blob: c5ac6b40e766b54590b47e32f6d4cf69bd8ce2ab [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
Eric Dumazetfa3536c2006-03-26 01:37:24 -0800145static kmem_cache_t *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{
150 return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
151}
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 */
202static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
203{
204 struct file_lock *lock = (struct file_lock *) foo;
205
206 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) !=
207 SLAB_CTOR_CONSTRUCTOR)
208 return;
209
210 locks_init_lock(lock);
211}
212
Trond Myklebust47831f32006-03-20 13:44:05 -0500213static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
214{
215 if (fl->fl_ops) {
216 if (fl->fl_ops->fl_copy_lock)
217 fl->fl_ops->fl_copy_lock(new, fl);
218 new->fl_ops = fl->fl_ops;
219 }
220 if (fl->fl_lmops) {
221 if (fl->fl_lmops->fl_copy_lock)
222 fl->fl_lmops->fl_copy_lock(new, fl);
223 new->fl_lmops = fl->fl_lmops;
224 }
225}
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227/*
228 * Initialize a new lock from an existing file_lock structure.
229 */
Trond Myklebust09969052006-03-20 13:44:38 -0500230static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231{
232 new->fl_owner = fl->fl_owner;
233 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500234 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 new->fl_flags = fl->fl_flags;
236 new->fl_type = fl->fl_type;
237 new->fl_start = fl->fl_start;
238 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500239 new->fl_ops = NULL;
240 new->fl_lmops = NULL;
241}
242
243void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
244{
245 locks_release_private(new);
246
247 __locks_copy_lock(new, fl);
248 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 new->fl_ops = fl->fl_ops;
250 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500251
252 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
255EXPORT_SYMBOL(locks_copy_lock);
256
257static inline int flock_translate_cmd(int cmd) {
258 if (cmd & LOCK_MAND)
259 return cmd & (LOCK_MAND | LOCK_RW);
260 switch (cmd) {
261 case LOCK_SH:
262 return F_RDLCK;
263 case LOCK_EX:
264 return F_WRLCK;
265 case LOCK_UN:
266 return F_UNLCK;
267 }
268 return -EINVAL;
269}
270
271/* Fill in a file_lock structure with an appropriate FLOCK lock. */
272static int flock_make_lock(struct file *filp, struct file_lock **lock,
273 unsigned int cmd)
274{
275 struct file_lock *fl;
276 int type = flock_translate_cmd(cmd);
277 if (type < 0)
278 return type;
279
280 fl = locks_alloc_lock();
281 if (fl == NULL)
282 return -ENOMEM;
283
284 fl->fl_file = filp;
285 fl->fl_pid = current->tgid;
286 fl->fl_flags = FL_FLOCK;
287 fl->fl_type = type;
288 fl->fl_end = OFFSET_MAX;
289
290 *lock = fl;
291 return 0;
292}
293
294static int assign_type(struct file_lock *fl, int type)
295{
296 switch (type) {
297 case F_RDLCK:
298 case F_WRLCK:
299 case F_UNLCK:
300 fl->fl_type = type;
301 break;
302 default:
303 return -EINVAL;
304 }
305 return 0;
306}
307
308/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
309 * style lock.
310 */
311static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
312 struct flock *l)
313{
314 off_t start, end;
315
316 switch (l->l_whence) {
317 case 0: /*SEEK_SET*/
318 start = 0;
319 break;
320 case 1: /*SEEK_CUR*/
321 start = filp->f_pos;
322 break;
323 case 2: /*SEEK_END*/
324 start = i_size_read(filp->f_dentry->d_inode);
325 break;
326 default:
327 return -EINVAL;
328 }
329
330 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
331 POSIX-2001 defines it. */
332 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (start < 0)
334 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700335 fl->fl_end = OFFSET_MAX;
336 if (l->l_len > 0) {
337 end = start + l->l_len - 1;
338 fl->fl_end = end;
339 } else if (l->l_len < 0) {
340 end = start - 1;
341 fl->fl_end = end;
342 start += l->l_len;
343 if (start < 0)
344 return -EINVAL;
345 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700347 if (fl->fl_end < fl->fl_start)
348 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349
350 fl->fl_owner = current->files;
351 fl->fl_pid = current->tgid;
352 fl->fl_file = filp;
353 fl->fl_flags = FL_POSIX;
354 fl->fl_ops = NULL;
355 fl->fl_lmops = NULL;
356
357 return assign_type(fl, l->l_type);
358}
359
360#if BITS_PER_LONG == 32
361static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
362 struct flock64 *l)
363{
364 loff_t start;
365
366 switch (l->l_whence) {
367 case 0: /*SEEK_SET*/
368 start = 0;
369 break;
370 case 1: /*SEEK_CUR*/
371 start = filp->f_pos;
372 break;
373 case 2: /*SEEK_END*/
374 start = i_size_read(filp->f_dentry->d_inode);
375 break;
376 default:
377 return -EINVAL;
378 }
379
Trond Myklebust4c780a42005-10-18 14:20:21 -0700380 start += l->l_start;
381 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700383 fl->fl_end = OFFSET_MAX;
384 if (l->l_len > 0) {
385 fl->fl_end = start + l->l_len - 1;
386 } else if (l->l_len < 0) {
387 fl->fl_end = start - 1;
388 start += l->l_len;
389 if (start < 0)
390 return -EINVAL;
391 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700393 if (fl->fl_end < fl->fl_start)
394 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
396 fl->fl_owner = current->files;
397 fl->fl_pid = current->tgid;
398 fl->fl_file = filp;
399 fl->fl_flags = FL_POSIX;
400 fl->fl_ops = NULL;
401 fl->fl_lmops = NULL;
402
403 switch (l->l_type) {
404 case F_RDLCK:
405 case F_WRLCK:
406 case F_UNLCK:
407 fl->fl_type = l->l_type;
408 break;
409 default:
410 return -EINVAL;
411 }
412
413 return (0);
414}
415#endif
416
417/* default lease lock manager operations */
418static void lease_break_callback(struct file_lock *fl)
419{
420 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
421}
422
423static void lease_release_private_callback(struct file_lock *fl)
424{
425 if (!fl->fl_file)
426 return;
427
428 f_delown(fl->fl_file);
429 fl->fl_file->f_owner.signum = 0;
430}
431
Adrian Bunk75c96f82005-05-05 16:16:09 -0700432static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 return fl->fl_file == try->fl_file;
435}
436
Adrian Bunk75c96f82005-05-05 16:16:09 -0700437static struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 .fl_break = lease_break_callback,
439 .fl_release_private = lease_release_private_callback,
440 .fl_mylease = lease_mylease_callback,
441 .fl_change = lease_modify,
442};
443
444/*
445 * Initialize a lease, use the default lock manager operations
446 */
447static int lease_init(struct file *filp, int type, struct file_lock *fl)
448 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400449 if (assign_type(fl, type) != 0)
450 return -EINVAL;
451
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 fl->fl_owner = current->files;
453 fl->fl_pid = current->tgid;
454
455 fl->fl_file = filp;
456 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 fl->fl_start = 0;
458 fl->fl_end = OFFSET_MAX;
459 fl->fl_ops = NULL;
460 fl->fl_lmops = &lease_manager_ops;
461 return 0;
462}
463
464/* Allocate a file_lock initialised to this type of lease */
465static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
466{
467 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400468 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
470 if (fl == NULL)
Trond Myklebust75dff552006-05-07 23:02:42 -0400471 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400474 if (error) {
475 locks_free_lock(fl);
476 fl = NULL;
477 }
478out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 *flp = fl;
Trond Myklebust75dff552006-05-07 23:02:42 -0400480 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481}
482
483/* Check if two locks overlap each other.
