blob: 924e4a7fb1d5e449ccb534581d9abb55568195dc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
119#include <linux/fs.h>
120#include <linux/init.h>
121#include <linux/module.h>
122#include <linux/security.h>
123#include <linux/slab.h>
124#include <linux/smp_lock.h>
125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129#include <asm/semaphore.h>
130#include <asm/uaccess.h>
131
132#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
135
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
144
Christoph Lametere18b8902006-12-06 20:33:20 -0800145static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147/* Allocate an empty lock structure. */
148static struct file_lock *locks_alloc_lock(void)
149{
Christoph Lametere94b1762006-12-06 20:33:17 -0800150 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Trond Myklebust47831f32006-03-20 13:44:05 -0500153static void locks_release_private(struct file_lock *fl)
154{
155 if (fl->fl_ops) {
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
158 fl->fl_ops = NULL;
159 }
160 if (fl->fl_lmops) {
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
163 fl->fl_lmops = NULL;
164 }
165
166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800169static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Trond Myklebust47831f32006-03-20 13:44:05 -0500175 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(filelock_cache, fl);
177}
178
179void locks_init_lock(struct file_lock *fl)
180{
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
184 fl->fl_next = NULL;
185 fl->fl_fasync = NULL;
186 fl->fl_owner = NULL;
187 fl->fl_pid = 0;
188 fl->fl_file = NULL;
189 fl->fl_flags = 0;
190 fl->fl_type = 0;
191 fl->fl_start = fl->fl_end = 0;
192 fl->fl_ops = NULL;
193 fl->fl_lmops = NULL;
194}
195
196EXPORT_SYMBOL(locks_init_lock);
197
198/*
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
201 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800202static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 struct file_lock *lock = (struct file_lock *) foo;
205
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 locks_init_lock(lock);
207}
208
Trond Myklebust47831f32006-03-20 13:44:05 -0500209static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
210{
211 if (fl->fl_ops) {
212 if (fl->fl_ops->fl_copy_lock)
213 fl->fl_ops->fl_copy_lock(new, fl);
214 new->fl_ops = fl->fl_ops;
215 }
216 if (fl->fl_lmops) {
217 if (fl->fl_lmops->fl_copy_lock)
218 fl->fl_lmops->fl_copy_lock(new, fl);
219 new->fl_lmops = fl->fl_lmops;
220 }
221}
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223/*
224 * Initialize a new lock from an existing file_lock structure.
225 */
Trond Myklebust09969052006-03-20 13:44:38 -0500226static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
228 new->fl_owner = fl->fl_owner;
229 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500230 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 new->fl_flags = fl->fl_flags;
232 new->fl_type = fl->fl_type;
233 new->fl_start = fl->fl_start;
234 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500235 new->fl_ops = NULL;
236 new->fl_lmops = NULL;
237}
238
239void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
240{
241 locks_release_private(new);
242
243 __locks_copy_lock(new, fl);
244 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 new->fl_ops = fl->fl_ops;
246 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500247
248 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
251EXPORT_SYMBOL(locks_copy_lock);
252
253static inline int flock_translate_cmd(int cmd) {
254 if (cmd & LOCK_MAND)
255 return cmd & (LOCK_MAND | LOCK_RW);
256 switch (cmd) {
257 case LOCK_SH:
258 return F_RDLCK;
259 case LOCK_EX:
260 return F_WRLCK;
261 case LOCK_UN:
262 return F_UNLCK;
263 }
264 return -EINVAL;
265}
266
267/* Fill in a file_lock structure with an appropriate FLOCK lock. */
268static int flock_make_lock(struct file *filp, struct file_lock **lock,
269 unsigned int cmd)
270{
271 struct file_lock *fl;
272 int type = flock_translate_cmd(cmd);
273 if (type < 0)
274 return type;
275
276 fl = locks_alloc_lock();
277 if (fl == NULL)
278 return -ENOMEM;
279
280 fl->fl_file = filp;
281 fl->fl_pid = current->tgid;
282 fl->fl_flags = FL_FLOCK;
283 fl->fl_type = type;
284 fl->fl_end = OFFSET_MAX;
285
286 *lock = fl;
287 return 0;
288}
289
290static int assign_type(struct file_lock *fl, int type)
291{
292 switch (type) {
293 case F_RDLCK:
294 case F_WRLCK:
295 case F_UNLCK:
296 fl->fl_type = type;
297 break;
298 default:
299 return -EINVAL;
300 }
301 return 0;
302}
303
304/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
305 * style lock.
306 */
307static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
308 struct flock *l)
309{
310 off_t start, end;
311
312 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700313 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 start = 0;
315 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700316 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 start = filp->f_pos;
318 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700319 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800320 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 break;
322 default:
323 return -EINVAL;
324 }
325
326 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
327 POSIX-2001 defines it. */
328 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 if (start < 0)
330 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700331 fl->fl_end = OFFSET_MAX;
332 if (l->l_len > 0) {
333 end = start + l->l_len - 1;
334 fl->fl_end = end;
335 } else if (l->l_len < 0) {
336 end = start - 1;
337 fl->fl_end = end;
338 start += l->l_len;
339 if (start < 0)
340 return -EINVAL;
341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700343 if (fl->fl_end < fl->fl_start)
344 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
346 fl->fl_owner = current->files;
347 fl->fl_pid = current->tgid;
348 fl->fl_file = filp;
349 fl->fl_flags = FL_POSIX;
350 fl->fl_ops = NULL;
351 fl->fl_lmops = NULL;
352
353 return assign_type(fl, l->l_type);
354}
355
356#if BITS_PER_LONG == 32
357static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
358 struct flock64 *l)
359{
360 loff_t start;
361
362 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700363 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 start = 0;
365 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700366 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 start = filp->f_pos;
368 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700369 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800370 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 break;
372 default:
373 return -EINVAL;
374 }
375
Trond Myklebust4c780a42005-10-18 14:20:21 -0700376 start += l->l_start;
377 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700379 fl->fl_end = OFFSET_MAX;
380 if (l->l_len > 0) {
381 fl->fl_end = start + l->l_len - 1;
382 } else if (l->l_len < 0) {
383 fl->fl_end = start - 1;
384 start += l->l_len;
385 if (start < 0)
386 return -EINVAL;
387 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700389 if (fl->fl_end < fl->fl_start)
390 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
392 fl->fl_owner = current->files;
393 fl->fl_pid = current->tgid;
394 fl->fl_file = filp;
395 fl->fl_flags = FL_POSIX;
396 fl->fl_ops = NULL;
397 fl->fl_lmops = NULL;
398
399 switch (l->l_type) {
400 case F_RDLCK:
401 case F_WRLCK:
402 case F_UNLCK:
403 fl->fl_type = l->l_type;
404 break;
405 default:
406 return -EINVAL;
407 }
408
409 return (0);
410}
411#endif
412
413/* default lease lock manager operations */
414static void lease_break_callback(struct file_lock *fl)
415{
416 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
417}
418
419static void lease_release_private_callback(struct file_lock *fl)
420{
421 if (!fl->fl_file)
422 return;
423
424 f_delown(fl->fl_file);
425 fl->fl_file->f_owner.signum = 0;
426}
427
Adrian Bunk75c96f82005-05-05 16:16:09 -0700428static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429{
430 return fl->fl_file == try->fl_file;
431}
432
Adrian Bunk75c96f82005-05-05 16:16:09 -0700433static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434 .fl_break = lease_break_callback,
435 .fl_release_private = lease_release_private_callback,
436 .fl_mylease = lease_mylease_callback,
437 .fl_change = lease_modify,
438};
439
440/*
441 * Initialize a lease, use the default lock manager operations
442 */
443static int lease_init(struct file *filp, int type, struct file_lock *fl)
444 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400445 if (assign_type(fl, type) != 0)
446 return -EINVAL;
447
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 fl->fl_owner = current->files;
449 fl->fl_pid = current->tgid;
450
451 fl->fl_file = filp;
452 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 fl->fl_start = 0;
454 fl->fl_end = OFFSET_MAX;
455 fl->fl_ops = NULL;
456 fl->fl_lmops = &lease_manager_ops;
457 return 0;
458}
459
460/* Allocate a file_lock initialised to this type of lease */
461static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
462{
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)
Trond Myklebust75dff552006-05-07 23:02:42 -0400467 goto out;
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);
472 fl = NULL;
473 }
474out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 *flp = fl;
Trond Myklebust75dff552006-05-07 23:02:42 -0400476 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477}
478
479/* Check if two locks overlap each other.
