }
221}
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223/*
224 * Initialize a new lock from an existing file_lock structure.
225 */
Trond Myklebust09969052006-03-20 13:44:38 -0500226static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
228 new->fl_owner = fl->fl_owner;
229 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500230 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 new->fl_flags = fl->fl_flags;
232 new->fl_type = fl->fl_type;
233 new->fl_start = fl->fl_start;
234 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500235 new->fl_ops = NULL;
236 new->fl_lmops = NULL;
237}
238
239void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
240{
241 locks_release_private(new);
242
243 __locks_copy_lock(new, fl);
244 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 new->fl_ops = fl->fl_ops;
246 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500247
248 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
251EXPORT_SYMBOL(locks_copy_lock);
252
253static inline int flock_translate_cmd(int cmd) {
254 if (cmd & LOCK_MAND)
255 return cmd & (LOCK_MAND | LOCK_RW);
256 switch (cmd) {
257 case LOCK_SH:
258 return F_RDLCK;
259 case LOCK_EX:
260 return F_WRLCK;
261 case LOCK_UN:
262 return F_UNLCK;
263 }
264 return -EINVAL;
265}
266
267/* Fill in a file_lock structure with an appropriate FLOCK lock. */
268static int flock_make_lock(struct file *filp, struct file_lock **lock,
269 unsigned int cmd)
270{
271 struct file_lock *fl;
272 int type = flock_translate_cmd(cmd);
273 if (type < 0)
274 return type;
275
276 fl = locks_alloc_lock();
277 if (fl == NULL)
278 return -ENOMEM;
279
280 fl->fl_file = filp;
281 fl->fl_pid = current->tgid;
282 fl->fl_flags = FL_FLOCK;
283 fl->fl_type = type;
284 fl->fl_end = OFFSET_MAX;
285
286 *lock = fl;
287 return 0;
288}
289
290static int assign_type(struct file_lock *fl, int type)
291{
292 switch (type) {
293 case F_RDLCK:
294 case F_WRLCK:
295 case F_UNLCK:
296 fl->fl_type = type;
297 break;
298 default:
299 return -EINVAL;
300 }
301 return 0;
302}
303
304/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
305 * style lock.
306 */
307static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
308 struct flock *l)
309{
310 off_t start, end;
311
312 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700313 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 start = 0;
315 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700316 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 start = filp->f_pos;
318 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700319 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800320 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 break;
322 default:
323 return -EINVAL;
324 }
325
326 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
327 POSIX-2001 defines it. */
328 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 if (start < 0)
330 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700331 fl->fl_end = OFFSET_MAX;
332 if (l->l_len > 0) {
333 end = start + l->l_len - 1;
334 fl->fl_end = end;
335 } else if (l->l_len < 0) {
336 end = start - 1;
337 fl->fl_end = end;
338 start += l->l_len;
339 if (start < 0)
340 return -EINVAL;
341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700343 if (fl->fl_end < fl->fl_start)
344 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 fl->fl_owner = current->files;
347 fl->fl_pid = current->tgid;
348 fl->fl_file = filp;
349 fl->fl_flags = FL_POSIX;
350 fl->fl_ops = NULL;
351 fl->fl_lmops = NULL;
352
353 return assign_type(fl, l->l_type);
354}
355
356#if BITS_PER_LONG == 32
357static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
358 struct flock64 *l)
359{
360 loff_t start;
361
362 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700363 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 start = 0;
365 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700366 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 start = filp->f_pos;
368 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700369 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800370 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 break;
372 default:
373 return -EINVAL;
374 }
375
Trond Myklebust4c780a42005-10-18 14:20:21 -0700376 start += l->l_start;
377 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700379 fl->fl_end = OFFSET_MAX;
380 if (l->l_len > 0) {
381 fl->fl_end = start + l->l_len - 1;
382 } else if (l->l_len < 0) {
383 fl->fl_end = start - 1;
384 start += l->l_len;
385 if (start < 0)
386 return -EINVAL;
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700389 if (fl->fl_end < fl->fl_start)
390 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 fl->fl_owner = current->files;
393 fl->fl_pid = current->tgid;
394 fl->fl_file = filp;
395 fl->fl_flags = FL_POSIX;
396 fl->fl_ops = NULL;
397 fl->fl_lmops = NULL;
398
399 switch (l->l_type) {
400 case F_RDLCK:
401 case F_WRLCK:
402 case F_UNLCK:
403 fl->fl_type = l->l_type;
404 break;
405 default:
406 return -EINVAL;
407 }
408
409 return (0);
410}
411#endif
412
413/* default lease lock manager operations */
414static void lease_break_callback(struct file_lock *fl)
415{
416 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
417}
418
419static void lease_release_private_callback(struct file_lock *fl)
420{
421 if (!fl->fl_file)
422 return;
423
424 f_delown(fl->fl_file);
425 fl->fl_file->f_owner.signum = 0;
426}
427
Adrian Bunk75c96f82005-05-05 16:16:09 -0700428static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
430 return fl->fl_file == try->fl_file;
431}
432
Adrian Bunk75c96f82005-05-05 16:16:09 -0700433static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 .fl_break = lease_break_callback,
435 .fl_release_private = lease_release_private_callback,
436 .fl_mylease = lease_mylease_callback,
437 .fl_change = lease_modify,
438};
439
440/*
441 * Initialize a lease, use the default lock manager operations
442 */
443static int lease_init(struct file *filp, int type, struct file_lock *fl)
444 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400445 if (assign_type(fl, type) != 0)
446 return -EINVAL;
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 fl->fl_owner = current->files;
449 fl->fl_pid = current->tgid;
450
451 fl->fl_file = filp;
452 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 fl->fl_start = 0;
454 fl->fl_end = OFFSET_MAX;
455 fl->fl_ops = NULL;
456 fl->fl_lmops = &lease_manager_ops;
457 return 0;
458}
459
460/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500461static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
463 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400464 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465
466 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500467 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400470 if (error) {
471 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500472 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400473 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500474 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}
476
477/* Check if two locks overlap each other.
478 */
479static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
480{
481 return ((fl1->fl_end >= fl2->fl_start) &&
482 (fl2->fl_end >= fl1->fl_start));
483}
484
485/*
486 * Check whether two locks have the same owner.
487 */
Matt Mackall33443c42006-01-08 01:05:22 -0800488static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
490 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
491 return fl2->fl_lmops == fl1->fl_lmops &&
492 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
493 return fl1->fl_owner == fl2->fl_owner;
494}
495
496/* Remove waiter from blocker's block list.
497 * When blocker ends up pointing to itself then the list is empty.
498 */
Matt Mackall33443c42006-01-08 01:05:22 -0800499static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
501 list_del_init(&waiter->fl_block);
502 list_del_init(&waiter->fl_link);
503 waiter->fl_next = NULL;
504}
505
506/*
507 */
508static void locks_delete_block(struct file_lock *waiter)
509{
510 lock_kernel();
511 __locks_delete_block(waiter);
512 unlock_kernel();
513}
514
515/* Insert waiter into blocker's block list.
516 * We use a circular list so that processes can be easily woken up in
517 * the order they blocked. The documentation doesn't require this but
518 * it seems like the reasonable thing to do.