484 */
485static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
486{
487 return ((fl1->fl_end >= fl2->fl_start) &&
488 (fl2->fl_end >= fl1->fl_start));
489}
490
491/*
492 * Check whether two locks have the same owner.
493 */
Matt Mackall33443c42006-01-08 01:05:22 -0800494static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495{
496 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
497 return fl2->fl_lmops == fl1->fl_lmops &&
498 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
499 return fl1->fl_owner == fl2->fl_owner;
500}
501
502/* Remove waiter from blocker's block list.
503 * When blocker ends up pointing to itself then the list is empty.
504 */
Matt Mackall33443c42006-01-08 01:05:22 -0800505static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506{
507 list_del_init(&waiter->fl_block);
508 list_del_init(&waiter->fl_link);
509 waiter->fl_next = NULL;
510}
511
512/*
513 */
514static void locks_delete_block(struct file_lock *waiter)
515{
516 lock_kernel();
517 __locks_delete_block(waiter);
518 unlock_kernel();
519}
520
521/* Insert waiter into blocker's block list.
522 * We use a circular list so that processes can be easily woken up in
523 * the order they blocked. The documentation doesn't require this but
524 * it seems like the reasonable thing to do.
525 */
526static void locks_insert_block(struct file_lock *blocker,
527 struct file_lock *waiter)
528{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800529 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 list_add_tail(&waiter->fl_block, &blocker->fl_block);
531 waiter->fl_next = blocker;
532 if (IS_POSIX(blocker))
533 list_add(&waiter->fl_link, &blocked_list);
534}
535
536/* Wake up processes blocked waiting for blocker.
537 * If told to wait then schedule the processes until the block list
538 * is empty, otherwise empty the block list ourselves.
539 */
540static void locks_wake_up_blocks(struct file_lock *blocker)
541{
542 while (!list_empty(&blocker->fl_block)) {
543 struct file_lock *waiter = list_entry(blocker->fl_block.next,
544 struct file_lock, fl_block);
545 __locks_delete_block(waiter);
546 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
547 waiter->fl_lmops->fl_notify(waiter);
548 else
549 wake_up(&waiter->fl_wait);
550 }
551}
552
553/* Insert file lock fl into an inode's lock list at the position indicated
554 * by pos. At the same time add the lock to the global file lock list.
555 */
556static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
557{
558 list_add(&fl->fl_link, &file_lock_list);
559
560 /* insert into file's list */
561 fl->fl_next = *pos;
562 *pos = fl;
563
564 if (fl->fl_ops && fl->fl_ops->fl_insert)
565 fl->fl_ops->fl_insert(fl);
566}
567
568/*
569 * Delete a lock and then free it.
570 * Wake up processes that are blocked waiting for this lock,
571 * notify the FS that the lock has been cleared and
572 * finally free the lock.
573 */
574static void locks_delete_lock(struct file_lock **thisfl_p)
575{
576 struct file_lock *fl = *thisfl_p;
577
578 *thisfl_p = fl->fl_next;
579 fl->fl_next = NULL;
580 list_del_init(&fl->fl_link);
581
582 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
583 if (fl->fl_fasync != NULL) {
584 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
585 fl->fl_fasync = NULL;
586 }
587
588 if (fl->fl_ops && fl->fl_ops->fl_remove)
589 fl->fl_ops->fl_remove(fl);
590
591 locks_wake_up_blocks(fl);
592 locks_free_lock(fl);
593}
594
595/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
596 * checks for shared/exclusive status of overlapping locks.
597 */
598static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
599{
600 if (sys_fl->fl_type == F_WRLCK)
601 return 1;
602 if (caller_fl->fl_type == F_WRLCK)
603 return 1;
604 return 0;
605}
606
607/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
608 * checking before calling the locks_conflict().
609 */
610static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
611{
612 /* POSIX locks owned by the same process do not conflict with
613 * each other.
614 */
615 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
616 return (0);
617
618 /* Check whether they overlap */
619 if (!locks_overlap(caller_fl, sys_fl))
620 return 0;
621
622 return (locks_conflict(caller_fl, sys_fl));
623}
624
625/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
626 * checking before calling the locks_conflict().
627 */
628static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
629{
630 /* FLOCK locks referring to the same filp do not conflict with
631 * each other.
632 */
633 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
634 return (0);
635 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
636 return 0;
637
638 return (locks_conflict(caller_fl, sys_fl));
639}
640
641static int interruptible_sleep_on_locked(wait_queue_head_t *fl_wait, int timeout)
642{
643 int result = 0;
644 DECLARE_WAITQUEUE(wait, current);
645
646 __set_current_state(TASK_INTERRUPTIBLE);
647 add_wait_queue(fl_wait, &wait);
648 if (timeout == 0)
649 schedule();
650 else
651 result = schedule_timeout(timeout);
652 if (signal_pending(current))
653 result = -ERESTARTSYS;
654 remove_wait_queue(fl_wait, &wait);
655 __set_current_state(TASK_RUNNING);
656 return result;
657}
658
659static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *waiter, int time)
660{
661 int result;
662 locks_insert_block(blocker, waiter);
663 result = interruptible_sleep_on_locked(&waiter->fl_wait, time);
664 __locks_delete_block(waiter);
665 return result;
666}
667
Andy Adamson8dc7c312006-03-20 13:44:26 -0500668int
669posix_test_lock(struct file *filp, struct file_lock *fl,
670 struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671{
672 struct file_lock *cfl;
673
674 lock_kernel();
675 for (cfl = filp->f_dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
676 if (!IS_POSIX(cfl))
677 continue;
678 if (posix_locks_conflict(cfl, fl))
679 break;
680 }
Andy Adamson8dc7c312006-03-20 13:44:26 -0500681 if (cfl) {
Trond Myklebust09969052006-03-20 13:44:38 -0500682 __locks_copy_lock(conflock, cfl);
Andy Adamson8dc7c312006-03-20 13:44:26 -0500683 unlock_kernel();
684 return 1;
685 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 unlock_kernel();
Andy Adamson8dc7c312006-03-20 13:44:26 -0500687 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688}
689
690EXPORT_SYMBOL(posix_test_lock);
691
692/* This function tests for deadlock condition before putting a process to
693 * sleep. The detection scheme is no longer recursive. Recursive was neat,
694 * but dangerous - we risked stack corruption if the lock data was bad, or
695 * if the recursion was too deep for any other reason.
696 *
697 * We rely on the fact that a task can only be on one lock's wait queue
698 * at a time. When we find blocked_task on a wait queue we can re-search
699 * with blocked_task equal to that queue's owner, until either blocked_task
700 * isn't found, or blocked_task is found on a queue owned by my_task.
