blob: c3eecb895acf1346944d2dd3af217cf11c9c0f5a [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 Lameter4ba9b9d2007-10-16 23:25:51 -0700202static void init_once(struct kmem_cache *cache, void *foo)
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
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400686/*
687 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400689 * We attempt to detect deadlocks that are due purely to posix file
690 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400692 * We assume that a task can be waiting for at most one lock at a time.
693 * So for any acquired lock, the process holding that lock may be
694 * waiting on at most one other lock. That lock in turns may be held by
695 * someone waiting for at most one other lock. Given a requested lock
696 * caller_fl which is about to wait for a conflicting lock block_fl, we
697 * follow this chain of waiters to ensure we are not about to create a
698 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400699 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400700 * Since we do this before we ever put a process to sleep on a lock, we
701 * are ensured that there is never a cycle; that is what guarantees that
702 * the while() loop in posix_locks_deadlock() eventually completes.
703 *
704 * Note: the above assumption may not be true when handling lock
705 * requests from a broken NFS client. It may also fail in the presence
706 * of tasks (such as posix threads) sharing the same open file table.
707 *
708 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400710
711#define MAX_DEADLK_ITERATIONS 10
712
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400713/* Find a lock that the owner of the given block_fl is blocking on. */
714static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
715{
716 struct file_lock *fl;
717
718 list_for_each_entry(fl, &blocked_list, fl_link) {
719 if (posix_same_owner(fl, block_fl))
720 return fl->fl_next;
721 }
722 return NULL;
723}
724
Adrian Bunkb0904e12006-06-23 02:05:13 -0700725static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 struct file_lock *block_fl)
727{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400728 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400730 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
731 if (i++ > MAX_DEADLK_ITERATIONS)
732 return 0;
733 if (posix_same_owner(caller_fl, block_fl))
734 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 }
736 return 0;
737}
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400740 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400741 *
742 * Note that if called with an FL_EXISTS argument, the caller may determine
743 * whether or not a lock was successfully freed by testing the return
744 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800746static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800748 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800750 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 int error = 0;
752 int found = 0;
753
754 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400755 if (request->fl_flags & FL_ACCESS)
756 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400757
758 if (request->fl_type != F_UNLCK) {
759 error = -ENOMEM;
760 new_fl = locks_alloc_lock();
761 if (new_fl == NULL)
762 goto out;
763 error = 0;
764 }
765
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 for_each_lock(inode, before) {
767 struct file_lock *fl = *before;
768 if (IS_POSIX(fl))
769 break;
770 if (IS_LEASE(fl))
771 continue;
772 if (filp != fl->fl_file)
773 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800774 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 goto out;
776 found = 1;
777 locks_delete_lock(before);
778 break;
779 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
Trond Myklebustf475ae92006-06-29 16:38:32 -0400781 if (request->fl_type == F_UNLCK) {
782 if ((request->fl_flags & FL_EXISTS) && !found)
783 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800784 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787 /*
788 * If a higher-priority process was blocked on the old file lock,
789 * give it the opportunity to lock the file.
790 */
791 if (found)
792 cond_resched();
793
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400794find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 for_each_lock(inode, before) {
796 struct file_lock *fl = *before;
797 if (IS_POSIX(fl))
798 break;
799 if (IS_LEASE(fl))
800 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800801 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 continue;
803 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800804 if (request->fl_flags & FL_SLEEP)
805 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 goto out;
807 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400808 if (request->fl_flags & FL_ACCESS)
809 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800810 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700811 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800812 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400813 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815out:
816 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800817 if (new_fl)
818 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 return error;
820}
821
Marc Eshel150b3932007-01-18 16:15:35 -0500822static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823{
824 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700825 struct file_lock *new_fl = NULL;
826 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 struct file_lock *left = NULL;
828 struct file_lock *right = NULL;
829 struct file_lock **before;
830 int error, added = 0;
831
832 /*
833 * We may need two file_lock structures for this operation,
834 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700835 *
836 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700838 if (!(request->fl_flags & FL_ACCESS) &&
839 (request->fl_type != F_UNLCK ||
840 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
841 new_fl = locks_alloc_lock();
842 new_fl2 = locks_alloc_lock();
843 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
845 lock_kernel();
846 if (request->fl_type != F_UNLCK) {
847 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500848 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 if (!IS_POSIX(fl))
850 continue;
851 if (!posix_locks_conflict(request, fl))
852 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800853 if (conflock)
854 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 error = -EAGAIN;
856 if (!(request->fl_flags & FL_SLEEP))
857 goto out;
858 error = -EDEADLK;
859 if (posix_locks_deadlock(request, fl))
860 goto out;
861 error = -EAGAIN;
862 locks_insert_block(fl, request);
863 goto out;
864 }
865 }
866
867 /* If we're just looking for a conflict, we're done. */
868 error = 0;
869 if (request->fl_flags & FL_ACCESS)
870 goto out;
871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 * Find the first old lock with the same owner as the new lock.
