blob: 282b6c11670aa1c2f80239d2524378f8cd7bc2b0 [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)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400537 struct file_lock *waiter;
538
539 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 struct file_lock, fl_block);
541 __locks_delete_block(waiter);
542 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
543 waiter->fl_lmops->fl_notify(waiter);
544 else
545 wake_up(&waiter->fl_wait);
546 }
547}
548
549/* Insert file lock fl into an inode's lock list at the position indicated
550 * by pos. At the same time add the lock to the global file lock list.
551 */
552static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
553{
554 list_add(&fl->fl_link, &file_lock_list);
555
556 /* insert into file's list */
557 fl->fl_next = *pos;
558 *pos = fl;
559
560 if (fl->fl_ops && fl->fl_ops->fl_insert)
561 fl->fl_ops->fl_insert(fl);
562}
563
564/*
565 * Delete a lock and then free it.
566 * Wake up processes that are blocked waiting for this lock,
567 * notify the FS that the lock has been cleared and
568 * finally free the lock.
569 */
570static void locks_delete_lock(struct file_lock **thisfl_p)
571{
572 struct file_lock *fl = *thisfl_p;
573
574 *thisfl_p = fl->fl_next;
575 fl->fl_next = NULL;
576 list_del_init(&fl->fl_link);
577
578 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
579 if (fl->fl_fasync != NULL) {
580 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
581 fl->fl_fasync = NULL;
582 }
583
584 if (fl->fl_ops && fl->fl_ops->fl_remove)
585 fl->fl_ops->fl_remove(fl);
586
587 locks_wake_up_blocks(fl);
588 locks_free_lock(fl);
589}
590
591/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
592 * checks for shared/exclusive status of overlapping locks.
593 */
594static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
595{
596 if (sys_fl->fl_type == F_WRLCK)
597 return 1;
598 if (caller_fl->fl_type == F_WRLCK)
599 return 1;
600 return 0;
601}
602
603/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
604 * checking before calling the locks_conflict().
605 */
606static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
607{
608 /* POSIX locks owned by the same process do not conflict with
609 * each other.
610 */
611 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
612 return (0);
613
614 /* Check whether they overlap */
615 if (!locks_overlap(caller_fl, sys_fl))
616 return 0;
617
618 return (locks_conflict(caller_fl, sys_fl));
619}
620
621/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
622 * checking before calling the locks_conflict().
623 */
624static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
625{
626 /* FLOCK locks referring to the same filp do not conflict with
627 * each other.
628 */
629 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
630 return (0);
631 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
632 return 0;
633
634 return (locks_conflict(caller_fl, sys_fl));
635}
636
637static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
638{
639 int result = 0;
640 DECLARE_WAITQUEUE(wait, current);
641
642 __set_current_state(TASK_INTERRUPTIBLE);
643 add_wait_queue(fl_wait, &wait);
644 if (timeout == 0)
645 schedule();
646 else
647 result = schedule_timeout(timeout);
648 if (signal_pending(current))
649 result = -ERESTARTSYS;
650 remove_wait_queue(fl_wait, &wait);
651 __set_current_state(TASK_RUNNING);
652 return result;
653}
654
655static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
656{
657 int result;
658 locks_insert_block(blocker, waiter);
659 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
660 __locks_delete_block(waiter);
661 return result;
662}
663
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400664void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500665posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 struct file_lock *cfl;
668
669 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800670 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 if (!IS_POSIX(cfl))
672 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400673 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 break;
675 }
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400676 if (cfl)
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500677 __locks_copy_lock(fl, cfl);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400678 else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400679 fl->fl_type = F_UNLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 unlock_kernel();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400681 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682}
683
684EXPORT_SYMBOL(posix_test_lock);
685
686/* This function tests for deadlock condition before putting a process to
687 * sleep. The detection scheme is no longer recursive. Recursive was neat,
688 * but dangerous - we risked stack corruption if the lock data was bad, or
689 * if the recursion was too deep for any other reason.
690 *
691 * We rely on the fact that a task can only be on one lock's wait queue
692 * at a time. When we find blocked_task on a wait queue we can re-search
693 * with blocked_task equal to that queue's owner, until either blocked_task
694 * isn't found, or blocked_task is found on a queue owned by my_task.
695 *
696 * Note: the above assumption may not be true when handling lock requests
697 * from a broken NFS client. But broken NFS clients have a lot more to
698 * worry about than proper deadlock detection anyway... --okir
699 */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700700static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 struct file_lock *block_fl)
702{
703 struct list_head *tmp;
704
705next_task:
706 if (posix_same_owner(caller_fl, block_fl))
707 return 1;
708 list_for_each(tmp, &blocked_list) {
709 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
710 if (posix_same_owner(fl, block_fl)) {
711 fl = fl->fl_next;
712 block_fl = fl;
713 goto next_task;
714 }
715 }
716 return 0;
717}
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400720 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400721 *
722 * Note that if called with an FL_EXISTS argument, the caller may determine
723 * whether or not a lock was successfully freed by testing the return
724 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800726static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800728 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800730 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 int error = 0;
732 int found = 0;
733
734 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400735 if (request->fl_flags & FL_ACCESS)
736 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400737
738 if (request->fl_type != F_UNLCK) {
739 error = -ENOMEM;
740 new_fl = locks_alloc_lock();
741 if (new_fl == NULL)
742 goto out;
743 error = 0;
744 }
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 for_each_lock(inode, before) {
747 struct file_lock *fl = *before;
748 if (IS_POSIX(fl))
749 break;
750 if (IS_LEASE(fl))
751 continue;
752 if (filp != fl->fl_file)
753 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800754 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 goto out;
756 found = 1;
757 locks_delete_lock(before);
758 break;
759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Trond Myklebustf475ae92006-06-29 16:38:32 -0400761 if (request->fl_type == F_UNLCK) {
762 if ((request->fl_flags & FL_EXISTS) && !found)
763 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800764 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
767 /*
768 * If a higher-priority process was blocked on the old file lock,
769 * give it the opportunity to lock the file.