519 */
520static void locks_insert_block(struct file_lock *blocker,
521 struct file_lock *waiter)
522{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800523 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 list_add_tail(&waiter->fl_block, &blocker->fl_block);
525 waiter->fl_next = blocker;
526 if (IS_POSIX(blocker))
527 list_add(&waiter->fl_link, &blocked_list);
528}
529
530/* Wake up processes blocked waiting for blocker.
531 * If told to wait then schedule the processes until the block list
532 * is empty, otherwise empty the block list ourselves.
533 */
534static void locks_wake_up_blocks(struct file_lock *blocker)
535{
536 while (!list_empty(&blocker->fl_block)) {
537 struct file_lock *waiter = list_entry(blocker->fl_block.next,
538 struct file_lock, fl_block);
539 __locks_delete_block(waiter);
540 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
541 waiter->fl_lmops->fl_notify(waiter);
542 else
543 wake_up(&waiter->fl_wait);
544 }
545}
546
547/* Insert file lock fl into an inode's lock list at the position indicated
548 * by pos. At the same time add the lock to the global file lock list.
549 */
550static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
551{
552 list_add(&fl->fl_link, &file_lock_list);
553
554 /* insert into file's list */
555 fl->fl_next = *pos;
556 *pos = fl;
557
558 if (fl->fl_ops && fl->fl_ops->fl_insert)
559 fl->fl_ops->fl_insert(fl);
560}
561
562/*
563 * Delete a lock and then free it.
564 * Wake up processes that are blocked waiting for this lock,
565 * notify the FS that the lock has been cleared and
566 * finally free the lock.
567 */
568static void locks_delete_lock(struct file_lock **thisfl_p)
569{
570 struct file_lock *fl = *thisfl_p;
571
572 *thisfl_p = fl->fl_next;
573 fl->fl_next = NULL;
574 list_del_init(&fl->fl_link);
575
576 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
577 if (fl->fl_fasync != NULL) {
578 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
579 fl->fl_fasync = NULL;
580 }
581
582 if (fl->fl_ops && fl->fl_ops->fl_remove)
583 fl->fl_ops->fl_remove(fl);
584
585 locks_wake_up_blocks(fl);
586 locks_free_lock(fl);
587}
588
589/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
590 * checks for shared/exclusive status of overlapping locks.
591 */
592static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
593{
594 if (sys_fl->fl_type == F_WRLCK)
595 return 1;
596 if (caller_fl->fl_type == F_WRLCK)
597 return 1;
598 return 0;
599}
600
601/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
602 * checking before calling the locks_conflict().
603 */
604static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
605{
606 /* POSIX locks owned by the same process do not conflict with
607 * each other.
608 */
609 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
610 return (0);
611
612 /* Check whether they overlap */
613 if (!locks_overlap(caller_fl, sys_fl))
614 return 0;
615
616 return (locks_conflict(caller_fl, sys_fl));
617}
618
619/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
620 * checking before calling the locks_conflict().
621 */
622static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
623{
624 /* FLOCK locks referring to the same filp do not conflict with
625 * each other.
626 */
627 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
628 return (0);
629 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
630 return 0;
631
632 return (locks_conflict(caller_fl, sys_fl));
633}
634
635static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
636{
637 int result = 0;
638 DECLARE_WAITQUEUE(wait, current);
639
640 __set_current_state(TASK_INTERRUPTIBLE);
641 add_wait_queue(fl_wait, &wait);
642 if (timeout == 0)
643 schedule();
644 else
645 result = schedule_timeout(timeout);
646 if (signal_pending(current))
647 result = -ERESTARTSYS;
648 remove_wait_queue(fl_wait, &wait);
649 __set_current_state(TASK_RUNNING);
650 return result;
651}
652
653static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
654{
655 int result;
656 locks_insert_block(blocker, waiter);
657 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
658 __locks_delete_block(waiter);
659 return result;
660}
661
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400662void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500663posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664{
665 struct file_lock *cfl;
666
667 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800668 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 if (!IS_POSIX(cfl))
670 continue;
671 if (posix_locks_conflict(cfl, fl))
672 break;
673 }
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400674 if (cfl)
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500675 __locks_copy_lock(fl, cfl);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400676 else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400677 fl->fl_type = F_UNLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 unlock_kernel();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400679 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680}
681
682EXPORT_SYMBOL(posix_test_lock);
683
684/* This function tests for deadlock condition before putting a process to
685 * sleep. The detection scheme is no longer recursive. Recursive was neat,
686 * but dangerous - we risked stack corruption if the lock data was bad, or
687 * if the recursion was too deep for any other reason.
688 *
689 * We rely on the fact that a task can only be on one lock's wait queue
690 * at a time. When we find blocked_task on a wait queue we can re-search
691 * with blocked_task equal to that queue's owner, until either blocked_task
692 * isn't found, or blocked_task is found on a queue owned by my_task.
693 *
694 * Note: the above assumption may not be true when handling lock requests
695 * from a broken NFS client. But broken NFS clients have a lot more to
696 * worry about than proper deadlock detection anyway... --okir
697 */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700698static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 struct file_lock *block_fl)
700{
701 struct list_head *tmp;
702
703next_task:
704 if (posix_same_owner(caller_fl, block_fl))
705 return 1;
706 list_for_each(tmp, &blocked_list) {
707 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
708 if (posix_same_owner(fl, block_fl)) {
709 fl = fl->fl_next;
710 block_fl = fl;
711 goto next_task;
712 }
713 }
714 return 0;
715}
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
718 * at the head of the list, but that's secret knowledge known only to
719 * flock_lock_file and posix_lock_file.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400720 *
721 * Note that if called with an FL_EXISTS argument, the caller may determine
722 * whether or not a lock was successfully freed by testing the return
723 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800725static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800727 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800729 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 int error = 0;
731 int found = 0;
732
733 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400734 if (request->fl_flags & FL_ACCESS)
735 goto find_conflict;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 for_each_lock(inode, before) {
737 struct file_lock *fl = *before;
738 if (IS_POSIX(fl))
739 break;
740 if (IS_LEASE(fl))
741 continue;
742 if (filp != fl->fl_file)
743 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800744 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 goto out;
746 found = 1;
747 locks_delete_lock(before);
748 break;
749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Trond Myklebustf475ae92006-06-29 16:38:32 -0400751 if (request->fl_type == F_UNLCK) {
752 if ((request->fl_flags & FL_EXISTS) && !found)
753 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800754 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400755 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400757 error = -ENOMEM;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800758 new_fl = locks_alloc_lock();
759 if (new_fl == NULL)
760 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 /*
762 * If a higher-priority process was blocked on the old file lock,
763 * give it the opportunity to lock the file.
