blob: 6f99c0a6f83629fbaa3eaaaf853fb6b7b9a9cfcc [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
Trond Myklebust993dfa82006-03-31 02:30:55 -0800758 new_fl = locks_alloc_lock();
759 if (new_fl == NULL)
760 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 /*
762 * If a higher-priority process was blocked on the old file lock,
763 * give it the opportunity to lock the file.
764 */
765 if (found)
766 cond_resched();
767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 for_each_lock(inode, before) {
769 struct file_lock *fl = *before;
770 if (IS_POSIX(fl))
771 break;
772 if (IS_LEASE(fl))
773 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800774 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 continue;
776 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800777 if (request->fl_flags & FL_SLEEP)
778 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 goto out;
780 }
Trond Myklebust993dfa82006-03-31 02:30:55 -0800781 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800783 new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784
785out:
786 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800787 if (new_fl)
788 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 return error;
790}
791
Andy Adamson5842add2006-03-26 01:37:26 -0800792static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793{
794 struct file_lock *fl;
795 struct file_lock *new_fl, *new_fl2;
796 struct file_lock *left = NULL;
797 struct file_lock *right = NULL;
798 struct file_lock **before;
799 int error, added = 0;
800
801 /*
802 * We may need two file_lock structures for this operation,
803 * so we get them in advance to avoid races.
804 */
805 new_fl = locks_alloc_lock();
806 new_fl2 = locks_alloc_lock();
807
808 lock_kernel();
809 if (request->fl_type != F_UNLCK) {
810 for_each_lock(inode, before) {
811 struct file_lock *fl = *before;
812 if (!IS_POSIX(fl))
813 continue;
814 if (!posix_locks_conflict(request, fl))
815 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800816 if (conflock)
817 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 error = -EAGAIN;
819 if (!(request->fl_flags & FL_SLEEP))
820 goto out;
821 error = -EDEADLK;
822 if (posix_locks_deadlock(request, fl))
823 goto out;
824 error = -EAGAIN;
825 locks_insert_block(fl, request);
826 goto out;
827 }
828 }
829
830 /* If we're just looking for a conflict, we're done. */
831 error = 0;
832 if (request->fl_flags & FL_ACCESS)
833 goto out;
834
835 error = -ENOLCK; /* "no luck" */
836 if (!(new_fl && new_fl2))
837 goto out;
838
839 /*
840 * We've allocated the new locks in advance, so there are no
841 * errors possible (and no blocking operations) from here on.
842 *
843 * Find the first old lock with the same owner as the new lock.
844 */
845
846 before = &inode->i_flock;
847
848 /* First skip locks owned by other processes. */
849 while ((fl = *before) && (!IS_POSIX(fl) ||
850 !posix_same_owner(request, fl))) {
851 before = &fl->fl_next;
852 }
853
854 /* Process locks with this owner. */
855 while ((fl = *before) && posix_same_owner(request, fl)) {
856 /* Detect adjacent or overlapping regions (if same lock type)
857 */
858 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700859 /* In all comparisons of start vs end, use
860 * "start - 1" rather than "end + 1". If end
861 * is OFFSET_MAX, end + 1 will become negative.
862 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (fl->fl_end < request->fl_start - 1)
864 goto next_lock;
865 /* If the next lock in the list has entirely bigger
866 * addresses than the new one, insert the lock here.
867 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700868 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 break;
870
871 /* If we come here, the new and old lock are of the
872 * same type and adjacent or overlapping. Make one
873 * lock yielding from the lower start address of both
874 * locks to the higher end address.
875 */
876 if (fl->fl_start > request->fl_start)
877 fl->fl_start = request->fl_start;
878 else
879 request->fl_start = fl->fl_start;
880 if (fl->fl_end < request->fl_end)
881 fl->fl_end = request->fl_end;
882 else
883 request->fl_end = fl->fl_end;
884 if (added) {
885 locks_delete_lock(before);
886 continue;
887 }
888 request = fl;
889 added = 1;
890 }
891 else {
892 /* Processing for different lock types is a bit
893 * more complex.
894 */
895 if (fl->fl_end < request->fl_start)
896 goto next_lock;
897 if (fl->fl_start > request->fl_end)
898 break;
899 if (request->fl_type == F_UNLCK)
900 added = 1;
901 if (fl->fl_start < request->fl_start)
902 left = fl;
903 /* If the next lock in the list has a higher end
904 * address than the new one, insert the new one here.
905 */
906 if (fl->fl_end > request->fl_end) {
907 right = fl;
908 break;
909 }
910 if (fl->fl_start >= request->fl_start) {
911 /* The new lock completely replaces an old
912 * one (This may happen several times).
913 */
914 if (added) {
915 locks_delete_lock(before);
916 continue;
917 }
918 /* Replace the old lock with the new one.
919 * Wake up anybody waiting for the old one,
920 * as the change in lock type might satisfy
921 * their needs.
922 */
923 locks_wake_up_blocks(fl);
924 fl->fl_start = request->fl_start;
925 fl->fl_end = request->fl_end;
926 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500927 locks_release_private(fl);
928 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 request = fl;
930 added = 1;
931 }
932 }
933 /* Go on to next lock.
934 */
935 next_lock:
936 before = &fl->fl_next;
937 }
938
939 error = 0;
940 if (!added) {
941 if (request->fl_type == F_UNLCK)
942 goto out;
943 locks_copy_lock(new_fl, request);
944 locks_insert_lock(before, new_fl);
945 new_fl = NULL;
946 }
947 if (right) {
948 if (left == right) {
949 /* The new lock breaks the old one in two pieces,
950 * so we have to use the second new lock.
