blob: 8b8388eca05e88bde827d97bc71193247c190e86 [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
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
J. Bruce Fields97855b42007-10-30 11:20:02 -0400699 *
700 * However, the failure of this assumption (also possible in the case of
701 * multiple tasks sharing the same open file table) also means there's no
702 * guarantee that the loop below will terminate. As a hack, we give up
703 * after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400705
706#define MAX_DEADLK_ITERATIONS 10
707
Adrian Bunkb0904e12006-06-23 02:05:13 -0700708static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 struct file_lock *block_fl)
710{
Matthias Kaehlcke094f2822007-10-02 11:21:34 -0700711 struct file_lock *fl;
J. Bruce Fields97855b42007-10-30 11:20:02 -0400712 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714next_task:
715 if (posix_same_owner(caller_fl, block_fl))
716 return 1;
Matthias Kaehlcke094f2822007-10-02 11:21:34 -0700717 list_for_each_entry(fl, &blocked_list, fl_link) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 if (posix_same_owner(fl, block_fl)) {
J. Bruce Fields97855b42007-10-30 11:20:02 -0400719 if (i++ > MAX_DEADLK_ITERATIONS)
720 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 fl = fl->fl_next;
722 block_fl = fl;
723 goto next_task;
724 }
725 }
726 return 0;
727}
728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400730 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400731 *
732 * Note that if called with an FL_EXISTS argument, the caller may determine
733 * whether or not a lock was successfully freed by testing the return
734 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800736static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800738 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800740 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 int error = 0;
742 int found = 0;
743
744 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400745 if (request->fl_flags & FL_ACCESS)
746 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400747
748 if (request->fl_type != F_UNLCK) {
749 error = -ENOMEM;
750 new_fl = locks_alloc_lock();
751 if (new_fl == NULL)
752 goto out;
753 error = 0;
754 }
755
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 for_each_lock(inode, before) {
757 struct file_lock *fl = *before;
758 if (IS_POSIX(fl))
759 break;
760 if (IS_LEASE(fl))
761 continue;
762 if (filp != fl->fl_file)
763 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800764 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 goto out;
766 found = 1;
767 locks_delete_lock(before);
768 break;
769 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Trond Myklebustf475ae92006-06-29 16:38:32 -0400771 if (request->fl_type == F_UNLCK) {
772 if ((request->fl_flags & FL_EXISTS) && !found)
773 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800774 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776
777 /*
778 * If a higher-priority process was blocked on the old file lock,
779 * give it the opportunity to lock the file.
780 */
781 if (found)
782 cond_resched();
783
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400784find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 for_each_lock(inode, before) {
786 struct file_lock *fl = *before;
787 if (IS_POSIX(fl))
788 break;
789 if (IS_LEASE(fl))
790 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800791 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 continue;
793 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800794 if (request->fl_flags & FL_SLEEP)
795 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 goto out;
797 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400798 if (request->fl_flags & FL_ACCESS)
799 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800800 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700801 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800802 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400803 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
805out:
806 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800807 if (new_fl)
808 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 return error;
810}
811
Marc Eshel150b3932007-01-18 16:15:35 -0500812static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
814 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700815 struct file_lock *new_fl = NULL;
816 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 struct file_lock *left = NULL;
818 struct file_lock *right = NULL;
819 struct file_lock **before;
820 int error, added = 0;
821
822 /*
823 * We may need two file_lock structures for this operation,
824 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700825 *
826 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700828 if (!(request->fl_flags & FL_ACCESS) &&
829 (request->fl_type != F_UNLCK ||
830 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
831 new_fl = locks_alloc_lock();
832 new_fl2 = locks_alloc_lock();
833 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
835 lock_kernel();
836 if (request->fl_type != F_UNLCK) {
837 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500838 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 if (!IS_POSIX(fl))
840 continue;
841 if (!posix_locks_conflict(request, fl))
842 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800843 if (conflock)
844 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 error = -EAGAIN;
846 if (!(request->fl_flags & FL_SLEEP))
847 goto out;
848 error = -EDEADLK;
849 if (posix_locks_deadlock(request, fl))
850 goto out;
851 error = -EAGAIN;
852 locks_insert_block(fl, request);
853 goto out;
854 }
855 }
856
857 /* If we're just looking for a conflict, we're done. */
858 error = 0;
859 if (request->fl_flags & FL_ACCESS)
860 goto out;
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 * Find the first old lock with the same owner as the new lock.
864 */
865
866 before = &inode->i_flock;
867
868 /* First skip locks owned by other processes. */
869 while ((fl = *before) && (!IS_POSIX(fl) ||
870 !posix_same_owner(request, fl))) {
871 before = &fl->fl_next;
872 }
873
874 /* Process locks with this owner. */
875 while ((fl = *before) && posix_same_owner(request, fl)) {
876 /* Detect adjacent or overlapping regions (if same lock type)
877 */
878 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700879 /* In all comparisons of start vs end, use
880 * "start - 1" rather than "end + 1". If end
881 * is OFFSET_MAX, end + 1 will become negative.
882 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 if (fl->fl_end < request->fl_start - 1)
884 goto next_lock;
885 /* If the next lock in the list has entirely bigger
886 * addresses than the new one, insert the lock here.
887 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700888 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 break;
890
891 /* If we come here, the new and old lock are of the
892 * same type and adjacent or overlapping. Make one
893 * lock yielding from the lower start address of both
894 * locks to the higher end address.