764 */
765 if (found)
766 cond_resched();
767
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400768find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 for_each_lock(inode, before) {
770 struct file_lock *fl = *before;
771 if (IS_POSIX(fl))
772 break;
773 if (IS_LEASE(fl))
774 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800775 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 continue;
777 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800778 if (request->fl_flags & FL_SLEEP)
779 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 goto out;
781 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400782 if (request->fl_flags & FL_ACCESS)
783 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800784 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800786 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400787 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
789out:
790 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800791 if (new_fl)
792 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return error;
794}
795
Marc Eshel150b3932007-01-18 16:15:35 -0500796static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797{
798 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700799 struct file_lock *new_fl = NULL;
800 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 struct file_lock *left = NULL;
802 struct file_lock *right = NULL;
803 struct file_lock **before;
804 int error, added = 0;
805
806 /*
807 * We may need two file_lock structures for this operation,
808 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700809 *
810 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700812 if (!(request->fl_flags & FL_ACCESS) &&
813 (request->fl_type != F_UNLCK ||
814 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
815 new_fl = locks_alloc_lock();
816 new_fl2 = locks_alloc_lock();
817 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 lock_kernel();
820 if (request->fl_type != F_UNLCK) {
821 for_each_lock(inode, before) {
822 struct file_lock *fl = *before;
823 if (!IS_POSIX(fl))
824 continue;
825 if (!posix_locks_conflict(request, fl))
826 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800827 if (conflock)
828 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 error = -EAGAIN;
830 if (!(request->fl_flags & FL_SLEEP))
831 goto out;
832 error = -EDEADLK;
833 if (posix_locks_deadlock(request, fl))
834 goto out;
835 error = -EAGAIN;
836 locks_insert_block(fl, request);
837 goto out;
838 }
839 }
840
841 /* If we're just looking for a conflict, we're done. */
842 error = 0;
843 if (request->fl_flags & FL_ACCESS)
844 goto out;
845
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 * Find the first old lock with the same owner as the new lock.
848 */
849
850 before = &inode->i_flock;
851
852 /* First skip locks owned by other processes. */
853 while ((fl = *before) && (!IS_POSIX(fl) ||
854 !posix_same_owner(request, fl))) {
855 before = &fl->fl_next;
856 }
857
858 /* Process locks with this owner. */
859 while ((fl = *before) && posix_same_owner(request, fl)) {
860 /* Detect adjacent or overlapping regions (if same lock type)
861 */
862 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700863 /* In all comparisons of start vs end, use
864 * "start - 1" rather than "end + 1". If end
865 * is OFFSET_MAX, end + 1 will become negative.
866 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 if (fl->fl_end < request->fl_start - 1)
868 goto next_lock;
869 /* If the next lock in the list has entirely bigger
870 * addresses than the new one, insert the lock here.
871 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700872 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 break;
874
875 /* If we come here, the new and old lock are of the
876 * same type and adjacent or overlapping. Make one
877 * lock yielding from the lower start address of both
878 * locks to the higher end address.
879 */
880 if (fl->fl_start > request->fl_start)
881 fl->fl_start = request->fl_start;
882 else
883 request->fl_start = fl->fl_start;
884 if (fl->fl_end < request->fl_end)
885 fl->fl_end = request->fl_end;
886 else
887 request->fl_end = fl->fl_end;
888 if (added) {
889 locks_delete_lock(before);
890 continue;
891 }
892 request = fl;
893 added = 1;
894 }
895 else {
896 /* Processing for different lock types is a bit
897 * more complex.
898 */
899 if (fl->fl_end < request->fl_start)
900 goto next_lock;
901 if (fl->fl_start > request->fl_end)
902 break;
903 if (request->fl_type == F_UNLCK)
904 added = 1;
905 if (fl->fl_start < request->fl_start)
906 left = fl;
907 /* If the next lock in the list has a higher end
908 * address than the new one, insert the new one here.
909 */
910 if (fl->fl_end > request->fl_end) {
911 right = fl;
912 break;
913 }
914 if (fl->fl_start >= request->fl_start) {
915 /* The new lock completely replaces an old
916 * one (This may happen several times).
917 */
918 if (added) {
919 locks_delete_lock(before);
920 continue;
921 }
922 /* Replace the old lock with the new one.
923 * Wake up anybody waiting for the old one,
924 * as the change in lock type might satisfy
925 * their needs.
926 */
927 locks_wake_up_blocks(fl);
928 fl->fl_start = request->fl_start;
929 fl->fl_end = request->fl_end;
930 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500931 locks_release_private(fl);
932 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 request = fl;
934 added = 1;
935 }
936 }
937 /* Go on to next lock.
938 */
939 next_lock:
940 before = &fl->fl_next;
941 }
942
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700943 /*
944 * The above code only modifies existing locks in case of
945 * merging or replacing. If new lock(s) need to be inserted
946 * all modifications are done bellow this, so it's safe yet to
947 * bail out.
948 */
949 error = -ENOLCK; /* "no luck" */
950 if (right && left == right && !new_fl2)
951 goto out;
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 error = 0;
954 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400955 if (request->fl_type == F_UNLCK) {
956 if (request->fl_flags & FL_EXISTS)
957 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400959 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700960
961 if (!new_fl) {
962 error = -ENOLCK;
963 goto out;
964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 locks_copy_lock(new_fl, request);
966 locks_insert_lock(before, new_fl);
967 new_fl = NULL;
968 }
969 if (right) {
970 if (left == right) {
971 /* The new lock breaks the old one in two pieces,
972 * so we have to use the second new lock.
973 */
974 left = new_fl2;
975 new_fl2 = NULL;
976 locks_copy_lock(left, right);
977 locks_insert_lock(before, left);
978 }
979 right->fl_start = request->fl_end + 1;
980 locks_wake_up_blocks(right);
981 }
982 if (left) {
983 left->fl_end = request->fl_start - 1;
984 locks_wake_up_blocks(left);
985 }
986 out:
987 unlock_kernel();
988 /*
989 * Free any unused locks.
990 */
991 if (new_fl)
992 locks_free_lock(new_fl);
993 if (new_fl2)
994 locks_free_lock(new_fl2);
995 return error;
996}
997
998/**
999 * posix_lock_file - Apply a POSIX-style lock to a file
1000 * @filp: The file to apply the lock to
1001 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001002 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 *
1004 * Add a POSIX style lock to a file.
1005 * We merge adjacent & overlapping locks whenever possible.
1006 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001007 *
1008 * Note that if called with an FL_EXISTS argument, the caller may determine
1009 * whether or not a lock was successfully freed by testing the return
1010 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 */
Marc Eshel150b3932007-01-18 16:15:35 -05001012int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001013 struct file_lock *conflock)
1014{
Marc Eshel150b3932007-01-18 16:15:35 -05001015 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001016}
Marc Eshel150b3932007-01-18 16:15:35 -05001017EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018
1019/**
1020 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1021 * @filp: The file to apply the lock to
1022 * @fl: The lock to be applied
1023 *
1024 * Add a POSIX style lock to a file.
1025 * We merge adjacent & overlapping locks whenever possible.
1026 * POSIX locks are sorted by owner task, then by starting address
1027 */
1028int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1029{
1030 int error;
1031 might_sleep ();
1032 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001033 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1035 break;
1036 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1037 if (!error)
1038 continue;
1039
1040 locks_delete_block(fl);
1041 break;
1042 }
1043 return error;
1044}
1045EXPORT_SYMBOL(posix_lock_file_wait);
1046
1047/**
1048 * locks_mandatory_locked - Check for an active lock
1049 * @inode: the file to check
1050 *
1051 * Searches the inode's list of locks to find any POSIX locks which conflict.