770 */
771 if (found)
772 cond_resched();
773
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400774find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 for_each_lock(inode, before) {
776 struct file_lock *fl = *before;
777 if (IS_POSIX(fl))
778 break;
779 if (IS_LEASE(fl))
780 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800781 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 continue;
783 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800784 if (request->fl_flags & FL_SLEEP)
785 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 goto out;
787 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400788 if (request->fl_flags & FL_ACCESS)
789 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800790 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700791 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800792 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400793 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794
795out:
796 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800797 if (new_fl)
798 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return error;
800}
801
Marc Eshel150b3932007-01-18 16:15:35 -0500802static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803{
804 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700805 struct file_lock *new_fl = NULL;
806 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 struct file_lock *left = NULL;
808 struct file_lock *right = NULL;
809 struct file_lock **before;
810 int error, added = 0;
811
812 /*
813 * We may need two file_lock structures for this operation,
814 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700815 *
816 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700818 if (!(request->fl_flags & FL_ACCESS) &&
819 (request->fl_type != F_UNLCK ||
820 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
821 new_fl = locks_alloc_lock();
822 new_fl2 = locks_alloc_lock();
823 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824
825 lock_kernel();
826 if (request->fl_type != F_UNLCK) {
827 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500828 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 if (!IS_POSIX(fl))
830 continue;
831 if (!posix_locks_conflict(request, fl))
832 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800833 if (conflock)
834 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 error = -EAGAIN;
836 if (!(request->fl_flags & FL_SLEEP))
837 goto out;
838 error = -EDEADLK;
839 if (posix_locks_deadlock(request, fl))
840 goto out;
841 error = -EAGAIN;
842 locks_insert_block(fl, request);
843 goto out;
844 }
845 }
846
847 /* If we're just looking for a conflict, we're done. */
848 error = 0;
849 if (request->fl_flags & FL_ACCESS)
850 goto out;
851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 * Find the first old lock with the same owner as the new lock.
854 */
855
856 before = &inode->i_flock;
857
858 /* First skip locks owned by other processes. */
859 while ((fl = *before) && (!IS_POSIX(fl) ||
860 !posix_same_owner(request, fl))) {
861 before = &fl->fl_next;
862 }
863
864 /* Process locks with this owner. */
865 while ((fl = *before) && posix_same_owner(request, fl)) {
866 /* Detect adjacent or overlapping regions (if same lock type)
867 */
868 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700869 /* In all comparisons of start vs end, use
870 * "start - 1" rather than "end + 1". If end
871 * is OFFSET_MAX, end + 1 will become negative.
872 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 if (fl->fl_end < request->fl_start - 1)
874 goto next_lock;
875 /* If the next lock in the list has entirely bigger
876 * addresses than the new one, insert the lock here.
877 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700878 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 break;
880
881 /* If we come here, the new and old lock are of the
882 * same type and adjacent or overlapping. Make one
883 * lock yielding from the lower start address of both
884 * locks to the higher end address.
885 */
886 if (fl->fl_start > request->fl_start)
887 fl->fl_start = request->fl_start;
888 else
889 request->fl_start = fl->fl_start;
890 if (fl->fl_end < request->fl_end)
891 fl->fl_end = request->fl_end;
892 else
893 request->fl_end = fl->fl_end;
894 if (added) {
895 locks_delete_lock(before);
896 continue;
897 }
898 request = fl;
899 added = 1;
900 }
901 else {
902 /* Processing for different lock types is a bit
903 * more complex.
904 */
905 if (fl->fl_end < request->fl_start)
906 goto next_lock;
907 if (fl->fl_start > request->fl_end)
908 break;
909 if (request->fl_type == F_UNLCK)
910 added = 1;
911 if (fl->fl_start < request->fl_start)
912 left = fl;
913 /* If the next lock in the list has a higher end
914 * address than the new one, insert the new one here.
915 */
916 if (fl->fl_end > request->fl_end) {
917 right = fl;
918 break;
919 }
920 if (fl->fl_start >= request->fl_start) {
921 /* The new lock completely replaces an old
922 * one (This may happen several times).
923 */
924 if (added) {
925 locks_delete_lock(before);
926 continue;
927 }
928 /* Replace the old lock with the new one.
929 * Wake up anybody waiting for the old one,
930 * as the change in lock type might satisfy
931 * their needs.
932 */
933 locks_wake_up_blocks(fl);
934 fl->fl_start = request->fl_start;
935 fl->fl_end = request->fl_end;
936 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500937 locks_release_private(fl);
938 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 request = fl;
940 added = 1;
941 }
942 }
943 /* Go on to next lock.
944 */
945 next_lock:
946 before = &fl->fl_next;
947 }
948
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700949 /*
950 * The above code only modifies existing locks in case of
951 * merging or replacing. If new lock(s) need to be inserted
952 * all modifications are done bellow this, so it's safe yet to
953 * bail out.
954 */
955 error = -ENOLCK; /* "no luck" */
956 if (right && left == right && !new_fl2)
957 goto out;
958
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 error = 0;
960 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400961 if (request->fl_type == F_UNLCK) {
962 if (request->fl_flags & FL_EXISTS)
963 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400965 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700966
967 if (!new_fl) {
968 error = -ENOLCK;
969 goto out;
970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 locks_copy_lock(new_fl, request);
972 locks_insert_lock(before, new_fl);
973 new_fl = NULL;
974 }
975 if (right) {
976 if (left == right) {
977 /* The new lock breaks the old one in two pieces,
978 * so we have to use the second new lock.
979 */
980 left = new_fl2;
981 new_fl2 = NULL;
982 locks_copy_lock(left, right);
983 locks_insert_lock(before, left);
984 }
985 right->fl_start = request->fl_end + 1;
986 locks_wake_up_blocks(right);
987 }
988 if (left) {
989 left->fl_end = request->fl_start - 1;
990 locks_wake_up_blocks(left);
991 }
992 out:
993 unlock_kernel();
994 /*
995 * Free any unused locks.
996 */
997 if (new_fl)
998 locks_free_lock(new_fl);
999 if (new_fl2)
1000 locks_free_lock(new_fl2);
1001 return error;
1002}
1003
1004/**
1005 * posix_lock_file - Apply a POSIX-style lock to a file
1006 * @filp: The file to apply the lock to
1007 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001008 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 *
1010 * Add a POSIX style lock to a file.