895 */
896 if (fl->fl_start > request->fl_start)
897 fl->fl_start = request->fl_start;
898 else
899 request->fl_start = fl->fl_start;
900 if (fl->fl_end < request->fl_end)
901 fl->fl_end = request->fl_end;
902 else
903 request->fl_end = fl->fl_end;
904 if (added) {
905 locks_delete_lock(before);
906 continue;
907 }
908 request = fl;
909 added = 1;
910 }
911 else {
912 /* Processing for different lock types is a bit
913 * more complex.
914 */
915 if (fl->fl_end < request->fl_start)
916 goto next_lock;
917 if (fl->fl_start > request->fl_end)
918 break;
919 if (request->fl_type == F_UNLCK)
920 added = 1;
921 if (fl->fl_start < request->fl_start)
922 left = fl;
923 /* If the next lock in the list has a higher end
924 * address than the new one, insert the new one here.
925 */
926 if (fl->fl_end > request->fl_end) {
927 right = fl;
928 break;
929 }
930 if (fl->fl_start >= request->fl_start) {
931 /* The new lock completely replaces an old
932 * one (This may happen several times).
933 */
934 if (added) {
935 locks_delete_lock(before);
936 continue;
937 }
938 /* Replace the old lock with the new one.
939 * Wake up anybody waiting for the old one,
940 * as the change in lock type might satisfy
941 * their needs.
942 */
943 locks_wake_up_blocks(fl);
944 fl->fl_start = request->fl_start;
945 fl->fl_end = request->fl_end;
946 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500947 locks_release_private(fl);
948 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 request = fl;
950 added = 1;
951 }
952 }
953 /* Go on to next lock.
954 */
955 next_lock:
956 before = &fl->fl_next;
957 }
958
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700959 /*
960 * The above code only modifies existing locks in case of
961 * merging or replacing. If new lock(s) need to be inserted
962 * all modifications are done bellow this, so it's safe yet to
963 * bail out.
964 */
965 error = -ENOLCK; /* "no luck" */
966 if (right && left == right && !new_fl2)
967 goto out;
968
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 error = 0;
970 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400971 if (request->fl_type == F_UNLCK) {
972 if (request->fl_flags & FL_EXISTS)
973 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400975 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700976
977 if (!new_fl) {
978 error = -ENOLCK;
979 goto out;
980 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 locks_copy_lock(new_fl, request);
982 locks_insert_lock(before, new_fl);
983 new_fl = NULL;
984 }
985 if (right) {
986 if (left == right) {
987 /* The new lock breaks the old one in two pieces,
988 * so we have to use the second new lock.
989 */
990 left = new_fl2;
991 new_fl2 = NULL;
992 locks_copy_lock(left, right);
993 locks_insert_lock(before, left);
994 }
995 right->fl_start = request->fl_end + 1;
996 locks_wake_up_blocks(right);
997 }
998 if (left) {
999 left->fl_end = request->fl_start - 1;
1000 locks_wake_up_blocks(left);
1001 }
1002 out:
1003 unlock_kernel();
1004 /*
1005 * Free any unused locks.
1006 */
1007 if (new_fl)
1008 locks_free_lock(new_fl);
1009 if (new_fl2)
1010 locks_free_lock(new_fl2);
1011 return error;
1012}
1013
1014/**
1015 * posix_lock_file - Apply a POSIX-style lock to a file
1016 * @filp: The file to apply the lock to
1017 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001018 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 *
1020 * Add a POSIX style lock to a file.
1021 * We merge adjacent & overlapping locks whenever possible.
1022 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001023 *
1024 * Note that if called with an FL_EXISTS argument, the caller may determine
1025 * whether or not a lock was successfully freed by testing the return
1026 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 */
Marc Eshel150b3932007-01-18 16:15:35 -05001028int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001029 struct file_lock *conflock)
1030{
Marc Eshel150b3932007-01-18 16:15:35 -05001031 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001032}
Marc Eshel150b3932007-01-18 16:15:35 -05001033EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
1035/**
1036 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1037 * @filp: The file to apply the lock to
1038 * @fl: The lock to be applied
1039 *
1040 * Add a POSIX style lock to a file.
1041 * We merge adjacent & overlapping locks whenever possible.
1042 * POSIX locks are sorted by owner task, then by starting address
1043 */
1044int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1045{
1046 int error;
1047 might_sleep ();
1048 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001049 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1051 break;
1052 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1053 if (!error)
1054 continue;
1055
1056 locks_delete_block(fl);
1057 break;
1058 }
1059 return error;
1060}
1061EXPORT_SYMBOL(posix_lock_file_wait);
1062
1063/**
1064 * locks_mandatory_locked - Check for an active lock
1065 * @inode: the file to check
1066 *
1067 * Searches the inode's list of locks to find any POSIX locks which conflict.
1068 * This function is called from locks_verify_locked() only.
1069 */
1070int locks_mandatory_locked(struct inode *inode)
1071{
1072 fl_owner_t owner = current->files;
1073 struct file_lock *fl;
1074
1075 /*
1076 * Search the lock list for this inode for any POSIX locks.
1077 */
1078 lock_kernel();
1079 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1080 if (!IS_POSIX(fl))
1081 continue;
1082 if (fl->fl_owner != owner)
1083 break;
1084 }
1085 unlock_kernel();
1086 return fl ? -EAGAIN : 0;
1087}
1088
1089/**
1090 * locks_mandatory_area - Check for a conflicting lock
1091 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1092 * for shared
1093 * @inode: the file to check
1094 * @filp: how the file was opened (if it was)
1095 * @offset: start of area to check
1096 * @count: length of area to check
1097 *
1098 * Searches the inode's list of locks to find any POSIX locks which conflict.