951 */
952 left = new_fl2;
953 new_fl2 = NULL;
954 locks_copy_lock(left, right);
955 locks_insert_lock(before, left);
956 }
957 right->fl_start = request->fl_end + 1;
958 locks_wake_up_blocks(right);
959 }
960 if (left) {
961 left->fl_end = request->fl_start - 1;
962 locks_wake_up_blocks(left);
963 }
964 out:
965 unlock_kernel();
966 /*
967 * Free any unused locks.
968 */
969 if (new_fl)
970 locks_free_lock(new_fl);
971 if (new_fl2)
972 locks_free_lock(new_fl2);
973 return error;
974}
975
976/**
977 * posix_lock_file - Apply a POSIX-style lock to a file
978 * @filp: The file to apply the lock to
979 * @fl: The lock to be applied
980 *
981 * Add a POSIX style lock to a file.
982 * We merge adjacent & overlapping locks whenever possible.
983 * POSIX locks are sorted by owner task, then by starting address
984 */
985int posix_lock_file(struct file *filp, struct file_lock *fl)
986{
Andy Adamson5842add2006-03-26 01:37:26 -0800987 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988}
Andy Adamson5842add2006-03-26 01:37:26 -0800989EXPORT_SYMBOL(posix_lock_file);
990
991/**
992 * posix_lock_file_conf - Apply a POSIX-style lock to a file
993 * @filp: The file to apply the lock to
994 * @fl: The lock to be applied
995 * @conflock: Place to return a copy of the conflicting lock, if found.
996 *
997 * Except for the conflock parameter, acts just like posix_lock_file.
998 */
999int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1000 struct file_lock *conflock)
1001{
1002 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1003}
1004EXPORT_SYMBOL(posix_lock_file_conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005
1006/**
1007 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1008 * @filp: The file to apply the lock to
1009 * @fl: The lock to be applied
1010 *
1011 * Add a POSIX style lock to a file.
1012 * We merge adjacent & overlapping locks whenever possible.
1013 * POSIX locks are sorted by owner task, then by starting address
1014 */
1015int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1016{
1017 int error;
1018 might_sleep ();
1019 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001020 error = posix_lock_file(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1022 break;
1023 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1024 if (!error)
1025 continue;
1026
1027 locks_delete_block(fl);
1028 break;
1029 }
1030 return error;
1031}
1032EXPORT_SYMBOL(posix_lock_file_wait);
1033
1034/**
1035 * locks_mandatory_locked - Check for an active lock
1036 * @inode: the file to check
1037 *
1038 * Searches the inode's list of locks to find any POSIX locks which conflict.
1039 * This function is called from locks_verify_locked() only.
1040 */
1041int locks_mandatory_locked(struct inode *inode)
1042{
1043 fl_owner_t owner = current->files;
1044 struct file_lock *fl;
1045
1046 /*
1047 * Search the lock list for this inode for any POSIX locks.
1048 */
1049 lock_kernel();
1050 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1051 if (!IS_POSIX(fl))
1052 continue;
1053 if (fl->fl_owner != owner)
1054 break;
1055 }
1056 unlock_kernel();
1057 return fl ? -EAGAIN : 0;
1058}
1059
1060/**
1061 * locks_mandatory_area - Check for a conflicting lock
1062 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1063 * for shared
1064 * @inode: the file to check
1065 * @filp: how the file was opened (if it was)
1066 * @offset: start of area to check
1067 * @count: length of area to check
1068 *
1069 * Searches the inode's list of locks to find any POSIX locks which conflict.
1070 * This function is called from rw_verify_area() and
1071 * locks_verify_truncate().
1072 */
1073int locks_mandatory_area(int read_write, struct inode *inode,
1074 struct file *filp, loff_t offset,
1075 size_t count)
1076{
1077 struct file_lock fl;
1078 int error;
1079
1080 locks_init_lock(&fl);
1081 fl.fl_owner = current->files;
1082 fl.fl_pid = current->tgid;
1083 fl.fl_file = filp;
1084 fl.fl_flags = FL_POSIX | FL_ACCESS;
1085 if (filp && !(filp->f_flags & O_NONBLOCK))
1086 fl.fl_flags |= FL_SLEEP;
1087 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1088 fl.fl_start = offset;
1089 fl.fl_end = offset + count - 1;
1090
1091 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001092 error = __posix_lock_file_conf(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 if (error != -EAGAIN)
1094 break;
1095 if (!(fl.fl_flags & FL_SLEEP))
1096 break;
1097 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1098 if (!error) {
1099 /*
1100 * If we've been sleeping someone might have
1101 * changed the permissions behind our back.
1102 */
1103 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1104 continue;
1105 }
1106
1107 locks_delete_block(&fl);
1108 break;
1109 }
1110
1111 return error;
1112}
1113
1114EXPORT_SYMBOL(locks_mandatory_area);
1115
1116/* We already had a lease on this file; just change its type */
1117int lease_modify(struct file_lock **before, int arg)
1118{
1119 struct file_lock *fl = *before;
1120 int error = assign_type(fl, arg);
1121
1122 if (error)
1123 return error;
1124 locks_wake_up_blocks(fl);
1125 if (arg == F_UNLCK)
1126 locks_delete_lock(before);
1127 return 0;
1128}
1129
1130EXPORT_SYMBOL(lease_modify);
1131
1132static void time_out_leases(struct inode *inode)
1133{
1134 struct file_lock **before;
1135 struct file_lock *fl;
1136
1137 before = &inode->i_flock;
1138 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1139 if ((fl->fl_break_time == 0)
1140 || time_before(jiffies, fl->fl_break_time)) {
1141 before = &fl->fl_next;
1142 continue;
1143 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1145 if (fl == *before) /* lease_modify may have freed fl */
1146 before = &fl->fl_next;
1147 }
1148}
1149
1150/**
1151 * __break_lease - revoke all outstanding leases on file
1152 * @inode: the inode of the file to return
1153 * @mode: the open mode (read or write)
1154 *
1155 * break_lease (inlined for speed) has checked there already
1156 * is a lease on this file. Leases are broken on a call to open()
1157 * or truncate(). This function can sleep unless you
1158 * specified %O_NONBLOCK to your open().