701 *
702 * Note: the above assumption may not be true when handling lock requests
703 * from a broken NFS client. But broken NFS clients have a lot more to
704 * worry about than proper deadlock detection anyway... --okir
705 */
706int posix_locks_deadlock(struct file_lock *caller_fl,
707 struct file_lock *block_fl)
708{
709 struct list_head *tmp;
710
711next_task:
712 if (posix_same_owner(caller_fl, block_fl))
713 return 1;
714 list_for_each(tmp, &blocked_list) {
715 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
716 if (posix_same_owner(fl, block_fl)) {
717 fl = fl->fl_next;
718 block_fl = fl;
719 goto next_task;
720 }
721 }
722 return 0;
723}
724
725EXPORT_SYMBOL(posix_locks_deadlock);
726
727/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
728 * at the head of the list, but that's secret knowledge known only to
729 * flock_lock_file and posix_lock_file.
730 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800731static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800733 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 struct file_lock **before;
735 struct inode * inode = filp->f_dentry->d_inode;
736 int error = 0;
737 int found = 0;
738
739 lock_kernel();
740 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 Myklebust993dfa82006-03-31 02:30:55 -0800755 if (request->fl_type == F_UNLCK)
756 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400758 error = -ENOMEM;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800759 new_fl = locks_alloc_lock();
760 if (new_fl == NULL)
761 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 /*
763 * If a higher-priority process was blocked on the old file lock,
764 * give it the opportunity to lock the file.
765 */
766 if (found)
767 cond_resched();
768
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 for_each_lock(inode, before) {
770 struct file_lock *fl = *before;
771 if (IS_POSIX(fl))
772 break;
773 if (IS_LEASE(fl))
774 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800775 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 continue;
777 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800778 if (request->fl_flags & FL_SLEEP)
779 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 goto out;
781 }
Trond Myklebust993dfa82006-03-31 02:30:55 -0800782 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800784 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400785 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787out:
788 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800789 if (new_fl)
790 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 return error;
792}
793
Andy Adamson5842add2006-03-26 01:37:26 -0800794static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795{
796 struct file_lock *fl;
797 struct file_lock *new_fl, *new_fl2;
798 struct file_lock *left = NULL;
799 struct file_lock *right = NULL;
800 struct file_lock **before;
801 int error, added = 0;
802
803 /*
804 * We may need two file_lock structures for this operation,
805 * so we get them in advance to avoid races.
806 */
807 new_fl = locks_alloc_lock();
808 new_fl2 = locks_alloc_lock();
809
810 lock_kernel();
811 if (request->fl_type != F_UNLCK) {
812 for_each_lock(inode, before) {
813 struct file_lock *fl = *before;
814 if (!IS_POSIX(fl))
815 continue;
816 if (!posix_locks_conflict(request, fl))
817 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800818 if (conflock)
819 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 error = -EAGAIN;
821 if (!(request->fl_flags & FL_SLEEP))
822 goto out;
823 error = -EDEADLK;
824 if (posix_locks_deadlock(request, fl))
825 goto out;
826 error = -EAGAIN;
827 locks_insert_block(fl, request);
828 goto out;
829 }
830 }
831
832 /* If we're just looking for a conflict, we're done. */
833 error = 0;
834 if (request->fl_flags & FL_ACCESS)
835 goto out;
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 * Find the first old lock with the same owner as the new lock.
839 */
840
841 before = &inode->i_flock;
842
843 /* First skip locks owned by other processes. */
844 while ((fl = *before) && (!IS_POSIX(fl) ||
845 !posix_same_owner(request, fl))) {
846 before = &fl->fl_next;
847 }
848
849 /* Process locks with this owner. */
850 while ((fl = *before) && posix_same_owner(request, fl)) {
851 /* Detect adjacent or overlapping regions (if same lock type)
852 */
853 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700854 /* In all comparisons of start vs end, use
855 * "start - 1" rather than "end + 1". If end
856 * is OFFSET_MAX, end + 1 will become negative.
857 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 if (fl->fl_end < request->fl_start - 1)
859 goto next_lock;
860 /* If the next lock in the list has entirely bigger
861 * addresses than the new one, insert the lock here.
862 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700863 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 break;
865
866 /* If we come here, the new and old lock are of the
867 * same type and adjacent or overlapping. Make one
868 * lock yielding from the lower start address of both
869 * locks to the higher end address.
870 */
871 if (fl->fl_start > request->fl_start)
872 fl->fl_start = request->fl_start;
873 else
874 request->fl_start = fl->fl_start;
875 if (fl->fl_end < request->fl_end)
876 fl->fl_end = request->fl_end;
877 else
878 request->fl_end = fl->fl_end;
879 if (added) {
880 locks_delete_lock(before);
881 continue;
882 }
883 request = fl;
884 added = 1;
885 }
886 else {
887 /* Processing for different lock types is a bit
888 * more complex.
889 */
890 if (fl->fl_end < request->fl_start)
891 goto next_lock;
892 if (fl->fl_start > request->fl_end)
893 break;
894 if (request->fl_type == F_UNLCK)
895 added = 1;
896 if (fl->fl_start < request->fl_start)
897 left = fl;
898 /* If the next lock in the list has a higher end
899 * address than the new one, insert the new one here.
900 */
901 if (fl->fl_end > request->fl_end) {
902 right = fl;
903 break;
904 }
905 if (fl->fl_start >= request->fl_start) {
906 /* The new lock completely replaces an old
907 * one (This may happen several times).
908 */
909 if (added) {
910 locks_delete_lock(before);
911 continue;
912 }
913 /* Replace the old lock with the new one.
914 * Wake up anybody waiting for the old one,
915 * as the change in lock type might satisfy
916 * their needs.
917 */
918 locks_wake_up_blocks(fl);
919 fl->fl_start = request->fl_start;
920 fl->fl_end = request->fl_end;
921 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500922 locks_release_private(fl);
923 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 request = fl;
925 added = 1;
926 }
927 }
928 /* Go on to next lock.
929 */
930 next_lock:
931 before = &fl->fl_next;
932 }
933
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700934 /*
935 * The above code only modifies existing locks in case of
936 * merging or replacing. If new lock(s) need to be inserted
937 * all modifications are done bellow this, so it's safe yet to
938 * bail out.
939 */
940 error = -ENOLCK; /* "no luck" */
941 if (right && left == right && !new_fl2)
942 goto out;
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 error = 0;
945 if (!added) {
946 if (request->fl_type == F_UNLCK)
947 goto out;
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700948
949 if (!new_fl) {
950 error = -ENOLCK;
951 goto out;
952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 locks_copy_lock(new_fl, request);
954 locks_insert_lock(before, new_fl);
955 new_fl = NULL;
956 }
957 if (right) {
958 if (left == right) {
959 /* The new lock breaks the old one in two pieces,
960 * so we have to use the second new lock.
961 */
962 left = new_fl2;
963 new_fl2 = NULL;
964 locks_copy_lock(left, right);
965 locks_insert_lock(before, left);
966 }
967 right->fl_start = request->fl_end + 1;
968 locks_wake_up_blocks(right);
969 }
970 if (left) {
971 left->fl_end = request->fl_start - 1;
972 locks_wake_up_blocks(left);
973 }
974 out:
975 unlock_kernel();
976 /*
977 * Free any unused locks.