1099 * This function is called from rw_verify_area() and
1100 * locks_verify_truncate().
1101 */
1102int locks_mandatory_area(int read_write, struct inode *inode,
1103 struct file *filp, loff_t offset,
1104 size_t count)
1105{
1106 struct file_lock fl;
1107 int error;
1108
1109 locks_init_lock(&fl);
1110 fl.fl_owner = current->files;
1111 fl.fl_pid = current->tgid;
1112 fl.fl_file = filp;
1113 fl.fl_flags = FL_POSIX | FL_ACCESS;
1114 if (filp && !(filp->f_flags & O_NONBLOCK))
1115 fl.fl_flags |= FL_SLEEP;
1116 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1117 fl.fl_start = offset;
1118 fl.fl_end = offset + count - 1;
1119
1120 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001121 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 if (error != -EAGAIN)
1123 break;
1124 if (!(fl.fl_flags & FL_SLEEP))
1125 break;
1126 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1127 if (!error) {
1128 /*
1129 * If we've been sleeping someone might have
1130 * changed the permissions behind our back.
1131 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001132 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 continue;
1134 }
1135
1136 locks_delete_block(&fl);
1137 break;
1138 }
1139
1140 return error;
1141}
1142
1143EXPORT_SYMBOL(locks_mandatory_area);
1144
1145/* We already had a lease on this file; just change its type */
1146int lease_modify(struct file_lock **before, int arg)
1147{
1148 struct file_lock *fl = *before;
1149 int error = assign_type(fl, arg);
1150
1151 if (error)
1152 return error;
1153 locks_wake_up_blocks(fl);
1154 if (arg == F_UNLCK)
1155 locks_delete_lock(before);
1156 return 0;
1157}
1158
1159EXPORT_SYMBOL(lease_modify);
1160
1161static void time_out_leases(struct inode *inode)
1162{
1163 struct file_lock **before;
1164 struct file_lock *fl;
1165
1166 before = &inode->i_flock;
1167 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1168 if ((fl->fl_break_time == 0)
1169 || time_before(jiffies, fl->fl_break_time)) {
1170 before = &fl->fl_next;
1171 continue;
1172 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1174 if (fl == *before) /* lease_modify may have freed fl */
1175 before = &fl->fl_next;
1176 }
1177}
1178
1179/**
1180 * __break_lease - revoke all outstanding leases on file
1181 * @inode: the inode of the file to return
1182 * @mode: the open mode (read or write)
1183 *
david m. richter87250dd2007-05-09 16:10:27 -04001184 * break_lease (inlined for speed) has checked there already is at least
1185 * some kind of lock (maybe a lease) on this file. Leases are broken on
1186 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 * specified %O_NONBLOCK to your open().
1188 */
1189int __break_lease(struct inode *inode, unsigned int mode)
1190{
1191 int error = 0, future;
1192 struct file_lock *new_fl, *flock;
1193 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 unsigned long break_time;
1195 int i_have_this_lease = 0;
1196
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001197 new_fl = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 lock_kernel();
1200
1201 time_out_leases(inode);
1202
1203 flock = inode->i_flock;
1204 if ((flock == NULL) || !IS_LEASE(flock))
1205 goto out;
1206
1207 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1208 if (fl->fl_owner == current->files)
1209 i_have_this_lease = 1;
1210
1211 if (mode & FMODE_WRITE) {
1212 /* If we want write access, we have to revoke any lease. */
1213 future = F_UNLCK | F_INPROGRESS;
1214 } else if (flock->fl_type & F_INPROGRESS) {
1215 /* If the lease is already being broken, we just leave it */
1216 future = flock->fl_type;
1217 } else if (flock->fl_type & F_WRLCK) {
1218 /* Downgrade the exclusive lease to a read-only lease. */
1219 future = F_RDLCK | F_INPROGRESS;
1220 } else {
1221 /* the existing lease was read-only, so we can read too. */
1222 goto out;
1223 }
1224
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001225 if (IS_ERR(new_fl) && !i_have_this_lease
1226 && ((mode & O_NONBLOCK) == 0)) {
1227 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 goto out;
1229 }
1230
1231 break_time = 0;
1232 if (lease_break_time > 0) {
1233 break_time = jiffies + lease_break_time * HZ;
1234 if (break_time == 0)
1235 break_time++; /* so that 0 means no break time */
1236 }
1237
1238 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1239 if (fl->fl_type != future) {
1240 fl->fl_type = future;
1241 fl->fl_break_time = break_time;
1242 /* lease must have lmops break callback */
1243 fl->fl_lmops->fl_break(fl);
1244 }
1245 }
1246
1247 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1248 error = -EWOULDBLOCK;
1249 goto out;
1250 }
1251
1252restart:
1253 break_time = flock->fl_break_time;
1254 if (break_time != 0) {
1255 break_time -= jiffies;
1256 if (break_time == 0)
1257 break_time++;
1258 }
1259 error = locks_block_on_timeout(flock, new_fl, break_time);
1260 if (error >= 0) {
1261 if (error == 0)
1262 time_out_leases(inode);
1263 /* Wait for the next lease that has not been broken yet */
1264 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1265 flock = flock->fl_next) {
1266 if (flock->fl_type & F_INPROGRESS)
1267 goto restart;
1268 }
1269 error = 0;
1270 }
1271
1272out:
1273 unlock_kernel();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001274 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 locks_free_lock(new_fl);
1276 return error;
1277}
1278
1279EXPORT_SYMBOL(__break_lease);
1280
1281/**
1282 * lease_get_mtime
1283 * @inode: the inode
1284 * @time: pointer to a timespec which will contain the last modified time
1285 *
1286 * This is to force NFS clients to flush their caches for files with
1287 * exclusive leases. The justification is that if someone has an
1288 * exclusive lease, then they could be modifiying it.