1159 */
1160int __break_lease(struct inode *inode, unsigned int mode)
1161{
1162 int error = 0, future;
1163 struct file_lock *new_fl, *flock;
1164 struct file_lock *fl;
1165 int alloc_err;
1166 unsigned long break_time;
1167 int i_have_this_lease = 0;
1168
1169 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1170 &new_fl);
1171
1172 lock_kernel();
1173
1174 time_out_leases(inode);
1175
1176 flock = inode->i_flock;
1177 if ((flock == NULL) || !IS_LEASE(flock))
1178 goto out;
1179
1180 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1181 if (fl->fl_owner == current->files)
1182 i_have_this_lease = 1;
1183
1184 if (mode & FMODE_WRITE) {
1185 /* If we want write access, we have to revoke any lease. */
1186 future = F_UNLCK | F_INPROGRESS;
1187 } else if (flock->fl_type & F_INPROGRESS) {
1188 /* If the lease is already being broken, we just leave it */
1189 future = flock->fl_type;
1190 } else if (flock->fl_type & F_WRLCK) {
1191 /* Downgrade the exclusive lease to a read-only lease. */
1192 future = F_RDLCK | F_INPROGRESS;
1193 } else {
1194 /* the existing lease was read-only, so we can read too. */
1195 goto out;
1196 }
1197
1198 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1199 error = alloc_err;
1200 goto out;
1201 }
1202
1203 break_time = 0;
1204 if (lease_break_time > 0) {
1205 break_time = jiffies + lease_break_time * HZ;
1206 if (break_time == 0)
1207 break_time++; /* so that 0 means no break time */
1208 }
1209
1210 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1211 if (fl->fl_type != future) {
1212 fl->fl_type = future;
1213 fl->fl_break_time = break_time;
1214 /* lease must have lmops break callback */
1215 fl->fl_lmops->fl_break(fl);
1216 }
1217 }
1218
1219 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1220 error = -EWOULDBLOCK;
1221 goto out;
1222 }
1223
1224restart:
1225 break_time = flock->fl_break_time;
1226 if (break_time != 0) {
1227 break_time -= jiffies;
1228 if (break_time == 0)
1229 break_time++;
1230 }
1231 error = locks_block_on_timeout(flock, new_fl, break_time);
1232 if (error >= 0) {
1233 if (error == 0)
1234 time_out_leases(inode);
1235 /* Wait for the next lease that has not been broken yet */
1236 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1237 flock = flock->fl_next) {
1238 if (flock->fl_type & F_INPROGRESS)
1239 goto restart;
1240 }
1241 error = 0;
1242 }
1243
1244out:
1245 unlock_kernel();
1246 if (!alloc_err)
1247 locks_free_lock(new_fl);
1248 return error;
1249}
1250
1251EXPORT_SYMBOL(__break_lease);
1252
1253/**
1254 * lease_get_mtime
1255 * @inode: the inode
1256 * @time: pointer to a timespec which will contain the last modified time
1257 *
1258 * This is to force NFS clients to flush their caches for files with
1259 * exclusive leases. The justification is that if someone has an
1260 * exclusive lease, then they could be modifiying it.
1261 */
1262void lease_get_mtime(struct inode *inode, struct timespec *time)
1263{
1264 struct file_lock *flock = inode->i_flock;
1265 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1266 *time = current_fs_time(inode->i_sb);
1267 else
1268 *time = inode->i_mtime;
1269}
1270
1271EXPORT_SYMBOL(lease_get_mtime);
1272
1273/**
1274 * fcntl_getlease - Enquire what lease is currently active
1275 * @filp: the file
1276 *
1277 * The value returned by this function will be one of
1278 * (if no lease break is pending):
1279 *
1280 * %F_RDLCK to indicate a shared lease is held.
1281 *
1282 * %F_WRLCK to indicate an exclusive lease is held.
1283 *
1284 * %F_UNLCK to indicate no lease is held.
1285 *
1286 * (if a lease break is pending):
1287 *
1288 * %F_RDLCK to indicate an exclusive lease needs to be
1289 * changed to a shared lease (or removed).
1290 *
1291 * %F_UNLCK to indicate the lease needs to be removed.
1292 *
1293 * XXX: sfr & willy disagree over whether F_INPROGRESS
1294 * should be returned to userspace.
1295 */
1296int fcntl_getlease(struct file *filp)
1297{
1298 struct file_lock *fl;
1299 int type = F_UNLCK;
1300
1301 lock_kernel();
1302 time_out_leases(filp->f_dentry->d_inode);
1303 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1304 fl = fl->fl_next) {
1305 if (fl->fl_file == filp) {
1306 type = fl->fl_type & ~F_INPROGRESS;
1307 break;
1308 }
1309 }
1310 unlock_kernel();
1311 return type;
1312}
1313
1314/**
1315 * __setlease - sets a lease on an open file
1316 * @filp: file pointer
1317 * @arg: type of lease to obtain
1318 * @flp: input - file_lock to use, output - file_lock inserted
1319 *
1320 * The (input) flp->fl_lmops->fl_break function is required
1321 * by break_lease().