978 */
979 if (new_fl)
980 locks_free_lock(new_fl);
981 if (new_fl2)
982 locks_free_lock(new_fl2);
983 return error;
984}
985
986/**
987 * posix_lock_file - Apply a POSIX-style lock to a file
988 * @filp: The file to apply the lock to
989 * @fl: The lock to be applied
990 *
991 * Add a POSIX style lock to a file.
992 * We merge adjacent & overlapping locks whenever possible.
993 * POSIX locks are sorted by owner task, then by starting address
994 */
995int posix_lock_file(struct file *filp, struct file_lock *fl)
996{
Andy Adamson5842add2006-03-26 01:37:26 -0800997 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998}
Andy Adamson5842add2006-03-26 01:37:26 -0800999EXPORT_SYMBOL(posix_lock_file);
1000
1001/**
1002 * posix_lock_file_conf - Apply a POSIX-style lock to a file
1003 * @filp: The file to apply the lock to
1004 * @fl: The lock to be applied
1005 * @conflock: Place to return a copy of the conflicting lock, if found.
1006 *
1007 * Except for the conflock parameter, acts just like posix_lock_file.
1008 */
1009int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1010 struct file_lock *conflock)
1011{
1012 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1013}
1014EXPORT_SYMBOL(posix_lock_file_conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016/**
1017 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1018 * @filp: The file to apply the lock to
1019 * @fl: The lock to be applied
1020 *
1021 * Add a POSIX style lock to a file.
1022 * We merge adjacent & overlapping locks whenever possible.
1023 * POSIX locks are sorted by owner task, then by starting address
1024 */
1025int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1026{
1027 int error;
1028 might_sleep ();
1029 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001030 error = posix_lock_file(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1032 break;
1033 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1034 if (!error)
1035 continue;
1036
1037 locks_delete_block(fl);
1038 break;
1039 }
1040 return error;
1041}
1042EXPORT_SYMBOL(posix_lock_file_wait);
1043
1044/**
1045 * locks_mandatory_locked - Check for an active lock
1046 * @inode: the file to check
1047 *
1048 * Searches the inode's list of locks to find any POSIX locks which conflict.
1049 * This function is called from locks_verify_locked() only.
1050 */
1051int locks_mandatory_locked(struct inode *inode)
1052{
1053 fl_owner_t owner = current->files;
1054 struct file_lock *fl;
1055
1056 /*
1057 * Search the lock list for this inode for any POSIX locks.
1058 */
1059 lock_kernel();
1060 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1061 if (!IS_POSIX(fl))
1062 continue;
1063 if (fl->fl_owner != owner)
1064 break;
1065 }
1066 unlock_kernel();
1067 return fl ? -EAGAIN : 0;
1068}
1069
1070/**
1071 * locks_mandatory_area - Check for a conflicting lock
1072 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1073 * for shared
1074 * @inode: the file to check
1075 * @filp: how the file was opened (if it was)
1076 * @offset: start of area to check
1077 * @count: length of area to check
1078 *
1079 * Searches the inode's list of locks to find any POSIX locks which conflict.
1080 * This function is called from rw_verify_area() and
1081 * locks_verify_truncate().
1082 */
1083int locks_mandatory_area(int read_write, struct inode *inode,
1084 struct file *filp, loff_t offset,
1085 size_t count)
1086{
1087 struct file_lock fl;
1088 int error;
1089
1090 locks_init_lock(&fl);
1091 fl.fl_owner = current->files;
1092 fl.fl_pid = current->tgid;
1093 fl.fl_file = filp;
1094 fl.fl_flags = FL_POSIX | FL_ACCESS;
1095 if (filp && !(filp->f_flags & O_NONBLOCK))
1096 fl.fl_flags |= FL_SLEEP;
1097 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1098 fl.fl_start = offset;
1099 fl.fl_end = offset + count - 1;
1100
1101 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001102 error = __posix_lock_file_conf(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 if (error != -EAGAIN)
1104 break;
1105 if (!(fl.fl_flags & FL_SLEEP))
1106 break;
1107 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1108 if (!error) {
1109 /*
1110 * If we've been sleeping someone might have
1111 * changed the permissions behind our back.
1112 */
1113 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1114 continue;
1115 }
1116
1117 locks_delete_block(&fl);
1118 break;
1119 }
1120
1121 return error;
1122}
1123
1124EXPORT_SYMBOL(locks_mandatory_area);
1125
1126/* We already had a lease on this file; just change its type */
1127int lease_modify(struct file_lock **before, int arg)
1128{
1129 struct file_lock *fl = *before;
1130 int error = assign_type(fl, arg);
1131
1132 if (error)
1133 return error;
1134 locks_wake_up_blocks(fl);
1135 if (arg == F_UNLCK)
1136 locks_delete_lock(before);
1137 return 0;
1138}
1139
1140EXPORT_SYMBOL(lease_modify);
1141
1142static void time_out_leases(struct inode *inode)
1143{
1144 struct file_lock **before;
1145 struct file_lock *fl;
1146
1147 before = &inode->i_flock;
1148 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1149 if ((fl->fl_break_time == 0)
1150 || time_before(jiffies, fl->fl_break_time)) {
1151 before = &fl->fl_next;
1152 continue;
1153 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1155 if (fl == *before) /* lease_modify may have freed fl */
1156 before = &fl->fl_next;
1157 }
1158}
1159
1160/**
1161 * __break_lease - revoke all outstanding leases on file
1162 * @inode: the inode of the file to return
1163 * @mode: the open mode (read or write)
1164 *
1165 * break_lease (inlined for speed) has checked there already
1166 * is a lease on this file. Leases are broken on a call to open()
1167 * or truncate(). This function can sleep unless you
1168 * specified %O_NONBLOCK to your open().