1289 */
1290void lease_get_mtime(struct inode *inode, struct timespec *time)
1291{
1292 struct file_lock *flock = inode->i_flock;
1293 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1294 *time = current_fs_time(inode->i_sb);
1295 else
1296 *time = inode->i_mtime;
1297}
1298
1299EXPORT_SYMBOL(lease_get_mtime);
1300
1301/**
1302 * fcntl_getlease - Enquire what lease is currently active
1303 * @filp: the file
1304 *
1305 * The value returned by this function will be one of
1306 * (if no lease break is pending):
1307 *
1308 * %F_RDLCK to indicate a shared lease is held.
1309 *
1310 * %F_WRLCK to indicate an exclusive lease is held.
1311 *
1312 * %F_UNLCK to indicate no lease is held.
1313 *
1314 * (if a lease break is pending):
1315 *
1316 * %F_RDLCK to indicate an exclusive lease needs to be
1317 * changed to a shared lease (or removed).
1318 *
1319 * %F_UNLCK to indicate the lease needs to be removed.
1320 *
1321 * XXX: sfr & willy disagree over whether F_INPROGRESS
1322 * should be returned to userspace.
1323 */
1324int fcntl_getlease(struct file *filp)
1325{
1326 struct file_lock *fl;
1327 int type = F_UNLCK;
1328
1329 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001330 time_out_leases(filp->f_path.dentry->d_inode);
1331 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 fl = fl->fl_next) {
1333 if (fl->fl_file == filp) {
1334 type = fl->fl_type & ~F_INPROGRESS;
1335 break;
1336 }
1337 }
1338 unlock_kernel();
1339 return type;
1340}
1341
1342/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001343 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 * @filp: file pointer
1345 * @arg: type of lease to obtain
1346 * @flp: input - file_lock to use, output - file_lock inserted
1347 *
1348 * The (input) flp->fl_lmops->fl_break function is required
1349 * by break_lease().
1350 *
1351 * Called with kernel lock held.
1352 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001353int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001355 struct file_lock *fl, **before, **my_before = NULL, *lease;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001356 struct file_lock *new_fl = NULL;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001357 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 struct inode *inode = dentry->d_inode;
1359 int error, rdlease_count = 0, wrlease_count = 0;
1360
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001361 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1362 return -EACCES;
1363 if (!S_ISREG(inode->i_mode))
1364 return -EINVAL;
1365 error = security_file_lock(filp, arg);
1366 if (error)
1367 return error;
1368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 time_out_leases(inode);
1370
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001371 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001373 lease = *flp;
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 error = -EAGAIN;
1376 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1377 goto out;
1378 if ((arg == F_WRLCK)
1379 && ((atomic_read(&dentry->d_count) > 1)
1380 || (atomic_read(&inode->i_count) > 1)))
1381 goto out;
1382
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001383 error = -ENOMEM;
1384 new_fl = locks_alloc_lock();
1385 if (new_fl == NULL)
1386 goto out;
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 /*
1389 * At this point, we know that if there is an exclusive
1390 * lease on this file, then we hold it on this filp
1391 * (otherwise our open of this file would have blocked).
1392 * And if we are trying to acquire an exclusive lease,
1393 * then the file is not open by anyone (including us)
1394 * except for this filp.
1395 */
1396 for (before = &inode->i_flock;
1397 ((fl = *before) != NULL) && IS_LEASE(fl);
1398 before = &fl->fl_next) {
1399 if (lease->fl_lmops->fl_mylease(fl, lease))
1400 my_before = before;
1401 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1402 /*
1403 * Someone is in the process of opening this
1404 * file for writing so we may not take an
1405 * exclusive lease on it.
1406 */
1407 wrlease_count++;
1408 else
1409 rdlease_count++;
1410 }
1411
1412 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1413 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1414 goto out;
1415
1416 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001417 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 error = lease->fl_lmops->fl_change(my_before, arg);
1419 goto out;
1420 }
1421
1422 error = 0;
1423 if (arg == F_UNLCK)
1424 goto out;
1425
1426 error = -EINVAL;
1427 if (!leases_enable)
1428 goto out;
1429
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001430 locks_copy_lock(new_fl, lease);
1431 locks_insert_lock(before, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001433 *flp = new_fl;
1434 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436out:
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001437 if (new_fl != NULL)
1438 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 return error;
1440}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001441EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
1443 /**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001444 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 * @filp: file pointer
1446 * @arg: type of lease to obtain
1447 * @lease: file_lock to use
1448 *
1449 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001450 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1451 * break_lease will oops!
1452 *
1453 * This will call the filesystem's setlease file method, if
1454 * defined. Note that there is no getlease method; instead, the
1455 * filesystem setlease method should call back to setlease() to
1456 * add a lease to the inode's lease list, where fcntl_getlease() can
1457 * find it. Since fcntl_getlease() only reports whether the current
1458 * task holds a lease, a cluster filesystem need only do this for
1459 * leases held by processes on this node.
