blob: e0b6a80649a01ff79c26206c039ece3d4faebd0c [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) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700317 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 start = 0;
319 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700320 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 start = filp->f_pos;
322 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700323 case SEEK_END:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 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) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700367 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 start = 0;
369 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700370 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 start = filp->f_pos;
372 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700373 case SEEK_END:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 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 */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700706static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
726 * at the head of the list, but that's secret knowledge known only to
727 * flock_lock_file and posix_lock_file.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400728 *
729 * Note that if called with an FL_EXISTS argument, the caller may determine
730 * whether or not a lock was successfully freed by testing the return
731 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800733static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800735 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 struct file_lock **before;
737 struct inode * inode = filp->f_dentry->d_inode;
738 int error = 0;
739 int found = 0;
740
741 lock_kernel();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400742 if (request->fl_flags & FL_ACCESS)
743 goto find_conflict;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 for_each_lock(inode, before) {
745 struct file_lock *fl = *before;
746 if (IS_POSIX(fl))
747 break;
748 if (IS_LEASE(fl))
749 continue;
750 if (filp != fl->fl_file)
751 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800752 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 goto out;
754 found = 1;
755 locks_delete_lock(before);
756 break;
757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Trond Myklebustf475ae92006-06-29 16:38:32 -0400759 if (request->fl_type == F_UNLCK) {
760 if ((request->fl_flags & FL_EXISTS) && !found)
761 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800762 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400765 error = -ENOMEM;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800766 new_fl = locks_alloc_lock();
767 if (new_fl == NULL)
768 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 /*
770 * If a higher-priority process was blocked on the old file lock,
771 * give it the opportunity to lock the file.
772 */
773 if (found)
774 cond_resched();
775
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400776find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 for_each_lock(inode, before) {
778 struct file_lock *fl = *before;
779 if (IS_POSIX(fl))
780 break;
781 if (IS_LEASE(fl))
782 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800783 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 continue;
785 error = -EAGAIN;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800786 if (request->fl_flags & FL_SLEEP)
787 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 goto out;
789 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400790 if (request->fl_flags & FL_ACCESS)
791 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800792 locks_copy_lock(new_fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 locks_insert_lock(&inode->i_flock, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800794 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400795 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796
797out:
798 unlock_kernel();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800799 if (new_fl)
800 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 return error;
802}
803
Andy Adamson5842add2006-03-26 01:37:26 -0800804static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805{
806 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700807 struct file_lock *new_fl = NULL;
808 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 struct file_lock *left = NULL;
810 struct file_lock *right = NULL;
811 struct file_lock **before;
812 int error, added = 0;
813
814 /*
815 * We may need two file_lock structures for this operation,
816 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700817 *
818 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700820 if (!(request->fl_flags & FL_ACCESS) &&
821 (request->fl_type != F_UNLCK ||
822 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
823 new_fl = locks_alloc_lock();
824 new_fl2 = locks_alloc_lock();
825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827 lock_kernel();
828 if (request->fl_type != F_UNLCK) {
829 for_each_lock(inode, before) {
830 struct file_lock *fl = *before;
831 if (!IS_POSIX(fl))
832 continue;
833 if (!posix_locks_conflict(request, fl))
834 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800835 if (conflock)
836 locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 error = -EAGAIN;
838 if (!(request->fl_flags & FL_SLEEP))
839 goto out;
840 error = -EDEADLK;
841 if (posix_locks_deadlock(request, fl))
842 goto out;
843 error = -EAGAIN;
844 locks_insert_block(fl, request);
845 goto out;
846 }
847 }
848
849 /* If we're just looking for a conflict, we're done. */
850 error = 0;
851 if (request->fl_flags & FL_ACCESS)
852 goto out;
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 * Find the first old lock with the same owner as the new lock.
856 */
857
858 before = &inode->i_flock;
859
860 /* First skip locks owned by other processes. */
861 while ((fl = *before) && (!IS_POSIX(fl) ||
862 !posix_same_owner(request, fl))) {
863 before = &fl->fl_next;
864 }
865
866 /* Process locks with this owner. */
867 while ((fl = *before) && posix_same_owner(request, fl)) {
868 /* Detect adjacent or overlapping regions (if same lock type)
869 */
870 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700871 /* In all comparisons of start vs end, use
872 * "start - 1" rather than "end + 1". If end
873 * is OFFSET_MAX, end + 1 will become negative.