1169 */
1170int __break_lease(struct inode *inode, unsigned int mode)
1171{
1172 int error = 0, future;
1173 struct file_lock *new_fl, *flock;
1174 struct file_lock *fl;
1175 int alloc_err;
1176 unsigned long break_time;
1177 int i_have_this_lease = 0;
1178
1179 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1180 &new_fl);
1181
1182 lock_kernel();
1183
1184 time_out_leases(inode);
1185
1186 flock = inode->i_flock;
1187 if ((flock == NULL) || !IS_LEASE(flock))
1188 goto out;
1189
1190 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1191 if (fl->fl_owner == current->files)
1192 i_have_this_lease = 1;
1193
1194 if (mode & FMODE_WRITE) {
1195 /* If we want write access, we have to revoke any lease. */
1196 future = F_UNLCK | F_INPROGRESS;
1197 } else if (flock->fl_type & F_INPROGRESS) {
1198 /* If the lease is already being broken, we just leave it */
1199 future = flock->fl_type;
1200 } else if (flock->fl_type & F_WRLCK) {
1201 /* Downgrade the exclusive lease to a read-only lease. */
1202 future = F_RDLCK | F_INPROGRESS;
1203 } else {
1204 /* the existing lease was read-only, so we can read too. */
1205 goto out;
1206 }
1207
1208 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1209 error = alloc_err;
1210 goto out;
1211 }
1212
1213 break_time = 0;
1214 if (lease_break_time > 0) {
1215 break_time = jiffies + lease_break_time * HZ;
1216 if (break_time == 0)
1217 break_time++; /* so that 0 means no break time */
1218 }
1219
1220 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1221 if (fl->fl_type != future) {
1222 fl->fl_type = future;
1223 fl->fl_break_time = break_time;
1224 /* lease must have lmops break callback */
1225 fl->fl_lmops->fl_break(fl);
1226 }
1227 }
1228
1229 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1230 error = -EWOULDBLOCK;
1231 goto out;
1232 }
1233
1234restart:
1235 break_time = flock->fl_break_time;
1236 if (break_time != 0) {
1237 break_time -= jiffies;
1238 if (break_time == 0)
1239 break_time++;
1240 }
1241 error = locks_block_on_timeout(flock, new_fl, break_time);
1242 if (error >= 0) {
1243 if (error == 0)
1244 time_out_leases(inode);
1245 /* Wait for the next lease that has not been broken yet */
1246 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1247 flock = flock->fl_next) {
1248 if (flock->fl_type & F_INPROGRESS)
1249 goto restart;
1250 }
1251 error = 0;
1252 }
1253
1254out:
1255 unlock_kernel();
1256 if (!alloc_err)
1257 locks_free_lock(new_fl);
1258 return error;
1259}
1260
1261EXPORT_SYMBOL(__break_lease);
1262
1263/**
1264 * lease_get_mtime
1265 * @inode: the inode
1266 * @time: pointer to a timespec which will contain the last modified time
1267 *
1268 * This is to force NFS clients to flush their caches for files with
1269 * exclusive leases. The justification is that if someone has an
1270 * exclusive lease, then they could be modifiying it.
1271 */
1272void lease_get_mtime(struct inode *inode, struct timespec *time)
1273{
1274 struct file_lock *flock = inode->i_flock;
1275 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1276 *time = current_fs_time(inode->i_sb);
1277 else
1278 *time = inode->i_mtime;
1279}
1280
1281EXPORT_SYMBOL(lease_get_mtime);
1282
1283/**
1284 * fcntl_getlease - Enquire what lease is currently active
1285 * @filp: the file
1286 *
1287 * The value returned by this function will be one of
1288 * (if no lease break is pending):
1289 *
1290 * %F_RDLCK to indicate a shared lease is held.
1291 *
1292 * %F_WRLCK to indicate an exclusive lease is held.
1293 *
1294 * %F_UNLCK to indicate no lease is held.
1295 *
1296 * (if a lease break is pending):
1297 *
1298 * %F_RDLCK to indicate an exclusive lease needs to be
1299 * changed to a shared lease (or removed).
1300 *
1301 * %F_UNLCK to indicate the lease needs to be removed.
1302 *
1303 * XXX: sfr & willy disagree over whether F_INPROGRESS
1304 * should be returned to userspace.
1305 */
1306int fcntl_getlease(struct file *filp)
1307{
1308 struct file_lock *fl;
1309 int type = F_UNLCK;
1310
1311 lock_kernel();
1312 time_out_leases(filp->f_dentry->d_inode);
1313 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1314 fl = fl->fl_next) {
1315 if (fl->fl_file == filp) {
1316 type = fl->fl_type & ~F_INPROGRESS;
1317 break;
1318 }
1319 }
1320 unlock_kernel();
1321 return type;
1322}
1323
1324/**
1325 * __setlease - sets a lease on an open file
1326 * @filp: file pointer
1327 * @arg: type of lease to obtain
1328 * @flp: input - file_lock to use, output - file_lock inserted
1329 *
1330 * The (input) flp->fl_lmops->fl_break function is required
1331 * by break_lease().
1332 *
1333 * Called with kernel lock held.
1334 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001335static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001337 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 struct dentry *dentry = filp->f_dentry;
1339 struct inode *inode = dentry->d_inode;
1340 int error, rdlease_count = 0, wrlease_count = 0;
1341
1342 time_out_leases(inode);
1343
1344 error = -EINVAL;
1345 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1346 goto out;
1347
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001348 lease = *flp;
1349
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 error = -EAGAIN;
1351 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1352 goto out;
1353 if ((arg == F_WRLCK)
1354 && ((atomic_read(&dentry->d_count) > 1)
1355 || (atomic_read(&inode->i_count) > 1)))
1356 goto out;
1357
1358 /*
1359 * At this point, we know that if there is an exclusive
1360 * lease on this file, then we hold it on this filp
1361 * (otherwise our open of this file would have blocked).
1362 * And if we are trying to acquire an exclusive lease,
1363 * then the file is not open by anyone (including us)
1364 * except for this filp.
1365 */
1366 for (before = &inode->i_flock;
1367 ((fl = *before) != NULL) && IS_LEASE(fl);
1368 before = &fl->fl_next) {
1369 if (lease->fl_lmops->fl_mylease(fl, lease))
1370 my_before = before;
1371 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1372 /*
1373 * Someone is in the process of opening this
1374 * file for writing so we may not take an
1375 * exclusive lease on it.
1376 */
1377 wrlease_count++;
1378 else
1379 rdlease_count++;
1380 }
1381
1382 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1383 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1384 goto out;
1385
1386 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001387 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 error = lease->fl_lmops->fl_change(my_before, arg);
1389 goto out;
1390 }
1391
1392 error = 0;
1393 if (arg == F_UNLCK)
1394 goto out;
1395
1396 error = -EINVAL;
1397 if (!leases_enable)
1398 goto out;
1399
1400 error = lease_alloc(filp, arg, &fl);
1401 if (error)
1402 goto out;
1403
1404 locks_copy_lock(fl, lease);
1405
1406 locks_insert_lock(before, fl);
1407
1408 *flp = fl;
1409out:
1410 return error;
1411}
1412
1413 /**
1414 * setlease - sets a lease on an open file
1415 * @filp: file pointer
1416 * @arg: type of lease to obtain
1417 * @lease: file_lock to use
1418 *
1419 * Call this to establish a lease on the file.
1420 * The fl_lmops fl_break function is required by break_lease
1421 */
1422
1423int setlease(struct file *filp, long arg, struct file_lock **lease)
1424{
1425 struct dentry *dentry = filp->f_dentry;
1426 struct inode *inode = dentry->d_inode;
1427 int error;
1428
1429 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1430 return -EACCES;
1431 if (!S_ISREG(inode->i_mode))
1432 return -EINVAL;
1433 error = security_file_lock(filp, arg);
1434 if (error)
1435 return error;
1436
1437 lock_kernel();
1438 error = __setlease(filp, arg, lease);
1439 unlock_kernel();
1440
1441 return error;
1442}
1443
1444EXPORT_SYMBOL(setlease);
1445
1446/**
1447 * fcntl_setlease - sets a lease on an open file
1448 * @fd: open file descriptor
1449 * @filp: file pointer
1450 * @arg: type of lease to obtain
1451 *
1452 * Call this fcntl to establish a lease on the file.
1453 * Note that you also need to call %F_SETSIG to
1454 * receive a signal when the lease is broken.