1460 *
1461 * There is also no break_lease method; filesystems that
1462 * handle their own leases shoud break leases themselves from the
1463 * filesystem's open, create, and (on truncate) setattr methods.
1464 *
1465 * Warning: the only current setlease methods exist only to disable
1466 * leases in certain cases. More vfs changes may be required to
1467 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 */
1469
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001470int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 int error;
1473
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 lock_kernel();
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001475 if (filp->f_op && filp->f_op->setlease)
1476 error = filp->f_op->setlease(filp, arg, lease);
1477 else
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001478 error = generic_setlease(filp, arg, lease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 unlock_kernel();
1480
1481 return error;
1482}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001483EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
1485/**
1486 * fcntl_setlease - sets a lease on an open file
1487 * @fd: open file descriptor
1488 * @filp: file pointer
1489 * @arg: type of lease to obtain
1490 *
1491 * Call this fcntl to establish a lease on the file.
1492 * Note that you also need to call %F_SETSIG to
1493 * receive a signal when the lease is broken.
1494 */
1495int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1496{
1497 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001498 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 struct inode *inode = dentry->d_inode;
1500 int error;
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 locks_init_lock(&fl);
1503 error = lease_init(filp, arg, &fl);
1504 if (error)
1505 return error;
1506
1507 lock_kernel();
1508
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001509 error = vfs_setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001510 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001511 goto out_unlock;
1512
1513 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1514 if (error < 0) {
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001515 /* remove lease just inserted by setlease */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 flp->fl_type = F_UNLCK | F_INPROGRESS;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001517 flp->fl_break_time = jiffies - 10;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 time_out_leases(inode);
1519 goto out_unlock;
1520 }
1521
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001522 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523out_unlock:
1524 unlock_kernel();
1525 return error;
1526}
1527
1528/**
1529 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1530 * @filp: The file to apply the lock to
1531 * @fl: The lock to be applied
1532 *
1533 * Add a FLOCK style lock to a file.
1534 */
1535int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1536{
1537 int error;
1538 might_sleep();
1539 for (;;) {
1540 error = flock_lock_file(filp, fl);
1541 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1542 break;
1543 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1544 if (!error)
1545 continue;
1546
1547 locks_delete_block(fl);
1548 break;
1549 }
1550 return error;
1551}
1552
1553EXPORT_SYMBOL(flock_lock_file_wait);
1554
1555/**
1556 * sys_flock: - flock() system call.
1557 * @fd: the file descriptor to lock.
1558 * @cmd: the type of lock to apply.
1559 *
1560 * Apply a %FL_FLOCK style lock to an open file descriptor.
1561 * The @cmd can be one of
1562 *
1563 * %LOCK_SH -- a shared lock.
1564 *
1565 * %LOCK_EX -- an exclusive lock.
1566 *
1567 * %LOCK_UN -- remove an existing lock.
1568 *
1569 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1570 *
1571 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1572 * processes read and write access respectively.
1573 */
1574asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1575{
1576 struct file *filp;
1577 struct file_lock *lock;
1578 int can_sleep, unlock;
1579 int error;
1580
1581 error = -EBADF;
1582 filp = fget(fd);
1583 if (!filp)
1584 goto out;
1585
1586 can_sleep = !(cmd & LOCK_NB);
1587 cmd &= ~LOCK_NB;
1588 unlock = (cmd == LOCK_UN);
1589
1590 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1591 goto out_putf;
1592
1593 error = flock_make_lock(filp, &lock, cmd);
1594 if (error)
1595 goto out_putf;
1596 if (can_sleep)
1597 lock->fl_flags |= FL_SLEEP;
1598
1599 error = security_file_lock(filp, cmd);
1600 if (error)
1601 goto out_free;
1602
1603 if (filp->f_op && filp->f_op->flock)
1604 error = filp->f_op->flock(filp,
1605 (can_sleep) ? F_SETLKW : F_SETLK,
1606 lock);
1607 else
1608 error = flock_lock_file_wait(filp, lock);
1609
1610 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001611 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
1613 out_putf:
1614 fput(filp);
1615 out:
1616 return error;
1617}
1618
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001619/**
1620 * vfs_test_lock - test file byte range lock
1621 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001622 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001623 *
1624 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1625 * setting conf->fl_type to something other than F_UNLCK.
1626 */
1627int vfs_test_lock(struct file *filp, struct file_lock *fl)
1628{
1629 if (filp->f_op && filp->f_op->lock)
1630 return filp->f_op->lock(filp, F_GETLK, fl);
1631 posix_test_lock(filp, fl);
1632 return 0;
1633}
1634EXPORT_SYMBOL_GPL(vfs_test_lock);
1635
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001636static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1637{
1638 flock->l_pid = fl->fl_pid;
1639#if BITS_PER_LONG == 32
1640 /*
1641 * Make sure we can represent the posix lock via
1642 * legacy 32bit flock.
1643 */
1644 if (fl->fl_start > OFFT_OFFSET_MAX)
1645 return -EOVERFLOW;
1646 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1647 return -EOVERFLOW;
1648#endif
1649 flock->l_start = fl->fl_start;
1650 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1651 fl->fl_end - fl->fl_start + 1;
1652 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001653 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001654 return 0;
1655}
1656
1657#if BITS_PER_LONG == 32
1658static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1659{
1660 flock->l_pid = fl->fl_pid;
1661 flock->l_start = fl->fl_start;
1662 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1663 fl->fl_end - fl->fl_start + 1;
1664 flock->l_whence = 0;
1665 flock->l_type = fl->fl_type;
1666}
1667#endif
1668
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669/* Report the first existing lock that would conflict with l.