1322 *
1323 * Called with kernel lock held.
1324 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001325static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001327 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 struct dentry *dentry = filp->f_dentry;
1329 struct inode *inode = dentry->d_inode;
1330 int error, rdlease_count = 0, wrlease_count = 0;
1331
1332 time_out_leases(inode);
1333
1334 error = -EINVAL;
1335 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1336 goto out;
1337
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001338 lease = *flp;
1339
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 error = -EAGAIN;
1341 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1342 goto out;
1343 if ((arg == F_WRLCK)
1344 && ((atomic_read(&dentry->d_count) > 1)
1345 || (atomic_read(&inode->i_count) > 1)))
1346 goto out;
1347
1348 /*
1349 * At this point, we know that if there is an exclusive
1350 * lease on this file, then we hold it on this filp
1351 * (otherwise our open of this file would have blocked).
1352 * And if we are trying to acquire an exclusive lease,
1353 * then the file is not open by anyone (including us)
1354 * except for this filp.
1355 */
1356 for (before = &inode->i_flock;
1357 ((fl = *before) != NULL) && IS_LEASE(fl);
1358 before = &fl->fl_next) {
1359 if (lease->fl_lmops->fl_mylease(fl, lease))
1360 my_before = before;
1361 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1362 /*
1363 * Someone is in the process of opening this
1364 * file for writing so we may not take an
1365 * exclusive lease on it.
1366 */
1367 wrlease_count++;
1368 else
1369 rdlease_count++;
1370 }
1371
1372 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1373 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1374 goto out;
1375
1376 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001377 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 error = lease->fl_lmops->fl_change(my_before, arg);
1379 goto out;
1380 }
1381
1382 error = 0;
1383 if (arg == F_UNLCK)
1384 goto out;
1385
1386 error = -EINVAL;
1387 if (!leases_enable)
1388 goto out;
1389
1390 error = lease_alloc(filp, arg, &fl);
1391 if (error)
1392 goto out;
1393
1394 locks_copy_lock(fl, lease);
1395
1396 locks_insert_lock(before, fl);
1397
1398 *flp = fl;
1399out:
1400 return error;
1401}
1402
1403 /**
1404 * setlease - sets a lease on an open file
1405 * @filp: file pointer
1406 * @arg: type of lease to obtain
1407 * @lease: file_lock to use
1408 *
1409 * Call this to establish a lease on the file.
1410 * The fl_lmops fl_break function is required by break_lease
1411 */
1412
1413int setlease(struct file *filp, long arg, struct file_lock **lease)
1414{
1415 struct dentry *dentry = filp->f_dentry;
1416 struct inode *inode = dentry->d_inode;
1417 int error;
1418
1419 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1420 return -EACCES;
1421 if (!S_ISREG(inode->i_mode))
1422 return -EINVAL;
1423 error = security_file_lock(filp, arg);
1424 if (error)
1425 return error;
1426
1427 lock_kernel();
1428 error = __setlease(filp, arg, lease);
1429 unlock_kernel();
1430
1431 return error;
1432}
1433
1434EXPORT_SYMBOL(setlease);
1435
1436/**
1437 * fcntl_setlease - sets a lease on an open file
1438 * @fd: open file descriptor
1439 * @filp: file pointer
1440 * @arg: type of lease to obtain
1441 *
1442 * Call this fcntl to establish a lease on the file.
1443 * Note that you also need to call %F_SETSIG to
1444 * receive a signal when the lease is broken.
1445 */
1446int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1447{
1448 struct file_lock fl, *flp = &fl;
1449 struct dentry *dentry = filp->f_dentry;
1450 struct inode *inode = dentry->d_inode;
1451 int error;
1452
1453 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1454 return -EACCES;
1455 if (!S_ISREG(inode->i_mode))
1456 return -EINVAL;
1457 error = security_file_lock(filp, arg);
1458 if (error)
1459 return error;
1460
1461 locks_init_lock(&fl);
1462 error = lease_init(filp, arg, &fl);
1463 if (error)
1464 return error;
1465
1466 lock_kernel();
1467
1468 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001469 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 goto out_unlock;
1471
1472 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1473 if (error < 0) {
1474 /* remove lease just inserted by __setlease */
1475 flp->fl_type = F_UNLCK | F_INPROGRESS;
1476 flp->fl_break_time = jiffies- 10;
1477 time_out_leases(inode);
1478 goto out_unlock;
1479 }
1480
1481 error = f_setown(filp, current->pid, 0);
1482out_unlock:
1483 unlock_kernel();
1484 return error;
1485}
1486
1487/**
1488 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1489 * @filp: The file to apply the lock to
1490 * @fl: The lock to be applied
1491 *
1492 * Add a FLOCK style lock to a file.
1493 */
1494int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1495{
1496 int error;
1497 might_sleep();
1498 for (;;) {
1499 error = flock_lock_file(filp, fl);
1500 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1501 break;
1502 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1503 if (!error)
1504 continue;
1505
1506 locks_delete_block(fl);
1507 break;
1508 }
1509 return error;
1510}
1511
1512EXPORT_SYMBOL(flock_lock_file_wait);
1513
1514/**
1515 * sys_flock: - flock() system call.
1516 * @fd: the file descriptor to lock.
1517 * @cmd: the type of lock to apply.