480 */
481static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
482{
483 return ((fl1->fl_end >= fl2->fl_start) &&
484 (fl2->fl_end >= fl1->fl_start));
485}
486
487/*
488 * Check whether two locks have the same owner.
489 */
Matt Mackall33443c42006-01-08 01:05:22 -0800490static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
493 return fl2->fl_lmops == fl1->fl_lmops &&
494 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
495 return fl1->fl_owner == fl2->fl_owner;
496}
497
498/* Remove waiter from blocker's block list.
499 * When blocker ends up pointing to itself then the list is empty.
500 */
Matt Mackall33443c42006-01-08 01:05:22 -0800501static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 list_del_init(&waiter->fl_block);
504 list_del_init(&waiter->fl_link);
505 waiter->fl_next = NULL;
506}
507
508/*
509 */
510static void locks_delete_block(struct file_lock *waiter)
511{
512 lock_kernel();
513 __locks_delete_block(waiter);
514 unlock_kernel();
515}
516
517/* Insert waiter into blocker's block list.
518 * We use a circular list so that processes can be easily woken up in
519 * the order they blocked. The documentation doesn't require this but
520 * it seems like the reasonable thing to do.
521 */
522static void locks_insert_block(struct file_lock *blocker,
523 struct file_lock *waiter)
524{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800525 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 list_add_tail(&waiter->fl_block, &blocker->fl_block);
527 waiter->fl_next = blocker;
528 if (IS_POSIX(blocker))
529 list_add(&waiter->fl_link, &blocked_list);
530}
531
532/* Wake up processes blocked waiting for blocker.
533 * If told to wait then schedule the processes until the block list
534 * is empty, otherwise empty the block list ourselves.
535 */
536static void locks_wake_up_blocks(struct file_lock *blocker)
537{
538 while (!list_empty(&blocker->fl_block)) {
539 struct file_lock *waiter = list_entry(blocker->fl_block.next,
540 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
Andy Adamson8dc7c312006-03-20 13:44:26 -0500664int
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;
673 if (posix_locks_conflict(cfl, fl))
674 break;
675 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500676 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500677 __locks_copy_lock(fl, cfl);
Andy Adamson8dc7c312006-03-20 13:44:26 -0500678 unlock_kernel();
679 return 1;
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400680 } else
681 fl->fl_type = F_UNLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500683 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
686EXPORT_SYMBOL(posix_test_lock);
687
688/* This function tests for deadlock condition before putting a process to
689 * sleep. The detection scheme is no longer recursive. Recursive was neat,
690 * but dangerous - we risked stack corruption if the lock data was bad, or
691 * if the recursion was too deep for any other reason.
692 *
693 * We rely on the fact that a task can only be on one lock's wait queue
694 * at a time. When we find blocked_task on a wait queue we can re-search
695 * with blocked_task equal to that queue's owner, until either blocked_task
696 * isn't found, or blocked_task is found on a queue owned by my_task.
697 *
698 * Note: the above assumption may not be true when handling lock requests
699 * from a broken NFS client. But broken NFS clients have a lot more to
700 * worry about than proper deadlock detection anyway... --okir
701 */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700702static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 struct file_lock *block_fl)
704{
705 struct list_head *tmp;
706
707next_task:
708 if (posix_same_owner(caller_fl, block_fl))
709 return 1;
710 list_for_each(tmp, &blocked_list) {
711 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
712 if (posix_same_owner(fl, block_fl)) {
713 fl = fl->fl_next;
714 block_fl = fl;
715 goto next_task;
716 }
717 }
718 return 0;
719}
720
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
722 * at the head of the list, but that's secret knowledge known only to
723 * flock_lock_file and posix_lock_file.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400724 *
725 * Note that if called with an FL_EXISTS argument, the caller may determine
726 * whether or not a lock was successfully freed by testing the return
727 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800729static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800731 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800733 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 int error = 0;
735 int found = 0;
736
737 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400738 if (request->fl_flags & FL_ACCESS)
739 goto find_conflict;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 for_each_lock(inode, before) {
741 struct file_lock *fl = *before;
742 if (IS_POSIX(fl))
743 break;
744 if (IS_LEASE(fl))
745 continue;
746 if (filp != fl->fl_file)
747 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800748 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 goto out;
750 found = 1;
751 locks_delete_lock(before);
752 break;
753 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
Trond Myklebustf475ae92006-06-29 16:38:32 -0400755 if (request->fl_type == F_UNLCK) {
756 if ((request->fl_flags & FL_EXISTS) && !found)
757 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800758 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400761 error = -ENOMEM;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800762 new_fl = locks_alloc_lock();
763 if (new_fl == NULL)
764 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 /*
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
768 */
769 if (found)
770 cond_resched();
771
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400772find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 for_each_lock(inode, before) {
774 struct file_lock *fl = *before;
775 if (IS_POSIX(fl))
776 break;
777 if (IS_LEASE(fl))
778 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800779 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 continue;
781 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800782 if (request->fl_flags & FL_SLEEP)
783 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 goto out;
785 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400786 if (request->fl_flags & FL_ACCESS)
787 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800788 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800790 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400791 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
793out:
794 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800795 if (new_fl)
796 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 return error;
798}
799
Marc Eshel150b3932007-01-18 16:15:35 -0500800static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801{
802 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700803 struct file_lock *new_fl = NULL;
804 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 struct file_lock *left = NULL;
806 struct file_lock *right = NULL;
807 struct file_lock **before;
808 int error, added = 0;
809
810 /*
811 * We may need two file_lock structures for this operation,
812 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700813 *
814 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700816 if (!(request->fl_flags & FL_ACCESS) &&
817 (request->fl_type != F_UNLCK ||
818 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
819 new_fl = locks_alloc_lock();
820 new_fl2 = locks_alloc_lock();
821 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822
823 lock_kernel();
824 if (request->fl_type != F_UNLCK) {
825 for_each_lock(inode, before) {
826 struct file_lock *fl = *before;
827 if (!IS_POSIX(fl))
828 continue;
829 if (!posix_locks_conflict(request, fl))
830 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800831 if (conflock)
832 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 error = -EAGAIN;
834 if (!(request->fl_flags & FL_SLEEP))
835 goto out;
836 error = -EDEADLK;
837 if (posix_locks_deadlock(request, fl))
838 goto out;
839 error = -EAGAIN;
840 locks_insert_block(fl, request);
841 goto out;
842 }
843 }
844
845 /* If we're just looking for a conflict, we're done. */
846 error = 0;
847 if (request->fl_flags & FL_ACCESS)
848 goto out;
849
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 * Find the first old lock with the same owner as the new lock.