1670 * This implements the F_GETLK command of fcntl().
1671 */
1672int fcntl_getlk(struct file *filp, struct flock __user *l)
1673{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001674 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 struct flock flock;
1676 int error;
1677
1678 error = -EFAULT;
1679 if (copy_from_user(&flock, l, sizeof(flock)))
1680 goto out;
1681 error = -EINVAL;
1682 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1683 goto out;
1684
1685 error = flock_to_posix_lock(filp, &file_lock, &flock);
1686 if (error)
1687 goto out;
1688
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001689 error = vfs_test_lock(filp, &file_lock);
1690 if (error)
1691 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001693 flock.l_type = file_lock.fl_type;
1694 if (file_lock.fl_type != F_UNLCK) {
1695 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001696 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 }
1699 error = -EFAULT;
1700 if (!copy_to_user(l, &flock, sizeof(flock)))
1701 error = 0;
1702out:
1703 return error;
1704}
1705
Marc Eshel7723ec92007-01-18 15:08:55 -05001706/**
1707 * vfs_lock_file - file byte range lock
1708 * @filp: The file to apply the lock to
1709 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1710 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001711 * @conf: Place to return a copy of the conflicting lock, if found.
1712 *
1713 * A caller that doesn't care about the conflicting lock may pass NULL
1714 * as the final argument.
1715 *
1716 * If the filesystem defines a private ->lock() method, then @conf will
1717 * be left unchanged; so a caller that cares should initialize it to
1718 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001719 *
1720 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1721 * locks, the ->lock() interface may return asynchronously, before the lock has
1722 * been granted or denied by the underlying filesystem, if (and only if)
1723 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1724 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1725 * the request is for a blocking lock. When ->lock() does return asynchronously,
1726 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1727 * request completes.
1728 * If the request is for non-blocking lock the file system should return
1729 * -EINPROGRESS then try to get the lock and call the callback routine with
1730 * the result. If the request timed out the callback routine will return a
1731 * nonzero return code and the file system should release the lock. The file
1732 * system is also responsible to keep a corresponding posix lock when it
1733 * grants a lock so the VFS can find out which locks are locally held and do
1734 * the correct lock cleanup when required.
1735 * The underlying filesystem must not drop the kernel lock or call
1736 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1737 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001738 */
Marc Eshel150b3932007-01-18 16:15:35 -05001739int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001740{
1741 if (filp->f_op && filp->f_op->lock)
1742 return filp->f_op->lock(filp, cmd, fl);
1743 else
Marc Eshel150b3932007-01-18 16:15:35 -05001744 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001745}
1746EXPORT_SYMBOL_GPL(vfs_lock_file);
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748/* Apply the lock described by l to an open file descriptor.
1749 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1750 */
Peter Staubachc2936212005-07-27 11:45:09 -07001751int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1752 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
1754 struct file_lock *file_lock = locks_alloc_lock();
1755 struct flock flock;
1756 struct inode *inode;
1757 int error;
1758
1759 if (file_lock == NULL)
1760 return -ENOLCK;
1761
1762 /*
1763 * This might block, so we do it before checking the inode.
1764 */
1765 error = -EFAULT;
1766 if (copy_from_user(&flock, l, sizeof(flock)))
1767 goto out;
1768
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001769 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770
1771 /* Don't allow mandatory locks on files that may be memory mapped
1772 * and shared.
1773 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001774 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 error = -EAGAIN;
1776 goto out;
1777 }
1778
Peter Staubachc2936212005-07-27 11:45:09 -07001779again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 error = flock_to_posix_lock(filp, file_lock, &flock);
1781 if (error)
1782 goto out;
1783 if (cmd == F_SETLKW) {
1784 file_lock->fl_flags |= FL_SLEEP;
1785 }
1786
1787 error = -EBADF;
1788 switch (flock.l_type) {
1789 case F_RDLCK:
1790 if (!(filp->f_mode & FMODE_READ))
1791 goto out;
1792 break;
1793 case F_WRLCK:
1794 if (!(filp->f_mode & FMODE_WRITE))
1795 goto out;
1796 break;
1797 case F_UNLCK:
1798 break;
1799 default:
1800 error = -EINVAL;
1801 goto out;
1802 }
1803
1804 error = security_file_lock(filp, file_lock->fl_type);
1805 if (error)
1806 goto out;
1807
Marc Eshel7723ec92007-01-18 15:08:55 -05001808 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001809 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001810 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001812 error = wait_event_interruptible(file_lock->fl_wait,
1813 !file_lock->fl_next);
1814 if (!error)
1815 continue;
1816
1817 locks_delete_block(file_lock);
1818 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 }
1820
Peter Staubachc2936212005-07-27 11:45:09 -07001821 /*
1822 * Attempt to detect a close/fcntl race and recover by
1823 * releasing the lock that was just acquired.
1824 */
1825 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1826 flock.l_type = F_UNLCK;
1827 goto again;
1828 }
1829
1830out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 locks_free_lock(file_lock);
1832 return error;
1833}
1834
1835#if BITS_PER_LONG == 32
1836/* Report the first existing lock that would conflict with l.
1837 * This implements the F_GETLK command of fcntl().