1011 * We merge adjacent & overlapping locks whenever possible.
1012 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001013 *
1014 * Note that if called with an FL_EXISTS argument, the caller may determine
1015 * whether or not a lock was successfully freed by testing the return
1016 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 */
Marc Eshel150b3932007-01-18 16:15:35 -05001018int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001019 struct file_lock *conflock)
1020{
Marc Eshel150b3932007-01-18 16:15:35 -05001021 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001022}
Marc Eshel150b3932007-01-18 16:15:35 -05001023EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024
1025/**
1026 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1027 * @filp: The file to apply the lock to
1028 * @fl: The lock to be applied
1029 *
1030 * Add a POSIX style lock to a file.
1031 * We merge adjacent & overlapping locks whenever possible.
1032 * POSIX locks are sorted by owner task, then by starting address
1033 */
1034int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1035{
1036 int error;
1037 might_sleep ();
1038 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001039 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1041 break;
1042 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1043 if (!error)
1044 continue;
1045
1046 locks_delete_block(fl);
1047 break;
1048 }
1049 return error;
1050}
1051EXPORT_SYMBOL(posix_lock_file_wait);
1052
1053/**
1054 * locks_mandatory_locked - Check for an active lock
1055 * @inode: the file to check
1056 *
1057 * Searches the inode's list of locks to find any POSIX locks which conflict.
1058 * This function is called from locks_verify_locked() only.
1059 */
1060int locks_mandatory_locked(struct inode *inode)
1061{
1062 fl_owner_t owner = current->files;
1063 struct file_lock *fl;
1064
1065 /*
1066 * Search the lock list for this inode for any POSIX locks.
1067 */
1068 lock_kernel();
1069 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1070 if (!IS_POSIX(fl))
1071 continue;
1072 if (fl->fl_owner != owner)
1073 break;
1074 }
1075 unlock_kernel();
1076 return fl ? -EAGAIN : 0;
1077}
1078
1079/**
1080 * locks_mandatory_area - Check for a conflicting lock
1081 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1082 * for shared
1083 * @inode: the file to check
1084 * @filp: how the file was opened (if it was)
1085 * @offset: start of area to check
1086 * @count: length of area to check
1087 *
1088 * Searches the inode's list of locks to find any POSIX locks which conflict.
1089 * This function is called from rw_verify_area() and
1090 * locks_verify_truncate().
1091 */
1092int locks_mandatory_area(int read_write, struct inode *inode,
1093 struct file *filp, loff_t offset,
1094 size_t count)
1095{
1096 struct file_lock fl;
1097 int error;
1098
1099 locks_init_lock(&fl);
1100 fl.fl_owner = current->files;
1101 fl.fl_pid = current->tgid;
1102 fl.fl_file = filp;
1103 fl.fl_flags = FL_POSIX | FL_ACCESS;
1104 if (filp && !(filp->f_flags & O_NONBLOCK))
1105 fl.fl_flags |= FL_SLEEP;
1106 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1107 fl.fl_start = offset;
1108 fl.fl_end = offset + count - 1;
1109
1110 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001111 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 if (error != -EAGAIN)
1113 break;
1114 if (!(fl.fl_flags & FL_SLEEP))
1115 break;
1116 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1117 if (!error) {
1118 /*
1119 * If we've been sleeping someone might have
1120 * changed the permissions behind our back.
1121 */
1122 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1123 continue;
1124 }
1125
1126 locks_delete_block(&fl);
1127 break;
1128 }
1129
1130 return error;
1131}
1132
1133EXPORT_SYMBOL(locks_mandatory_area);
1134
1135/* We already had a lease on this file; just change its type */
1136int lease_modify(struct file_lock **before, int arg)
1137{
1138 struct file_lock *fl = *before;
1139 int error = assign_type(fl, arg);
1140
1141 if (error)
1142 return error;
1143 locks_wake_up_blocks(fl);
1144 if (arg == F_UNLCK)
1145 locks_delete_lock(before);
1146 return 0;
1147}
1148
1149EXPORT_SYMBOL(lease_modify);
1150
1151static void time_out_leases(struct inode *inode)
1152{
1153 struct file_lock **before;
1154 struct file_lock *fl;
1155
1156 before = &inode->i_flock;
1157 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1158 if ((fl->fl_break_time == 0)
1159 || time_before(jiffies, fl->fl_break_time)) {
1160 before = &fl->fl_next;
1161 continue;
1162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1164 if (fl == *before) /* lease_modify may have freed fl */
1165 before = &fl->fl_next;
1166 }
1167}
1168
1169/**
1170 * __break_lease - revoke all outstanding leases on file
1171 * @inode: the inode of the file to return
1172 * @mode: the open mode (read or write)
1173 *
david m. richter87250dd2007-05-09 16:10:27 -04001174 * break_lease (inlined for speed) has checked there already is at least
1175 * some kind of lock (maybe a lease) on this file. Leases are broken on
1176 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 * specified %O_NONBLOCK to your open().