852 */
853
854 before = &inode->i_flock;
855
856 /* First skip locks owned by other processes. */
857 while ((fl = *before) && (!IS_POSIX(fl) ||
858 !posix_same_owner(request, fl))) {
859 before = &fl->fl_next;
860 }
861
862 /* Process locks with this owner. */
863 while ((fl = *before) && posix_same_owner(request, fl)) {
864 /* Detect adjacent or overlapping regions (if same lock type)
865 */
866 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700867 /* In all comparisons of start vs end, use
868 * "start - 1" rather than "end + 1". If end
869 * is OFFSET_MAX, end + 1 will become negative.
870 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (fl->fl_end < request->fl_start - 1)
872 goto next_lock;
873 /* If the next lock in the list has entirely bigger
874 * addresses than the new one, insert the lock here.
875 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700876 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 break;
878
879 /* If we come here, the new and old lock are of the
880 * same type and adjacent or overlapping. Make one
881 * lock yielding from the lower start address of both
882 * locks to the higher end address.
883 */
884 if (fl->fl_start > request->fl_start)
885 fl->fl_start = request->fl_start;
886 else
887 request->fl_start = fl->fl_start;
888 if (fl->fl_end < request->fl_end)
889 fl->fl_end = request->fl_end;
890 else
891 request->fl_end = fl->fl_end;
892 if (added) {
893 locks_delete_lock(before);
894 continue;
895 }
896 request = fl;
897 added = 1;
898 }
899 else {
900 /* Processing for different lock types is a bit
901 * more complex.
902 */
903 if (fl->fl_end < request->fl_start)
904 goto next_lock;
905 if (fl->fl_start > request->fl_end)
906 break;
907 if (request->fl_type == F_UNLCK)
908 added = 1;
909 if (fl->fl_start < request->fl_start)
910 left = fl;
911 /* If the next lock in the list has a higher end
912 * address than the new one, insert the new one here.
913 */
914 if (fl->fl_end > request->fl_end) {
915 right = fl;
916 break;
917 }
918 if (fl->fl_start >= request->fl_start) {
919 /* The new lock completely replaces an old
920 * one (This may happen several times).
921 */
922 if (added) {
923 locks_delete_lock(before);
924 continue;
925 }
926 /* Replace the old lock with the new one.
927 * Wake up anybody waiting for the old one,
928 * as the change in lock type might satisfy
929 * their needs.
930 */
931 locks_wake_up_blocks(fl);
932 fl->fl_start = request->fl_start;
933 fl->fl_end = request->fl_end;
934 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500935 locks_release_private(fl);
936 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 request = fl;
938 added = 1;
939 }
940 }
941 /* Go on to next lock.
942 */
943 next_lock:
944 before = &fl->fl_next;
945 }
946
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700947 /*
948 * The above code only modifies existing locks in case of
949 * merging or replacing. If new lock(s) need to be inserted
950 * all modifications are done bellow this, so it's safe yet to
951 * bail out.
952 */
953 error = -ENOLCK; /* "no luck" */
954 if (right && left == right && !new_fl2)
955 goto out;
956
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 error = 0;
958 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400959 if (request->fl_type == F_UNLCK) {
960 if (request->fl_flags & FL_EXISTS)
961 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400963 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700964
965 if (!new_fl) {
966 error = -ENOLCK;
967 goto out;
968 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 locks_copy_lock(new_fl, request);
970 locks_insert_lock(before, new_fl);
971 new_fl = NULL;
972 }
973 if (right) {
974 if (left == right) {
975 /* The new lock breaks the old one in two pieces,
976 * so we have to use the second new lock.
977 */
978 left = new_fl2;
979 new_fl2 = NULL;
980 locks_copy_lock(left, right);
981 locks_insert_lock(before, left);
982 }
983 right->fl_start = request->fl_end + 1;
984 locks_wake_up_blocks(right);
985 }
986 if (left) {
987 left->fl_end = request->fl_start - 1;
988 locks_wake_up_blocks(left);
989 }
990 out:
991 unlock_kernel();
992 /*
993 * Free any unused locks.
994 */
995 if (new_fl)
996 locks_free_lock(new_fl);
997 if (new_fl2)
998 locks_free_lock(new_fl2);
999 return error;
1000}
1001
1002/**
1003 * posix_lock_file - Apply a POSIX-style lock to a file
1004 * @filp: The file to apply the lock to
1005 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001006 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 *
1008 * Add a POSIX style lock to a file.
1009 * We merge adjacent & overlapping locks whenever possible.
1010 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001011 *
1012 * Note that if called with an FL_EXISTS argument, the caller may determine
1013 * whether or not a lock was successfully freed by testing the return
1014 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 */
Marc Eshel150b3932007-01-18 16:15:35 -05001016int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001017 struct file_lock *conflock)
1018{
Marc Eshel150b3932007-01-18 16:15:35 -05001019 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001020}
Marc Eshel150b3932007-01-18 16:15:35 -05001021EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022
1023/**
1024 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1025 * @filp: The file to apply the lock to
1026 * @fl: The lock to be applied
1027 *
1028 * Add a POSIX style lock to a file.
1029 * We merge adjacent & overlapping locks whenever possible.
1030 * POSIX locks are sorted by owner task, then by starting address
1031 */
1032int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1033{
1034 int error;
1035 might_sleep ();
1036 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001037 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1039 break;
1040 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1041 if (!error)
1042 continue;
1043
1044 locks_delete_block(fl);
1045 break;
1046 }
1047 return error;
1048}
1049EXPORT_SYMBOL(posix_lock_file_wait);
1050
1051/**
1052 * locks_mandatory_locked - Check for an active lock
1053 * @inode: the file to check
1054 *
1055 * Searches the inode's list of locks to find any POSIX locks which conflict.
1056 * This function is called from locks_verify_locked() only.
1057 */
1058int locks_mandatory_locked(struct inode *inode)
1059{
1060 fl_owner_t owner = current->files;
1061 struct file_lock *fl;
1062
1063 /*
1064 * Search the lock list for this inode for any POSIX locks.