1838 */
1839int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1840{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001841 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 struct flock64 flock;
1843 int error;
1844
1845 error = -EFAULT;
1846 if (copy_from_user(&flock, l, sizeof(flock)))
1847 goto out;
1848 error = -EINVAL;
1849 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1850 goto out;
1851
1852 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1853 if (error)
1854 goto out;
1855
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001856 error = vfs_test_lock(filp, &file_lock);
1857 if (error)
1858 goto out;
1859
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001860 flock.l_type = file_lock.fl_type;
1861 if (file_lock.fl_type != F_UNLCK)
1862 posix_lock_to_flock64(&flock, &file_lock);
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 error = -EFAULT;
1865 if (!copy_to_user(l, &flock, sizeof(flock)))
1866 error = 0;
1867
1868out:
1869 return error;
1870}
1871
1872/* Apply the lock described by l to an open file descriptor.
1873 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1874 */
Peter Staubachc2936212005-07-27 11:45:09 -07001875int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1876 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877{
1878 struct file_lock *file_lock = locks_alloc_lock();
1879 struct flock64 flock;
1880 struct inode *inode;
1881 int error;
1882
1883 if (file_lock == NULL)
1884 return -ENOLCK;
1885
1886 /*
1887 * This might block, so we do it before checking the inode.
1888 */
1889 error = -EFAULT;
1890 if (copy_from_user(&flock, l, sizeof(flock)))
1891 goto out;
1892
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001893 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894
1895 /* Don't allow mandatory locks on files that may be memory mapped
1896 * and shared.
1897 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001898 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 error = -EAGAIN;
1900 goto out;
1901 }
1902
Peter Staubachc2936212005-07-27 11:45:09 -07001903again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 error = flock64_to_posix_lock(filp, file_lock, &flock);
1905 if (error)
1906 goto out;
1907 if (cmd == F_SETLKW64) {
1908 file_lock->fl_flags |= FL_SLEEP;
1909 }
1910
1911 error = -EBADF;
1912 switch (flock.l_type) {
1913 case F_RDLCK:
1914 if (!(filp->f_mode & FMODE_READ))
1915 goto out;
1916 break;
1917 case F_WRLCK:
1918 if (!(filp->f_mode & FMODE_WRITE))
1919 goto out;
1920 break;
1921 case F_UNLCK:
1922 break;
1923 default:
1924 error = -EINVAL;
1925 goto out;
1926 }
1927
1928 error = security_file_lock(filp, file_lock->fl_type);
1929 if (error)
1930 goto out;
1931
Marc Eshel7723ec92007-01-18 15:08:55 -05001932 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001933 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001934 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001935 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001936 error = wait_event_interruptible(file_lock->fl_wait,
1937 !file_lock->fl_next);
1938 if (!error)
1939 continue;
1940
1941 locks_delete_block(file_lock);
1942 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 }
1944
Peter Staubachc2936212005-07-27 11:45:09 -07001945 /*
1946 * Attempt to detect a close/fcntl race and recover by
1947 * releasing the lock that was just acquired.
1948 */
1949 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1950 flock.l_type = F_UNLCK;
1951 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 }
1953
1954out:
1955 locks_free_lock(file_lock);
1956 return error;
1957}
1958#endif /* BITS_PER_LONG == 32 */
1959
1960/*
1961 * This function is called when the file is being removed
1962 * from the task's fd array. POSIX locks belonging to this task
1963 * are deleted at this time.
1964 */
1965void locks_remove_posix(struct file *filp, fl_owner_t owner)
1966{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001967 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
1969 /*
1970 * If there are no locks held on this file, we don't need to call
1971 * posix_lock_file(). Another process could be setting a lock on this
1972 * file at the same time, but we wouldn't remove that lock anyway.
1973 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001974 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 return;
1976
1977 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001978 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 lock.fl_start = 0;
1980 lock.fl_end = OFFSET_MAX;
1981 lock.fl_owner = owner;
1982 lock.fl_pid = current->tgid;
1983 lock.fl_file = filp;
1984 lock.fl_ops = NULL;
1985 lock.fl_lmops = NULL;
1986
Marc Eshel150b3932007-01-18 16:15:35 -05001987 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1990 lock.fl_ops->fl_release_private(&lock);
1991}
1992
1993EXPORT_SYMBOL(locks_remove_posix);
1994
1995/*
1996 * This function is called on the last close of an open file.
1997 */
1998void locks_remove_flock(struct file *filp)
1999{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002000 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 struct file_lock *fl;
2002 struct file_lock **before;
2003
2004 if (!inode->i_flock)
2005 return;
2006
2007 if (filp->f_op && filp->f_op->flock) {
2008 struct file_lock fl = {
2009 .fl_pid = current->tgid,
2010 .fl_file = filp,
2011 .fl_flags = FL_FLOCK,
2012 .fl_type = F_UNLCK,
2013 .fl_end = OFFSET_MAX,
2014 };
2015 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002016 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2017 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 }
2019
2020 lock_kernel();
2021 before = &inode->i_flock;
2022
2023 while ((fl = *before) != NULL) {
2024 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002025 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026 locks_delete_lock(before);
2027 continue;
2028 }
2029 if (IS_LEASE(fl)) {
2030 lease_modify(before, F_UNLCK);
2031 continue;
2032 }
2033 /* What? */
2034 BUG();
2035 }
2036 before = &fl->fl_next;
2037 }
2038 unlock_kernel();
2039}
2040
2041/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 * posix_unblock_lock - stop waiting for a file lock
2043 * @filp: how the file was opened
2044 * @waiter: the lock which was waiting
2045 *
2046 * lockd needs to block waiting for locks.
