blob: 2344f241c6870f19f634401c07e7f0bdffdf5507 [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;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700797 struct file_lock *new_fl = NULL;
798 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 struct file_lock *left = NULL;
800 struct file_lock *right = NULL;
801 struct file_lock **before;
802 int error, added = 0;
803
804 /*
805 * We may need two file_lock structures for this operation,
806 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700807 *
808 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700810 if (!(request->fl_flags & FL_ACCESS) &&
811 (request->fl_type != F_UNLCK ||
812 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
813 new_fl = locks_alloc_lock();
814 new_fl2 = locks_alloc_lock();
815 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
817 lock_kernel();
818 if (request->fl_type != F_UNLCK) {
819 for_each_lock(inode, before) {
820 struct file_lock *fl = *before;
821 if (!IS_POSIX(fl))
822 continue;
823 if (!posix_locks_conflict(request, fl))
824 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800825 if (conflock)
826 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 error = -EAGAIN;
828 if (!(request->fl_flags & FL_SLEEP))
829 goto out;
830 error = -EDEADLK;
831 if (posix_locks_deadlock(request, fl))
832 goto out;
833 error = -EAGAIN;
834 locks_insert_block(fl, request);
835 goto out;
836 }
837 }
838
839 /* If we're just looking for a conflict, we're done. */
840 error = 0;
841 if (request->fl_flags & FL_ACCESS)
842 goto out;
843
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 * Find the first old lock with the same owner as the new lock.
846 */
847
848 before = &inode->i_flock;
849
850 /* First skip locks owned by other processes. */
851 while ((fl = *before) && (!IS_POSIX(fl) ||
852 !posix_same_owner(request, fl))) {
853 before = &fl->fl_next;
854 }
855
856 /* Process locks with this owner. */
857 while ((fl = *before) && posix_same_owner(request, fl)) {
858 /* Detect adjacent or overlapping regions (if same lock type)
859 */
860 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700861 /* In all comparisons of start vs end, use
862 * "start - 1" rather than "end + 1". If end
863 * is OFFSET_MAX, end + 1 will become negative.
864 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 if (fl->fl_end < request->fl_start - 1)
866 goto next_lock;
867 /* If the next lock in the list has entirely bigger
868 * addresses than the new one, insert the lock here.
869 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700870 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 break;
872
873 /* If we come here, the new and old lock are of the
874 * same type and adjacent or overlapping. Make one
875 * lock yielding from the lower start address of both
876 * locks to the higher end address.
877 */
878 if (fl->fl_start > request->fl_start)
879 fl->fl_start = request->fl_start;
880 else
881 request->fl_start = fl->fl_start;
882 if (fl->fl_end < request->fl_end)
883 fl->fl_end = request->fl_end;
884 else
885 request->fl_end = fl->fl_end;
886 if (added) {
887 locks_delete_lock(before);
888 continue;
889 }
890 request = fl;
891 added = 1;
892 }
893 else {
894 /* Processing for different lock types is a bit
895 * more complex.
896 */
897 if (fl->fl_end < request->fl_start)
898 goto next_lock;
899 if (fl->fl_start > request->fl_end)
900 break;
901 if (request->fl_type == F_UNLCK)
902 added = 1;
903 if (fl->fl_start < request->fl_start)
904 left = fl;
905 /* If the next lock in the list has a higher end
906 * address than the new one, insert the new one here.
907 */
908 if (fl->fl_end > request->fl_end) {
909 right = fl;
910 break;
911 }
912 if (fl->fl_start >= request->fl_start) {
913 /* The new lock completely replaces an old
914 * one (This may happen several times).
915 */
916 if (added) {
917 locks_delete_lock(before);
918 continue;
919 }
920 /* Replace the old lock with the new one.
921 * Wake up anybody waiting for the old one,
922 * as the change in lock type might satisfy
923 * their needs.
924 */
925 locks_wake_up_blocks(fl);
926 fl->fl_start = request->fl_start;
927 fl->fl_end = request->fl_end;
928 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500929 locks_release_private(fl);
930 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 request = fl;
932 added = 1;
933 }
934 }
935 /* Go on to next lock.
936 */
937 next_lock:
938 before = &fl->fl_next;
939 }
940
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700941 /*
942 * The above code only modifies existing locks in case of
943 * merging or replacing. If new lock(s) need to be inserted
944 * all modifications are done bellow this, so it's safe yet to
945 * bail out.