874 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (fl->fl_end < request->fl_start - 1)
876 goto next_lock;
877 /* If the next lock in the list has entirely bigger
878 * addresses than the new one, insert the lock here.
879 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700880 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 break;
882
883 /* If we come here, the new and old lock are of the
884 * same type and adjacent or overlapping. Make one
885 * lock yielding from the lower start address of both
886 * locks to the higher end address.
887 */
888 if (fl->fl_start > request->fl_start)
889 fl->fl_start = request->fl_start;
890 else
891 request->fl_start = fl->fl_start;
892 if (fl->fl_end < request->fl_end)
893 fl->fl_end = request->fl_end;
894 else
895 request->fl_end = fl->fl_end;
896 if (added) {
897 locks_delete_lock(before);
898 continue;
899 }
900 request = fl;
901 added = 1;
902 }
903 else {
904 /* Processing for different lock types is a bit
905 * more complex.
906 */
907 if (fl->fl_end < request->fl_start)
908 goto next_lock;
909 if (fl->fl_start > request->fl_end)
910 break;
911 if (request->fl_type == F_UNLCK)
912 added = 1;
913 if (fl->fl_start < request->fl_start)
914 left = fl;
915 /* If the next lock in the list has a higher end
916 * address than the new one, insert the new one here.
917 */
918 if (fl->fl_end > request->fl_end) {
919 right = fl;
920 break;
921 }
922 if (fl->fl_start >= request->fl_start) {
923 /* The new lock completely replaces an old
924 * one (This may happen several times).
925 */
926 if (added) {
927 locks_delete_lock(before);
928 continue;
929 }
930 /* Replace the old lock with the new one.
931 * Wake up anybody waiting for the old one,
932 * as the change in lock type might satisfy
933 * their needs.
934 */
935 locks_wake_up_blocks(fl);
936 fl->fl_start = request->fl_start;
937 fl->fl_end = request->fl_end;
938 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500939 locks_release_private(fl);
940 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 request = fl;
942 added = 1;
943 }
944 }
945 /* Go on to next lock.
946 */
947 next_lock:
948 before = &fl->fl_next;
949 }
950
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700951 /*
952 * The above code only modifies existing locks in case of
953 * merging or replacing. If new lock(s) need to be inserted
954 * all modifications are done bellow this, so it's safe yet to
955 * bail out.
956 */
957 error = -ENOLCK; /* "no luck" */
958 if (right && left == right && !new_fl2)
959 goto out;
960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 error = 0;
962 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400963 if (request->fl_type == F_UNLCK) {
964 if (request->fl_flags & FL_EXISTS)
965 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400967 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700968
969 if (!new_fl) {
970 error = -ENOLCK;
971 goto out;
972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 locks_copy_lock(new_fl, request);
974 locks_insert_lock(before, new_fl);
975 new_fl = NULL;
976 }
977 if (right) {
978 if (left == right) {
979 /* The new lock breaks the old one in two pieces,
980 * so we have to use the second new lock.
981 */
982 left = new_fl2;
983 new_fl2 = NULL;
984 locks_copy_lock(left, right);
985 locks_insert_lock(before, left);
986 }
987 right->fl_start = request->fl_end + 1;
988 locks_wake_up_blocks(right);
989 }
990 if (left) {
991 left->fl_end = request->fl_start - 1;
992 locks_wake_up_blocks(left);
993 }
994 out:
995 unlock_kernel();
996 /*
997 * Free any unused locks.
998 */
999 if (new_fl)
1000 locks_free_lock(new_fl);
1001 if (new_fl2)
1002 locks_free_lock(new_fl2);
1003 return error;
1004}
1005
1006/**
1007 * posix_lock_file - 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
Trond Myklebustf475ae92006-06-29 16:38:32 -04001014 *
1015 * Note that if called with an FL_EXISTS argument, the caller may determine
1016 * whether or not a lock was successfully freed by testing the return
1017 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 */
1019int posix_lock_file(struct file *filp, struct file_lock *fl)
1020{
Andy Adamson5842add2006-03-26 01:37:26 -08001021 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022}
Andy Adamson5842add2006-03-26 01:37:26 -08001023EXPORT_SYMBOL(posix_lock_file);
1024
1025/**
1026 * posix_lock_file_conf - Apply a POSIX-style lock to a file
1027 * @filp: The file to apply the lock to
1028 * @fl: The lock to be applied
1029 * @conflock: Place to return a copy of the conflicting lock, if found.