874 */
875
876 before = &inode->i_flock;
877
878 /* First skip locks owned by other processes. */
879 while ((fl = *before) && (!IS_POSIX(fl) ||
880 !posix_same_owner(request, fl))) {
881 before = &fl->fl_next;
882 }
883
884 /* Process locks with this owner. */
885 while ((fl = *before) && posix_same_owner(request, fl)) {
886 /* Detect adjacent or overlapping regions (if same lock type)
887 */
888 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700889 /* In all comparisons of start vs end, use
890 * "start - 1" rather than "end + 1". If end
891 * is OFFSET_MAX, end + 1 will become negative.
892 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (fl->fl_end < request->fl_start - 1)
894 goto next_lock;
895 /* If the next lock in the list has entirely bigger
896 * addresses than the new one, insert the lock here.
897 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700898 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 break;
900
901 /* If we come here, the new and old lock are of the
902 * same type and adjacent or overlapping. Make one
903 * lock yielding from the lower start address of both
904 * locks to the higher end address.
905 */
906 if (fl->fl_start > request->fl_start)
907 fl->fl_start = request->fl_start;
908 else
909 request->fl_start = fl->fl_start;
910 if (fl->fl_end < request->fl_end)
911 fl->fl_end = request->fl_end;
912 else
913 request->fl_end = fl->fl_end;
914 if (added) {
915 locks_delete_lock(before);
916 continue;
917 }
918 request = fl;
919 added = 1;
920 }
921 else {
922 /* Processing for different lock types is a bit
923 * more complex.
924 */
925 if (fl->fl_end < request->fl_start)
926 goto next_lock;
927 if (fl->fl_start > request->fl_end)
928 break;
929 if (request->fl_type == F_UNLCK)
930 added = 1;
931 if (fl->fl_start < request->fl_start)
932 left = fl;
933 /* If the next lock in the list has a higher end
934 * address than the new one, insert the new one here.
935 */
936 if (fl->fl_end > request->fl_end) {
937 right = fl;
938 break;
939 }
940 if (fl->fl_start >= request->fl_start) {
941 /* The new lock completely replaces an old
942 * one (This may happen several times).
943 */
944 if (added) {
945 locks_delete_lock(before);
946 continue;
947 }
948 /* Replace the old lock with the new one.
949 * Wake up anybody waiting for the old one,
950 * as the change in lock type might satisfy
951 * their needs.
952 */
953 locks_wake_up_blocks(fl);
954 fl->fl_start = request->fl_start;
955 fl->fl_end = request->fl_end;
956 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500957 locks_release_private(fl);
958 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 request = fl;
960 added = 1;
961 }
962 }
963 /* Go on to next lock.
964 */
965 next_lock:
966 before = &fl->fl_next;
967 }
968
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700969 /*
970 * The above code only modifies existing locks in case of
971 * merging or replacing. If new lock(s) need to be inserted
972 * all modifications are done bellow this, so it's safe yet to
973 * bail out.
974 */
975 error = -ENOLCK; /* "no luck" */
976 if (right && left == right && !new_fl2)
977 goto out;
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 error = 0;
980 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400981 if (request->fl_type == F_UNLCK) {
982 if (request->fl_flags & FL_EXISTS)
983 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400985 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700986
987 if (!new_fl) {
988 error = -ENOLCK;
989 goto out;
990 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 locks_copy_lock(new_fl, request);
992 locks_insert_lock(before, new_fl);
993 new_fl = NULL;
994 }
995 if (right) {
996 if (left == right) {
997 /* The new lock breaks the old one in two pieces,
998 * so we have to use the second new lock.
999 */
1000 left = new_fl2;
1001 new_fl2 = NULL;
1002 locks_copy_lock(left, right);
1003 locks_insert_lock(before, left);
1004 }
1005 right->fl_start = request->fl_end + 1;
1006 locks_wake_up_blocks(right);
1007 }
1008 if (left) {
1009 left->fl_end = request->fl_start - 1;
1010 locks_wake_up_blocks(left);
1011 }
1012 out:
1013 unlock_kernel();
1014 /*
1015 * Free any unused locks.
1016 */
1017 if (new_fl)
1018 locks_free_lock(new_fl);
1019 if (new_fl2)
1020 locks_free_lock(new_fl2);
1021 return error;
1022}
1023
1024/**
1025 * posix_lock_file - Apply a POSIX-style lock to a file
1026 * @filp: The file to apply the lock to
1027 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001028 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 *
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
Trond Myklebustf475ae92006-06-29 16:38:32 -04001033 *
1034 * Note that if called with an FL_EXISTS argument, the caller may determine
1035 * whether or not a lock was successfully freed by testing the return
1036 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 */
Marc Eshel150b3932007-01-18 16:15:35 -05001038int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001039 struct file_lock *conflock)
1040{
Marc Eshel150b3932007-01-18 16:15:35 -05001041 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001042}
Marc Eshel150b3932007-01-18 16:15:35 -05001043EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044
1045/**
1046 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1047 * @filp: The file to apply the lock to
1048 * @fl: The lock to be applied
1049 *
1050 * Add a POSIX style lock to a file.
1051 * We merge adjacent & overlapping locks whenever possible.
1052 * POSIX locks are sorted by owner task, then by starting address
1053 */
1054int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1055{
1056 int error;
1057 might_sleep ();
1058 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001059 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1061 break;
1062 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1063 if (!error)
1064 continue;
1065
1066 locks_delete_block(fl);
1067 break;
1068 }
1069 return error;
1070}
1071EXPORT_SYMBOL(posix_lock_file_wait);
1072
1073/**
1074 * locks_mandatory_locked - Check for an active lock
1075 * @inode: the file to check
1076 *
1077 * Searches the inode's list of locks to find any POSIX locks which conflict.