1178 */
1179int __break_lease(struct inode *inode, unsigned int mode)
1180{
1181 int error = 0, future;
1182 struct file_lock *new_fl, *flock;
1183 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 unsigned long break_time;
1185 int i_have_this_lease = 0;
1186
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001187 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189 lock_kernel();
1190
1191 time_out_leases(inode);
1192
1193 flock = inode->i_flock;
1194 if ((flock == NULL) || !IS_LEASE(flock))
1195 goto out;
1196
1197 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1198 if (fl->fl_owner == current->files)
1199 i_have_this_lease = 1;
1200
1201 if (mode & FMODE_WRITE) {
1202 /* If we want write access, we have to revoke any lease. */
1203 future = F_UNLCK | F_INPROGRESS;
1204 } else if (flock->fl_type & F_INPROGRESS) {
1205 /* If the lease is already being broken, we just leave it */
1206 future = flock->fl_type;
1207 } else if (flock->fl_type & F_WRLCK) {
1208 /* Downgrade the exclusive lease to a read-only lease. */
1209 future = F_RDLCK | F_INPROGRESS;
1210 } else {
1211 /* the existing lease was read-only, so we can read too. */
1212 goto out;
1213 }
1214
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001215 if (IS_ERR(new_fl) && !i_have_this_lease
1216 && ((mode & O_NONBLOCK) == 0)) {
1217 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 goto out;
1219 }
1220
1221 break_time = 0;
1222 if (lease_break_time > 0) {
1223 break_time = jiffies + lease_break_time * HZ;
1224 if (break_time == 0)
1225 break_time++; /* so that 0 means no break time */
1226 }
1227
1228 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1229 if (fl->fl_type != future) {
1230 fl->fl_type = future;
1231 fl->fl_break_time = break_time;
1232 /* lease must have lmops break callback */
1233 fl->fl_lmops->fl_break(fl);
1234 }
1235 }
1236
1237 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1238 error = -EWOULDBLOCK;
1239 goto out;
1240 }
1241
1242restart:
1243 break_time = flock->fl_break_time;
1244 if (break_time != 0) {
1245 break_time -= jiffies;
1246 if (break_time == 0)
1247 break_time++;
1248 }
1249 error = locks_block_on_timeout(flock, new_fl, break_time);
1250 if (error >= 0) {
1251 if (error == 0)
1252 time_out_leases(inode);
1253 /* Wait for the next lease that has not been broken yet */
1254 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1255 flock = flock->fl_next) {
1256 if (flock->fl_type & F_INPROGRESS)
1257 goto restart;
1258 }
1259 error = 0;
1260 }
1261
1262out:
1263 unlock_kernel();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001264 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 locks_free_lock(new_fl);
1266 return error;
1267}
1268
1269EXPORT_SYMBOL(__break_lease);
1270
1271/**
1272 * lease_get_mtime
1273 * @inode: the inode
1274 * @time: pointer to a timespec which will contain the last modified time
1275 *
1276 * This is to force NFS clients to flush their caches for files with
1277 * exclusive leases. The justification is that if someone has an
1278 * exclusive lease, then they could be modifiying it.
1279 */
1280void lease_get_mtime(struct inode *inode, struct timespec *time)
1281{
1282 struct file_lock *flock = inode->i_flock;
1283 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1284 *time = current_fs_time(inode->i_sb);
1285 else
1286 *time = inode->i_mtime;
1287}
1288
1289EXPORT_SYMBOL(lease_get_mtime);
1290
1291/**
1292 * fcntl_getlease - Enquire what lease is currently active
1293 * @filp: the file
1294 *
1295 * The value returned by this function will be one of
1296 * (if no lease break is pending):
1297 *
1298 * %F_RDLCK to indicate a shared lease is held.
1299 *
1300 * %F_WRLCK to indicate an exclusive lease is held.
1301 *
1302 * %F_UNLCK to indicate no lease is held.
1303 *
1304 * (if a lease break is pending):
1305 *
1306 * %F_RDLCK to indicate an exclusive lease needs to be
1307 * changed to a shared lease (or removed).
1308 *
1309 * %F_UNLCK to indicate the lease needs to be removed.
1310 *
1311 * XXX: sfr & willy disagree over whether F_INPROGRESS
1312 * should be returned to userspace.
1313 */
1314int fcntl_getlease(struct file *filp)
1315{
1316 struct file_lock *fl;
1317 int type = F_UNLCK;
1318
1319 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001320 time_out_leases(filp->f_path.dentry->d_inode);
1321 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 fl = fl->fl_next) {
1323 if (fl->fl_file == filp) {
1324 type = fl->fl_type & ~F_INPROGRESS;
1325 break;
1326 }
1327 }
1328 unlock_kernel();
1329 return type;
1330}
1331
1332/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001333 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 * @filp: file pointer
1335 * @arg: type of lease to obtain
1336 * @flp: input - file_lock to use, output - file_lock inserted
1337 *
1338 * The (input) flp->fl_lmops->fl_break function is required
1339 * by break_lease().
1340 *
1341 * Called with kernel lock held.
1342 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001343int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001345 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001346 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 struct inode *inode = dentry->d_inode;
1348 int error, rdlease_count = 0, wrlease_count = 0;
1349
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001350 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1351 return -EACCES;
1352 if (!S_ISREG(inode->i_mode))
1353 return -EINVAL;
1354 error = security_file_lock(filp, arg);
1355 if (error)
1356 return error;
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 time_out_leases(inode);
1359
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001360 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001362 lease = *flp;
1363
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 error = -EAGAIN;
1365 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1366 goto out;
1367 if ((arg == F_WRLCK)
1368 && ((atomic_read(&dentry->d_count) > 1)
1369 || (atomic_read(&inode->i_count) > 1)))
1370 goto out;
1371
1372 /*
1373 * At this point, we know that if there is an exclusive
1374 * lease on this file, then we hold it on this filp
1375 * (otherwise our open of this file would have blocked).
1376 * And if we are trying to acquire an exclusive lease,
1377 * then the file is not open by anyone (including us)
1378 * except for this filp.
1379 */
1380 for (before = &inode->i_flock;
1381 ((fl = *before) != NULL) && IS_LEASE(fl);
1382 before = &fl->fl_next) {
1383 if (lease->fl_lmops->fl_mylease(fl, lease))
1384 my_before = before;
1385 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1386 /*
1387 * Someone is in the process of opening this
1388 * file for writing so we may not take an
1389 * exclusive lease on it.
1390 */
1391 wrlease_count++;
1392 else
1393 rdlease_count++;
1394 }
1395
1396 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1397 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1398 goto out;
1399
1400 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001401 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 error = lease->fl_lmops->fl_change(my_before, arg);
1403 goto out;
1404 }
1405
1406 error = 0;
1407 if (arg == F_UNLCK)
1408 goto out;
1409
1410 error = -EINVAL;
1411 if (!leases_enable)
1412 goto out;
1413
Trond Myklebust74361cb2006-08-14 08:54:48 -07001414 error = -ENOMEM;
1415 fl = locks_alloc_lock();
1416 if (fl == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 goto out;
1418
1419 locks_copy_lock(fl, lease);
1420
1421 locks_insert_lock(before, fl);
1422
1423 *flp = fl;
Trond Myklebust74361cb2006-08-14 08:54:48 -07001424 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425out:
1426 return error;
1427}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001428EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
1430 /**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001431 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 * @filp: file pointer
1433 * @arg: type of lease to obtain
1434 * @lease: file_lock to use
1435 *
1436 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001437 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1438 * break_lease will oops!