946 */
947 error = -ENOLCK; /* "no luck" */
948 if (right && left == right && !new_fl2)
949 goto out;
950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 error = 0;
952 if (!added) {
953 if (request->fl_type == F_UNLCK)
954 goto out;
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700955
956 if (!new_fl) {
957 error = -ENOLCK;
958 goto out;
959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 locks_copy_lock(new_fl, request);
961 locks_insert_lock(before, new_fl);
962 new_fl = NULL;
963 }
964 if (right) {
965 if (left == right) {
966 /* The new lock breaks the old one in two pieces,
967 * so we have to use the second new lock.
968 */
969 left = new_fl2;
970 new_fl2 = NULL;
971 locks_copy_lock(left, right);
972 locks_insert_lock(before, left);
973 }
974 right->fl_start = request->fl_end + 1;
975 locks_wake_up_blocks(right);
976 }
977 if (left) {
978 left->fl_end = request->fl_start - 1;
979 locks_wake_up_blocks(left);
980 }
981 out:
982 unlock_kernel();
983 /*
984 * Free any unused locks.
985 */
986 if (new_fl)
987 locks_free_lock(new_fl);
988 if (new_fl2)
989 locks_free_lock(new_fl2);
990 return error;
991}
992
993/**
994 * posix_lock_file - Apply a POSIX-style lock to a file
995 * @filp: The file to apply the lock to
996 * @fl: The lock to be applied
997 *
998 * Add a POSIX style lock to a file.
999 * We merge adjacent & overlapping locks whenever possible.
1000 * POSIX locks are sorted by owner task, then by starting address
1001 */
1002int posix_lock_file(struct file *filp, struct file_lock *fl)
1003{
Andy Adamson5842add2006-03-26 01:37:26 -08001004 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
Andy Adamson5842add2006-03-26 01:37:26 -08001006EXPORT_SYMBOL(posix_lock_file);
1007
1008/**
1009 * posix_lock_file_conf - Apply a POSIX-style lock to a file
1010 * @filp: The file to apply the lock to
1011 * @fl: The lock to be applied
1012 * @conflock: Place to return a copy of the conflicting lock, if found.
1013 *
1014 * Except for the conflock parameter, acts just like posix_lock_file.
1015 */
1016int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1017 struct file_lock *conflock)
1018{
1019 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1020}
1021EXPORT_SYMBOL(posix_lock_file_conf);
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 (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001037 error = posix_lock_file(filp, fl);
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 (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001109 error = __posix_lock_file_conf(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 *
1172 * break_lease (inlined for speed) has checked there already
1173 * is a lease on this file. Leases are broken on a call to open()
1174 * or truncate(). This function can sleep unless you
1175 * 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();
1319 time_out_leases(filp->f_dentry->d_inode);
1320 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1321 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;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 struct dentry *dentry = filp->f_dentry;
1346 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
1407 error = lease_alloc(filp, arg, &fl);
1408 if (error)
1409 goto out;
1410
1411 locks_copy_lock(fl, lease);
1412
1413 locks_insert_lock(before, fl);
1414
1415 *flp = fl;
1416out:
1417 return error;
1418}
1419
1420 /**
1421 * setlease - sets a lease on an open file
1422 * @filp: file pointer
1423 * @arg: type of lease to obtain
1424 * @lease: file_lock to use
1425 *
1426 * Call this to establish a lease on the file.
1427 * The fl_lmops fl_break function is required by break_lease
1428 */
1429
1430int setlease(struct file *filp, long arg, struct file_lock **lease)
1431{
1432 struct dentry *dentry = filp->f_dentry;
1433 struct inode *inode = dentry->d_inode;
1434 int error;
1435
1436 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1437 return -EACCES;
1438 if (!S_ISREG(inode->i_mode))
1439 return -EINVAL;
1440 error = security_file_lock(filp, arg);
1441 if (error)
1442 return error;
1443
1444 lock_kernel();
1445 error = __setlease(filp, arg, lease);
1446 unlock_kernel();
1447
1448 return error;
1449}
1450
1451EXPORT_SYMBOL(setlease);
1452
1453/**
1454 * fcntl_setlease - sets a lease on an open file
1455 * @fd: open file descriptor
1456 * @filp: file pointer
1457 * @arg: type of lease to obtain
1458 *
1459 * Call this fcntl to establish a lease on the file.
1460 * Note that you also need to call %F_SETSIG to
1461 * receive a signal when the lease is broken.