1078 * This function is called from locks_verify_locked() only.
1079 */
1080int locks_mandatory_locked(struct inode *inode)
1081{
1082 fl_owner_t owner = current->files;
1083 struct file_lock *fl;
1084
1085 /*
1086 * Search the lock list for this inode for any POSIX locks.
1087 */
1088 lock_kernel();
1089 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1090 if (!IS_POSIX(fl))
1091 continue;
1092 if (fl->fl_owner != owner)
1093 break;
1094 }
1095 unlock_kernel();
1096 return fl ? -EAGAIN : 0;
1097}
1098
1099/**
1100 * locks_mandatory_area - Check for a conflicting lock
1101 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1102 * for shared
1103 * @inode: the file to check
1104 * @filp: how the file was opened (if it was)
1105 * @offset: start of area to check
1106 * @count: length of area to check
1107 *
1108 * Searches the inode's list of locks to find any POSIX locks which conflict.
1109 * This function is called from rw_verify_area() and
1110 * locks_verify_truncate().
1111 */
1112int locks_mandatory_area(int read_write, struct inode *inode,
1113 struct file *filp, loff_t offset,
1114 size_t count)
1115{
1116 struct file_lock fl;
1117 int error;
1118
1119 locks_init_lock(&fl);
1120 fl.fl_owner = current->files;
1121 fl.fl_pid = current->tgid;
1122 fl.fl_file = filp;
1123 fl.fl_flags = FL_POSIX | FL_ACCESS;
1124 if (filp && !(filp->f_flags & O_NONBLOCK))
1125 fl.fl_flags |= FL_SLEEP;
1126 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1127 fl.fl_start = offset;
1128 fl.fl_end = offset + count - 1;
1129
1130 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001131 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 if (error != -EAGAIN)
1133 break;
1134 if (!(fl.fl_flags & FL_SLEEP))
1135 break;
1136 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1137 if (!error) {
1138 /*
1139 * If we've been sleeping someone might have
1140 * changed the permissions behind our back.
1141 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001142 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 continue;
1144 }
1145
1146 locks_delete_block(&fl);
1147 break;
1148 }
1149
1150 return error;
1151}
1152
1153EXPORT_SYMBOL(locks_mandatory_area);
1154
1155/* We already had a lease on this file; just change its type */
1156int lease_modify(struct file_lock **before, int arg)
1157{
1158 struct file_lock *fl = *before;
1159 int error = assign_type(fl, arg);
1160
1161 if (error)
1162 return error;
1163 locks_wake_up_blocks(fl);
1164 if (arg == F_UNLCK)
1165 locks_delete_lock(before);
1166 return 0;
1167}
1168
1169EXPORT_SYMBOL(lease_modify);
1170
1171static void time_out_leases(struct inode *inode)
1172{
1173 struct file_lock **before;
1174 struct file_lock *fl;
1175
1176 before = &inode->i_flock;
1177 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1178 if ((fl->fl_break_time == 0)
1179 || time_before(jiffies, fl->fl_break_time)) {
1180 before = &fl->fl_next;
1181 continue;
1182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1184 if (fl == *before) /* lease_modify may have freed fl */
1185 before = &fl->fl_next;
1186 }
1187}
1188
1189/**
1190 * __break_lease - revoke all outstanding leases on file
1191 * @inode: the inode of the file to return
1192 * @mode: the open mode (read or write)
1193 *
david m. richter87250dd2007-05-09 16:10:27 -04001194 * break_lease (inlined for speed) has checked there already is at least
1195 * some kind of lock (maybe a lease) on this file. Leases are broken on
1196 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 * specified %O_NONBLOCK to your open().