1439 *
1440 * This will call the filesystem's setlease file method, if
1441 * defined. Note that there is no getlease method; instead, the
1442 * filesystem setlease method should call back to setlease() to
1443 * add a lease to the inode's lease list, where fcntl_getlease() can
1444 * find it. Since fcntl_getlease() only reports whether the current
1445 * task holds a lease, a cluster filesystem need only do this for
1446 * leases held by processes on this node.
1447 *
1448 * There is also no break_lease method; filesystems that
1449 * handle their own leases shoud break leases themselves from the
1450 * filesystem's open, create, and (on truncate) setattr methods.
1451 *
1452 * Warning: the only current setlease methods exist only to disable
1453 * leases in certain cases. More vfs changes may be required to
1454 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 */
1456
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001457int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 int error;
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 lock_kernel();
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001462 if (filp->f_op && filp->f_op->setlease)
1463 error = filp->f_op->setlease(filp, arg, lease);
1464 else
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001465 error = generic_setlease(filp, arg, lease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 unlock_kernel();
1467
1468 return error;
1469}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001470EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
1472/**
1473 * fcntl_setlease - sets a lease on an open file
1474 * @fd: open file descriptor
1475 * @filp: file pointer
1476 * @arg: type of lease to obtain
1477 *
1478 * Call this fcntl to establish a lease on the file.
1479 * Note that you also need to call %F_SETSIG to
1480 * receive a signal when the lease is broken.
1481 */
1482int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1483{
1484 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001485 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 struct inode *inode = dentry->d_inode;
1487 int error;
1488
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 locks_init_lock(&fl);
1490 error = lease_init(filp, arg, &fl);
1491 if (error)
1492 return error;
1493
1494 lock_kernel();
1495
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001496 error = vfs_setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001497 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 goto out_unlock;
1499
1500 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1501 if (error < 0) {
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001502 /* remove lease just inserted by setlease */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 flp->fl_type = F_UNLCK | F_INPROGRESS;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001504 flp->fl_break_time = jiffies - 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 time_out_leases(inode);
1506 goto out_unlock;
1507 }
1508
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001509 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510out_unlock:
1511 unlock_kernel();
1512 return error;
1513}
1514
1515/**
1516 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1517 * @filp: The file to apply the lock to
1518 * @fl: The lock to be applied
1519 *
1520 * Add a FLOCK style lock to a file.
1521 */
1522int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1523{
1524 int error;
1525 might_sleep();
1526 for (;;) {
1527 error = flock_lock_file(filp, fl);
1528 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1529 break;
1530 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1531 if (!error)
1532 continue;
1533
1534 locks_delete_block(fl);
1535 break;
1536 }
1537 return error;
1538}
1539
1540EXPORT_SYMBOL(flock_lock_file_wait);
1541
1542/**
1543 * sys_flock: - flock() system call.
1544 * @fd: the file descriptor to lock.
1545 * @cmd: the type of lock to apply.
1546 *
1547 * Apply a %FL_FLOCK style lock to an open file descriptor.
1548 * The @cmd can be one of
1549 *
1550 * %LOCK_SH -- a shared lock.
1551 *
1552 * %LOCK_EX -- an exclusive lock.
1553 *
1554 * %LOCK_UN -- remove an existing lock.
1555 *
1556 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1557 *
1558 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1559 * processes read and write access respectively.
1560 */
1561asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1562{
1563 struct file *filp;
1564 struct file_lock *lock;
1565 int can_sleep, unlock;
1566 int error;
1567
1568 error = -EBADF;
1569 filp = fget(fd);
1570 if (!filp)
1571 goto out;
1572
1573 can_sleep = !(cmd & LOCK_NB);
1574 cmd &= ~LOCK_NB;
1575 unlock = (cmd == LOCK_UN);
1576
1577 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1578 goto out_putf;
1579
1580 error = flock_make_lock(filp, &lock, cmd);
1581 if (error)
1582 goto out_putf;
1583 if (can_sleep)
1584 lock->fl_flags |= FL_SLEEP;
1585
1586 error = security_file_lock(filp, cmd);
1587 if (error)
1588 goto out_free;
1589
1590 if (filp->f_op && filp->f_op->flock)
1591 error = filp->f_op->flock(filp,
1592 (can_sleep) ? F_SETLKW : F_SETLK,
1593 lock);
1594 else
1595 error = flock_lock_file_wait(filp, lock);
1596
1597 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001598 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
1600 out_putf:
1601 fput(filp);
1602 out:
1603 return error;
1604}
1605
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001606/**
1607 * vfs_test_lock - test file byte range lock
1608 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001609 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001610 *
1611 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1612 * setting conf->fl_type to something other than F_UNLCK.
1613 */
1614int vfs_test_lock(struct file *filp, struct file_lock *fl)
1615{
1616 if (filp->f_op && filp->f_op->lock)
1617 return filp->f_op->lock(filp, F_GETLK, fl);
1618 posix_test_lock(filp, fl);
1619 return 0;
1620}
1621EXPORT_SYMBOL_GPL(vfs_test_lock);
1622
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001623static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1624{
1625 flock->l_pid = fl->fl_pid;
1626#if BITS_PER_LONG == 32
1627 /*
1628 * Make sure we can represent the posix lock via
1629 * legacy 32bit flock.
1630 */
1631 if (fl->fl_start > OFFT_OFFSET_MAX)
1632 return -EOVERFLOW;
1633 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1634 return -EOVERFLOW;
1635#endif
1636 flock->l_start = fl->fl_start;
1637 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1638 fl->fl_end - fl->fl_start + 1;
1639 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001640 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001641 return 0;
1642}
1643
1644#if BITS_PER_LONG == 32
1645static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1646{
1647 flock->l_pid = fl->fl_pid;
1648 flock->l_start = fl->fl_start;
1649 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1650 fl->fl_end - fl->fl_start + 1;
1651 flock->l_whence = 0;
1652 flock->l_type = fl->fl_type;
1653}
1654#endif
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656/* Report the first existing lock that would conflict with l.