1462 */
1463int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1464{
1465 struct file_lock fl, *flp = &fl;
1466 struct dentry *dentry = filp->f_dentry;
1467 struct inode *inode = dentry->d_inode;
1468 int error;
1469
1470 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1471 return -EACCES;
1472 if (!S_ISREG(inode->i_mode))
1473 return -EINVAL;
1474 error = security_file_lock(filp, arg);
1475 if (error)
1476 return error;
1477
1478 locks_init_lock(&fl);
1479 error = lease_init(filp, arg, &fl);
1480 if (error)
1481 return error;
1482
1483 lock_kernel();
1484
1485 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001486 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 goto out_unlock;
1488
1489 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1490 if (error < 0) {
1491 /* remove lease just inserted by __setlease */
1492 flp->fl_type = F_UNLCK | F_INPROGRESS;
1493 flp->fl_break_time = jiffies- 10;
1494 time_out_leases(inode);
1495 goto out_unlock;
1496 }
1497
1498 error = f_setown(filp, current->pid, 0);
1499out_unlock:
1500 unlock_kernel();
1501 return error;
1502}
1503
1504/**
1505 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1506 * @filp: The file to apply the lock to
1507 * @fl: The lock to be applied
1508 *
1509 * Add a FLOCK style lock to a file.
1510 */
1511int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1512{
1513 int error;
1514 might_sleep();
1515 for (;;) {
1516 error = flock_lock_file(filp, fl);
1517 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1518 break;
1519 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1520 if (!error)
1521 continue;
1522
1523 locks_delete_block(fl);
1524 break;
1525 }
1526 return error;
1527}
1528
1529EXPORT_SYMBOL(flock_lock_file_wait);
1530
1531/**
1532 * sys_flock: - flock() system call.
1533 * @fd: the file descriptor to lock.
1534 * @cmd: the type of lock to apply.
1535 *
1536 * Apply a %FL_FLOCK style lock to an open file descriptor.
1537 * The @cmd can be one of
1538 *
1539 * %LOCK_SH -- a shared lock.
1540 *
1541 * %LOCK_EX -- an exclusive lock.
1542 *
1543 * %LOCK_UN -- remove an existing lock.
1544 *
1545 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1546 *
1547 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1548 * processes read and write access respectively.
1549 */
1550asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1551{
1552 struct file *filp;
1553 struct file_lock *lock;
1554 int can_sleep, unlock;
1555 int error;
1556
1557 error = -EBADF;
1558 filp = fget(fd);
1559 if (!filp)
1560 goto out;
1561
1562 can_sleep = !(cmd & LOCK_NB);
1563 cmd &= ~LOCK_NB;
1564 unlock = (cmd == LOCK_UN);
1565
1566 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1567 goto out_putf;
1568
1569 error = flock_make_lock(filp, &lock, cmd);
1570 if (error)
1571 goto out_putf;
1572 if (can_sleep)
1573 lock->fl_flags |= FL_SLEEP;
1574
1575 error = security_file_lock(filp, cmd);
1576 if (error)
1577 goto out_free;
1578
1579 if (filp->f_op && filp->f_op->flock)
1580 error = filp->f_op->flock(filp,
1581 (can_sleep) ? F_SETLKW : F_SETLK,
1582 lock);
1583 else
1584 error = flock_lock_file_wait(filp, lock);
1585
1586 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001587 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 out_putf:
1590 fput(filp);
1591 out:
1592 return error;
1593}
1594
1595/* Report the first existing lock that would conflict with l.
1596 * This implements the F_GETLK command of fcntl().
1597 */
1598int fcntl_getlk(struct file *filp, struct flock __user *l)
1599{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001600 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 struct flock flock;
1602 int error;
1603
1604 error = -EFAULT;
1605 if (copy_from_user(&flock, l, sizeof(flock)))
1606 goto out;
1607 error = -EINVAL;
1608 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1609 goto out;
1610
1611 error = flock_to_posix_lock(filp, &file_lock, &flock);
1612 if (error)
1613 goto out;
1614
1615 if (filp->f_op && filp->f_op->lock) {
1616 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001617 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1618 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 if (error < 0)
1620 goto out;
1621 else
1622 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1623 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001624 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 }
1626
1627 flock.l_type = F_UNLCK;
1628 if (fl != NULL) {
1629 flock.l_pid = fl->fl_pid;
1630#if BITS_PER_LONG == 32
1631 /*
1632 * Make sure we can represent the posix lock via
1633 * legacy 32bit flock.