1030 *
1031 * Except for the conflock parameter, acts just like posix_lock_file.
1032 */
1033int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
1034 struct file_lock *conflock)
1035{
1036 return __posix_lock_file_conf(filp->f_dentry->d_inode, fl, conflock);
1037}
1038EXPORT_SYMBOL(posix_lock_file_conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
1040/**
1041 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1042 * @filp: The file to apply the lock to
1043 * @fl: The lock to be applied
1044 *
1045 * Add a POSIX style lock to a file.
1046 * We merge adjacent & overlapping locks whenever possible.
1047 * POSIX locks are sorted by owner task, then by starting address
1048 */
1049int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1050{
1051 int error;
1052 might_sleep ();
1053 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001054 error = posix_lock_file(filp, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1056 break;
1057 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1058 if (!error)
1059 continue;
1060
1061 locks_delete_block(fl);
1062 break;
1063 }
1064 return error;
1065}
1066EXPORT_SYMBOL(posix_lock_file_wait);
1067
1068/**
1069 * locks_mandatory_locked - Check for an active lock
1070 * @inode: the file to check
1071 *
1072 * Searches the inode's list of locks to find any POSIX locks which conflict.
1073 * This function is called from locks_verify_locked() only.
1074 */
1075int locks_mandatory_locked(struct inode *inode)
1076{
1077 fl_owner_t owner = current->files;
1078 struct file_lock *fl;
1079
1080 /*
1081 * Search the lock list for this inode for any POSIX locks.
1082 */
1083 lock_kernel();
1084 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1085 if (!IS_POSIX(fl))
1086 continue;
1087 if (fl->fl_owner != owner)
1088 break;
1089 }
1090 unlock_kernel();
1091 return fl ? -EAGAIN : 0;
1092}
1093
1094/**
1095 * locks_mandatory_area - Check for a conflicting lock
1096 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1097 * for shared
1098 * @inode: the file to check
1099 * @filp: how the file was opened (if it was)
1100 * @offset: start of area to check
1101 * @count: length of area to check
1102 *
1103 * Searches the inode's list of locks to find any POSIX locks which conflict.
1104 * This function is called from rw_verify_area() and
1105 * locks_verify_truncate().
1106 */
1107int locks_mandatory_area(int read_write, struct inode *inode,
1108 struct file *filp, loff_t offset,
1109 size_t count)
1110{
1111 struct file_lock fl;
1112 int error;
1113
1114 locks_init_lock(&fl);
1115 fl.fl_owner = current->files;
1116 fl.fl_pid = current->tgid;
1117 fl.fl_file = filp;
1118 fl.fl_flags = FL_POSIX | FL_ACCESS;
1119 if (filp && !(filp->f_flags & O_NONBLOCK))
1120 fl.fl_flags |= FL_SLEEP;
1121 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1122 fl.fl_start = offset;
1123 fl.fl_end = offset + count - 1;
1124
1125 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001126 error = __posix_lock_file_conf(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (error != -EAGAIN)
1128 break;
1129 if (!(fl.fl_flags & FL_SLEEP))
1130 break;
1131 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1132 if (!error) {
1133 /*
1134 * If we've been sleeping someone might have
1135 * changed the permissions behind our back.
1136 */
1137 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1138 continue;
1139 }
1140
1141 locks_delete_block(&fl);
1142 break;
1143 }
1144
1145 return error;
1146}
1147
1148EXPORT_SYMBOL(locks_mandatory_area);
1149
1150/* We already had a lease on this file; just change its type */
1151int lease_modify(struct file_lock **before, int arg)
1152{
1153 struct file_lock *fl = *before;
1154 int error = assign_type(fl, arg);
1155
1156 if (error)
1157 return error;
1158 locks_wake_up_blocks(fl);
1159 if (arg == F_UNLCK)
1160 locks_delete_lock(before);
1161 return 0;
1162}
1163
1164EXPORT_SYMBOL(lease_modify);
1165
1166static void time_out_leases(struct inode *inode)
1167{
1168 struct file_lock **before;
1169 struct file_lock *fl;
1170
1171 before = &inode->i_flock;
1172 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1173 if ((fl->fl_break_time == 0)
1174 || time_before(jiffies, fl->fl_break_time)) {
1175 before = &fl->fl_next;
1176 continue;
1177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1179 if (fl == *before) /* lease_modify may have freed fl */
1180 before = &fl->fl_next;
1181 }
1182}
1183
1184/**
1185 * __break_lease - revoke all outstanding leases on file
1186 * @inode: the inode of the file to return
1187 * @mode: the open mode (read or write)
1188 *
1189 * break_lease (inlined for speed) has checked there already
1190 * is a lease on this file. Leases are broken on a call to open()
1191 * or truncate(). This function can sleep unless you
1192 * specified %O_NONBLOCK to your open().