1198 */
1199int __break_lease(struct inode *inode, unsigned int mode)
1200{
1201 int error = 0, future;
1202 struct file_lock *new_fl, *flock;
1203 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 unsigned long break_time;
1205 int i_have_this_lease = 0;
1206
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001207 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208
1209 lock_kernel();
1210
1211 time_out_leases(inode);
1212
1213 flock = inode->i_flock;
1214 if ((flock == NULL) || !IS_LEASE(flock))
1215 goto out;
1216
1217 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1218 if (fl->fl_owner == current->files)
1219 i_have_this_lease = 1;
1220
1221 if (mode & FMODE_WRITE) {
1222 /* If we want write access, we have to revoke any lease. */
1223 future = F_UNLCK | F_INPROGRESS;
1224 } else if (flock->fl_type & F_INPROGRESS) {
1225 /* If the lease is already being broken, we just leave it */
1226 future = flock->fl_type;
1227 } else if (flock->fl_type & F_WRLCK) {
1228 /* Downgrade the exclusive lease to a read-only lease. */
1229 future = F_RDLCK | F_INPROGRESS;
1230 } else {
1231 /* the existing lease was read-only, so we can read too. */
1232 goto out;
1233 }
1234
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001235 if (IS_ERR(new_fl) && !i_have_this_lease
1236 && ((mode & O_NONBLOCK) == 0)) {
1237 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 goto out;
1239 }
1240
1241 break_time = 0;
1242 if (lease_break_time > 0) {
1243 break_time = jiffies + lease_break_time * HZ;
1244 if (break_time == 0)
1245 break_time++; /* so that 0 means no break time */
1246 }
1247
1248 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1249 if (fl->fl_type != future) {
1250 fl->fl_type = future;
1251 fl->fl_break_time = break_time;
1252 /* lease must have lmops break callback */
1253 fl->fl_lmops->fl_break(fl);
1254 }
1255 }
1256
1257 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1258 error = -EWOULDBLOCK;
1259 goto out;
1260 }
1261
1262restart:
1263 break_time = flock->fl_break_time;
1264 if (break_time != 0) {
1265 break_time -= jiffies;
1266 if (break_time == 0)
1267 break_time++;
1268 }
1269 error = locks_block_on_timeout(flock, new_fl, break_time);
1270 if (error >= 0) {
1271 if (error == 0)
1272 time_out_leases(inode);
1273 /* Wait for the next lease that has not been broken yet */
1274 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1275 flock = flock->fl_next) {
1276 if (flock->fl_type & F_INPROGRESS)
1277 goto restart;
1278 }
1279 error = 0;
1280 }
1281
1282out:
1283 unlock_kernel();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001284 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 locks_free_lock(new_fl);
1286 return error;
1287}
1288
1289EXPORT_SYMBOL(__break_lease);
1290
1291/**
1292 * lease_get_mtime
1293 * @inode: the inode
1294 * @time: pointer to a timespec which will contain the last modified time
1295 *
1296 * This is to force NFS clients to flush their caches for files with
1297 * exclusive leases. The justification is that if someone has an
1298 * exclusive lease, then they could be modifiying it.
1299 */
1300void lease_get_mtime(struct inode *inode, struct timespec *time)
1301{
1302 struct file_lock *flock = inode->i_flock;
1303 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1304 *time = current_fs_time(inode->i_sb);
1305 else
1306 *time = inode->i_mtime;
1307}
1308
1309EXPORT_SYMBOL(lease_get_mtime);
1310
1311/**
1312 * fcntl_getlease - Enquire what lease is currently active
1313 * @filp: the file
1314 *
1315 * The value returned by this function will be one of
1316 * (if no lease break is pending):
1317 *
1318 * %F_RDLCK to indicate a shared lease is held.
1319 *
1320 * %F_WRLCK to indicate an exclusive lease is held.
1321 *
1322 * %F_UNLCK to indicate no lease is held.
1323 *
1324 * (if a lease break is pending):
1325 *
1326 * %F_RDLCK to indicate an exclusive lease needs to be
1327 * changed to a shared lease (or removed).
1328 *
1329 * %F_UNLCK to indicate the lease needs to be removed.
1330 *
1331 * XXX: sfr & willy disagree over whether F_INPROGRESS
1332 * should be returned to userspace.
1333 */
1334int fcntl_getlease(struct file *filp)
1335{
1336 struct file_lock *fl;
1337 int type = F_UNLCK;
1338
1339 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001340 time_out_leases(filp->f_path.dentry->d_inode);
1341 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 fl = fl->fl_next) {
1343 if (fl->fl_file == filp) {
1344 type = fl->fl_type & ~F_INPROGRESS;
1345 break;
1346 }
1347 }
1348 unlock_kernel();
1349 return type;
1350}
1351
1352/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001353 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 * @filp: file pointer
1355 * @arg: type of lease to obtain
1356 * @flp: input - file_lock to use, output - file_lock inserted
1357 *
1358 * The (input) flp->fl_lmops->fl_break function is required
1359 * by break_lease().
1360 *
1361 * Called with kernel lock held.
1362 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001363int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001365 struct file_lock *fl, **before, **my_before = NULL, *lease;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001366 struct file_lock *new_fl = NULL;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001367 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 struct inode *inode = dentry->d_inode;
1369 int error, rdlease_count = 0, wrlease_count = 0;
1370
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001371 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1372 return -EACCES;
1373 if (!S_ISREG(inode->i_mode))
1374 return -EINVAL;
1375 error = security_file_lock(filp, arg);
1376 if (error)
1377 return error;
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 time_out_leases(inode);
1380
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001381 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001383 lease = *flp;
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 error = -EAGAIN;
1386 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1387 goto out;
1388 if ((arg == F_WRLCK)
1389 && ((atomic_read(&dentry->d_count) > 1)
1390 || (atomic_read(&inode->i_count) > 1)))
1391 goto out;
1392
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001393 error = -ENOMEM;
1394 new_fl = locks_alloc_lock();
1395 if (new_fl == NULL)
1396 goto out;
1397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 /*
1399 * At this point, we know that if there is an exclusive
1400 * lease on this file, then we hold it on this filp
1401 * (otherwise our open of this file would have blocked).
1402 * And if we are trying to acquire an exclusive lease,
1403 * then the file is not open by anyone (including us)
1404 * except for this filp.
1405 */
1406 for (before = &inode->i_flock;
1407 ((fl = *before) != NULL) && IS_LEASE(fl);
1408 before = &fl->fl_next) {
1409 if (lease->fl_lmops->fl_mylease(fl, lease))
1410 my_before = before;
1411 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1412 /*
1413 * Someone is in the process of opening this
1414 * file for writing so we may not take an
1415 * exclusive lease on it.