1634 */
1635 error = -EOVERFLOW;
1636 if (fl->fl_start > OFFT_OFFSET_MAX)
1637 goto out;
1638 if ((fl->fl_end != OFFSET_MAX)
1639 && (fl->fl_end > OFFT_OFFSET_MAX))
1640 goto out;
1641#endif
1642 flock.l_start = fl->fl_start;
1643 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1644 fl->fl_end - fl->fl_start + 1;
1645 flock.l_whence = 0;
1646 flock.l_type = fl->fl_type;
1647 }
1648 error = -EFAULT;
1649 if (!copy_to_user(l, &flock, sizeof(flock)))
1650 error = 0;
1651out:
1652 return error;
1653}
1654
1655/* Apply the lock described by l to an open file descriptor.
1656 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1657 */
Peter Staubachc2936212005-07-27 11:45:09 -07001658int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1659 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660{
1661 struct file_lock *file_lock = locks_alloc_lock();
1662 struct flock flock;
1663 struct inode *inode;
1664 int error;
1665
1666 if (file_lock == NULL)
1667 return -ENOLCK;
1668
1669 /*
1670 * This might block, so we do it before checking the inode.
1671 */
1672 error = -EFAULT;
1673 if (copy_from_user(&flock, l, sizeof(flock)))
1674 goto out;
1675
1676 inode = filp->f_dentry->d_inode;
1677
1678 /* Don't allow mandatory locks on files that may be memory mapped
1679 * and shared.
1680 */
1681 if (IS_MANDLOCK(inode) &&
1682 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1683 mapping_writably_mapped(filp->f_mapping)) {
1684 error = -EAGAIN;
1685 goto out;
1686 }
1687
Peter Staubachc2936212005-07-27 11:45:09 -07001688again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 error = flock_to_posix_lock(filp, file_lock, &flock);
1690 if (error)
1691 goto out;
1692 if (cmd == F_SETLKW) {
1693 file_lock->fl_flags |= FL_SLEEP;
1694 }
1695
1696 error = -EBADF;
1697 switch (flock.l_type) {
1698 case F_RDLCK:
1699 if (!(filp->f_mode & FMODE_READ))
1700 goto out;
1701 break;
1702 case F_WRLCK:
1703 if (!(filp->f_mode & FMODE_WRITE))
1704 goto out;
1705 break;
1706 case F_UNLCK:
1707 break;
1708 default:
1709 error = -EINVAL;
1710 goto out;
1711 }
1712
1713 error = security_file_lock(filp, file_lock->fl_type);
1714 if (error)
1715 goto out;
1716
Peter Staubachc2936212005-07-27 11:45:09 -07001717 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001719 else {
1720 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001721 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001722 if ((error != -EAGAIN) || (cmd == F_SETLK))
1723 break;
1724 error = wait_event_interruptible(file_lock->fl_wait,
1725 !file_lock->fl_next);
1726 if (!error)
1727 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001728
Peter Staubachc2936212005-07-27 11:45:09 -07001729 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001731 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 }
1733
Peter Staubachc2936212005-07-27 11:45:09 -07001734 /*
1735 * Attempt to detect a close/fcntl race and recover by
1736 * releasing the lock that was just acquired.
1737 */
1738 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1739 flock.l_type = F_UNLCK;
1740 goto again;
1741 }
1742
1743out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001744 locks_free_lock(file_lock);
1745 return error;
1746}
1747
1748#if BITS_PER_LONG == 32
1749/* Report the first existing lock that would conflict with l.
1750 * This implements the F_GETLK command of fcntl().
1751 */
1752int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1753{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001754 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 struct flock64 flock;
1756 int error;
1757
1758 error = -EFAULT;
1759 if (copy_from_user(&flock, l, sizeof(flock)))
1760 goto out;
1761 error = -EINVAL;
1762 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1763 goto out;
1764
1765 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1766 if (error)
1767 goto out;
1768
1769 if (filp->f_op && filp->f_op->lock) {
1770 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001771 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1772 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 if (error < 0)
1774 goto out;
1775 else
1776 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1777 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001778 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779 }
1780
1781 flock.l_type = F_UNLCK;
1782 if (fl != NULL) {
1783 flock.l_pid = fl->fl_pid;
1784 flock.l_start = fl->fl_start;
1785 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1786 fl->fl_end - fl->fl_start + 1;
1787 flock.l_whence = 0;
1788 flock.l_type = fl->fl_type;
1789 }
1790 error = -EFAULT;
1791 if (!copy_to_user(l, &flock, sizeof(flock)))
1792 error = 0;
1793
1794out:
1795 return error;
1796}
1797
1798/* Apply the lock described by l to an open file descriptor.