1518 *
1519 * Apply a %FL_FLOCK style lock to an open file descriptor.
1520 * The @cmd can be one of
1521 *
1522 * %LOCK_SH -- a shared lock.
1523 *
1524 * %LOCK_EX -- an exclusive lock.
1525 *
1526 * %LOCK_UN -- remove an existing lock.
1527 *
1528 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1529 *
1530 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1531 * processes read and write access respectively.
1532 */
1533asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1534{
1535 struct file *filp;
1536 struct file_lock *lock;
1537 int can_sleep, unlock;
1538 int error;
1539
1540 error = -EBADF;
1541 filp = fget(fd);
1542 if (!filp)
1543 goto out;
1544
1545 can_sleep = !(cmd & LOCK_NB);
1546 cmd &= ~LOCK_NB;
1547 unlock = (cmd == LOCK_UN);
1548
1549 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1550 goto out_putf;
1551
1552 error = flock_make_lock(filp, &lock, cmd);
1553 if (error)
1554 goto out_putf;
1555 if (can_sleep)
1556 lock->fl_flags |= FL_SLEEP;
1557
1558 error = security_file_lock(filp, cmd);
1559 if (error)
1560 goto out_free;
1561
1562 if (filp->f_op && filp->f_op->flock)
1563 error = filp->f_op->flock(filp,
1564 (can_sleep) ? F_SETLKW : F_SETLK,
1565 lock);
1566 else
1567 error = flock_lock_file_wait(filp, lock);
1568
1569 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001570 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571
1572 out_putf:
1573 fput(filp);
1574 out:
1575 return error;
1576}
1577
1578/* Report the first existing lock that would conflict with l.
1579 * This implements the F_GETLK command of fcntl().
1580 */
1581int fcntl_getlk(struct file *filp, struct flock __user *l)
1582{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001583 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 struct flock flock;
1585 int error;
1586
1587 error = -EFAULT;
1588 if (copy_from_user(&flock, l, sizeof(flock)))
1589 goto out;
1590 error = -EINVAL;
1591 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1592 goto out;
1593
1594 error = flock_to_posix_lock(filp, &file_lock, &flock);
1595 if (error)
1596 goto out;
1597
1598 if (filp->f_op && filp->f_op->lock) {
1599 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001600 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1601 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 if (error < 0)
1603 goto out;
1604 else
1605 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1606 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001607 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 }
1609
1610 flock.l_type = F_UNLCK;
1611 if (fl != NULL) {
1612 flock.l_pid = fl->fl_pid;
1613#if BITS_PER_LONG == 32
1614 /*
1615 * Make sure we can represent the posix lock via
1616 * legacy 32bit flock.
1617 */
1618 error = -EOVERFLOW;
1619 if (fl->fl_start > OFFT_OFFSET_MAX)
1620 goto out;
1621 if ((fl->fl_end != OFFSET_MAX)
1622 && (fl->fl_end > OFFT_OFFSET_MAX))
1623 goto out;
1624#endif
1625 flock.l_start = fl->fl_start;
1626 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1627 fl->fl_end - fl->fl_start + 1;
1628 flock.l_whence = 0;
1629 flock.l_type = fl->fl_type;
1630 }
1631 error = -EFAULT;
1632 if (!copy_to_user(l, &flock, sizeof(flock)))
1633 error = 0;
1634out:
1635 return error;
1636}
1637
1638/* Apply the lock described by l to an open file descriptor.
1639 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1640 */
Peter Staubachc2936212005-07-27 11:45:09 -07001641int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1642 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643{
1644 struct file_lock *file_lock = locks_alloc_lock();
1645 struct flock flock;
1646 struct inode *inode;
1647 int error;
1648
1649 if (file_lock == NULL)
1650 return -ENOLCK;
1651
1652 /*
1653 * This might block, so we do it before checking the inode.
1654 */
1655 error = -EFAULT;
1656 if (copy_from_user(&flock, l, sizeof(flock)))
1657 goto out;
1658
1659 inode = filp->f_dentry->d_inode;
1660
1661 /* Don't allow mandatory locks on files that may be memory mapped
1662 * and shared.
1663 */
1664 if (IS_MANDLOCK(inode) &&
1665 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1666 mapping_writably_mapped(filp->f_mapping)) {
1667 error = -EAGAIN;
1668 goto out;
1669 }
1670
Peter Staubachc2936212005-07-27 11:45:09 -07001671again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 error = flock_to_posix_lock(filp, file_lock, &flock);
1673 if (error)
1674 goto out;
1675 if (cmd == F_SETLKW) {
1676 file_lock->fl_flags |= FL_SLEEP;
1677 }
1678
1679 error = -EBADF;
1680 switch (flock.l_type) {
1681 case F_RDLCK:
1682 if (!(filp->f_mode & FMODE_READ))
1683 goto out;
1684 break;
1685 case F_WRLCK:
1686 if (!(filp->f_mode & FMODE_WRITE))
1687 goto out;
1688 break;
1689 case F_UNLCK:
1690 break;
1691 default:
1692 error = -EINVAL;
1693 goto out;
1694 }
1695
1696 error = security_file_lock(filp, file_lock->fl_type);
1697 if (error)
1698 goto out;
1699
Peter Staubachc2936212005-07-27 11:45:09 -07001700 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001702 else {
1703 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001704 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001705 if ((error != -EAGAIN) || (cmd == F_SETLK))
1706 break;
1707 error = wait_event_interruptible(file_lock->fl_wait,
1708 !file_lock->fl_next);
1709 if (!error)
1710 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
Peter Staubachc2936212005-07-27 11:45:09 -07001712 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001714 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 }
1716
Peter Staubachc2936212005-07-27 11:45:09 -07001717 /*
1718 * Attempt to detect a close/fcntl race and recover by
1719 * releasing the lock that was just acquired.