1455 */
1456int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1457{
1458 struct file_lock fl, *flp = &fl;
1459 struct dentry *dentry = filp->f_dentry;
1460 struct inode *inode = dentry->d_inode;
1461 int error;
1462
1463 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1464 return -EACCES;
1465 if (!S_ISREG(inode->i_mode))
1466 return -EINVAL;
1467 error = security_file_lock(filp, arg);
1468 if (error)
1469 return error;
1470
1471 locks_init_lock(&fl);
1472 error = lease_init(filp, arg, &fl);
1473 if (error)
1474 return error;
1475
1476 lock_kernel();
1477
1478 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001479 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 goto out_unlock;
1481
1482 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1483 if (error < 0) {
1484 /* remove lease just inserted by __setlease */
1485 flp->fl_type = F_UNLCK | F_INPROGRESS;
1486 flp->fl_break_time = jiffies- 10;
1487 time_out_leases(inode);
1488 goto out_unlock;
1489 }
1490
1491 error = f_setown(filp, current->pid, 0);
1492out_unlock:
1493 unlock_kernel();
1494 return error;
1495}
1496
1497/**
1498 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1499 * @filp: The file to apply the lock to
1500 * @fl: The lock to be applied
1501 *
1502 * Add a FLOCK style lock to a file.
1503 */
1504int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1505{
1506 int error;
1507 might_sleep();
1508 for (;;) {
1509 error = flock_lock_file(filp, fl);
1510 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1511 break;
1512 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1513 if (!error)
1514 continue;
1515
1516 locks_delete_block(fl);
1517 break;
1518 }
1519 return error;
1520}
1521
1522EXPORT_SYMBOL(flock_lock_file_wait);
1523
1524/**
1525 * sys_flock: - flock() system call.
1526 * @fd: the file descriptor to lock.
1527 * @cmd: the type of lock to apply.
1528 *
1529 * Apply a %FL_FLOCK style lock to an open file descriptor.
1530 * The @cmd can be one of
1531 *
1532 * %LOCK_SH -- a shared lock.
1533 *
1534 * %LOCK_EX -- an exclusive lock.
1535 *
1536 * %LOCK_UN -- remove an existing lock.
1537 *
1538 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1539 *
1540 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1541 * processes read and write access respectively.
1542 */
1543asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1544{
1545 struct file *filp;
1546 struct file_lock *lock;
1547 int can_sleep, unlock;
1548 int error;
1549
1550 error = -EBADF;
1551 filp = fget(fd);
1552 if (!filp)
1553 goto out;
1554
1555 can_sleep = !(cmd & LOCK_NB);
1556 cmd &= ~LOCK_NB;
1557 unlock = (cmd == LOCK_UN);
1558
1559 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1560 goto out_putf;
1561
1562 error = flock_make_lock(filp, &lock, cmd);
1563 if (error)
1564 goto out_putf;
1565 if (can_sleep)
1566 lock->fl_flags |= FL_SLEEP;
1567
1568 error = security_file_lock(filp, cmd);
1569 if (error)
1570 goto out_free;
1571
1572 if (filp->f_op && filp->f_op->flock)
1573 error = filp->f_op->flock(filp,
1574 (can_sleep) ? F_SETLKW : F_SETLK,
1575 lock);
1576 else
1577 error = flock_lock_file_wait(filp, lock);
1578
1579 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001580 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 out_putf:
1583 fput(filp);
1584 out:
1585 return error;
1586}
1587
1588/* Report the first existing lock that would conflict with l.
1589 * This implements the F_GETLK command of fcntl().
1590 */
1591int fcntl_getlk(struct file *filp, struct flock __user *l)
1592{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001593 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 struct flock flock;
1595 int error;
1596
1597 error = -EFAULT;
1598 if (copy_from_user(&flock, l, sizeof(flock)))
1599 goto out;
1600 error = -EINVAL;
1601 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1602 goto out;
1603
1604 error = flock_to_posix_lock(filp, &file_lock, &flock);
1605 if (error)
1606 goto out;
1607
1608 if (filp->f_op && filp->f_op->lock) {
1609 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001610 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1611 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (error < 0)
1613 goto out;
1614 else
1615 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1616 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001617 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 }
1619
1620 flock.l_type = F_UNLCK;
1621 if (fl != NULL) {
1622 flock.l_pid = fl->fl_pid;
1623#if BITS_PER_LONG == 32
1624 /*
1625 * Make sure we can represent the posix lock via
1626 * legacy 32bit flock.
1627 */
1628 error = -EOVERFLOW;
1629 if (fl->fl_start > OFFT_OFFSET_MAX)
1630 goto out;
1631 if ((fl->fl_end != OFFSET_MAX)
1632 && (fl->fl_end > OFFT_OFFSET_MAX))
1633 goto out;
1634#endif
1635 flock.l_start = fl->fl_start;
1636 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1637 fl->fl_end - fl->fl_start + 1;
1638 flock.l_whence = 0;
1639 flock.l_type = fl->fl_type;
1640 }
1641 error = -EFAULT;
1642 if (!copy_to_user(l, &flock, sizeof(flock)))
1643 error = 0;
1644out:
1645 return error;
1646}
1647
1648/* Apply the lock described by l to an open file descriptor.
1649 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1650 */
Peter Staubachc2936212005-07-27 11:45:09 -07001651int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1652 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653{
1654 struct file_lock *file_lock = locks_alloc_lock();
1655 struct flock flock;
1656 struct inode *inode;
1657 int error;
1658
1659 if (file_lock == NULL)
1660 return -ENOLCK;
1661
1662 /*
1663 * This might block, so we do it before checking the inode.
1664 */
1665 error = -EFAULT;
1666 if (copy_from_user(&flock, l, sizeof(flock)))
1667 goto out;
1668
1669 inode = filp->f_dentry->d_inode;
1670
1671 /* Don't allow mandatory locks on files that may be memory mapped
1672 * and shared.
1673 */
1674 if (IS_MANDLOCK(inode) &&
1675 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1676 mapping_writably_mapped(filp->f_mapping)) {
1677 error = -EAGAIN;
1678 goto out;
1679 }
1680
Peter Staubachc2936212005-07-27 11:45:09 -07001681again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 error = flock_to_posix_lock(filp, file_lock, &flock);
1683 if (error)
1684 goto out;
1685 if (cmd == F_SETLKW) {
1686 file_lock->fl_flags |= FL_SLEEP;
1687 }
1688
1689 error = -EBADF;
1690 switch (flock.l_type) {
1691 case F_RDLCK:
1692 if (!(filp->f_mode & FMODE_READ))
1693 goto out;
1694 break;
1695 case F_WRLCK:
1696 if (!(filp->f_mode & FMODE_WRITE))
1697 goto out;
1698 break;
1699 case F_UNLCK:
1700 break;
1701 default:
1702 error = -EINVAL;
1703 goto out;
1704 }
1705
1706 error = security_file_lock(filp, file_lock->fl_type);
1707 if (error)
1708 goto out;
1709
Peter Staubachc2936212005-07-27 11:45:09 -07001710 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001712 else {
1713 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001714 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001715 if ((error != -EAGAIN) || (cmd == F_SETLK))
1716 break;
1717 error = wait_event_interruptible(file_lock->fl_wait,
1718 !file_lock->fl_next);
1719 if (!error)
1720 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
Peter Staubachc2936212005-07-27 11:45:09 -07001722 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001724 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 }
1726
Peter Staubachc2936212005-07-27 11:45:09 -07001727 /*
1728 * Attempt to detect a close/fcntl race and recover by
1729 * releasing the lock that was just acquired.