1065 */
1066 lock_kernel();
1067 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1068 if (!IS_POSIX(fl))
1069 continue;
1070 if (fl->fl_owner != owner)
1071 break;
1072 }
1073 unlock_kernel();
1074 return fl ? -EAGAIN : 0;
1075}
1076
1077/**
1078 * locks_mandatory_area - Check for a conflicting lock
1079 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1080 * for shared
1081 * @inode: the file to check
1082 * @filp: how the file was opened (if it was)
1083 * @offset: start of area to check
1084 * @count: length of area to check
1085 *
1086 * Searches the inode's list of locks to find any POSIX locks which conflict.
1087 * This function is called from rw_verify_area() and
1088 * locks_verify_truncate().
1089 */
1090int locks_mandatory_area(int read_write, struct inode *inode,
1091 struct file *filp, loff_t offset,
1092 size_t count)
1093{
1094 struct file_lock fl;
1095 int error;
1096
1097 locks_init_lock(&fl);
1098 fl.fl_owner = current->files;
1099 fl.fl_pid = current->tgid;
1100 fl.fl_file = filp;
1101 fl.fl_flags = FL_POSIX | FL_ACCESS;
1102 if (filp && !(filp->f_flags & O_NONBLOCK))
1103 fl.fl_flags |= FL_SLEEP;
1104 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1105 fl.fl_start = offset;
1106 fl.fl_end = offset + count - 1;
1107
1108 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001109 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 if (error != -EAGAIN)
1111 break;
1112 if (!(fl.fl_flags & FL_SLEEP))
1113 break;
1114 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1115 if (!error) {
1116 /*
1117 * If we've been sleeping someone might have
1118 * changed the permissions behind our back.
1119 */
1120 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1121 continue;
1122 }
1123
1124 locks_delete_block(&fl);
1125 break;
1126 }
1127
1128 return error;
1129}
1130
1131EXPORT_SYMBOL(locks_mandatory_area);
1132
1133/* We already had a lease on this file; just change its type */
1134int lease_modify(struct file_lock **before, int arg)
1135{
1136 struct file_lock *fl = *before;
1137 int error = assign_type(fl, arg);
1138
1139 if (error)
1140 return error;
1141 locks_wake_up_blocks(fl);
1142 if (arg == F_UNLCK)
1143 locks_delete_lock(before);
1144 return 0;
1145}
1146
1147EXPORT_SYMBOL(lease_modify);
1148
1149static void time_out_leases(struct inode *inode)
1150{
1151 struct file_lock **before;
1152 struct file_lock *fl;
1153
1154 before = &inode->i_flock;
1155 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1156 if ((fl->fl_break_time == 0)
1157 || time_before(jiffies, fl->fl_break_time)) {
1158 before = &fl->fl_next;
1159 continue;
1160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1162 if (fl == *before) /* lease_modify may have freed fl */
1163 before = &fl->fl_next;
1164 }
1165}
1166
1167/**
1168 * __break_lease - revoke all outstanding leases on file
1169 * @inode: the inode of the file to return
1170 * @mode: the open mode (read or write)
1171 *
david m. richter87250dd2007-05-09 16:10:27 -04001172 * break_lease (inlined for speed) has checked there already is at least
1173 * some kind of lock (maybe a lease) on this file. Leases are broken on
1174 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 * specified %O_NONBLOCK to your open().
1176 */
1177int __break_lease(struct inode *inode, unsigned int mode)
1178{
1179 int error = 0, future;
1180 struct file_lock *new_fl, *flock;
1181 struct file_lock *fl;
1182 int alloc_err;
1183 unsigned long break_time;
1184 int i_have_this_lease = 0;
1185
1186 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1187 &new_fl);
1188
1189 lock_kernel();
1190
1191 time_out_leases(inode);
1192
1193 flock = inode->i_flock;
1194 if ((flock == NULL) || !IS_LEASE(flock))
1195 goto out;
1196
1197 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1198 if (fl->fl_owner == current->files)
1199 i_have_this_lease = 1;
1200
1201 if (mode & FMODE_WRITE) {
1202 /* If we want write access, we have to revoke any lease. */
1203 future = F_UNLCK | F_INPROGRESS;
1204 } else if (flock->fl_type & F_INPROGRESS) {
1205 /* If the lease is already being broken, we just leave it */
1206 future = flock->fl_type;
1207 } else if (flock->fl_type & F_WRLCK) {
1208 /* Downgrade the exclusive lease to a read-only lease. */
1209 future = F_RDLCK | F_INPROGRESS;
1210 } else {
1211 /* the existing lease was read-only, so we can read too. */
1212 goto out;
1213 }
1214
1215 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1216 error = alloc_err;
1217 goto out;
1218 }
1219
1220 break_time = 0;
1221 if (lease_break_time > 0) {
1222 break_time = jiffies + lease_break_time * HZ;
1223 if (break_time == 0)
1224 break_time++; /* so that 0 means no break time */
1225 }
1226
1227 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1228 if (fl->fl_type != future) {
1229 fl->fl_type = future;
1230 fl->fl_break_time = break_time;
1231 /* lease must have lmops break callback */
1232 fl->fl_lmops->fl_break(fl);
1233 }
1234 }
1235
1236 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1237 error = -EWOULDBLOCK;
1238 goto out;
1239 }
1240
1241restart:
1242 break_time = flock->fl_break_time;
1243 if (break_time != 0) {
1244 break_time -= jiffies;
1245 if (break_time == 0)
1246 break_time++;
1247 }
1248 error = locks_block_on_timeout(flock, new_fl, break_time);
1249 if (error >= 0) {
1250 if (error == 0)
1251 time_out_leases(inode);
1252 /* Wait for the next lease that has not been broken yet */
1253 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1254 flock = flock->fl_next) {
1255 if (flock->fl_type & F_INPROGRESS)
1256 goto restart;
1257 }
1258 error = 0;
1259 }
1260
1261out:
1262 unlock_kernel();
1263 if (!alloc_err)
1264 locks_free_lock(new_fl);
1265 return error;
1266}
1267
1268EXPORT_SYMBOL(__break_lease);
1269
1270/**
1271 * lease_get_mtime
1272 * @inode: the inode
1273 * @time: pointer to a timespec which will contain the last modified time
1274 *
1275 * This is to force NFS clients to flush their caches for files with
1276 * exclusive leases. The justification is that if someone has an
1277 * exclusive lease, then they could be modifiying it.
1278 */
1279void lease_get_mtime(struct inode *inode, struct timespec *time)
1280{
1281 struct file_lock *flock = inode->i_flock;
1282 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1283 *time = current_fs_time(inode->i_sb);
1284 else
1285 *time = inode->i_mtime;
1286}
1287
1288EXPORT_SYMBOL(lease_get_mtime);
1289
1290/**
1291 * fcntl_getlease - Enquire what lease is currently active
1292 * @filp: the file
1293 *
1294 * The value returned by this function will be one of
1295 * (if no lease break is pending):
1296 *
1297 * %F_RDLCK to indicate a shared lease is held.