1657 * This implements the F_GETLK command of fcntl().
1658 */
1659int fcntl_getlk(struct file *filp, struct flock __user *l)
1660{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001661 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 struct flock flock;
1663 int error;
1664
1665 error = -EFAULT;
1666 if (copy_from_user(&flock, l, sizeof(flock)))
1667 goto out;
1668 error = -EINVAL;
1669 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1670 goto out;
1671
1672 error = flock_to_posix_lock(filp, &file_lock, &flock);
1673 if (error)
1674 goto out;
1675
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001676 error = vfs_test_lock(filp, &file_lock);
1677 if (error)
1678 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001680 flock.l_type = file_lock.fl_type;
1681 if (file_lock.fl_type != F_UNLCK) {
1682 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001683 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 }
1686 error = -EFAULT;
1687 if (!copy_to_user(l, &flock, sizeof(flock)))
1688 error = 0;
1689out:
1690 return error;
1691}
1692
Marc Eshel7723ec92007-01-18 15:08:55 -05001693/**
1694 * vfs_lock_file - file byte range lock
1695 * @filp: The file to apply the lock to
1696 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1697 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001698 * @conf: Place to return a copy of the conflicting lock, if found.
1699 *
1700 * A caller that doesn't care about the conflicting lock may pass NULL
1701 * as the final argument.
1702 *
1703 * If the filesystem defines a private ->lock() method, then @conf will
1704 * be left unchanged; so a caller that cares should initialize it to
1705 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001706 *
1707 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1708 * locks, the ->lock() interface may return asynchronously, before the lock has
1709 * been granted or denied by the underlying filesystem, if (and only if)
1710 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1711 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1712 * the request is for a blocking lock. When ->lock() does return asynchronously,
1713 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1714 * request completes.
1715 * If the request is for non-blocking lock the file system should return
1716 * -EINPROGRESS then try to get the lock and call the callback routine with
1717 * the result. If the request timed out the callback routine will return a
1718 * nonzero return code and the file system should release the lock. The file
1719 * system is also responsible to keep a corresponding posix lock when it
1720 * grants a lock so the VFS can find out which locks are locally held and do
1721 * the correct lock cleanup when required.
1722 * The underlying filesystem must not drop the kernel lock or call
1723 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1724 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001725 */
Marc Eshel150b3932007-01-18 16:15:35 -05001726int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001727{
1728 if (filp->f_op && filp->f_op->lock)
1729 return filp->f_op->lock(filp, cmd, fl);
1730 else
Marc Eshel150b3932007-01-18 16:15:35 -05001731 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001732}
1733EXPORT_SYMBOL_GPL(vfs_lock_file);
1734
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735/* Apply the lock described by l to an open file descriptor.
1736 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1737 */
Peter Staubachc2936212005-07-27 11:45:09 -07001738int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1739 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740{
1741 struct file_lock *file_lock = locks_alloc_lock();
1742 struct flock flock;
1743 struct inode *inode;
1744 int error;
1745
1746 if (file_lock == NULL)
1747 return -ENOLCK;
1748
1749 /*
1750 * This might block, so we do it before checking the inode.
1751 */
1752 error = -EFAULT;
1753 if (copy_from_user(&flock, l, sizeof(flock)))
1754 goto out;
1755
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001756 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
1758 /* Don't allow mandatory locks on files that may be memory mapped
1759 * and shared.
1760 */
1761 if (IS_MANDLOCK(inode) &&
1762 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1763 mapping_writably_mapped(filp->f_mapping)) {
1764 error = -EAGAIN;
1765 goto out;
1766 }
1767
Peter Staubachc2936212005-07-27 11:45:09 -07001768again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 error = flock_to_posix_lock(filp, file_lock, &flock);
1770 if (error)
1771 goto out;
1772 if (cmd == F_SETLKW) {
1773 file_lock->fl_flags |= FL_SLEEP;
1774 }
1775
1776 error = -EBADF;
1777 switch (flock.l_type) {
1778 case F_RDLCK:
1779 if (!(filp->f_mode & FMODE_READ))
1780 goto out;
1781 break;
1782 case F_WRLCK:
1783 if (!(filp->f_mode & FMODE_WRITE))
1784 goto out;
1785 break;
1786 case F_UNLCK:
1787 break;
1788 default:
1789 error = -EINVAL;
1790 goto out;
1791 }
1792
1793 error = security_file_lock(filp, file_lock->fl_type);
1794 if (error)
1795 goto out;
1796
Marc Eshel7723ec92007-01-18 15:08:55 -05001797 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001798 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001799 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001801 error = wait_event_interruptible(file_lock->fl_wait,
1802 !file_lock->fl_next);
1803 if (!error)
1804 continue;
1805
1806 locks_delete_block(file_lock);
1807 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 }
1809
Peter Staubachc2936212005-07-27 11:45:09 -07001810 /*
1811 * Attempt to detect a close/fcntl race and recover by
1812 * releasing the lock that was just acquired.
1813 */
1814 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1815 flock.l_type = F_UNLCK;
1816 goto again;
1817 }
1818
1819out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 locks_free_lock(file_lock);
1821 return error;
1822}
1823
1824#if BITS_PER_LONG == 32
1825/* Report the first existing lock that would conflict with l.
1826 * This implements the F_GETLK command of fcntl().
1827 */
1828int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1829{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001830 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 struct flock64 flock;
1832 int error;
1833
1834 error = -EFAULT;
1835 if (copy_from_user(&flock, l, sizeof(flock)))
1836 goto out;
1837 error = -EINVAL;
1838 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1839 goto out;
1840
1841 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1842 if (error)
1843 goto out;
1844
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001845 error = vfs_test_lock(filp, &file_lock);
1846 if (error)
1847 goto out;
1848
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001849 flock.l_type = file_lock.fl_type;
1850 if (file_lock.fl_type != F_UNLCK)
1851 posix_lock_to_flock64(&flock, &file_lock);
1852
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 error = -EFAULT;
1854 if (!copy_to_user(l, &flock, sizeof(flock)))
1855 error = 0;
1856
1857out:
1858 return error;
1859}
1860
1861/* Apply the lock described by l to an open file descriptor.