1416 */
1417 wrlease_count++;
1418 else
1419 rdlease_count++;
1420 }
1421
1422 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1423 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1424 goto out;
1425
1426 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001427 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428 error = lease->fl_lmops->fl_change(my_before, arg);
1429 goto out;
1430 }
1431
1432 error = 0;
1433 if (arg == F_UNLCK)
1434 goto out;
1435
1436 error = -EINVAL;
1437 if (!leases_enable)
1438 goto out;
1439
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001440 locks_copy_lock(new_fl, lease);
1441 locks_insert_lock(before, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001443 *flp = new_fl;
1444 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446out:
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001447 if (new_fl != NULL)
1448 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 return error;
1450}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001451EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452
1453 /**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001454 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 * @filp: file pointer
1456 * @arg: type of lease to obtain
1457 * @lease: file_lock to use
1458 *
1459 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001460 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1461 * break_lease will oops!
1462 *
1463 * This will call the filesystem's setlease file method, if
1464 * defined. Note that there is no getlease method; instead, the
1465 * filesystem setlease method should call back to setlease() to
1466 * add a lease to the inode's lease list, where fcntl_getlease() can
1467 * find it. Since fcntl_getlease() only reports whether the current
1468 * task holds a lease, a cluster filesystem need only do this for
1469 * leases held by processes on this node.
1470 *
1471 * There is also no break_lease method; filesystems that
1472 * handle their own leases shoud break leases themselves from the
1473 * filesystem's open, create, and (on truncate) setattr methods.
1474 *
1475 * Warning: the only current setlease methods exist only to disable
1476 * leases in certain cases. More vfs changes may be required to
1477 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478 */
1479
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001480int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 int error;
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 lock_kernel();
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001485 if (filp->f_op && filp->f_op->setlease)
1486 error = filp->f_op->setlease(filp, arg, lease);
1487 else
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001488 error = generic_setlease(filp, arg, lease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 unlock_kernel();
1490
1491 return error;
1492}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001493EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
1495/**
1496 * fcntl_setlease - sets a lease on an open file
1497 * @fd: open file descriptor
1498 * @filp: file pointer
1499 * @arg: type of lease to obtain
1500 *
1501 * Call this fcntl to establish a lease on the file.
1502 * Note that you also need to call %F_SETSIG to
1503 * receive a signal when the lease is broken.
1504 */
1505int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1506{
1507 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001508 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 struct inode *inode = dentry->d_inode;
1510 int error;
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 locks_init_lock(&fl);
1513 error = lease_init(filp, arg, &fl);
1514 if (error)
1515 return error;
1516
1517 lock_kernel();
1518
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001519 error = vfs_setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001520 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 goto out_unlock;
1522
1523 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1524 if (error < 0) {
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001525 /* remove lease just inserted by setlease */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 flp->fl_type = F_UNLCK | F_INPROGRESS;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001527 flp->fl_break_time = jiffies - 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 time_out_leases(inode);
1529 goto out_unlock;
1530 }
1531
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001532 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533out_unlock:
1534 unlock_kernel();
1535 return error;
1536}
1537
1538/**
1539 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1540 * @filp: The file to apply the lock to
1541 * @fl: The lock to be applied
1542 *
1543 * Add a FLOCK style lock to a file.
1544 */
1545int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1546{
1547 int error;
1548 might_sleep();
1549 for (;;) {
1550 error = flock_lock_file(filp, fl);
1551 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1552 break;
1553 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1554 if (!error)
1555 continue;
1556
1557 locks_delete_block(fl);
1558 break;
1559 }
1560 return error;
1561}
1562
1563EXPORT_SYMBOL(flock_lock_file_wait);
1564
1565/**
1566 * sys_flock: - flock() system call.
1567 * @fd: the file descriptor to lock.
1568 * @cmd: the type of lock to apply.
1569 *
1570 * Apply a %FL_FLOCK style lock to an open file descriptor.
1571 * The @cmd can be one of
1572 *
1573 * %LOCK_SH -- a shared lock.
1574 *
1575 * %LOCK_EX -- an exclusive lock.
1576 *
1577 * %LOCK_UN -- remove an existing lock.
1578 *
1579 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1580 *
1581 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1582 * processes read and write access respectively.
1583 */
1584asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1585{
1586 struct file *filp;
1587 struct file_lock *lock;
1588 int can_sleep, unlock;
1589 int error;
1590
1591 error = -EBADF;
1592 filp = fget(fd);
1593 if (!filp)
1594 goto out;
1595
1596 can_sleep = !(cmd & LOCK_NB);
1597 cmd &= ~LOCK_NB;
1598 unlock = (cmd == LOCK_UN);
1599
1600 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1601 goto out_putf;
1602
1603 error = flock_make_lock(filp, &lock, cmd);
1604 if (error)
1605 goto out_putf;
1606 if (can_sleep)
1607 lock->fl_flags |= FL_SLEEP;
1608
1609 error = security_file_lock(filp, cmd);
1610 if (error)
1611 goto out_free;
1612
1613 if (filp->f_op && filp->f_op->flock)
1614 error = filp->f_op->flock(filp,
1615 (can_sleep) ? F_SETLKW : F_SETLK,
1616 lock);
1617 else
1618 error = flock_lock_file_wait(filp, lock);
1619
1620 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001621 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
1623 out_putf:
1624 fput(filp);
1625 out:
1626 return error;
1627}
1628
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001629/**
1630 * vfs_test_lock - test file byte range lock
1631 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001632 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001633 *
1634 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1635 * setting conf->fl_type to something other than F_UNLCK.