1730 */
1731 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1732 flock.l_type = F_UNLCK;
1733 goto again;
1734 }
1735
1736out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 locks_free_lock(file_lock);
1738 return error;
1739}
1740
1741#if BITS_PER_LONG == 32
1742/* Report the first existing lock that would conflict with l.
1743 * This implements the F_GETLK command of fcntl().
1744 */
1745int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1746{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001747 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 struct flock64 flock;
1749 int error;
1750
1751 error = -EFAULT;
1752 if (copy_from_user(&flock, l, sizeof(flock)))
1753 goto out;
1754 error = -EINVAL;
1755 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1756 goto out;
1757
1758 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1759 if (error)
1760 goto out;
1761
1762 if (filp->f_op && filp->f_op->lock) {
1763 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001764 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1765 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (error < 0)
1767 goto out;
1768 else
1769 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1770 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001771 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772 }
1773
1774 flock.l_type = F_UNLCK;
1775 if (fl != NULL) {
1776 flock.l_pid = fl->fl_pid;
1777 flock.l_start = fl->fl_start;
1778 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1779 fl->fl_end - fl->fl_start + 1;
1780 flock.l_whence = 0;
1781 flock.l_type = fl->fl_type;
1782 }
1783 error = -EFAULT;
1784 if (!copy_to_user(l, &flock, sizeof(flock)))
1785 error = 0;
1786
1787out:
1788 return error;
1789}
1790
1791/* Apply the lock described by l to an open file descriptor.
1792 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1793 */
Peter Staubachc2936212005-07-27 11:45:09 -07001794int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1795 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796{
1797 struct file_lock *file_lock = locks_alloc_lock();
1798 struct flock64 flock;
1799 struct inode *inode;
1800 int error;
1801
1802 if (file_lock == NULL)
1803 return -ENOLCK;
1804
1805 /*
1806 * This might block, so we do it before checking the inode.
1807 */
1808 error = -EFAULT;
1809 if (copy_from_user(&flock, l, sizeof(flock)))
1810 goto out;
1811
1812 inode = filp->f_dentry->d_inode;
1813
1814 /* Don't allow mandatory locks on files that may be memory mapped
1815 * and shared.
1816 */
1817 if (IS_MANDLOCK(inode) &&
1818 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1819 mapping_writably_mapped(filp->f_mapping)) {
1820 error = -EAGAIN;
1821 goto out;
1822 }
1823
Peter Staubachc2936212005-07-27 11:45:09 -07001824again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 error = flock64_to_posix_lock(filp, file_lock, &flock);
1826 if (error)
1827 goto out;
1828 if (cmd == F_SETLKW64) {
1829 file_lock->fl_flags |= FL_SLEEP;
1830 }
1831
1832 error = -EBADF;
1833 switch (flock.l_type) {
1834 case F_RDLCK:
1835 if (!(filp->f_mode & FMODE_READ))
1836 goto out;
1837 break;
1838 case F_WRLCK:
1839 if (!(filp->f_mode & FMODE_WRITE))
1840 goto out;
1841 break;
1842 case F_UNLCK:
1843 break;
1844 default:
1845 error = -EINVAL;
1846 goto out;
1847 }
1848
1849 error = security_file_lock(filp, file_lock->fl_type);
1850 if (error)
1851 goto out;
1852
Peter Staubachc2936212005-07-27 11:45:09 -07001853 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001855 else {
1856 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001857 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001858 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1859 break;
1860 error = wait_event_interruptible(file_lock->fl_wait,
1861 !file_lock->fl_next);
1862 if (!error)
1863 continue;
1864
1865 locks_delete_block(file_lock);
1866 break;
1867 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 }
1869
Peter Staubachc2936212005-07-27 11:45:09 -07001870 /*
1871 * Attempt to detect a close/fcntl race and recover by
1872 * releasing the lock that was just acquired.
1873 */
1874 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1875 flock.l_type = F_UNLCK;
1876 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 }
1878
1879out:
1880 locks_free_lock(file_lock);
1881 return error;
1882}
1883#endif /* BITS_PER_LONG == 32 */
1884
1885/*
1886 * This function is called when the file is being removed
1887 * from the task's fd array. POSIX locks belonging to this task
1888 * are deleted at this time.
1889 */
1890void locks_remove_posix(struct file *filp, fl_owner_t owner)
1891{
1892 struct file_lock lock, **before;
1893
1894 /*
1895 * If there are no locks held on this file, we don't need to call
1896 * posix_lock_file(). Another process could be setting a lock on this
1897 * file at the same time, but we wouldn't remove that lock anyway.
1898 */
1899 before = &filp->f_dentry->d_inode->i_flock;
1900 if (*before == NULL)
1901 return;
1902
1903 lock.fl_type = F_UNLCK;
1904 lock.fl_flags = FL_POSIX;
1905 lock.fl_start = 0;
1906 lock.fl_end = OFFSET_MAX;
1907 lock.fl_owner = owner;
1908 lock.fl_pid = current->tgid;
1909 lock.fl_file = filp;
1910 lock.fl_ops = NULL;
1911 lock.fl_lmops = NULL;
1912
1913 if (filp->f_op && filp->f_op->lock != NULL) {
1914 filp->f_op->lock(filp, F_SETLK, &lock);
1915 goto out;
1916 }
1917
1918 /* Can't use posix_lock_file here; we need to remove it no matter
1919 * which pid we have.
1920 */
1921 lock_kernel();
1922 while (*before != NULL) {
1923 struct file_lock *fl = *before;
1924 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1925 locks_delete_lock(before);
1926 continue;
1927 }
1928 before = &fl->fl_next;
1929 }
1930 unlock_kernel();
1931out:
1932 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1933 lock.fl_ops->fl_release_private(&lock);
1934}
1935
1936EXPORT_SYMBOL(locks_remove_posix);
1937
1938/*
1939 * This function is called on the last close of an open file.
1940 */
1941void locks_remove_flock(struct file *filp)
1942{
1943 struct inode * inode = filp->f_dentry->d_inode;
1944 struct file_lock *fl;
1945 struct file_lock **before;
1946
1947 if (!inode->i_flock)
1948 return;
1949
1950 if (filp->f_op && filp->f_op->flock) {
1951 struct file_lock fl = {
1952 .fl_pid = current->tgid,
1953 .fl_file = filp,
1954 .fl_flags = FL_FLOCK,
1955 .fl_type = F_UNLCK,
1956 .fl_end = OFFSET_MAX,
1957 };
1958 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001959 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1960 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 }
1962
1963 lock_kernel();
1964 before = &inode->i_flock;
1965
1966 while ((fl = *before) != NULL) {
1967 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001968 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 locks_delete_lock(before);
1970 continue;
1971 }
1972 if (IS_LEASE(fl)) {
1973 lease_modify(before, F_UNLCK);
1974 continue;
1975 }
1976 /* What? */
1977 BUG();
1978 }
1979 before = &fl->fl_next;
1980 }
1981 unlock_kernel();
1982}
1983
1984/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 * posix_unblock_lock - stop waiting for a file lock
1986 * @filp: how the file was opened
1987 * @waiter: the lock which was waiting
1988 *
1989 * lockd needs to block waiting for locks.