1193 */
1194int __break_lease(struct inode *inode, unsigned int mode)
1195{
1196 int error = 0, future;
1197 struct file_lock *new_fl, *flock;
1198 struct file_lock *fl;
1199 int alloc_err;
1200 unsigned long break_time;
1201 int i_have_this_lease = 0;
1202
1203 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1204 &new_fl);
1205
1206 lock_kernel();
1207
1208 time_out_leases(inode);
1209
1210 flock = inode->i_flock;
1211 if ((flock == NULL) || !IS_LEASE(flock))
1212 goto out;
1213
1214 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1215 if (fl->fl_owner == current->files)
1216 i_have_this_lease = 1;
1217
1218 if (mode & FMODE_WRITE) {
1219 /* If we want write access, we have to revoke any lease. */
1220 future = F_UNLCK | F_INPROGRESS;
1221 } else if (flock->fl_type & F_INPROGRESS) {
1222 /* If the lease is already being broken, we just leave it */
1223 future = flock->fl_type;
1224 } else if (flock->fl_type & F_WRLCK) {
1225 /* Downgrade the exclusive lease to a read-only lease. */
1226 future = F_RDLCK | F_INPROGRESS;
1227 } else {
1228 /* the existing lease was read-only, so we can read too. */
1229 goto out;
1230 }
1231
1232 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1233 error = alloc_err;
1234 goto out;
1235 }
1236
1237 break_time = 0;
1238 if (lease_break_time > 0) {
1239 break_time = jiffies + lease_break_time * HZ;
1240 if (break_time == 0)
1241 break_time++; /* so that 0 means no break time */
1242 }
1243
1244 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1245 if (fl->fl_type != future) {
1246 fl->fl_type = future;
1247 fl->fl_break_time = break_time;
1248 /* lease must have lmops break callback */
1249 fl->fl_lmops->fl_break(fl);
1250 }
1251 }
1252
1253 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1254 error = -EWOULDBLOCK;
1255 goto out;
1256 }
1257
1258restart:
1259 break_time = flock->fl_break_time;
1260 if (break_time != 0) {
1261 break_time -= jiffies;
1262 if (break_time == 0)
1263 break_time++;
1264 }
1265 error = locks_block_on_timeout(flock, new_fl, break_time);
1266 if (error >= 0) {
1267 if (error == 0)
1268 time_out_leases(inode);
1269 /* Wait for the next lease that has not been broken yet */
1270 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1271 flock = flock->fl_next) {
1272 if (flock->fl_type & F_INPROGRESS)
1273 goto restart;
1274 }
1275 error = 0;
1276 }
1277
1278out:
1279 unlock_kernel();
1280 if (!alloc_err)
1281 locks_free_lock(new_fl);
1282 return error;
1283}
1284
1285EXPORT_SYMBOL(__break_lease);
1286
1287/**
1288 * lease_get_mtime
1289 * @inode: the inode
1290 * @time: pointer to a timespec which will contain the last modified time
1291 *
1292 * This is to force NFS clients to flush their caches for files with
1293 * exclusive leases. The justification is that if someone has an
1294 * exclusive lease, then they could be modifiying it.
1295 */
1296void lease_get_mtime(struct inode *inode, struct timespec *time)
1297{
1298 struct file_lock *flock = inode->i_flock;
1299 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1300 *time = current_fs_time(inode->i_sb);
1301 else
1302 *time = inode->i_mtime;
1303}
1304
1305EXPORT_SYMBOL(lease_get_mtime);
1306
1307/**
1308 * fcntl_getlease - Enquire what lease is currently active
1309 * @filp: the file
1310 *
1311 * The value returned by this function will be one of
1312 * (if no lease break is pending):
1313 *
1314 * %F_RDLCK to indicate a shared lease is held.
1315 *
1316 * %F_WRLCK to indicate an exclusive lease is held.
1317 *
1318 * %F_UNLCK to indicate no lease is held.
1319 *
1320 * (if a lease break is pending):
1321 *
1322 * %F_RDLCK to indicate an exclusive lease needs to be
1323 * changed to a shared lease (or removed).