1636 */
1637int vfs_test_lock(struct file *filp, struct file_lock *fl)
1638{
1639 if (filp->f_op && filp->f_op->lock)
1640 return filp->f_op->lock(filp, F_GETLK, fl);
1641 posix_test_lock(filp, fl);
1642 return 0;
1643}
1644EXPORT_SYMBOL_GPL(vfs_test_lock);
1645
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001646static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1647{
1648 flock->l_pid = fl->fl_pid;
1649#if BITS_PER_LONG == 32
1650 /*
1651 * Make sure we can represent the posix lock via
1652 * legacy 32bit flock.
1653 */
1654 if (fl->fl_start > OFFT_OFFSET_MAX)
1655 return -EOVERFLOW;
1656 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1657 return -EOVERFLOW;
1658#endif
1659 flock->l_start = fl->fl_start;
1660 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1661 fl->fl_end - fl->fl_start + 1;
1662 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001663 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001664 return 0;
1665}
1666
1667#if BITS_PER_LONG == 32
1668static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1669{
1670 flock->l_pid = fl->fl_pid;
1671 flock->l_start = fl->fl_start;
1672 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1673 fl->fl_end - fl->fl_start + 1;
1674 flock->l_whence = 0;
1675 flock->l_type = fl->fl_type;
1676}
1677#endif
1678
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679/* Report the first existing lock that would conflict with l.
1680 * This implements the F_GETLK command of fcntl().
1681 */
1682int fcntl_getlk(struct file *filp, struct flock __user *l)
1683{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001684 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 struct flock flock;
1686 int error;
1687
1688 error = -EFAULT;
1689 if (copy_from_user(&flock, l, sizeof(flock)))
1690 goto out;
1691 error = -EINVAL;
1692 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1693 goto out;
1694
1695 error = flock_to_posix_lock(filp, &file_lock, &flock);
1696 if (error)
1697 goto out;
1698
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001699 error = vfs_test_lock(filp, &file_lock);
1700 if (error)
1701 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001703 flock.l_type = file_lock.fl_type;
1704 if (file_lock.fl_type != F_UNLCK) {
1705 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001706 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
1709 error = -EFAULT;
1710 if (!copy_to_user(l, &flock, sizeof(flock)))
1711 error = 0;
1712out:
1713 return error;
1714}
1715
Marc Eshel7723ec92007-01-18 15:08:55 -05001716/**
1717 * vfs_lock_file - file byte range lock
1718 * @filp: The file to apply the lock to
1719 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1720 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001721 * @conf: Place to return a copy of the conflicting lock, if found.
1722 *
1723 * A caller that doesn't care about the conflicting lock may pass NULL
1724 * as the final argument.
1725 *
1726 * If the filesystem defines a private ->lock() method, then @conf will
1727 * be left unchanged; so a caller that cares should initialize it to
1728 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001729 *
1730 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1731 * locks, the ->lock() interface may return asynchronously, before the lock has
1732 * been granted or denied by the underlying filesystem, if (and only if)
1733 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1734 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1735 * the request is for a blocking lock. When ->lock() does return asynchronously,
1736 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1737 * request completes.
1738 * If the request is for non-blocking lock the file system should return
1739 * -EINPROGRESS then try to get the lock and call the callback routine with
1740 * the result. If the request timed out the callback routine will return a
1741 * nonzero return code and the file system should release the lock. The file
1742 * system is also responsible to keep a corresponding posix lock when it
1743 * grants a lock so the VFS can find out which locks are locally held and do
1744 * the correct lock cleanup when required.
1745 * The underlying filesystem must not drop the kernel lock or call
1746 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1747 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001748 */
Marc Eshel150b3932007-01-18 16:15:35 -05001749int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001750{
1751 if (filp->f_op && filp->f_op->lock)
1752 return filp->f_op->lock(filp, cmd, fl);
1753 else
Marc Eshel150b3932007-01-18 16:15:35 -05001754 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001755}
1756EXPORT_SYMBOL_GPL(vfs_lock_file);
1757
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758/* Apply the lock described by l to an open file descriptor.
1759 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1760 */
Peter Staubachc2936212005-07-27 11:45:09 -07001761int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1762 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763{
1764 struct file_lock *file_lock = locks_alloc_lock();
1765 struct flock flock;
1766 struct inode *inode;
1767 int error;
1768
1769 if (file_lock == NULL)
1770 return -ENOLCK;
1771
1772 /*
1773 * This might block, so we do it before checking the inode.
1774 */
1775 error = -EFAULT;
1776 if (copy_from_user(&flock, l, sizeof(flock)))
1777 goto out;
1778
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001779 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
1781 /* Don't allow mandatory locks on files that may be memory mapped
1782 * and shared.