1052 * This function is called from locks_verify_locked() only.
1053 */
1054int locks_mandatory_locked(struct inode *inode)
1055{
1056 fl_owner_t owner = current->files;
1057 struct file_lock *fl;
1058
1059 /*
1060 * Search the lock list for this inode for any POSIX locks.
1061 */
1062 lock_kernel();
1063 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1064 if (!IS_POSIX(fl))
1065 continue;
1066 if (fl->fl_owner != owner)
1067 break;
1068 }
1069 unlock_kernel();
1070 return fl ? -EAGAIN : 0;
1071}
1072
1073/**
1074 * locks_mandatory_area - Check for a conflicting lock
1075 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1076 * for shared
1077 * @inode: the file to check
1078 * @filp: how the file was opened (if it was)
1079 * @offset: start of area to check
1080 * @count: length of area to check
1081 *
1082 * Searches the inode's list of locks to find any POSIX locks which conflict.
1083 * This function is called from rw_verify_area() and
1084 * locks_verify_truncate().
1085 */
1086int locks_mandatory_area(int read_write, struct inode *inode,
1087 struct file *filp, loff_t offset,
1088 size_t count)
1089{
1090 struct file_lock fl;
1091 int error;
1092
1093 locks_init_lock(&fl);
1094 fl.fl_owner = current->files;
1095 fl.fl_pid = current->tgid;
1096 fl.fl_file = filp;
1097 fl.fl_flags = FL_POSIX | FL_ACCESS;
1098 if (filp && !(filp->f_flags & O_NONBLOCK))
1099 fl.fl_flags |= FL_SLEEP;
1100 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1101 fl.fl_start = offset;
1102 fl.fl_end = offset + count - 1;
1103
1104 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001105 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 if (error != -EAGAIN)
1107 break;
1108 if (!(fl.fl_flags & FL_SLEEP))
1109 break;
1110 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1111 if (!error) {
1112 /*
1113 * If we've been sleeping someone might have
1114 * changed the permissions behind our back.
1115 */
1116 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1117 continue;
1118 }
1119
1120 locks_delete_block(&fl);
1121 break;
1122 }
1123
1124 return error;
1125}
1126
1127EXPORT_SYMBOL(locks_mandatory_area);
1128
1129/* We already had a lease on this file; just change its type */
1130int lease_modify(struct file_lock **before, int arg)
1131{
1132 struct file_lock *fl = *before;
1133 int error = assign_type(fl, arg);
1134
1135 if (error)
1136 return error;
1137 locks_wake_up_blocks(fl);
1138 if (arg == F_UNLCK)
1139 locks_delete_lock(before);
1140 return 0;
1141}
1142
1143EXPORT_SYMBOL(lease_modify);
1144
1145static void time_out_leases(struct inode *inode)
1146{
1147 struct file_lock **before;
1148 struct file_lock *fl;
1149
1150 before = &inode->i_flock;
1151 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1152 if ((fl->fl_break_time == 0)
1153 || time_before(jiffies, fl->fl_break_time)) {
1154 before = &fl->fl_next;
1155 continue;
1156 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1158 if (fl == *before) /* lease_modify may have freed fl */
1159 before = &fl->fl_next;
1160 }
1161}
1162
1163/**
1164 * __break_lease - revoke all outstanding leases on file
1165 * @inode: the inode of the file to return
1166 * @mode: the open mode (read or write)
1167 *
david m. richter87250dd2007-05-09 16:10:27 -04001168 * break_lease (inlined for speed) has checked there already is at least
1169 * some kind of lock (maybe a lease) on this file. Leases are broken on
1170 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 * specified %O_NONBLOCK to your open().
1172 */
1173int __break_lease(struct inode *inode, unsigned int mode)
1174{
1175 int error = 0, future;
1176 struct file_lock *new_fl, *flock;
1177 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 unsigned long break_time;
1179 int i_have_this_lease = 0;
1180
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001181 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183 lock_kernel();
1184
1185 time_out_leases(inode);
1186
1187 flock = inode->i_flock;
1188 if ((flock == NULL) || !IS_LEASE(flock))
1189 goto out;
1190
1191 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1192 if (fl->fl_owner == current->files)
1193 i_have_this_lease = 1;
1194
1195 if (mode & FMODE_WRITE) {
1196 /* If we want write access, we have to revoke any lease. */
1197 future = F_UNLCK | F_INPROGRESS;
1198 } else if (flock->fl_type & F_INPROGRESS) {
1199 /* If the lease is already being broken, we just leave it */
1200 future = flock->fl_type;
1201 } else if (flock->fl_type & F_WRLCK) {
1202 /* Downgrade the exclusive lease to a read-only lease. */
1203 future = F_RDLCK | F_INPROGRESS;
1204 } else {
1205 /* the existing lease was read-only, so we can read too. */
1206 goto out;
1207 }
1208
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001209 if (IS_ERR(new_fl) && !i_have_this_lease
1210 && ((mode & O_NONBLOCK) == 0)) {
1211 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 goto out;
1213 }
1214
1215 break_time = 0;
1216 if (lease_break_time > 0) {
1217 break_time = jiffies + lease_break_time * HZ;
1218 if (break_time == 0)
1219 break_time++; /* so that 0 means no break time */
1220 }
1221
1222 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1223 if (fl->fl_type != future) {
1224 fl->fl_type = future;
1225 fl->fl_break_time = break_time;
1226 /* lease must have lmops break callback */
1227 fl->fl_lmops->fl_break(fl);
1228 }
1229 }
1230
1231 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1232 error = -EWOULDBLOCK;
1233 goto out;
1234 }
1235
1236restart:
1237 break_time = flock->fl_break_time;
1238 if (break_time != 0) {
1239 break_time -= jiffies;
1240 if (break_time == 0)
1241 break_time++;
1242 }
1243 error = locks_block_on_timeout(flock, new_fl, break_time);
1244 if (error >= 0) {
1245 if (error == 0)
1246 time_out_leases(inode);
1247 /* Wait for the next lease that has not been broken yet */
1248 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1249 flock = flock->fl_next) {
1250 if (flock->fl_type & F_INPROGRESS)
1251 goto restart;
1252 }
1253 error = 0;
1254 }
1255
1256out:
1257 unlock_kernel();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001258 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 locks_free_lock(new_fl);
1260 return error;
1261}
1262
1263EXPORT_SYMBOL(__break_lease);
1264
1265/**
1266 * lease_get_mtime
1267 * @inode: the inode
1268 * @time: pointer to a timespec which will contain the last modified time
1269 *
1270 * This is to force NFS clients to flush their caches for files with
1271 * exclusive leases. The justification is that if someone has an
1272 * exclusive lease, then they could be modifiying it.
1273 */
1274void lease_get_mtime(struct inode *inode, struct timespec *time)
1275{
1276 struct file_lock *flock = inode->i_flock;
1277 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1278 *time = current_fs_time(inode->i_sb);
1279 else
1280 *time = inode->i_mtime;
1281}
1282
1283EXPORT_SYMBOL(lease_get_mtime);
1284
1285/**
1286 * fcntl_getlease - Enquire what lease is currently active
1287 * @filp: the file
1288 *
1289 * The value returned by this function will be one of
1290 * (if no lease break is pending):
1291 *
1292 * %F_RDLCK to indicate a shared lease is held.