1862 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1863 */
Peter Staubachc2936212005-07-27 11:45:09 -07001864int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1865 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866{
1867 struct file_lock *file_lock = locks_alloc_lock();
1868 struct flock64 flock;
1869 struct inode *inode;
1870 int error;
1871
1872 if (file_lock == NULL)
1873 return -ENOLCK;
1874
1875 /*
1876 * This might block, so we do it before checking the inode.
1877 */
1878 error = -EFAULT;
1879 if (copy_from_user(&flock, l, sizeof(flock)))
1880 goto out;
1881
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001882 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
1884 /* Don't allow mandatory locks on files that may be memory mapped
1885 * and shared.
1886 */
1887 if (IS_MANDLOCK(inode) &&
1888 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1889 mapping_writably_mapped(filp->f_mapping)) {
1890 error = -EAGAIN;
1891 goto out;
1892 }
1893
Peter Staubachc2936212005-07-27 11:45:09 -07001894again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 error = flock64_to_posix_lock(filp, file_lock, &flock);
1896 if (error)
1897 goto out;
1898 if (cmd == F_SETLKW64) {
1899 file_lock->fl_flags |= FL_SLEEP;
1900 }
1901
1902 error = -EBADF;
1903 switch (flock.l_type) {
1904 case F_RDLCK:
1905 if (!(filp->f_mode & FMODE_READ))
1906 goto out;
1907 break;
1908 case F_WRLCK:
1909 if (!(filp->f_mode & FMODE_WRITE))
1910 goto out;
1911 break;
1912 case F_UNLCK:
1913 break;
1914 default:
1915 error = -EINVAL;
1916 goto out;
1917 }
1918
1919 error = security_file_lock(filp, file_lock->fl_type);
1920 if (error)
1921 goto out;
1922
Marc Eshel7723ec92007-01-18 15:08:55 -05001923 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001924 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001925 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001926 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001927 error = wait_event_interruptible(file_lock->fl_wait,
1928 !file_lock->fl_next);
1929 if (!error)
1930 continue;
1931
1932 locks_delete_block(file_lock);
1933 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 }
1935
Peter Staubachc2936212005-07-27 11:45:09 -07001936 /*
1937 * Attempt to detect a close/fcntl race and recover by
1938 * releasing the lock that was just acquired.
1939 */
1940 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1941 flock.l_type = F_UNLCK;
1942 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 }
1944
1945out:
1946 locks_free_lock(file_lock);
1947 return error;
1948}
1949#endif /* BITS_PER_LONG == 32 */
1950
1951/*
1952 * This function is called when the file is being removed
1953 * from the task's fd array. POSIX locks belonging to this task
1954 * are deleted at this time.
1955 */
1956void locks_remove_posix(struct file *filp, fl_owner_t owner)
1957{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001958 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959
1960 /*
1961 * If there are no locks held on this file, we don't need to call
1962 * posix_lock_file(). Another process could be setting a lock on this
1963 * file at the same time, but we wouldn't remove that lock anyway.
1964 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001965 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 return;
1967
1968 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001969 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970 lock.fl_start = 0;
1971 lock.fl_end = OFFSET_MAX;
1972 lock.fl_owner = owner;
1973 lock.fl_pid = current->tgid;
1974 lock.fl_file = filp;
1975 lock.fl_ops = NULL;
1976 lock.fl_lmops = NULL;
1977
Marc Eshel150b3932007-01-18 16:15:35 -05001978 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1981 lock.fl_ops->fl_release_private(&lock);
1982}
1983
1984EXPORT_SYMBOL(locks_remove_posix);
1985
1986/*
1987 * This function is called on the last close of an open file.
1988 */
1989void locks_remove_flock(struct file *filp)
1990{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001991 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 struct file_lock *fl;
1993 struct file_lock **before;
1994
1995 if (!inode->i_flock)
1996 return;
1997
1998 if (filp->f_op && filp->f_op->flock) {
1999 struct file_lock fl = {
2000 .fl_pid = current->tgid,
2001 .fl_file = filp,
2002 .fl_flags = FL_FLOCK,
2003 .fl_type = F_UNLCK,
2004 .fl_end = OFFSET_MAX,
2005 };
2006 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002007 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2008 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 }
2010
2011 lock_kernel();
2012 before = &inode->i_flock;
2013
2014 while ((fl = *before) != NULL) {
2015 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002016 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 locks_delete_lock(before);
2018 continue;
2019 }
2020 if (IS_LEASE(fl)) {
2021 lease_modify(before, F_UNLCK);
2022 continue;
2023 }
2024 /* What? */
2025 BUG();
2026 }
2027 before = &fl->fl_next;
2028 }
2029 unlock_kernel();
2030}
2031
2032/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 * posix_unblock_lock - stop waiting for a file lock
2034 * @filp: how the file was opened
2035 * @waiter: the lock which was waiting
2036 *
2037 * lockd needs to block waiting for locks.