1720 */
1721 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1722 flock.l_type = F_UNLCK;
1723 goto again;
1724 }
1725
1726out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 locks_free_lock(file_lock);
1728 return error;
1729}
1730
1731#if BITS_PER_LONG == 32
1732/* Report the first existing lock that would conflict with l.
1733 * This implements the F_GETLK command of fcntl().
1734 */
1735int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1736{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001737 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 struct flock64 flock;
1739 int error;
1740
1741 error = -EFAULT;
1742 if (copy_from_user(&flock, l, sizeof(flock)))
1743 goto out;
1744 error = -EINVAL;
1745 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1746 goto out;
1747
1748 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1749 if (error)
1750 goto out;
1751
1752 if (filp->f_op && filp->f_op->lock) {
1753 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001754 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1755 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 if (error < 0)
1757 goto out;
1758 else
1759 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1760 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001761 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 }
1763
1764 flock.l_type = F_UNLCK;
1765 if (fl != NULL) {
1766 flock.l_pid = fl->fl_pid;
1767 flock.l_start = fl->fl_start;
1768 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1769 fl->fl_end - fl->fl_start + 1;
1770 flock.l_whence = 0;
1771 flock.l_type = fl->fl_type;
1772 }
1773 error = -EFAULT;
1774 if (!copy_to_user(l, &flock, sizeof(flock)))
1775 error = 0;
1776
1777out:
1778 return error;
1779}
1780
1781/* Apply the lock described by l to an open file descriptor.
1782 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1783 */
Peter Staubachc2936212005-07-27 11:45:09 -07001784int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1785 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786{
1787 struct file_lock *file_lock = locks_alloc_lock();
1788 struct flock64 flock;
1789 struct inode *inode;
1790 int error;
1791
1792 if (file_lock == NULL)
1793 return -ENOLCK;
1794
1795 /*
1796 * This might block, so we do it before checking the inode.
1797 */
1798 error = -EFAULT;
1799 if (copy_from_user(&flock, l, sizeof(flock)))
1800 goto out;
1801
1802 inode = filp->f_dentry->d_inode;
1803
1804 /* Don't allow mandatory locks on files that may be memory mapped
1805 * and shared.
1806 */
1807 if (IS_MANDLOCK(inode) &&
1808 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1809 mapping_writably_mapped(filp->f_mapping)) {
1810 error = -EAGAIN;
1811 goto out;
1812 }
1813
Peter Staubachc2936212005-07-27 11:45:09 -07001814again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 error = flock64_to_posix_lock(filp, file_lock, &flock);
1816 if (error)
1817 goto out;
1818 if (cmd == F_SETLKW64) {
1819 file_lock->fl_flags |= FL_SLEEP;
1820 }
1821
1822 error = -EBADF;
1823 switch (flock.l_type) {
1824 case F_RDLCK:
1825 if (!(filp->f_mode & FMODE_READ))
1826 goto out;
1827 break;
1828 case F_WRLCK:
1829 if (!(filp->f_mode & FMODE_WRITE))
1830 goto out;
1831 break;
1832 case F_UNLCK:
1833 break;
1834 default:
1835 error = -EINVAL;
1836 goto out;
1837 }
1838
1839 error = security_file_lock(filp, file_lock->fl_type);
1840 if (error)
1841 goto out;
1842
Peter Staubachc2936212005-07-27 11:45:09 -07001843 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001845 else {
1846 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001847 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001848 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1849 break;
1850 error = wait_event_interruptible(file_lock->fl_wait,
1851 !file_lock->fl_next);
1852 if (!error)
1853 continue;
1854
1855 locks_delete_block(file_lock);
1856 break;
1857 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 }
1859
Peter Staubachc2936212005-07-27 11:45:09 -07001860 /*
1861 * Attempt to detect a close/fcntl race and recover by
1862 * releasing the lock that was just acquired.
1863 */
1864 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1865 flock.l_type = F_UNLCK;
1866 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 }
1868
1869out:
1870 locks_free_lock(file_lock);
1871 return error;
1872}
1873#endif /* BITS_PER_LONG == 32 */
1874
1875/*
1876 * This function is called when the file is being removed
1877 * from the task's fd array. POSIX locks belonging to this task
1878 * are deleted at this time.
1879 */
1880void locks_remove_posix(struct file *filp, fl_owner_t owner)
1881{
1882 struct file_lock lock, **before;
1883
1884 /*
1885 * If there are no locks held on this file, we don't need to call
1886 * posix_lock_file(). Another process could be setting a lock on this
1887 * file at the same time, but we wouldn't remove that lock anyway.
1888 */
1889 before = &filp->f_dentry->d_inode->i_flock;
1890 if (*before == NULL)
1891 return;
1892
1893 lock.fl_type = F_UNLCK;
1894 lock.fl_flags = FL_POSIX;
1895 lock.fl_start = 0;
1896 lock.fl_end = OFFSET_MAX;
1897 lock.fl_owner = owner;
1898 lock.fl_pid = current->tgid;
1899 lock.fl_file = filp;
1900 lock.fl_ops = NULL;
1901 lock.fl_lmops = NULL;
1902
1903 if (filp->f_op && filp->f_op->lock != NULL) {
1904 filp->f_op->lock(filp, F_SETLK, &lock);
1905 goto out;
1906 }
1907
1908 /* Can't use posix_lock_file here; we need to remove it no matter
1909 * which pid we have.