1293 *
1294 * %F_WRLCK to indicate an exclusive lease is held.
1295 *
1296 * %F_UNLCK to indicate no lease is held.
1297 *
1298 * (if a lease break is pending):
1299 *
1300 * %F_RDLCK to indicate an exclusive lease needs to be
1301 * changed to a shared lease (or removed).
1302 *
1303 * %F_UNLCK to indicate the lease needs to be removed.
1304 *
1305 * XXX: sfr & willy disagree over whether F_INPROGRESS
1306 * should be returned to userspace.
1307 */
1308int fcntl_getlease(struct file *filp)
1309{
1310 struct file_lock *fl;
1311 int type = F_UNLCK;
1312
1313 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001314 time_out_leases(filp->f_path.dentry->d_inode);
1315 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 fl = fl->fl_next) {
1317 if (fl->fl_file == filp) {
1318 type = fl->fl_type & ~F_INPROGRESS;
1319 break;
1320 }
1321 }
1322 unlock_kernel();
1323 return type;
1324}
1325
1326/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001327 * setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 * @filp: file pointer
1329 * @arg: type of lease to obtain
1330 * @flp: input - file_lock to use, output - file_lock inserted
1331 *
1332 * The (input) flp->fl_lmops->fl_break function is required
1333 * by break_lease().
1334 *
1335 * Called with kernel lock held.
1336 */
J. Bruce Fields4698afe2007-07-04 17:21:37 -04001337int setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001339 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001340 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 struct inode *inode = dentry->d_inode;
1342 int error, rdlease_count = 0, wrlease_count = 0;
1343
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001344 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1345 return -EACCES;
1346 if (!S_ISREG(inode->i_mode))
1347 return -EINVAL;
1348 error = security_file_lock(filp, arg);
1349 if (error)
1350 return error;
1351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 time_out_leases(inode);
1353
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001354 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001356 lease = *flp;
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 error = -EAGAIN;
1359 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1360 goto out;
1361 if ((arg == F_WRLCK)
1362 && ((atomic_read(&dentry->d_count) > 1)
1363 || (atomic_read(&inode->i_count) > 1)))
1364 goto out;
1365
1366 /*
1367 * At this point, we know that if there is an exclusive
1368 * lease on this file, then we hold it on this filp
1369 * (otherwise our open of this file would have blocked).
1370 * And if we are trying to acquire an exclusive lease,
1371 * then the file is not open by anyone (including us)
1372 * except for this filp.
1373 */
1374 for (before = &inode->i_flock;
1375 ((fl = *before) != NULL) && IS_LEASE(fl);
1376 before = &fl->fl_next) {
1377 if (lease->fl_lmops->fl_mylease(fl, lease))
1378 my_before = before;
1379 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1380 /*
1381 * Someone is in the process of opening this
1382 * file for writing so we may not take an
1383 * exclusive lease on it.
1384 */
1385 wrlease_count++;
1386 else
1387 rdlease_count++;
1388 }
1389
1390 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1391 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1392 goto out;
1393
1394 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001395 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 error = lease->fl_lmops->fl_change(my_before, arg);
1397 goto out;
1398 }
1399
1400 error = 0;
1401 if (arg == F_UNLCK)
1402 goto out;
1403
1404 error = -EINVAL;
1405 if (!leases_enable)
1406 goto out;
1407
Trond Myklebust74361cb2006-08-14 08:54:48 -07001408 error = -ENOMEM;
1409 fl = locks_alloc_lock();
1410 if (fl == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 goto out;
1412
1413 locks_copy_lock(fl, lease);
1414
1415 locks_insert_lock(before, fl);
1416
1417 *flp = fl;
Trond Myklebust74361cb2006-08-14 08:54:48 -07001418 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419out:
1420 return error;
1421}
J. Bruce Fields4698afe2007-07-04 17:21:37 -04001422EXPORT_SYMBOL(setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423
1424 /**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001425 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 * @filp: file pointer
1427 * @arg: type of lease to obtain
1428 * @lease: file_lock to use
1429 *
1430 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001431 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1432 * break_lease will oops!
1433 *
1434 * This will call the filesystem's setlease file method, if
1435 * defined. Note that there is no getlease method; instead, the
1436 * filesystem setlease method should call back to setlease() to
1437 * add a lease to the inode's lease list, where fcntl_getlease() can
1438 * find it. Since fcntl_getlease() only reports whether the current
1439 * task holds a lease, a cluster filesystem need only do this for
1440 * leases held by processes on this node.
1441 *
1442 * There is also no break_lease method; filesystems that
1443 * handle their own leases shoud break leases themselves from the
1444 * filesystem's open, create, and (on truncate) setattr methods.
1445 *
1446 * Warning: the only current setlease methods exist only to disable
1447 * leases in certain cases. More vfs changes may be required to
1448 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 */
1450
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001451int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 int error;
1454
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 lock_kernel();
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001456 if (filp->f_op && filp->f_op->setlease)
1457 error = filp->f_op->setlease(filp, arg, lease);
1458 else
1459 error = setlease(filp, arg, lease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 unlock_kernel();
1461
1462 return error;
1463}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001464EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
1466/**
1467 * fcntl_setlease - sets a lease on an open file
1468 * @fd: open file descriptor
1469 * @filp: file pointer
1470 * @arg: type of lease to obtain
1471 *
1472 * Call this fcntl to establish a lease on the file.
1473 * Note that you also need to call %F_SETSIG to
1474 * receive a signal when the lease is broken.
1475 */
1476int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1477{
1478 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001479 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 struct inode *inode = dentry->d_inode;
1481 int error;
1482
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 locks_init_lock(&fl);
1484 error = lease_init(filp, arg, &fl);
1485 if (error)
1486 return error;
1487
1488 lock_kernel();
1489
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001490 error = vfs_setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001491 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 goto out_unlock;
1493
1494 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1495 if (error < 0) {
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001496 /* remove lease just inserted by setlease */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 flp->fl_type = F_UNLCK | F_INPROGRESS;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001498 flp->fl_break_time = jiffies - 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 time_out_leases(inode);
1500 goto out_unlock;
1501 }
1502
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001503 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504out_unlock:
1505 unlock_kernel();
1506 return error;
1507}
1508
1509/**
1510 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1511 * @filp: The file to apply the lock to
1512 * @fl: The lock to be applied
1513 *
1514 * Add a FLOCK style lock to a file.
1515 */
1516int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1517{
1518 int error;
1519 might_sleep();
1520 for (;;) {
1521 error = flock_lock_file(filp, fl);
1522 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1523 break;
1524 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1525 if (!error)
1526 continue;
1527
1528 locks_delete_block(fl);
1529 break;
1530 }
1531 return error;
1532}
1533
1534EXPORT_SYMBOL(flock_lock_file_wait);
1535
1536/**
1537 * sys_flock: - flock() system call.
1538 * @fd: the file descriptor to lock.
1539 * @cmd: the type of lock to apply.