1298 *
1299 * %F_WRLCK to indicate an exclusive lease is held.
1300 *
1301 * %F_UNLCK to indicate no lease is held.
1302 *
1303 * (if a lease break is pending):
1304 *
1305 * %F_RDLCK to indicate an exclusive lease needs to be
1306 * changed to a shared lease (or removed).
1307 *
1308 * %F_UNLCK to indicate the lease needs to be removed.
1309 *
1310 * XXX: sfr & willy disagree over whether F_INPROGRESS
1311 * should be returned to userspace.
1312 */
1313int fcntl_getlease(struct file *filp)
1314{
1315 struct file_lock *fl;
1316 int type = F_UNLCK;
1317
1318 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001319 time_out_leases(filp->f_path.dentry->d_inode);
1320 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 fl = fl->fl_next) {
1322 if (fl->fl_file == filp) {
1323 type = fl->fl_type & ~F_INPROGRESS;
1324 break;
1325 }
1326 }
1327 unlock_kernel();
1328 return type;
1329}
1330
1331/**
1332 * __setlease - sets a lease on an open file
1333 * @filp: file pointer
1334 * @arg: type of lease to obtain
1335 * @flp: input - file_lock to use, output - file_lock inserted
1336 *
1337 * The (input) flp->fl_lmops->fl_break function is required
1338 * by break_lease().
1339 *
1340 * Called with kernel lock held.
1341 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001342static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001344 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001345 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 struct inode *inode = dentry->d_inode;
1347 int error, rdlease_count = 0, wrlease_count = 0;
1348
1349 time_out_leases(inode);
1350
1351 error = -EINVAL;
1352 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1353 goto out;
1354
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001355 lease = *flp;
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 error = -EAGAIN;
1358 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1359 goto out;
1360 if ((arg == F_WRLCK)
1361 && ((atomic_read(&dentry->d_count) > 1)
1362 || (atomic_read(&inode->i_count) > 1)))
1363 goto out;
1364
1365 /*
1366 * At this point, we know that if there is an exclusive
1367 * lease on this file, then we hold it on this filp
1368 * (otherwise our open of this file would have blocked).
1369 * And if we are trying to acquire an exclusive lease,
1370 * then the file is not open by anyone (including us)
1371 * except for this filp.
1372 */
1373 for (before = &inode->i_flock;
1374 ((fl = *before) != NULL) && IS_LEASE(fl);
1375 before = &fl->fl_next) {
1376 if (lease->fl_lmops->fl_mylease(fl, lease))
1377 my_before = before;
1378 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1379 /*
1380 * Someone is in the process of opening this
1381 * file for writing so we may not take an
1382 * exclusive lease on it.
1383 */
1384 wrlease_count++;
1385 else
1386 rdlease_count++;
1387 }
1388
1389 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1390 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1391 goto out;
1392
1393 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001394 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 error = lease->fl_lmops->fl_change(my_before, arg);
1396 goto out;
1397 }
1398
1399 error = 0;
1400 if (arg == F_UNLCK)
1401 goto out;
1402
1403 error = -EINVAL;
1404 if (!leases_enable)
1405 goto out;
1406
Trond Myklebust74361cb2006-08-14 08:54:48 -07001407 error = -ENOMEM;
1408 fl = locks_alloc_lock();
1409 if (fl == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 goto out;
1411
1412 locks_copy_lock(fl, lease);
1413
1414 locks_insert_lock(before, fl);
1415
1416 *flp = fl;
Trond Myklebust74361cb2006-08-14 08:54:48 -07001417 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418out:
1419 return error;
1420}
1421
1422 /**
1423 * setlease - sets a lease on an open file
1424 * @filp: file pointer
1425 * @arg: type of lease to obtain
1426 * @lease: file_lock to use
1427 *
1428 * Call this to establish a lease on the file.
1429 * The fl_lmops fl_break function is required by break_lease
1430 */
1431
1432int setlease(struct file *filp, long arg, struct file_lock **lease)
1433{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001434 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 struct inode *inode = dentry->d_inode;
1436 int error;
1437
1438 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1439 return -EACCES;
1440 if (!S_ISREG(inode->i_mode))
1441 return -EINVAL;
1442 error = security_file_lock(filp, arg);
1443 if (error)
1444 return error;
1445
1446 lock_kernel();
1447 error = __setlease(filp, arg, lease);
1448 unlock_kernel();
1449
1450 return error;
1451}
1452
1453EXPORT_SYMBOL(setlease);
1454
1455/**
1456 * fcntl_setlease - sets a lease on an open file
1457 * @fd: open file descriptor
1458 * @filp: file pointer
1459 * @arg: type of lease to obtain
1460 *
1461 * Call this fcntl to establish a lease on the file.
1462 * Note that you also need to call %F_SETSIG to
1463 * receive a signal when the lease is broken.
1464 */
1465int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1466{
1467 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001468 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 struct inode *inode = dentry->d_inode;
1470 int error;
1471
1472 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1473 return -EACCES;
1474 if (!S_ISREG(inode->i_mode))
1475 return -EINVAL;
1476 error = security_file_lock(filp, arg);
1477 if (error)
1478 return error;
1479
1480 locks_init_lock(&fl);
1481 error = lease_init(filp, arg, &fl);
1482 if (error)
1483 return error;
1484
1485 lock_kernel();
1486
1487 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001488 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 goto out_unlock;
1490
1491 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1492 if (error < 0) {
1493 /* remove lease just inserted by __setlease */
1494 flp->fl_type = F_UNLCK | F_INPROGRESS;
1495 flp->fl_break_time = jiffies- 10;
1496 time_out_leases(inode);
1497 goto out_unlock;
1498 }
1499
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001500 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501out_unlock:
1502 unlock_kernel();
1503 return error;
1504}
1505
1506/**
1507 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1508 * @filp: The file to apply the lock to
1509 * @fl: The lock to be applied
1510 *
1511 * Add a FLOCK style lock to a file.
1512 */
1513int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1514{
1515 int error;
1516 might_sleep();
1517 for (;;) {
1518 error = flock_lock_file(filp, fl);
1519 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1520 break;
1521 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1522 if (!error)
1523 continue;
1524
1525 locks_delete_block(fl);
1526 break;
1527 }
1528 return error;
1529}
1530
1531EXPORT_SYMBOL(flock_lock_file_wait);
1532
1533/**
1534 * sys_flock: - flock() system call.
1535 * @fd: the file descriptor to lock.
1536 * @cmd: the type of lock to apply.
1537 *
1538 * Apply a %FL_FLOCK style lock to an open file descriptor.
1539 * The @cmd can be one of
1540 *
1541 * %LOCK_SH -- a shared lock.
1542 *
1543 * %LOCK_EX -- an exclusive lock.
1544 *
1545 * %LOCK_UN -- remove an existing lock.
1546 *
1547 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1548 *
1549 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1550 * processes read and write access respectively.