1799 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1800 */
Peter Staubachc2936212005-07-27 11:45:09 -07001801int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1802 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
1804 struct file_lock *file_lock = locks_alloc_lock();
1805 struct flock64 flock;
1806 struct inode *inode;
1807 int error;
1808
1809 if (file_lock == NULL)
1810 return -ENOLCK;
1811
1812 /*
1813 * This might block, so we do it before checking the inode.
1814 */
1815 error = -EFAULT;
1816 if (copy_from_user(&flock, l, sizeof(flock)))
1817 goto out;
1818
1819 inode = filp->f_dentry->d_inode;
1820
1821 /* Don't allow mandatory locks on files that may be memory mapped
1822 * and shared.
1823 */
1824 if (IS_MANDLOCK(inode) &&
1825 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1826 mapping_writably_mapped(filp->f_mapping)) {
1827 error = -EAGAIN;
1828 goto out;
1829 }
1830
Peter Staubachc2936212005-07-27 11:45:09 -07001831again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 error = flock64_to_posix_lock(filp, file_lock, &flock);
1833 if (error)
1834 goto out;
1835 if (cmd == F_SETLKW64) {
1836 file_lock->fl_flags |= FL_SLEEP;
1837 }
1838
1839 error = -EBADF;
1840 switch (flock.l_type) {
1841 case F_RDLCK:
1842 if (!(filp->f_mode & FMODE_READ))
1843 goto out;
1844 break;
1845 case F_WRLCK:
1846 if (!(filp->f_mode & FMODE_WRITE))
1847 goto out;
1848 break;
1849 case F_UNLCK:
1850 break;
1851 default:
1852 error = -EINVAL;
1853 goto out;
1854 }
1855
1856 error = security_file_lock(filp, file_lock->fl_type);
1857 if (error)
1858 goto out;
1859
Peter Staubachc2936212005-07-27 11:45:09 -07001860 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001862 else {
1863 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001864 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001865 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1866 break;
1867 error = wait_event_interruptible(file_lock->fl_wait,
1868 !file_lock->fl_next);
1869 if (!error)
1870 continue;
1871
1872 locks_delete_block(file_lock);
1873 break;
1874 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 }
1876
Peter Staubachc2936212005-07-27 11:45:09 -07001877 /*
1878 * Attempt to detect a close/fcntl race and recover by
1879 * releasing the lock that was just acquired.
1880 */
1881 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1882 flock.l_type = F_UNLCK;
1883 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 }
1885
1886out:
1887 locks_free_lock(file_lock);
1888 return error;
1889}
1890#endif /* BITS_PER_LONG == 32 */
1891
1892/*
1893 * This function is called when the file is being removed
1894 * from the task's fd array. POSIX locks belonging to this task
1895 * are deleted at this time.
1896 */
1897void locks_remove_posix(struct file *filp, fl_owner_t owner)
1898{
1899 struct file_lock lock, **before;
1900
1901 /*
1902 * If there are no locks held on this file, we don't need to call
1903 * posix_lock_file(). Another process could be setting a lock on this
1904 * file at the same time, but we wouldn't remove that lock anyway.
1905 */
1906 before = &filp->f_dentry->d_inode->i_flock;
1907 if (*before == NULL)
1908 return;
1909
1910 lock.fl_type = F_UNLCK;
1911 lock.fl_flags = FL_POSIX;
1912 lock.fl_start = 0;
1913 lock.fl_end = OFFSET_MAX;
1914 lock.fl_owner = owner;
1915 lock.fl_pid = current->tgid;
1916 lock.fl_file = filp;
1917 lock.fl_ops = NULL;
1918 lock.fl_lmops = NULL;
1919
1920 if (filp->f_op && filp->f_op->lock != NULL) {
1921 filp->f_op->lock(filp, F_SETLK, &lock);
1922 goto out;
1923 }
1924
1925 /* Can't use posix_lock_file here; we need to remove it no matter
1926 * which pid we have.