1540 *
1541 * Apply a %FL_FLOCK style lock to an open file descriptor.
1542 * The @cmd can be one of
1543 *
1544 * %LOCK_SH -- a shared lock.
1545 *
1546 * %LOCK_EX -- an exclusive lock.
1547 *
1548 * %LOCK_UN -- remove an existing lock.
1549 *
1550 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1551 *
1552 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1553 * processes read and write access respectively.
1554 */
1555asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1556{
1557 struct file *filp;
1558 struct file_lock *lock;
1559 int can_sleep, unlock;
1560 int error;
1561
1562 error = -EBADF;
1563 filp = fget(fd);
1564 if (!filp)
1565 goto out;
1566
1567 can_sleep = !(cmd & LOCK_NB);
1568 cmd &= ~LOCK_NB;
1569 unlock = (cmd == LOCK_UN);
1570
1571 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1572 goto out_putf;
1573
1574 error = flock_make_lock(filp, &lock, cmd);
1575 if (error)
1576 goto out_putf;
1577 if (can_sleep)
1578 lock->fl_flags |= FL_SLEEP;
1579
1580 error = security_file_lock(filp, cmd);
1581 if (error)
1582 goto out_free;
1583
1584 if (filp->f_op && filp->f_op->flock)
1585 error = filp->f_op->flock(filp,
1586 (can_sleep) ? F_SETLKW : F_SETLK,
1587 lock);
1588 else
1589 error = flock_lock_file_wait(filp, lock);
1590
1591 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001592 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593
1594 out_putf:
1595 fput(filp);
1596 out:
1597 return error;
1598}
1599
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001600/**
1601 * vfs_test_lock - test file byte range lock
1602 * @filp: The file to test lock for
1603 * @fl: The lock to test
1604 * @conf: Place to return a copy of the conflicting lock, if found
1605 *
1606 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1607 * setting conf->fl_type to something other than F_UNLCK.
1608 */
1609int vfs_test_lock(struct file *filp, struct file_lock *fl)
1610{
1611 if (filp->f_op && filp->f_op->lock)
1612 return filp->f_op->lock(filp, F_GETLK, fl);
1613 posix_test_lock(filp, fl);
1614 return 0;
1615}
1616EXPORT_SYMBOL_GPL(vfs_test_lock);
1617
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001618static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1619{
1620 flock->l_pid = fl->fl_pid;
1621#if BITS_PER_LONG == 32
1622 /*
1623 * Make sure we can represent the posix lock via
1624 * legacy 32bit flock.
1625 */
1626 if (fl->fl_start > OFFT_OFFSET_MAX)
1627 return -EOVERFLOW;
1628 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1629 return -EOVERFLOW;
1630#endif
1631 flock->l_start = fl->fl_start;
1632 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1633 fl->fl_end - fl->fl_start + 1;
1634 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001635 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001636 return 0;
1637}
1638
1639#if BITS_PER_LONG == 32
1640static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1641{
1642 flock->l_pid = fl->fl_pid;
1643 flock->l_start = fl->fl_start;
1644 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1645 fl->fl_end - fl->fl_start + 1;
1646 flock->l_whence = 0;
1647 flock->l_type = fl->fl_type;
1648}
1649#endif
1650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651/* Report the first existing lock that would conflict with l.
1652 * This implements the F_GETLK command of fcntl().
1653 */
1654int fcntl_getlk(struct file *filp, struct flock __user *l)
1655{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001656 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 struct flock flock;
1658 int error;
1659
1660 error = -EFAULT;
1661 if (copy_from_user(&flock, l, sizeof(flock)))
1662 goto out;
1663 error = -EINVAL;
1664 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1665 goto out;
1666
1667 error = flock_to_posix_lock(filp, &file_lock, &flock);
1668 if (error)
1669 goto out;
1670
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001671 error = vfs_test_lock(filp, &file_lock);
1672 if (error)
1673 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001675 flock.l_type = file_lock.fl_type;
1676 if (file_lock.fl_type != F_UNLCK) {
1677 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001678 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
1681 error = -EFAULT;
1682 if (!copy_to_user(l, &flock, sizeof(flock)))
1683 error = 0;
1684out:
1685 return error;
1686}
1687
Marc Eshel7723ec92007-01-18 15:08:55 -05001688/**
1689 * vfs_lock_file - file byte range lock
1690 * @filp: The file to apply the lock to
1691 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1692 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001693 * @conf: Place to return a copy of the conflicting lock, if found.
1694 *
1695 * A caller that doesn't care about the conflicting lock may pass NULL
1696 * as the final argument.
1697 *
1698 * If the filesystem defines a private ->lock() method, then @conf will
1699 * be left unchanged; so a caller that cares should initialize it to
1700 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001701 *
1702 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1703 * locks, the ->lock() interface may return asynchronously, before the lock has
1704 * been granted or denied by the underlying filesystem, if (and only if)
1705 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1706 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1707 * the request is for a blocking lock. When ->lock() does return asynchronously,
1708 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1709 * request completes.
1710 * If the request is for non-blocking lock the file system should return
1711 * -EINPROGRESS then try to get the lock and call the callback routine with
1712 * the result. If the request timed out the callback routine will return a
1713 * nonzero return code and the file system should release the lock. The file
1714 * system is also responsible to keep a corresponding posix lock when it
1715 * grants a lock so the VFS can find out which locks are locally held and do
1716 * the correct lock cleanup when required.
1717 * The underlying filesystem must not drop the kernel lock or call
1718 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1719 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001720 */
Marc Eshel150b3932007-01-18 16:15:35 -05001721int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001722{
1723 if (filp->f_op && filp->f_op->lock)
1724 return filp->f_op->lock(filp, cmd, fl);
1725 else
Marc Eshel150b3932007-01-18 16:15:35 -05001726 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001727}
1728EXPORT_SYMBOL_GPL(vfs_lock_file);
1729
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730/* Apply the lock described by l to an open file descriptor.
1731 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1732 */
Peter Staubachc2936212005-07-27 11:45:09 -07001733int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1734 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735{
1736 struct file_lock *file_lock = locks_alloc_lock();
1737 struct flock flock;
1738 struct inode *inode;
1739 int error;
1740
1741 if (file_lock == NULL)
1742 return -ENOLCK;
1743
1744 /*
1745 * This might block, so we do it before checking the inode.
1746 */
1747 error = -EFAULT;
1748 if (copy_from_user(&flock, l, sizeof(flock)))
1749 goto out;
1750
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001751 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
1753 /* Don't allow mandatory locks on files that may be memory mapped
1754 * and shared.