1551 */
1552asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1553{
1554 struct file *filp;
1555 struct file_lock *lock;
1556 int can_sleep, unlock;
1557 int error;
1558
1559 error = -EBADF;
1560 filp = fget(fd);
1561 if (!filp)
1562 goto out;
1563
1564 can_sleep = !(cmd & LOCK_NB);
1565 cmd &= ~LOCK_NB;
1566 unlock = (cmd == LOCK_UN);
1567
1568 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1569 goto out_putf;
1570
1571 error = flock_make_lock(filp, &lock, cmd);
1572 if (error)
1573 goto out_putf;
1574 if (can_sleep)
1575 lock->fl_flags |= FL_SLEEP;
1576
1577 error = security_file_lock(filp, cmd);
1578 if (error)
1579 goto out_free;
1580
1581 if (filp->f_op && filp->f_op->flock)
1582 error = filp->f_op->flock(filp,
1583 (can_sleep) ? F_SETLKW : F_SETLK,
1584 lock);
1585 else
1586 error = flock_lock_file_wait(filp, lock);
1587
1588 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001589 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590
1591 out_putf:
1592 fput(filp);
1593 out:
1594 return error;
1595}
1596
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001597/**
1598 * vfs_test_lock - test file byte range lock
1599 * @filp: The file to test lock for
1600 * @fl: The lock to test
1601 * @conf: Place to return a copy of the conflicting lock, if found
1602 *
1603 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1604 * setting conf->fl_type to something other than F_UNLCK.
1605 */
1606int vfs_test_lock(struct file *filp, struct file_lock *fl)
1607{
1608 if (filp->f_op && filp->f_op->lock)
1609 return filp->f_op->lock(filp, F_GETLK, fl);
1610 posix_test_lock(filp, fl);
1611 return 0;
1612}
1613EXPORT_SYMBOL_GPL(vfs_test_lock);
1614
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001615static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1616{
1617 flock->l_pid = fl->fl_pid;
1618#if BITS_PER_LONG == 32
1619 /*
1620 * Make sure we can represent the posix lock via
1621 * legacy 32bit flock.
1622 */
1623 if (fl->fl_start > OFFT_OFFSET_MAX)
1624 return -EOVERFLOW;
1625 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1626 return -EOVERFLOW;
1627#endif
1628 flock->l_start = fl->fl_start;
1629 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1630 fl->fl_end - fl->fl_start + 1;
1631 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001632 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001633 return 0;
1634}
1635
1636#if BITS_PER_LONG == 32
1637static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1638{
1639 flock->l_pid = fl->fl_pid;
1640 flock->l_start = fl->fl_start;
1641 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1642 fl->fl_end - fl->fl_start + 1;
1643 flock->l_whence = 0;
1644 flock->l_type = fl->fl_type;
1645}
1646#endif
1647
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648/* Report the first existing lock that would conflict with l.
1649 * This implements the F_GETLK command of fcntl().
1650 */
1651int fcntl_getlk(struct file *filp, struct flock __user *l)
1652{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001653 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 struct flock flock;
1655 int error;
1656
1657 error = -EFAULT;
1658 if (copy_from_user(&flock, l, sizeof(flock)))
1659 goto out;
1660 error = -EINVAL;
1661 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1662 goto out;
1663
1664 error = flock_to_posix_lock(filp, &file_lock, &flock);
1665 if (error)
1666 goto out;
1667
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001668 error = vfs_test_lock(filp, &file_lock);
1669 if (error)
1670 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001672 flock.l_type = file_lock.fl_type;
1673 if (file_lock.fl_type != F_UNLCK) {
1674 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001675 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
1678 error = -EFAULT;
1679 if (!copy_to_user(l, &flock, sizeof(flock)))
1680 error = 0;
1681out:
1682 return error;
1683}
1684
Marc Eshel7723ec92007-01-18 15:08:55 -05001685/**
1686 * vfs_lock_file - file byte range lock
1687 * @filp: The file to apply the lock to
1688 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1689 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001690 * @conf: Place to return a copy of the conflicting lock, if found.
1691 *
1692 * A caller that doesn't care about the conflicting lock may pass NULL
1693 * as the final argument.
1694 *
1695 * If the filesystem defines a private ->lock() method, then @conf will
1696 * be left unchanged; so a caller that cares should initialize it to
1697 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001698 *
1699 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1700 * locks, the ->lock() interface may return asynchronously, before the lock has
1701 * been granted or denied by the underlying filesystem, if (and only if)
1702 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1703 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1704 * the request is for a blocking lock. When ->lock() does return asynchronously,
1705 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1706 * request completes.
1707 * If the request is for non-blocking lock the file system should return
1708 * -EINPROGRESS then try to get the lock and call the callback routine with
1709 * the result. If the request timed out the callback routine will return a
1710 * nonzero return code and the file system should release the lock. The file
1711 * system is also responsible to keep a corresponding posix lock when it
1712 * grants a lock so the VFS can find out which locks are locally held and do
1713 * the correct lock cleanup when required.
1714 * The underlying filesystem must not drop the kernel lock or call
1715 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1716 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001717 */
Marc Eshel150b3932007-01-18 16:15:35 -05001718int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001719{
1720 if (filp->f_op && filp->f_op->lock)
1721 return filp->f_op->lock(filp, cmd, fl);
1722 else
Marc Eshel150b3932007-01-18 16:15:35 -05001723 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001724}
1725EXPORT_SYMBOL_GPL(vfs_lock_file);
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727/* Apply the lock described by l to an open file descriptor.
1728 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1729 */
Peter Staubachc2936212005-07-27 11:45:09 -07001730int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1731 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732{
1733 struct file_lock *file_lock = locks_alloc_lock();
1734 struct flock flock;
1735 struct inode *inode;
1736 int error;
1737
1738 if (file_lock == NULL)
1739 return -ENOLCK;
1740
1741 /*
1742 * This might block, so we do it before checking the inode.
1743 */
1744 error = -EFAULT;
1745 if (copy_from_user(&flock, l, sizeof(flock)))
1746 goto out;
1747
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001748 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
1750 /* Don't allow mandatory locks on files that may be memory mapped
1751 * and shared.
1752 */
1753 if (IS_MANDLOCK(inode) &&
1754 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1755 mapping_writably_mapped(filp->f_mapping)) {
1756 error = -EAGAIN;
1757 goto out;
1758 }
1759
Peter Staubachc2936212005-07-27 11:45:09 -07001760again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 error = flock_to_posix_lock(filp, file_lock, &flock);
1762 if (error)
1763 goto out;
1764 if (cmd == F_SETLKW) {
1765 file_lock->fl_flags |= FL_SLEEP;
1766 }
1767
1768 error = -EBADF;
1769 switch (flock.l_type) {
1770 case F_RDLCK:
1771 if (!(filp->f_mode & FMODE_READ))
1772 goto out;
1773 break;
1774 case F_WRLCK:
1775 if (!(filp->f_mode & FMODE_WRITE))
1776 goto out;
1777 break;
1778 case F_UNLCK:
1779 break;
1780 default:
1781 error = -EINVAL;
1782 goto out;
1783 }
1784
1785 error = security_file_lock(filp, file_lock->fl_type);
1786 if (error)
1787 goto out;
1788
Marc Eshel7723ec92007-01-18 15:08:55 -05001789 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001790 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001791 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001793 error = wait_event_interruptible(file_lock->fl_wait,
1794 !file_lock->fl_next);
1795 if (!error)
1796 continue;
1797
1798 locks_delete_block(file_lock);
1799 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 }
1801
Peter Staubachc2936212005-07-27 11:45:09 -07001802 /*
1803 * Attempt to detect a close/fcntl race and recover by
1804 * releasing the lock that was just acquired.