1783 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001784 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 error = -EAGAIN;
1786 goto out;
1787 }
1788
Peter Staubachc2936212005-07-27 11:45:09 -07001789again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 error = flock_to_posix_lock(filp, file_lock, &flock);
1791 if (error)
1792 goto out;
1793 if (cmd == F_SETLKW) {
1794 file_lock->fl_flags |= FL_SLEEP;
1795 }
1796
1797 error = -EBADF;
1798 switch (flock.l_type) {
1799 case F_RDLCK:
1800 if (!(filp->f_mode & FMODE_READ))
1801 goto out;
1802 break;
1803 case F_WRLCK:
1804 if (!(filp->f_mode & FMODE_WRITE))
1805 goto out;
1806 break;
1807 case F_UNLCK:
1808 break;
1809 default:
1810 error = -EINVAL;
1811 goto out;
1812 }
1813
1814 error = security_file_lock(filp, file_lock->fl_type);
1815 if (error)
1816 goto out;
1817
Marc Eshel7723ec92007-01-18 15:08:55 -05001818 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001819 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001820 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001822 error = wait_event_interruptible(file_lock->fl_wait,
1823 !file_lock->fl_next);
1824 if (!error)
1825 continue;
1826
1827 locks_delete_block(file_lock);
1828 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 }
1830
Peter Staubachc2936212005-07-27 11:45:09 -07001831 /*
1832 * Attempt to detect a close/fcntl race and recover by
1833 * releasing the lock that was just acquired.
1834 */
1835 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1836 flock.l_type = F_UNLCK;
1837 goto again;
1838 }
1839
1840out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 locks_free_lock(file_lock);
1842 return error;
1843}
1844
1845#if BITS_PER_LONG == 32
1846/* Report the first existing lock that would conflict with l.
1847 * This implements the F_GETLK command of fcntl().
1848 */
1849int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1850{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001851 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 struct flock64 flock;
1853 int error;
1854
1855 error = -EFAULT;
1856 if (copy_from_user(&flock, l, sizeof(flock)))
1857 goto out;
1858 error = -EINVAL;
1859 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1860 goto out;
1861
1862 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1863 if (error)
1864 goto out;
1865
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001866 error = vfs_test_lock(filp, &file_lock);
1867 if (error)
1868 goto out;
1869
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001870 flock.l_type = file_lock.fl_type;
1871 if (file_lock.fl_type != F_UNLCK)
1872 posix_lock_to_flock64(&flock, &file_lock);
1873
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 error = -EFAULT;
1875 if (!copy_to_user(l, &flock, sizeof(flock)))
1876 error = 0;
1877
1878out:
1879 return error;
1880}
1881
1882/* Apply the lock described by l to an open file descriptor.
1883 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1884 */
Peter Staubachc2936212005-07-27 11:45:09 -07001885int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1886 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887{
1888 struct file_lock *file_lock = locks_alloc_lock();
1889 struct flock64 flock;
1890 struct inode *inode;
1891 int error;
1892
1893 if (file_lock == NULL)
1894 return -ENOLCK;
1895
1896 /*
1897 * This might block, so we do it before checking the inode.
1898 */
1899 error = -EFAULT;
1900 if (copy_from_user(&flock, l, sizeof(flock)))
1901 goto out;
1902
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001903 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904
1905 /* Don't allow mandatory locks on files that may be memory mapped
1906 * and shared.
1907 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001908 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 error = -EAGAIN;
1910 goto out;
1911 }
1912
Peter Staubachc2936212005-07-27 11:45:09 -07001913again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 error = flock64_to_posix_lock(filp, file_lock, &flock);
1915 if (error)
1916 goto out;
1917 if (cmd == F_SETLKW64) {
1918 file_lock->fl_flags |= FL_SLEEP;
1919 }
1920
1921 error = -EBADF;
1922 switch (flock.l_type) {
1923 case F_RDLCK:
1924 if (!(filp->f_mode & FMODE_READ))
1925 goto out;
1926 break;
1927 case F_WRLCK:
1928 if (!(filp->f_mode & FMODE_WRITE))
1929 goto out;
1930 break;
1931 case F_UNLCK:
1932 break;
1933 default:
1934 error = -EINVAL;
1935 goto out;
1936 }
1937
1938 error = security_file_lock(filp, file_lock->fl_type);
1939 if (error)
1940 goto out;
1941
Marc Eshel7723ec92007-01-18 15:08:55 -05001942 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001943 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001944 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001945 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001946 error = wait_event_interruptible(file_lock->fl_wait,
1947 !file_lock->fl_next);
1948 if (!error)
1949 continue;
1950
1951 locks_delete_block(file_lock);
1952 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 }
1954
Peter Staubachc2936212005-07-27 11:45:09 -07001955 /*
1956 * Attempt to detect a close/fcntl race and recover by
1957 * releasing the lock that was just acquired.
1958 */
1959 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1960 flock.l_type = F_UNLCK;
1961 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 }
1963
1964out:
1965 locks_free_lock(file_lock);
1966 return error;
1967}
1968#endif /* BITS_PER_LONG == 32 */
1969
1970/*
1971 * This function is called when the file is being removed
1972 * from the task's fd array. POSIX locks belonging to this task
1973 * are deleted at this time.