1927 */
1928 lock_kernel();
1929 while (*before != NULL) {
1930 struct file_lock *fl = *before;
1931 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1932 locks_delete_lock(before);
1933 continue;
1934 }
1935 before = &fl->fl_next;
1936 }
1937 unlock_kernel();
1938out:
1939 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1940 lock.fl_ops->fl_release_private(&lock);
1941}
1942
1943EXPORT_SYMBOL(locks_remove_posix);
1944
1945/*
1946 * This function is called on the last close of an open file.
1947 */
1948void locks_remove_flock(struct file *filp)
1949{
1950 struct inode * inode = filp->f_dentry->d_inode;
1951 struct file_lock *fl;
1952 struct file_lock **before;
1953
1954 if (!inode->i_flock)
1955 return;
1956
1957 if (filp->f_op && filp->f_op->flock) {
1958 struct file_lock fl = {
1959 .fl_pid = current->tgid,
1960 .fl_file = filp,
1961 .fl_flags = FL_FLOCK,
1962 .fl_type = F_UNLCK,
1963 .fl_end = OFFSET_MAX,
1964 };
1965 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001966 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1967 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 }
1969
1970 lock_kernel();
1971 before = &inode->i_flock;
1972
1973 while ((fl = *before) != NULL) {
1974 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001975 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 locks_delete_lock(before);
1977 continue;
1978 }
1979 if (IS_LEASE(fl)) {
1980 lease_modify(before, F_UNLCK);
1981 continue;
1982 }
1983 /* What? */
1984 BUG();
1985 }
1986 before = &fl->fl_next;
1987 }
1988 unlock_kernel();
1989}
1990
1991/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 * posix_unblock_lock - stop waiting for a file lock
1993 * @filp: how the file was opened
1994 * @waiter: the lock which was waiting
1995 *
1996 * lockd needs to block waiting for locks.
1997 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001998int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2000{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002001 int status = 0;
2002
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002004 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002006 else
2007 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002008 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002009 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010}
2011
2012EXPORT_SYMBOL(posix_unblock_lock);
2013
2014static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2015{
2016 struct inode *inode = NULL;
2017
2018 if (fl->fl_file != NULL)
2019 inode = fl->fl_file->f_dentry->d_inode;
2020
2021 out += sprintf(out, "%d:%s ", id, pfx);
2022 if (IS_POSIX(fl)) {
2023 out += sprintf(out, "%6s %s ",
2024 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2025 (inode == NULL) ? "*NOINODE*" :
2026 (IS_MANDLOCK(inode) &&
2027 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2028 "MANDATORY" : "ADVISORY ");
2029 } else if (IS_FLOCK(fl)) {
2030 if (fl->fl_type & LOCK_MAND) {
2031 out += sprintf(out, "FLOCK MSNFS ");
2032 } else {
2033 out += sprintf(out, "FLOCK ADVISORY ");
2034 }
2035 } else if (IS_LEASE(fl)) {
2036 out += sprintf(out, "LEASE ");
2037 if (fl->fl_type & F_INPROGRESS)
2038 out += sprintf(out, "BREAKING ");
2039 else if (fl->fl_file)
2040 out += sprintf(out, "ACTIVE ");
2041 else
2042 out += sprintf(out, "BREAKER ");
2043 } else {
2044 out += sprintf(out, "UNKNOWN UNKNOWN ");
2045 }
2046 if (fl->fl_type & LOCK_MAND) {
2047 out += sprintf(out, "%s ",
2048 (fl->fl_type & LOCK_READ)
2049 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2050 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2051 } else {
2052 out += sprintf(out, "%s ",
2053 (fl->fl_type & F_INPROGRESS)
2054 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2055 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2056 }
2057 if (inode) {
2058#ifdef WE_CAN_BREAK_LSLK_NOW
2059 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2060 inode->i_sb->s_id, inode->i_ino);
2061#else
2062 /* userspace relies on this representation of dev_t ;-( */
2063 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2064 MAJOR(inode->i_sb->s_dev),
2065 MINOR(inode->i_sb->s_dev), inode->i_ino);
2066#endif
2067 } else {
2068 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2069 }
2070 if (IS_POSIX(fl)) {
2071 if (fl->fl_end == OFFSET_MAX)
2072 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2073 else
2074 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2075 fl->fl_end);
2076 } else {
2077 out += sprintf(out, "0 EOF\n");
2078 }
2079}
2080
2081static void move_lock_status(char **p, off_t* pos, off_t offset)
2082{
2083 int len;
2084 len = strlen(*p);
2085 if(*pos >= offset) {
2086 /* the complete line is valid */
2087 *p += len;
2088 *pos += len;
2089 return;
2090 }
2091 if(*pos+len > offset) {
2092 /* use the second part of the line */
2093 int i = offset-*pos;
2094 memmove(*p,*p+i,len-i);
2095 *p += len-i;
2096 *pos += len;
2097 return;
2098 }
2099 /* discard the complete line */
2100 *pos += len;
2101}
2102
2103/**
2104 * get_locks_status - reports lock usage in /proc/locks
2105 * @buffer: address in userspace to write into
2106 * @start: ?