1755 */
1756 if (IS_MANDLOCK(inode) &&
1757 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1758 mapping_writably_mapped(filp->f_mapping)) {
1759 error = -EAGAIN;
1760 goto out;
1761 }
1762
Peter Staubachc2936212005-07-27 11:45:09 -07001763again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 error = flock_to_posix_lock(filp, file_lock, &flock);
1765 if (error)
1766 goto out;
1767 if (cmd == F_SETLKW) {
1768 file_lock->fl_flags |= FL_SLEEP;
1769 }
1770
1771 error = -EBADF;
1772 switch (flock.l_type) {
1773 case F_RDLCK:
1774 if (!(filp->f_mode & FMODE_READ))
1775 goto out;
1776 break;
1777 case F_WRLCK:
1778 if (!(filp->f_mode & FMODE_WRITE))
1779 goto out;
1780 break;
1781 case F_UNLCK:
1782 break;
1783 default:
1784 error = -EINVAL;
1785 goto out;
1786 }
1787
1788 error = security_file_lock(filp, file_lock->fl_type);
1789 if (error)
1790 goto out;
1791
Marc Eshel7723ec92007-01-18 15:08:55 -05001792 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001793 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001794 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001796 error = wait_event_interruptible(file_lock->fl_wait,
1797 !file_lock->fl_next);
1798 if (!error)
1799 continue;
1800
1801 locks_delete_block(file_lock);
1802 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804
Peter Staubachc2936212005-07-27 11:45:09 -07001805 /*
1806 * Attempt to detect a close/fcntl race and recover by
1807 * releasing the lock that was just acquired.
1808 */
1809 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1810 flock.l_type = F_UNLCK;
1811 goto again;
1812 }
1813
1814out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 locks_free_lock(file_lock);
1816 return error;
1817}
1818
1819#if BITS_PER_LONG == 32
1820/* Report the first existing lock that would conflict with l.
1821 * This implements the F_GETLK command of fcntl().
1822 */
1823int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1824{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001825 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 struct flock64 flock;
1827 int error;
1828
1829 error = -EFAULT;
1830 if (copy_from_user(&flock, l, sizeof(flock)))
1831 goto out;
1832 error = -EINVAL;
1833 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1834 goto out;
1835
1836 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1837 if (error)
1838 goto out;
1839
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001840 error = vfs_test_lock(filp, &file_lock);
1841 if (error)
1842 goto out;
1843
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001844 flock.l_type = file_lock.fl_type;
1845 if (file_lock.fl_type != F_UNLCK)
1846 posix_lock_to_flock64(&flock, &file_lock);
1847
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 error = -EFAULT;
1849 if (!copy_to_user(l, &flock, sizeof(flock)))
1850 error = 0;
1851
1852out:
1853 return error;
1854}
1855
1856/* Apply the lock described by l to an open file descriptor.
1857 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1858 */
Peter Staubachc2936212005-07-27 11:45:09 -07001859int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1860 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
1862 struct file_lock *file_lock = locks_alloc_lock();
1863 struct flock64 flock;
1864 struct inode *inode;
1865 int error;
1866
1867 if (file_lock == NULL)
1868 return -ENOLCK;
1869
1870 /*
1871 * This might block, so we do it before checking the inode.
1872 */
1873 error = -EFAULT;
1874 if (copy_from_user(&flock, l, sizeof(flock)))
1875 goto out;
1876
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001877 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
1879 /* Don't allow mandatory locks on files that may be memory mapped
1880 * and shared.
1881 */
1882 if (IS_MANDLOCK(inode) &&
1883 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1884 mapping_writably_mapped(filp->f_mapping)) {
1885 error = -EAGAIN;
1886 goto out;
1887 }
1888
Peter Staubachc2936212005-07-27 11:45:09 -07001889again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 error = flock64_to_posix_lock(filp, file_lock, &flock);
1891 if (error)
1892 goto out;
1893 if (cmd == F_SETLKW64) {
1894 file_lock->fl_flags |= FL_SLEEP;
1895 }
1896
1897 error = -EBADF;
1898 switch (flock.l_type) {
1899 case F_RDLCK:
1900 if (!(filp->f_mode & FMODE_READ))
1901 goto out;
1902 break;
1903 case F_WRLCK:
1904 if (!(filp->f_mode & FMODE_WRITE))
1905 goto out;
1906 break;
1907 case F_UNLCK:
1908 break;
1909 default:
1910 error = -EINVAL;
1911 goto out;
1912 }
1913
1914 error = security_file_lock(filp, file_lock->fl_type);
1915 if (error)
1916 goto out;
1917
Marc Eshel7723ec92007-01-18 15:08:55 -05001918 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001919 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001920 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001921 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001922 error = wait_event_interruptible(file_lock->fl_wait,
1923 !file_lock->fl_next);
1924 if (!error)
1925 continue;
1926
1927 locks_delete_block(file_lock);
1928 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 }
1930
Peter Staubachc2936212005-07-27 11:45:09 -07001931 /*
1932 * Attempt to detect a close/fcntl race and recover by
1933 * releasing the lock that was just acquired.
1934 */
1935 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1936 flock.l_type = F_UNLCK;
1937 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
1939
1940out:
1941 locks_free_lock(file_lock);
1942 return error;
1943}
1944#endif /* BITS_PER_LONG == 32 */
1945
1946/*
1947 * This function is called when the file is being removed
1948 * from the task's fd array. POSIX locks belonging to this task
1949 * are deleted at this time.
1950 */
1951void locks_remove_posix(struct file *filp, fl_owner_t owner)
1952{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001953 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955 /*
1956 * If there are no locks held on this file, we don't need to call
1957 * posix_lock_file(). Another process could be setting a lock on this
1958 * file at the same time, but we wouldn't remove that lock anyway.
1959 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001960 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 return;
1962
1963 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001964 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 lock.fl_start = 0;
1966 lock.fl_end = OFFSET_MAX;
1967 lock.fl_owner = owner;
1968 lock.fl_pid = current->tgid;
1969 lock.fl_file = filp;
1970 lock.fl_ops = NULL;
1971 lock.fl_lmops = NULL;
1972
Marc Eshel150b3932007-01-18 16:15:35 -05001973 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1976 lock.fl_ops->fl_release_private(&lock);
1977}
1978
1979EXPORT_SYMBOL(locks_remove_posix);
1980
1981/*
1982 * This function is called on the last close of an open file.
1983 */
1984void locks_remove_flock(struct file *filp)
1985{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001986 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 struct file_lock *fl;
1988 struct file_lock **before;
1989
1990 if (!inode->i_flock)
1991 return;
1992
1993 if (filp->f_op && filp->f_op->flock) {
1994 struct file_lock fl = {
1995 .fl_pid = current->tgid,
1996 .fl_file = filp,
1997 .fl_flags = FL_FLOCK,
1998 .fl_type = F_UNLCK,
1999 .fl_end = OFFSET_MAX,
2000 };
2001 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002002 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2003 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 }
2005
2006 lock_kernel();
2007 before = &inode->i_flock;
2008
2009 while ((fl = *before) != NULL) {
2010 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002011 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 locks_delete_lock(before);
2013 continue;
2014 }
2015 if (IS_LEASE(fl)) {
2016 lease_modify(before, F_UNLCK);
2017 continue;
2018 }
2019 /* What? */
2020 BUG();
2021 }
2022 before = &fl->fl_next;
2023 }
2024 unlock_kernel();
2025}
2026
2027/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 * posix_unblock_lock - stop waiting for a file lock
2029 * @filp: how the file was opened
2030 * @waiter: the lock which was waiting
2031 *
2032 * lockd needs to block waiting for locks.