1910 */
1911 lock_kernel();
1912 while (*before != NULL) {
1913 struct file_lock *fl = *before;
1914 if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
1915 locks_delete_lock(before);
1916 continue;
1917 }
1918 before = &fl->fl_next;
1919 }
1920 unlock_kernel();
1921out:
1922 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1923 lock.fl_ops->fl_release_private(&lock);
1924}
1925
1926EXPORT_SYMBOL(locks_remove_posix);
1927
1928/*
1929 * This function is called on the last close of an open file.
1930 */
1931void locks_remove_flock(struct file *filp)
1932{
1933 struct inode * inode = filp->f_dentry->d_inode;
1934 struct file_lock *fl;
1935 struct file_lock **before;
1936
1937 if (!inode->i_flock)
1938 return;
1939
1940 if (filp->f_op && filp->f_op->flock) {
1941 struct file_lock fl = {
1942 .fl_pid = current->tgid,
1943 .fl_file = filp,
1944 .fl_flags = FL_FLOCK,
1945 .fl_type = F_UNLCK,
1946 .fl_end = OFFSET_MAX,
1947 };
1948 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001949 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1950 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 }
1952
1953 lock_kernel();
1954 before = &inode->i_flock;
1955
1956 while ((fl = *before) != NULL) {
1957 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001958 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 locks_delete_lock(before);
1960 continue;
1961 }
1962 if (IS_LEASE(fl)) {
1963 lease_modify(before, F_UNLCK);
1964 continue;
1965 }
1966 /* What? */
1967 BUG();
1968 }
1969 before = &fl->fl_next;
1970 }
1971 unlock_kernel();
1972}
1973
1974/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 * posix_unblock_lock - stop waiting for a file lock
1976 * @filp: how the file was opened
1977 * @waiter: the lock which was waiting
1978 *
1979 * lockd needs to block waiting for locks.
1980 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001981int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982posix_unblock_lock(struct file *filp, struct file_lock *waiter)
1983{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001984 int status = 0;
1985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01001987 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001989 else
1990 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01001991 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01001992 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993}
1994
1995EXPORT_SYMBOL(posix_unblock_lock);
1996
1997static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
1998{
1999 struct inode *inode = NULL;
2000
2001 if (fl->fl_file != NULL)
2002 inode = fl->fl_file->f_dentry->d_inode;
2003
2004 out += sprintf(out, "%d:%s ", id, pfx);
2005 if (IS_POSIX(fl)) {
2006 out += sprintf(out, "%6s %s ",
2007 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2008 (inode == NULL) ? "*NOINODE*" :
2009 (IS_MANDLOCK(inode) &&
2010 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2011 "MANDATORY" : "ADVISORY ");
2012 } else if (IS_FLOCK(fl)) {
2013 if (fl->fl_type & LOCK_MAND) {
2014 out += sprintf(out, "FLOCK MSNFS ");
2015 } else {
2016 out += sprintf(out, "FLOCK ADVISORY ");
2017 }
2018 } else if (IS_LEASE(fl)) {
2019 out += sprintf(out, "LEASE ");
2020 if (fl->fl_type & F_INPROGRESS)
2021 out += sprintf(out, "BREAKING ");
2022 else if (fl->fl_file)
2023 out += sprintf(out, "ACTIVE ");
2024 else
2025 out += sprintf(out, "BREAKER ");
2026 } else {
2027 out += sprintf(out, "UNKNOWN UNKNOWN ");
2028 }
2029 if (fl->fl_type & LOCK_MAND) {
2030 out += sprintf(out, "%s ",
2031 (fl->fl_type & LOCK_READ)
2032 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2033 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2034 } else {
2035 out += sprintf(out, "%s ",
2036 (fl->fl_type & F_INPROGRESS)
2037 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2038 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2039 }
2040 if (inode) {
2041#ifdef WE_CAN_BREAK_LSLK_NOW
2042 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2043 inode->i_sb->s_id, inode->i_ino);
2044#else
2045 /* userspace relies on this representation of dev_t ;-( */
2046 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2047 MAJOR(inode->i_sb->s_dev),
2048 MINOR(inode->i_sb->s_dev), inode->i_ino);
2049#endif
2050 } else {
2051 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2052 }
2053 if (IS_POSIX(fl)) {
2054 if (fl->fl_end == OFFSET_MAX)
2055 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2056 else
2057 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2058 fl->fl_end);
2059 } else {
2060 out += sprintf(out, "0 EOF\n");
2061 }
2062}
2063
2064static void move_lock_status(char **p, off_t* pos, off_t offset)
2065{
2066 int len;
2067 len = strlen(*p);
2068 if(*pos >= offset) {
2069 /* the complete line is valid */
2070 *p += len;
2071 *pos += len;
2072 return;
2073 }
2074 if(*pos+len > offset) {
2075 /* use the second part of the line */
2076 int i = offset-*pos;
2077 memmove(*p,*p+i,len-i);
2078 *p += len-i;
2079 *pos += len;
2080 return;
2081 }
2082 /* discard the complete line */
2083 *pos += len;
2084}
2085
2086/**
2087 * get_locks_status - reports lock usage in /proc/locks
2088 * @buffer: address in userspace to write into
2089 * @start: ?