2107 * @offset: how far we are through the buffer
2108 * @length: how much to read
2109 */
2110
2111int get_locks_status(char *buffer, char **start, off_t offset, int length)
2112{
2113 struct list_head *tmp;
2114 char *q = buffer;
2115 off_t pos = 0;
2116 int i = 0;
2117
2118 lock_kernel();
2119 list_for_each(tmp, &file_lock_list) {
2120 struct list_head *btmp;
2121 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2122 lock_get_status(q, fl, ++i, "");
2123 move_lock_status(&q, &pos, offset);
2124
2125 if(pos >= offset+length)
2126 goto done;
2127
2128 list_for_each(btmp, &fl->fl_block) {
2129 struct file_lock *bfl = list_entry(btmp,
2130 struct file_lock, fl_block);
2131 lock_get_status(q, bfl, i, " ->");
2132 move_lock_status(&q, &pos, offset);
2133
2134 if(pos >= offset+length)
2135 goto done;
2136 }
2137 }
2138done:
2139 unlock_kernel();
2140 *start = buffer;
2141 if(q-buffer < length)
2142 return (q-buffer);
2143 return length;
2144}
2145
2146/**
2147 * lock_may_read - checks that the region is free of locks
2148 * @inode: the inode that is being read
2149 * @start: the first byte to read
2150 * @len: the number of bytes to read
2151 *
2152 * Emulates Windows locking requirements. Whole-file
2153 * mandatory locks (share modes) can prohibit a read and
2154 * byte-range POSIX locks can prohibit a read if they overlap.
2155 *
2156 * N.B. this function is only ever called
2157 * from knfsd and ownership of locks is never checked.
2158 */
2159int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2160{
2161 struct file_lock *fl;
2162 int result = 1;
2163 lock_kernel();
2164 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2165 if (IS_POSIX(fl)) {
2166 if (fl->fl_type == F_RDLCK)
2167 continue;
2168 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2169 continue;
2170 } else if (IS_FLOCK(fl)) {
2171 if (!(fl->fl_type & LOCK_MAND))
2172 continue;
2173 if (fl->fl_type & LOCK_READ)
2174 continue;
2175 } else
2176 continue;
2177 result = 0;
2178 break;
2179 }
2180 unlock_kernel();
2181 return result;
2182}
2183
2184EXPORT_SYMBOL(lock_may_read);
2185
2186/**
2187 * lock_may_write - checks that the region is free of locks
2188 * @inode: the inode that is being written
2189 * @start: the first byte to write
2190 * @len: the number of bytes to write
2191 *
2192 * Emulates Windows locking requirements. Whole-file
2193 * mandatory locks (share modes) can prohibit a write and
2194 * byte-range POSIX locks can prohibit a write if they overlap.
2195 *
2196 * N.B. this function is only ever called
2197 * from knfsd and ownership of locks is never checked.
2198 */
2199int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2200{
2201 struct file_lock *fl;
2202 int result = 1;
2203 lock_kernel();
2204 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2205 if (IS_POSIX(fl)) {
2206 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2207 continue;
2208 } else if (IS_FLOCK(fl)) {
2209 if (!(fl->fl_type & LOCK_MAND))
2210 continue;
2211 if (fl->fl_type & LOCK_WRITE)
2212 continue;
2213 } else
2214 continue;
2215 result = 0;
2216 break;
2217 }
2218 unlock_kernel();
2219 return result;
2220}
2221
2222EXPORT_SYMBOL(lock_may_write);
2223
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224static int __init filelock_init(void)
2225{
2226 filelock_cache = kmem_cache_create("file_lock_cache",
2227 sizeof(struct file_lock), 0, SLAB_PANIC,
2228 init_once, NULL);
2229 return 0;
2230}
2231
2232core_initcall(filelock_init);