1805 */
1806 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1807 flock.l_type = F_UNLCK;
1808 goto again;
1809 }
1810
1811out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 locks_free_lock(file_lock);
1813 return error;
1814}
1815
1816#if BITS_PER_LONG == 32
1817/* Report the first existing lock that would conflict with l.
1818 * This implements the F_GETLK command of fcntl().
1819 */
1820int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1821{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001822 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823 struct flock64 flock;
1824 int error;
1825
1826 error = -EFAULT;
1827 if (copy_from_user(&flock, l, sizeof(flock)))
1828 goto out;
1829 error = -EINVAL;
1830 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1831 goto out;
1832
1833 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1834 if (error)
1835 goto out;
1836
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001837 error = vfs_test_lock(filp, &file_lock);
1838 if (error)
1839 goto out;
1840
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001841 flock.l_type = file_lock.fl_type;
1842 if (file_lock.fl_type != F_UNLCK)
1843 posix_lock_to_flock64(&flock, &file_lock);
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 error = -EFAULT;
1846 if (!copy_to_user(l, &flock, sizeof(flock)))
1847 error = 0;
1848
1849out:
1850 return error;
1851}
1852
1853/* Apply the lock described by l to an open file descriptor.
1854 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1855 */
Peter Staubachc2936212005-07-27 11:45:09 -07001856int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1857 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858{
1859 struct file_lock *file_lock = locks_alloc_lock();
1860 struct flock64 flock;
1861 struct inode *inode;
1862 int error;
1863
1864 if (file_lock == NULL)
1865 return -ENOLCK;
1866
1867 /*
1868 * This might block, so we do it before checking the inode.
1869 */
1870 error = -EFAULT;
1871 if (copy_from_user(&flock, l, sizeof(flock)))
1872 goto out;
1873
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001874 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875
1876 /* Don't allow mandatory locks on files that may be memory mapped
1877 * and shared.
1878 */
1879 if (IS_MANDLOCK(inode) &&
1880 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1881 mapping_writably_mapped(filp->f_mapping)) {
1882 error = -EAGAIN;
1883 goto out;
1884 }
1885
Peter Staubachc2936212005-07-27 11:45:09 -07001886again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 error = flock64_to_posix_lock(filp, file_lock, &flock);
1888 if (error)
1889 goto out;
1890 if (cmd == F_SETLKW64) {
1891 file_lock->fl_flags |= FL_SLEEP;
1892 }
1893
1894 error = -EBADF;
1895 switch (flock.l_type) {
1896 case F_RDLCK:
1897 if (!(filp->f_mode & FMODE_READ))
1898 goto out;
1899 break;
1900 case F_WRLCK:
1901 if (!(filp->f_mode & FMODE_WRITE))
1902 goto out;
1903 break;
1904 case F_UNLCK:
1905 break;
1906 default:
1907 error = -EINVAL;
1908 goto out;
1909 }
1910
1911 error = security_file_lock(filp, file_lock->fl_type);
1912 if (error)
1913 goto out;
1914
Marc Eshel7723ec92007-01-18 15:08:55 -05001915 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001916 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001917 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001918 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001919 error = wait_event_interruptible(file_lock->fl_wait,
1920 !file_lock->fl_next);
1921 if (!error)
1922 continue;
1923
1924 locks_delete_block(file_lock);
1925 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 }
1927
Peter Staubachc2936212005-07-27 11:45:09 -07001928 /*
1929 * Attempt to detect a close/fcntl race and recover by
1930 * releasing the lock that was just acquired.
1931 */
1932 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1933 flock.l_type = F_UNLCK;
1934 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 }
1936
1937out:
1938 locks_free_lock(file_lock);
1939 return error;
1940}
1941#endif /* BITS_PER_LONG == 32 */
1942
1943/*
1944 * This function is called when the file is being removed
1945 * from the task's fd array. POSIX locks belonging to this task
1946 * are deleted at this time.
1947 */
1948void locks_remove_posix(struct file *filp, fl_owner_t owner)
1949{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001950 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
1952 /*
1953 * If there are no locks held on this file, we don't need to call
1954 * posix_lock_file(). Another process could be setting a lock on this
1955 * file at the same time, but we wouldn't remove that lock anyway.
1956 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001957 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 return;
1959
1960 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001961 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 lock.fl_start = 0;
1963 lock.fl_end = OFFSET_MAX;
1964 lock.fl_owner = owner;
1965 lock.fl_pid = current->tgid;
1966 lock.fl_file = filp;
1967 lock.fl_ops = NULL;
1968 lock.fl_lmops = NULL;
1969
Marc Eshel150b3932007-01-18 16:15:35 -05001970 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1973 lock.fl_ops->fl_release_private(&lock);
1974}
1975
1976EXPORT_SYMBOL(locks_remove_posix);
1977
1978/*
1979 * This function is called on the last close of an open file.
1980 */
1981void locks_remove_flock(struct file *filp)
1982{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001983 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 struct file_lock *fl;
1985 struct file_lock **before;
1986
1987 if (!inode->i_flock)
1988 return;
1989
1990 if (filp->f_op && filp->f_op->flock) {
1991 struct file_lock fl = {
1992 .fl_pid = current->tgid,
1993 .fl_file = filp,
1994 .fl_flags = FL_FLOCK,
1995 .fl_type = F_UNLCK,
1996 .fl_end = OFFSET_MAX,
1997 };
1998 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001999 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2000 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 }
2002
2003 lock_kernel();
2004 before = &inode->i_flock;
2005
2006 while ((fl = *before) != NULL) {
2007 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002008 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 locks_delete_lock(before);
2010 continue;
2011 }
2012 if (IS_LEASE(fl)) {
2013 lease_modify(before, F_UNLCK);
2014 continue;
2015 }
2016 /* What? */
2017 BUG();
2018 }
2019 before = &fl->fl_next;
2020 }
2021 unlock_kernel();
2022}
2023
2024/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 * posix_unblock_lock - stop waiting for a file lock
2026 * @filp: how the file was opened
2027 * @waiter: the lock which was waiting
2028 *
2029 * lockd needs to block waiting for locks.