1990 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001991int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1993{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001994 int status = 0;
1995
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001997 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001999 else
2000 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002001 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002002 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003}
2004
2005EXPORT_SYMBOL(posix_unblock_lock);
2006
2007static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2008{
2009 struct inode *inode = NULL;
2010
2011 if (fl->fl_file != NULL)
2012 inode = fl->fl_file->f_dentry->d_inode;
2013
2014 out += sprintf(out, "%d:%s ", id, pfx);
2015 if (IS_POSIX(fl)) {
2016 out += sprintf(out, "%6s %s ",
2017 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2018 (inode == NULL) ? "*NOINODE*" :
2019 (IS_MANDLOCK(inode) &&
2020 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2021 "MANDATORY" : "ADVISORY ");
2022 } else if (IS_FLOCK(fl)) {
2023 if (fl->fl_type & LOCK_MAND) {
2024 out += sprintf(out, "FLOCK MSNFS ");
2025 } else {
2026 out += sprintf(out, "FLOCK ADVISORY ");
2027 }
2028 } else if (IS_LEASE(fl)) {
2029 out += sprintf(out, "LEASE ");
2030 if (fl->fl_type & F_INPROGRESS)
2031 out += sprintf(out, "BREAKING ");
2032 else if (fl->fl_file)
2033 out += sprintf(out, "ACTIVE ");
2034 else
2035 out += sprintf(out, "BREAKER ");
2036 } else {
2037 out += sprintf(out, "UNKNOWN UNKNOWN ");
2038 }
2039 if (fl->fl_type & LOCK_MAND) {
2040 out += sprintf(out, "%s ",
2041 (fl->fl_type & LOCK_READ)
2042 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2043 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2044 } else {
2045 out += sprintf(out, "%s ",
2046 (fl->fl_type & F_INPROGRESS)
2047 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2048 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2049 }
2050 if (inode) {
2051#ifdef WE_CAN_BREAK_LSLK_NOW
2052 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2053 inode->i_sb->s_id, inode->i_ino);
2054#else
2055 /* userspace relies on this representation of dev_t ;-( */
2056 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2057 MAJOR(inode->i_sb->s_dev),
2058 MINOR(inode->i_sb->s_dev), inode->i_ino);
2059#endif
2060 } else {
2061 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2062 }
2063 if (IS_POSIX(fl)) {
2064 if (fl->fl_end == OFFSET_MAX)
2065 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2066 else
2067 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2068 fl->fl_end);
2069 } else {
2070 out += sprintf(out, "0 EOF\n");
2071 }
2072}
2073
2074static void move_lock_status(char **p, off_t* pos, off_t offset)
2075{
2076 int len;
2077 len = strlen(*p);
2078 if(*pos >= offset) {
2079 /* the complete line is valid */
2080 *p += len;
2081 *pos += len;
2082 return;
2083 }
2084 if(*pos+len > offset) {
2085 /* use the second part of the line */
2086 int i = offset-*pos;
2087 memmove(*p,*p+i,len-i);
2088 *p += len-i;
2089 *pos += len;
2090 return;
2091 }
2092 /* discard the complete line */
2093 *pos += len;
2094}
2095
2096/**
2097 * get_locks_status - reports lock usage in /proc/locks
2098 * @buffer: address in userspace to write into
2099 * @start: ?
2100 * @offset: how far we are through the buffer
2101 * @length: how much to read
2102 */
2103
2104int get_locks_status(char *buffer, char **start, off_t offset, int length)
2105{
2106 struct list_head *tmp;
2107 char *q = buffer;
2108 off_t pos = 0;
2109 int i = 0;
2110
2111 lock_kernel();
2112 list_for_each(tmp, &file_lock_list) {
2113 struct list_head *btmp;
2114 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2115 lock_get_status(q, fl, ++i, "");
2116 move_lock_status(&q, &pos, offset);
2117
2118 if(pos >= offset+length)
2119 goto done;
2120
2121 list_for_each(btmp, &fl->fl_block) {
2122 struct file_lock *bfl = list_entry(btmp,
2123 struct file_lock, fl_block);
2124 lock_get_status(q, bfl, i, " ->");
2125 move_lock_status(&q, &pos, offset);
2126
2127 if(pos >= offset+length)
2128 goto done;
2129 }
2130 }
2131done:
2132 unlock_kernel();
2133 *start = buffer;
2134 if(q-buffer < length)
2135 return (q-buffer);
2136 return length;
2137}
2138
2139/**
2140 * lock_may_read - checks that the region is free of locks
2141 * @inode: the inode that is being read
2142 * @start: the first byte to read
2143 * @len: the number of bytes to read
2144 *
2145 * Emulates Windows locking requirements. Whole-file
2146 * mandatory locks (share modes) can prohibit a read and
2147 * byte-range POSIX locks can prohibit a read if they overlap.
2148 *
2149 * N.B. this function is only ever called
2150 * from knfsd and ownership of locks is never checked.
2151 */
2152int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2153{
2154 struct file_lock *fl;
2155 int result = 1;
2156 lock_kernel();
2157 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2158 if (IS_POSIX(fl)) {
2159 if (fl->fl_type == F_RDLCK)
2160 continue;
2161 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2162 continue;
2163 } else if (IS_FLOCK(fl)) {
2164 if (!(fl->fl_type & LOCK_MAND))
2165 continue;
2166 if (fl->fl_type & LOCK_READ)
2167 continue;
2168 } else
2169 continue;
2170 result = 0;
2171 break;
2172 }
2173 unlock_kernel();
2174 return result;
2175}
2176
2177EXPORT_SYMBOL(lock_may_read);
2178
2179/**
2180 * lock_may_write - checks that the region is free of locks
2181 * @inode: the inode that is being written
2182 * @start: the first byte to write
2183 * @len: the number of bytes to write
2184 *
2185 * Emulates Windows locking requirements. Whole-file
2186 * mandatory locks (share modes) can prohibit a write and
2187 * byte-range POSIX locks can prohibit a write if they overlap.
2188 *
2189 * N.B. this function is only ever called
2190 * from knfsd and ownership of locks is never checked.
2191 */
2192int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2193{
2194 struct file_lock *fl;
2195 int result = 1;
2196 lock_kernel();
2197 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2198 if (IS_POSIX(fl)) {
2199 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2200 continue;
2201 } else if (IS_FLOCK(fl)) {
2202 if (!(fl->fl_type & LOCK_MAND))
2203 continue;
2204 if (fl->fl_type & LOCK_WRITE)
2205 continue;
2206 } else
2207 continue;
2208 result = 0;
2209 break;
2210 }
2211 unlock_kernel();
2212 return result;
2213}
2214
2215EXPORT_SYMBOL(lock_may_write);
2216
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217static int __init filelock_init(void)
2218{
2219 filelock_cache = kmem_cache_create("file_lock_cache",
2220 sizeof(struct file_lock), 0, SLAB_PANIC,
2221 init_once, NULL);
2222 return 0;
2223}
2224
2225core_initcall(filelock_init);