2090 * @offset: how far we are through the buffer
2091 * @length: how much to read
2092 */
2093
2094int get_locks_status(char *buffer, char **start, off_t offset, int length)
2095{
2096 struct list_head *tmp;
2097 char *q = buffer;
2098 off_t pos = 0;
2099 int i = 0;
2100
2101 lock_kernel();
2102 list_for_each(tmp, &file_lock_list) {
2103 struct list_head *btmp;
2104 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2105 lock_get_status(q, fl, ++i, "");
2106 move_lock_status(&q, &pos, offset);
2107
2108 if(pos >= offset+length)
2109 goto done;
2110
2111 list_for_each(btmp, &fl->fl_block) {
2112 struct file_lock *bfl = list_entry(btmp,
2113 struct file_lock, fl_block);
2114 lock_get_status(q, bfl, i, " ->");
2115 move_lock_status(&q, &pos, offset);
2116
2117 if(pos >= offset+length)
2118 goto done;
2119 }
2120 }
2121done:
2122 unlock_kernel();
2123 *start = buffer;
2124 if(q-buffer < length)
2125 return (q-buffer);
2126 return length;
2127}
2128
2129/**
2130 * lock_may_read - checks that the region is free of locks
2131 * @inode: the inode that is being read
2132 * @start: the first byte to read
2133 * @len: the number of bytes to read
2134 *
2135 * Emulates Windows locking requirements. Whole-file
2136 * mandatory locks (share modes) can prohibit a read and
2137 * byte-range POSIX locks can prohibit a read if they overlap.
2138 *
2139 * N.B. this function is only ever called
2140 * from knfsd and ownership of locks is never checked.
2141 */
2142int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2143{
2144 struct file_lock *fl;
2145 int result = 1;
2146 lock_kernel();
2147 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2148 if (IS_POSIX(fl)) {
2149 if (fl->fl_type == F_RDLCK)
2150 continue;
2151 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2152 continue;
2153 } else if (IS_FLOCK(fl)) {
2154 if (!(fl->fl_type & LOCK_MAND))
2155 continue;
2156 if (fl->fl_type & LOCK_READ)
2157 continue;
2158 } else
2159 continue;
2160 result = 0;
2161 break;
2162 }
2163 unlock_kernel();
2164 return result;
2165}
2166
2167EXPORT_SYMBOL(lock_may_read);
2168
2169/**
2170 * lock_may_write - checks that the region is free of locks
2171 * @inode: the inode that is being written
2172 * @start: the first byte to write
2173 * @len: the number of bytes to write
2174 *
2175 * Emulates Windows locking requirements. Whole-file
2176 * mandatory locks (share modes) can prohibit a write and
2177 * byte-range POSIX locks can prohibit a write if they overlap.
2178 *
2179 * N.B. this function is only ever called
2180 * from knfsd and ownership of locks is never checked.
2181 */
2182int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2183{
2184 struct file_lock *fl;
2185 int result = 1;
2186 lock_kernel();
2187 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2188 if (IS_POSIX(fl)) {
2189 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2190 continue;
2191 } else if (IS_FLOCK(fl)) {
2192 if (!(fl->fl_type & LOCK_MAND))
2193 continue;
2194 if (fl->fl_type & LOCK_WRITE)
2195 continue;
2196 } else
2197 continue;
2198 result = 0;
2199 break;
2200 }
2201 unlock_kernel();
2202 return result;
2203}
2204
2205EXPORT_SYMBOL(lock_may_write);
2206
2207static inline void __steal_locks(struct file *file, fl_owner_t from)
2208{
2209 struct inode *inode = file->f_dentry->d_inode;
2210 struct file_lock *fl = inode->i_flock;
2211
2212 while (fl) {
2213 if (fl->fl_file == file && fl->fl_owner == from)
2214 fl->fl_owner = current->files;
2215 fl = fl->fl_next;
2216 }
2217}
2218
2219/* When getting ready for executing a binary, we make sure that current
2220 * has a files_struct on its own. Before dropping the old files_struct,
2221 * we take over ownership of all locks for all file descriptors we own.
2222 * Note that we may accidentally steal a lock for a file that a sibling
2223 * has created since the unshare_files() call.
2224 */
2225void steal_locks(fl_owner_t from)
2226{
2227 struct files_struct *files = current->files;
2228 int i, j;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002229 struct fdtable *fdt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230
2231 if (from == files)
2232 return;
2233
2234 lock_kernel();
2235 j = 0;
Dipankar Sarmaca99c1d2006-04-18 22:21:46 -07002236
2237 /*
2238 * We are not taking a ref to the file structures, so
2239 * we need to acquire ->file_lock.
2240 */
2241 spin_lock(&files->file_lock);
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002242 fdt = files_fdtable(files);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243 for (;;) {
2244 unsigned long set;
2245 i = j * __NFDBITS;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002246 if (i >= fdt->max_fdset || i >= fdt->max_fds)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 break;
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002248 set = fdt->open_fds->fds_bits[j++];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 while (set) {
2250 if (set & 1) {
Dipankar Sarmabadf1662005-09-09 13:04:10 -07002251 struct file *file = fdt->fd[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 if (file)
2253 __steal_locks(file, from);
2254 }
2255 i++;
2256 set >>= 1;
2257 }
2258 }
Dipankar Sarmaca99c1d2006-04-18 22:21:46 -07002259 spin_unlock(&files->file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 unlock_kernel();
2261}
2262EXPORT_SYMBOL(steal_locks);
2263
2264static int __init filelock_init(void)
2265{
2266 filelock_cache = kmem_cache_create("file_lock_cache",
2267 sizeof(struct file_lock), 0, SLAB_PANIC,
2268 init_once, NULL);
2269 return 0;
2270}
2271
2272core_initcall(filelock_init);