2047 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002048int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2050{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002051 int status = 0;
2052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002054 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002056 else
2057 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002058 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002059 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060}
2061
2062EXPORT_SYMBOL(posix_unblock_lock);
2063
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002064/**
2065 * vfs_cancel_lock - file byte range unblock lock
2066 * @filp: The file to apply the unblock to
2067 * @fl: The lock to be unblocked
2068 *
2069 * Used by lock managers to cancel blocked requests
2070 */
2071int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2072{
2073 if (filp->f_op && filp->f_op->lock)
2074 return filp->f_op->lock(filp, F_CANCELLK, fl);
2075 return 0;
2076}
2077
2078EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2079
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002080#ifdef CONFIG_PROC_FS
2081#include <linux/seq_file.h>
2082
2083static void lock_get_status(struct seq_file *f, struct file_lock *fl,
2084 int id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085{
2086 struct inode *inode = NULL;
2087
2088 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002089 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002091 seq_printf(f, "%d:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002093 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2095 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002096 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 } else if (IS_FLOCK(fl)) {
2098 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002099 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002101 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 }
2103 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002104 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002106 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002108 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002110 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002112 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 }
2114 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002115 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 (fl->fl_type & LOCK_READ)
2117 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2118 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2119 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002120 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 (fl->fl_type & F_INPROGRESS)
2122 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2123 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2124 }
2125 if (inode) {
2126#ifdef WE_CAN_BREAK_LSLK_NOW
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002127 seq_printf(f, "%d %s:%ld ", fl->fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 inode->i_sb->s_id, inode->i_ino);
2129#else
2130 /* userspace relies on this representation of dev_t ;-( */
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002131 seq_printf(f, "%d %02x:%02x:%ld ", fl->fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 MAJOR(inode->i_sb->s_dev),
2133 MINOR(inode->i_sb->s_dev), inode->i_ino);
2134#endif
2135 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002136 seq_printf(f, "%d <none>:0 ", fl->fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 }
2138 if (IS_POSIX(fl)) {
2139 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002140 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002142 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002144 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 }
2146}
2147
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002148static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002150 struct file_lock *fl, *bfl;
2151
2152 fl = list_entry(v, struct file_lock, fl_link);
2153
2154 lock_get_status(f, fl, (long)f->private, "");
2155
2156 list_for_each_entry(bfl, &fl->fl_block, fl_block)
2157 lock_get_status(f, bfl, (long)f->private, " ->");
2158
2159 f->private++;
2160 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161}
2162
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002163static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 lock_kernel();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002166 f->private = (void *)1;
2167 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168}
2169
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002170static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2171{
2172 return seq_list_next(v, &file_lock_list, pos);
2173}
2174
2175static void locks_stop(struct seq_file *f, void *v)
2176{
2177 unlock_kernel();
2178}
2179
2180struct seq_operations locks_seq_operations = {
2181 .start = locks_start,
2182 .next = locks_next,
2183 .stop = locks_stop,
2184 .show = locks_show,
2185};
2186#endif
2187
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188/**
2189 * lock_may_read - checks that the region is free of locks
2190 * @inode: the inode that is being read
2191 * @start: the first byte to read
2192 * @len: the number of bytes to read
2193 *
2194 * Emulates Windows locking requirements. Whole-file
2195 * mandatory locks (share modes) can prohibit a read and
2196 * byte-range POSIX locks can prohibit a read if they overlap.
2197 *
2198 * N.B. this function is only ever called
2199 * from knfsd and ownership of locks is never checked.
2200 */
2201int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2202{
2203 struct file_lock *fl;
2204 int result = 1;
2205 lock_kernel();
2206 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2207 if (IS_POSIX(fl)) {
2208 if (fl->fl_type == F_RDLCK)
2209 continue;
2210 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2211 continue;
2212 } else if (IS_FLOCK(fl)) {
2213 if (!(fl->fl_type & LOCK_MAND))
2214 continue;
2215 if (fl->fl_type & LOCK_READ)
2216 continue;
2217 } else
2218 continue;
2219 result = 0;
2220 break;
2221 }
2222 unlock_kernel();
2223 return result;
2224}
2225
2226EXPORT_SYMBOL(lock_may_read);
2227
2228/**
2229 * lock_may_write - checks that the region is free of locks
2230 * @inode: the inode that is being written
2231 * @start: the first byte to write
2232 * @len: the number of bytes to write
2233 *
2234 * Emulates Windows locking requirements. Whole-file
2235 * mandatory locks (share modes) can prohibit a write and
2236 * byte-range POSIX locks can prohibit a write if they overlap.
2237 *
2238 * N.B. this function is only ever called
2239 * from knfsd and ownership of locks is never checked.
2240 */
2241int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2242{
2243 struct file_lock *fl;
2244 int result = 1;
2245 lock_kernel();
2246 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2247 if (IS_POSIX(fl)) {
2248 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2249 continue;
2250 } else if (IS_FLOCK(fl)) {
2251 if (!(fl->fl_type & LOCK_MAND))
2252 continue;
2253 if (fl->fl_type & LOCK_WRITE)
2254 continue;
2255 } else
2256 continue;
2257 result = 0;
2258 break;
2259 }
2260 unlock_kernel();
2261 return result;
2262}
2263
2264EXPORT_SYMBOL(lock_may_write);
2265
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266static int __init filelock_init(void)
2267{
2268 filelock_cache = kmem_cache_create("file_lock_cache",
2269 sizeof(struct file_lock), 0, SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002270 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 return 0;
2272}
2273
2274core_initcall(filelock_init);