2038 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002039int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2041{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002042 int status = 0;
2043
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002045 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002047 else
2048 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002049 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002050 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051}
2052
2053EXPORT_SYMBOL(posix_unblock_lock);
2054
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002055/**
2056 * vfs_cancel_lock - file byte range unblock lock
2057 * @filp: The file to apply the unblock to
2058 * @fl: The lock to be unblocked
2059 *
2060 * Used by lock managers to cancel blocked requests
2061 */
2062int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2063{
2064 if (filp->f_op && filp->f_op->lock)
2065 return filp->f_op->lock(filp, F_CANCELLK, fl);
2066 return 0;
2067}
2068
2069EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2070
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2072{
2073 struct inode *inode = NULL;
2074
2075 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002076 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077
2078 out += sprintf(out, "%d:%s ", id, pfx);
2079 if (IS_POSIX(fl)) {
2080 out += sprintf(out, "%6s %s ",
2081 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2082 (inode == NULL) ? "*NOINODE*" :
2083 (IS_MANDLOCK(inode) &&
2084 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2085 "MANDATORY" : "ADVISORY ");
2086 } else if (IS_FLOCK(fl)) {
2087 if (fl->fl_type & LOCK_MAND) {
2088 out += sprintf(out, "FLOCK MSNFS ");
2089 } else {
2090 out += sprintf(out, "FLOCK ADVISORY ");
2091 }
2092 } else if (IS_LEASE(fl)) {
2093 out += sprintf(out, "LEASE ");
2094 if (fl->fl_type & F_INPROGRESS)
2095 out += sprintf(out, "BREAKING ");
2096 else if (fl->fl_file)
2097 out += sprintf(out, "ACTIVE ");
2098 else
2099 out += sprintf(out, "BREAKER ");
2100 } else {
2101 out += sprintf(out, "UNKNOWN UNKNOWN ");
2102 }
2103 if (fl->fl_type & LOCK_MAND) {
2104 out += sprintf(out, "%s ",
2105 (fl->fl_type & LOCK_READ)
2106 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2107 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2108 } else {
2109 out += sprintf(out, "%s ",
2110 (fl->fl_type & F_INPROGRESS)
2111 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2112 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2113 }
2114 if (inode) {
2115#ifdef WE_CAN_BREAK_LSLK_NOW
2116 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2117 inode->i_sb->s_id, inode->i_ino);
2118#else
2119 /* userspace relies on this representation of dev_t ;-( */
2120 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2121 MAJOR(inode->i_sb->s_dev),
2122 MINOR(inode->i_sb->s_dev), inode->i_ino);
2123#endif
2124 } else {
2125 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2126 }
2127 if (IS_POSIX(fl)) {
2128 if (fl->fl_end == OFFSET_MAX)
2129 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2130 else
2131 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2132 fl->fl_end);
2133 } else {
2134 out += sprintf(out, "0 EOF\n");
2135 }
2136}
2137
2138static void move_lock_status(char **p, off_t* pos, off_t offset)
2139{
2140 int len;
2141 len = strlen(*p);
2142 if(*pos >= offset) {
2143 /* the complete line is valid */
2144 *p += len;
2145 *pos += len;
2146 return;
2147 }
2148 if(*pos+len > offset) {
2149 /* use the second part of the line */
2150 int i = offset-*pos;
2151 memmove(*p,*p+i,len-i);
2152 *p += len-i;
2153 *pos += len;
2154 return;
2155 }
2156 /* discard the complete line */
2157 *pos += len;
2158}
2159
2160/**
2161 * get_locks_status - reports lock usage in /proc/locks
2162 * @buffer: address in userspace to write into
2163 * @start: ?
2164 * @offset: how far we are through the buffer
2165 * @length: how much to read
2166 */
2167
2168int get_locks_status(char *buffer, char **start, off_t offset, int length)
2169{
2170 struct list_head *tmp;
2171 char *q = buffer;
2172 off_t pos = 0;
2173 int i = 0;
2174
2175 lock_kernel();
2176 list_for_each(tmp, &file_lock_list) {
2177 struct list_head *btmp;
2178 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2179 lock_get_status(q, fl, ++i, "");
2180 move_lock_status(&q, &pos, offset);
2181
2182 if(pos >= offset+length)
2183 goto done;
2184
2185 list_for_each(btmp, &fl->fl_block) {
2186 struct file_lock *bfl = list_entry(btmp,
2187 struct file_lock, fl_block);
2188 lock_get_status(q, bfl, i, " ->");
2189 move_lock_status(&q, &pos, offset);
2190
2191 if(pos >= offset+length)
2192 goto done;
2193 }
2194 }
2195done:
2196 unlock_kernel();
2197 *start = buffer;
2198 if(q-buffer < length)
2199 return (q-buffer);
2200 return length;
2201}
2202
2203/**
2204 * lock_may_read - checks that the region is free of locks
2205 * @inode: the inode that is being read
2206 * @start: the first byte to read
2207 * @len: the number of bytes to read
2208 *
2209 * Emulates Windows locking requirements. Whole-file
2210 * mandatory locks (share modes) can prohibit a read and
2211 * byte-range POSIX locks can prohibit a read if they overlap.
2212 *
2213 * N.B. this function is only ever called
2214 * from knfsd and ownership of locks is never checked.
2215 */
2216int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2217{
2218 struct file_lock *fl;
2219 int result = 1;
2220 lock_kernel();
2221 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2222 if (IS_POSIX(fl)) {
2223 if (fl->fl_type == F_RDLCK)
2224 continue;
2225 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2226 continue;
2227 } else if (IS_FLOCK(fl)) {
2228 if (!(fl->fl_type & LOCK_MAND))
2229 continue;
2230 if (fl->fl_type & LOCK_READ)
2231 continue;
2232 } else
2233 continue;
2234 result = 0;
2235 break;
2236 }
2237 unlock_kernel();
2238 return result;
2239}
2240
2241EXPORT_SYMBOL(lock_may_read);
2242
2243/**
2244 * lock_may_write - checks that the region is free of locks
2245 * @inode: the inode that is being written
2246 * @start: the first byte to write
2247 * @len: the number of bytes to write
2248 *
2249 * Emulates Windows locking requirements. Whole-file
2250 * mandatory locks (share modes) can prohibit a write and
2251 * byte-range POSIX locks can prohibit a write if they overlap.
2252 *
2253 * N.B. this function is only ever called
2254 * from knfsd and ownership of locks is never checked.
2255 */
2256int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2257{
2258 struct file_lock *fl;
2259 int result = 1;
2260 lock_kernel();
2261 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2262 if (IS_POSIX(fl)) {
2263 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2264 continue;
2265 } else if (IS_FLOCK(fl)) {
2266 if (!(fl->fl_type & LOCK_MAND))
2267 continue;
2268 if (fl->fl_type & LOCK_WRITE)
2269 continue;
2270 } else
2271 continue;
2272 result = 0;
2273 break;
2274 }
2275 unlock_kernel();
2276 return result;
2277}
2278
2279EXPORT_SYMBOL(lock_may_write);
2280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281static int __init filelock_init(void)
2282{
2283 filelock_cache = kmem_cache_create("file_lock_cache",
2284 sizeof(struct file_lock), 0, SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002285 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 return 0;
2287}
2288
2289core_initcall(filelock_init);