1974 */
1975void locks_remove_posix(struct file *filp, fl_owner_t owner)
1976{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001977 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 /*
1980 * If there are no locks held on this file, we don't need to call
1981 * posix_lock_file(). Another process could be setting a lock on this
1982 * file at the same time, but we wouldn't remove that lock anyway.
1983 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001984 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 return;
1986
1987 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001988 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 lock.fl_start = 0;
1990 lock.fl_end = OFFSET_MAX;
1991 lock.fl_owner = owner;
1992 lock.fl_pid = current->tgid;
1993 lock.fl_file = filp;
1994 lock.fl_ops = NULL;
1995 lock.fl_lmops = NULL;
1996
Marc Eshel150b3932007-01-18 16:15:35 -05001997 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2000 lock.fl_ops->fl_release_private(&lock);
2001}
2002
2003EXPORT_SYMBOL(locks_remove_posix);
2004
2005/*
2006 * This function is called on the last close of an open file.
2007 */
2008void locks_remove_flock(struct file *filp)
2009{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002010 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011 struct file_lock *fl;
2012 struct file_lock **before;
2013
2014 if (!inode->i_flock)
2015 return;
2016
2017 if (filp->f_op && filp->f_op->flock) {
2018 struct file_lock fl = {
2019 .fl_pid = current->tgid,
2020 .fl_file = filp,
2021 .fl_flags = FL_FLOCK,
2022 .fl_type = F_UNLCK,
2023 .fl_end = OFFSET_MAX,
2024 };
2025 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002026 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2027 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 }
2029
2030 lock_kernel();
2031 before = &inode->i_flock;
2032
2033 while ((fl = *before) != NULL) {
2034 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002035 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 locks_delete_lock(before);
2037 continue;
2038 }
2039 if (IS_LEASE(fl)) {
2040 lease_modify(before, F_UNLCK);
2041 continue;
2042 }
2043 /* What? */
2044 BUG();
2045 }
2046 before = &fl->fl_next;
2047 }
2048 unlock_kernel();
2049}
2050
2051/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 * posix_unblock_lock - stop waiting for a file lock
2053 * @filp: how the file was opened
2054 * @waiter: the lock which was waiting
2055 *
2056 * lockd needs to block waiting for locks.
2057 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002058int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2060{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002061 int status = 0;
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002064 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002066 else
2067 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002068 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002069 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070}
2071
2072EXPORT_SYMBOL(posix_unblock_lock);
2073
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002074/**
2075 * vfs_cancel_lock - file byte range unblock lock
2076 * @filp: The file to apply the unblock to
2077 * @fl: The lock to be unblocked
2078 *
2079 * Used by lock managers to cancel blocked requests
2080 */
2081int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2082{
2083 if (filp->f_op && filp->f_op->lock)
2084 return filp->f_op->lock(filp, F_CANCELLK, fl);
2085 return 0;
2086}
2087
2088EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2089
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002090#ifdef CONFIG_PROC_FS
2091#include <linux/seq_file.h>
2092
2093static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2094 int id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095{
2096 struct inode *inode = NULL;
2097
2098 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002099 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002101 seq_printf(f, "%d:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002103 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2105 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002106 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 } else if (IS_FLOCK(fl)) {
2108 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002109 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002111 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 }
2113 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002114 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002116 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002118 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002120 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002122 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 }
2124 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002125 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 (fl->fl_type & LOCK_READ)
2127 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2128 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2129 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002130 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 (fl->fl_type & F_INPROGRESS)
2132 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2133 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2134 }
2135 if (inode) {
2136#ifdef WE_CAN_BREAK_LSLK_NOW
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002137 seq_printf(f, "%d %s:%ld ", fl->fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 inode->i_sb->s_id, inode->i_ino);
2139#else
2140 /* userspace relies on this representation of dev_t ;-( */
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002141 seq_printf(f, "%d %02x:%02x:%ld ", fl->fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 MAJOR(inode->i_sb->s_dev),
2143 MINOR(inode->i_sb->s_dev), inode->i_ino);
2144#endif
2145 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002146 seq_printf(f, "%d <none>:0 ", fl->fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 }
2148 if (IS_POSIX(fl)) {
2149 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002150 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002152 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002154 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 }
2156}
2157
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002158static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002160 struct file_lock *fl, *bfl;
2161
2162 fl = list_entry(v, struct file_lock, fl_link);
2163
2164 lock_get_status(f, fl, (long)f->private, "");
2165
2166 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2167 lock_get_status(f, bfl, (long)f->private, " ->");
2168
2169 f->private++;
2170 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171}
2172
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002173static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 lock_kernel();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002176 f->private = (void *)1;
2177 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178}
2179
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002180static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2181{
2182 return seq_list_next(v, &file_lock_list, pos);
2183}
2184
2185static void locks_stop(struct seq_file *f, void *v)
2186{
2187 unlock_kernel();
2188}
2189
2190struct seq_operations locks_seq_operations = {
2191 .start = locks_start,
2192 .next = locks_next,
2193 .stop = locks_stop,
2194 .show = locks_show,
2195};
2196#endif
2197
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198/**
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,
Paul Mundt20c2df82007-07-20 10:11:58 +09002280 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 return 0;
2282}
2283
2284core_initcall(filelock_init);