2033 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002034int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2036{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002037 int status = 0;
2038
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002040 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002042 else
2043 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002044 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002045 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
2048EXPORT_SYMBOL(posix_unblock_lock);
2049
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002050/**
2051 * vfs_cancel_lock - file byte range unblock lock
2052 * @filp: The file to apply the unblock to
2053 * @fl: The lock to be unblocked
2054 *
2055 * Used by lock managers to cancel blocked requests
2056 */
2057int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2058{
2059 if (filp->f_op && filp->f_op->lock)
2060 return filp->f_op->lock(filp, F_CANCELLK, fl);
2061 return 0;
2062}
2063
2064EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2065
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2067{
2068 struct inode *inode = NULL;
2069
2070 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002071 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
2073 out += sprintf(out, "%d:%s ", id, pfx);
2074 if (IS_POSIX(fl)) {
2075 out += sprintf(out, "%6s %s ",
2076 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2077 (inode == NULL) ? "*NOINODE*" :
2078 (IS_MANDLOCK(inode) &&
2079 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2080 "MANDATORY" : "ADVISORY ");
2081 } else if (IS_FLOCK(fl)) {
2082 if (fl->fl_type & LOCK_MAND) {
2083 out += sprintf(out, "FLOCK MSNFS ");
2084 } else {
2085 out += sprintf(out, "FLOCK ADVISORY ");
2086 }
2087 } else if (IS_LEASE(fl)) {
2088 out += sprintf(out, "LEASE ");
2089 if (fl->fl_type & F_INPROGRESS)
2090 out += sprintf(out, "BREAKING ");
2091 else if (fl->fl_file)
2092 out += sprintf(out, "ACTIVE ");
2093 else
2094 out += sprintf(out, "BREAKER ");
2095 } else {
2096 out += sprintf(out, "UNKNOWN UNKNOWN ");
2097 }
2098 if (fl->fl_type & LOCK_MAND) {
2099 out += sprintf(out, "%s ",
2100 (fl->fl_type & LOCK_READ)
2101 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2102 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2103 } else {
2104 out += sprintf(out, "%s ",
2105 (fl->fl_type & F_INPROGRESS)
2106 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2107 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2108 }
2109 if (inode) {
2110#ifdef WE_CAN_BREAK_LSLK_NOW
2111 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2112 inode->i_sb->s_id, inode->i_ino);
2113#else
2114 /* userspace relies on this representation of dev_t ;-( */
2115 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2116 MAJOR(inode->i_sb->s_dev),
2117 MINOR(inode->i_sb->s_dev), inode->i_ino);
2118#endif
2119 } else {
2120 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2121 }
2122 if (IS_POSIX(fl)) {
2123 if (fl->fl_end == OFFSET_MAX)
2124 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2125 else
2126 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2127 fl->fl_end);
2128 } else {
2129 out += sprintf(out, "0 EOF\n");
2130 }
2131}
2132
2133static void move_lock_status(char **p, off_t* pos, off_t offset)
2134{
2135 int len;
2136 len = strlen(*p);
2137 if(*pos >= offset) {
2138 /* the complete line is valid */
2139 *p += len;
2140 *pos += len;
2141 return;
2142 }
2143 if(*pos+len > offset) {
2144 /* use the second part of the line */
2145 int i = offset-*pos;
2146 memmove(*p,*p+i,len-i);
2147 *p += len-i;
2148 *pos += len;
2149 return;
2150 }
2151 /* discard the complete line */
2152 *pos += len;
2153}
2154
2155/**
2156 * get_locks_status - reports lock usage in /proc/locks
2157 * @buffer: address in userspace to write into
2158 * @start: ?
2159 * @offset: how far we are through the buffer
2160 * @length: how much to read
2161 */
2162
2163int get_locks_status(char *buffer, char **start, off_t offset, int length)
2164{
2165 struct list_head *tmp;
2166 char *q = buffer;
2167 off_t pos = 0;
2168 int i = 0;
2169
2170 lock_kernel();
2171 list_for_each(tmp, &file_lock_list) {
2172 struct list_head *btmp;
2173 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2174 lock_get_status(q, fl, ++i, "");
2175 move_lock_status(&q, &pos, offset);
2176
2177 if(pos >= offset+length)
2178 goto done;
2179
2180 list_for_each(btmp, &fl->fl_block) {
2181 struct file_lock *bfl = list_entry(btmp,
2182 struct file_lock, fl_block);
2183 lock_get_status(q, bfl, i, " ->");
2184 move_lock_status(&q, &pos, offset);
2185
2186 if(pos >= offset+length)
2187 goto done;
2188 }
2189 }
2190done:
2191 unlock_kernel();
2192 *start = buffer;
2193 if(q-buffer < length)
2194 return (q-buffer);
2195 return length;
2196}
2197
2198/**
2199 * lock_may_read - checks that the region is free of locks
2200 * @inode: the inode that is being read
2201 * @start: the first byte to read
2202 * @len: the number of bytes to read
2203 *
2204 * Emulates Windows locking requirements. Whole-file
2205 * mandatory locks (share modes) can prohibit a read and
2206 * byte-range POSIX locks can prohibit a read if they overlap.
2207 *
2208 * N.B. this function is only ever called
2209 * from knfsd and ownership of locks is never checked.
2210 */
2211int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2212{
2213 struct file_lock *fl;
2214 int result = 1;
2215 lock_kernel();
2216 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2217 if (IS_POSIX(fl)) {
2218 if (fl->fl_type == F_RDLCK)
2219 continue;
2220 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2221 continue;
2222 } else if (IS_FLOCK(fl)) {
2223 if (!(fl->fl_type & LOCK_MAND))
2224 continue;
2225 if (fl->fl_type & LOCK_READ)
2226 continue;
2227 } else
2228 continue;
2229 result = 0;
2230 break;
2231 }
2232 unlock_kernel();
2233 return result;
2234}
2235
2236EXPORT_SYMBOL(lock_may_read);
2237
2238/**
2239 * lock_may_write - checks that the region is free of locks
2240 * @inode: the inode that is being written
2241 * @start: the first byte to write
2242 * @len: the number of bytes to write
2243 *
2244 * Emulates Windows locking requirements. Whole-file
2245 * mandatory locks (share modes) can prohibit a write and
2246 * byte-range POSIX locks can prohibit a write if they overlap.
2247 *
2248 * N.B. this function is only ever called
2249 * from knfsd and ownership of locks is never checked.
2250 */
2251int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2252{
2253 struct file_lock *fl;
2254 int result = 1;
2255 lock_kernel();
2256 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2257 if (IS_POSIX(fl)) {
2258 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2259 continue;
2260 } else if (IS_FLOCK(fl)) {
2261 if (!(fl->fl_type & LOCK_MAND))
2262 continue;
2263 if (fl->fl_type & LOCK_WRITE)
2264 continue;
2265 } else
2266 continue;
2267 result = 0;
2268 break;
2269 }
2270 unlock_kernel();
2271 return result;
2272}
2273
2274EXPORT_SYMBOL(lock_may_write);
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276static int __init filelock_init(void)
2277{
2278 filelock_cache = kmem_cache_create("file_lock_cache",
2279 sizeof(struct file_lock), 0, SLAB_PANIC,
2280 init_once, NULL);
2281 return 0;
2282}
2283
2284core_initcall(filelock_init);