2030 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002031int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2033{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002034 int status = 0;
2035
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002037 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002039 else
2040 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002041 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002042 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043}
2044
2045EXPORT_SYMBOL(posix_unblock_lock);
2046
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002047/**
2048 * vfs_cancel_lock - file byte range unblock lock
2049 * @filp: The file to apply the unblock to
2050 * @fl: The lock to be unblocked
2051 *
2052 * Used by lock managers to cancel blocked requests
2053 */
2054int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2055{
2056 if (filp->f_op && filp->f_op->lock)
2057 return filp->f_op->lock(filp, F_CANCELLK, fl);
2058 return 0;
2059}
2060
2061EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2062
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2064{
2065 struct inode *inode = NULL;
2066
2067 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002068 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070 out += sprintf(out, "%d:%s ", id, pfx);
2071 if (IS_POSIX(fl)) {
2072 out += sprintf(out, "%6s %s ",
2073 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2074 (inode == NULL) ? "*NOINODE*" :
2075 (IS_MANDLOCK(inode) &&
2076 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2077 "MANDATORY" : "ADVISORY ");
2078 } else if (IS_FLOCK(fl)) {
2079 if (fl->fl_type & LOCK_MAND) {
2080 out += sprintf(out, "FLOCK MSNFS ");
2081 } else {
2082 out += sprintf(out, "FLOCK ADVISORY ");
2083 }
2084 } else if (IS_LEASE(fl)) {
2085 out += sprintf(out, "LEASE ");
2086 if (fl->fl_type & F_INPROGRESS)
2087 out += sprintf(out, "BREAKING ");
2088 else if (fl->fl_file)
2089 out += sprintf(out, "ACTIVE ");
2090 else
2091 out += sprintf(out, "BREAKER ");
2092 } else {
2093 out += sprintf(out, "UNKNOWN UNKNOWN ");
2094 }
2095 if (fl->fl_type & LOCK_MAND) {
2096 out += sprintf(out, "%s ",
2097 (fl->fl_type & LOCK_READ)
2098 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2099 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2100 } else {
2101 out += sprintf(out, "%s ",
2102 (fl->fl_type & F_INPROGRESS)
2103 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2104 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2105 }
2106 if (inode) {
2107#ifdef WE_CAN_BREAK_LSLK_NOW
2108 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2109 inode->i_sb->s_id, inode->i_ino);
2110#else
2111 /* userspace relies on this representation of dev_t ;-( */
2112 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2113 MAJOR(inode->i_sb->s_dev),
2114 MINOR(inode->i_sb->s_dev), inode->i_ino);
2115#endif
2116 } else {
2117 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2118 }
2119 if (IS_POSIX(fl)) {
2120 if (fl->fl_end == OFFSET_MAX)
2121 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2122 else
2123 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2124 fl->fl_end);
2125 } else {
2126 out += sprintf(out, "0 EOF\n");
2127 }
2128}
2129
2130static void move_lock_status(char **p, off_t* pos, off_t offset)
2131{
2132 int len;
2133 len = strlen(*p);
2134 if(*pos >= offset) {
2135 /* the complete line is valid */
2136 *p += len;
2137 *pos += len;
2138 return;
2139 }
2140 if(*pos+len > offset) {
2141 /* use the second part of the line */
2142 int i = offset-*pos;
2143 memmove(*p,*p+i,len-i);
2144 *p += len-i;
2145 *pos += len;
2146 return;
2147 }
2148 /* discard the complete line */
2149 *pos += len;
2150}
2151
2152/**
2153 * get_locks_status - reports lock usage in /proc/locks
2154 * @buffer: address in userspace to write into
2155 * @start: ?
2156 * @offset: how far we are through the buffer
2157 * @length: how much to read
2158 */
2159
2160int get_locks_status(char *buffer, char **start, off_t offset, int length)
2161{
2162 struct list_head *tmp;
2163 char *q = buffer;
2164 off_t pos = 0;
2165 int i = 0;
2166
2167 lock_kernel();
2168 list_for_each(tmp, &file_lock_list) {
2169 struct list_head *btmp;
2170 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2171 lock_get_status(q, fl, ++i, "");
2172 move_lock_status(&q, &pos, offset);
2173
2174 if(pos >= offset+length)
2175 goto done;
2176
2177 list_for_each(btmp, &fl->fl_block) {
2178 struct file_lock *bfl = list_entry(btmp,
2179 struct file_lock, fl_block);
2180 lock_get_status(q, bfl, i, " ->");
2181 move_lock_status(&q, &pos, offset);
2182
2183 if(pos >= offset+length)
2184 goto done;
2185 }
2186 }
2187done:
2188 unlock_kernel();
2189 *start = buffer;
2190 if(q-buffer < length)
2191 return (q-buffer);
2192 return length;
2193}
2194
2195/**
2196 * lock_may_read - checks that the region is free of locks
2197 * @inode: the inode that is being read
2198 * @start: the first byte to read
2199 * @len: the number of bytes to read
2200 *
2201 * Emulates Windows locking requirements. Whole-file
2202 * mandatory locks (share modes) can prohibit a read and
2203 * byte-range POSIX locks can prohibit a read if they overlap.
2204 *
2205 * N.B. this function is only ever called
2206 * from knfsd and ownership of locks is never checked.
2207 */
2208int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2209{
2210 struct file_lock *fl;
2211 int result = 1;
2212 lock_kernel();
2213 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2214 if (IS_POSIX(fl)) {
2215 if (fl->fl_type == F_RDLCK)
2216 continue;
2217 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2218 continue;
2219 } else if (IS_FLOCK(fl)) {
2220 if (!(fl->fl_type & LOCK_MAND))
2221 continue;
2222 if (fl->fl_type & LOCK_READ)
2223 continue;
2224 } else
2225 continue;
2226 result = 0;
2227 break;
2228 }
2229 unlock_kernel();
2230 return result;
2231}
2232
2233EXPORT_SYMBOL(lock_may_read);
2234
2235/**
2236 * lock_may_write - checks that the region is free of locks
2237 * @inode: the inode that is being written
2238 * @start: the first byte to write
2239 * @len: the number of bytes to write
2240 *
2241 * Emulates Windows locking requirements. Whole-file
2242 * mandatory locks (share modes) can prohibit a write and
2243 * byte-range POSIX locks can prohibit a write if they overlap.
2244 *
2245 * N.B. this function is only ever called
2246 * from knfsd and ownership of locks is never checked.
2247 */
2248int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2249{
2250 struct file_lock *fl;
2251 int result = 1;
2252 lock_kernel();
2253 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2254 if (IS_POSIX(fl)) {
2255 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2256 continue;
2257 } else if (IS_FLOCK(fl)) {
2258 if (!(fl->fl_type & LOCK_MAND))
2259 continue;
2260 if (fl->fl_type & LOCK_WRITE)
2261 continue;
2262 } else
2263 continue;
2264 result = 0;
2265 break;
2266 }
2267 unlock_kernel();
2268 return result;
2269}
2270
2271EXPORT_SYMBOL(lock_may_write);
2272
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273static int __init filelock_init(void)
2274{
2275 filelock_cache = kmem_cache_create("file_lock_cache",
2276 sizeof(struct file_lock), 0, SLAB_PANIC,
2277 init_once, NULL);
2278 return 0;
2279}
2280
2281core_initcall(filelock_init);