1324 *
1325 * %F_UNLCK to indicate the lease needs to be removed.
1326 *
1327 * XXX: sfr & willy disagree over whether F_INPROGRESS
1328 * should be returned to userspace.
1329 */
1330int fcntl_getlease(struct file *filp)
1331{
1332 struct file_lock *fl;
1333 int type = F_UNLCK;
1334
1335 lock_kernel();
1336 time_out_leases(filp->f_dentry->d_inode);
1337 for (fl = filp->f_dentry->d_inode->i_flock; fl && IS_LEASE(fl);
1338 fl = fl->fl_next) {
1339 if (fl->fl_file == filp) {
1340 type = fl->fl_type & ~F_INPROGRESS;
1341 break;
1342 }
1343 }
1344 unlock_kernel();
1345 return type;
1346}
1347
1348/**
1349 * __setlease - sets a lease on an open file
1350 * @filp: file pointer
1351 * @arg: type of lease to obtain
1352 * @flp: input - file_lock to use, output - file_lock inserted
1353 *
1354 * The (input) flp->fl_lmops->fl_break function is required
1355 * by break_lease().
1356 *
1357 * Called with kernel lock held.
1358 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001359static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001361 struct file_lock *fl, **before, **my_before = NULL, *lease;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 struct dentry *dentry = filp->f_dentry;
1363 struct inode *inode = dentry->d_inode;
1364 int error, rdlease_count = 0, wrlease_count = 0;
1365
1366 time_out_leases(inode);
1367
1368 error = -EINVAL;
1369 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1370 goto out;
1371
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001372 lease = *flp;
1373
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 error = -EAGAIN;
1375 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1376 goto out;
1377 if ((arg == F_WRLCK)
1378 && ((atomic_read(&dentry->d_count) > 1)
1379 || (atomic_read(&inode->i_count) > 1)))
1380 goto out;
1381
1382 /*
1383 * At this point, we know that if there is an exclusive
1384 * lease on this file, then we hold it on this filp
1385 * (otherwise our open of this file would have blocked).
1386 * And if we are trying to acquire an exclusive lease,
1387 * then the file is not open by anyone (including us)
1388 * except for this filp.
1389 */
1390 for (before = &inode->i_flock;
1391 ((fl = *before) != NULL) && IS_LEASE(fl);
1392 before = &fl->fl_next) {
1393 if (lease->fl_lmops->fl_mylease(fl, lease))
1394 my_before = before;
1395 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1396 /*
1397 * Someone is in the process of opening this
1398 * file for writing so we may not take an
1399 * exclusive lease on it.
1400 */
1401 wrlease_count++;
1402 else
1403 rdlease_count++;
1404 }
1405
1406 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1407 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1408 goto out;
1409
1410 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001411 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 error = lease->fl_lmops->fl_change(my_before, arg);
1413 goto out;
1414 }
1415
1416 error = 0;
1417 if (arg == F_UNLCK)
1418 goto out;
1419
1420 error = -EINVAL;
1421 if (!leases_enable)
1422 goto out;
1423
Trond Myklebust74361cb2006-08-14 08:54:48 -07001424 error = -ENOMEM;
1425 fl = locks_alloc_lock();
1426 if (fl == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 goto out;
1428
1429 locks_copy_lock(fl, lease);
1430
1431 locks_insert_lock(before, fl);
1432
1433 *flp = fl;
Trond Myklebust74361cb2006-08-14 08:54:48 -07001434 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435out:
1436 return error;
1437}
1438
1439 /**
1440 * setlease - sets a lease on an open file
1441 * @filp: file pointer
1442 * @arg: type of lease to obtain
1443 * @lease: file_lock to use
1444 *
1445 * Call this to establish a lease on the file.
1446 * The fl_lmops fl_break function is required by break_lease
1447 */
1448
1449int setlease(struct file *filp, long arg, struct file_lock **lease)
1450{
1451 struct dentry *dentry = filp->f_dentry;
1452 struct inode *inode = dentry->d_inode;
1453 int error;
1454
1455 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1456 return -EACCES;
1457 if (!S_ISREG(inode->i_mode))
1458 return -EINVAL;
1459 error = security_file_lock(filp, arg);
1460 if (error)
1461 return error;
1462
1463 lock_kernel();
1464 error = __setlease(filp, arg, lease);
1465 unlock_kernel();
1466
1467 return error;
1468}
1469
1470EXPORT_SYMBOL(setlease);
1471
1472/**
1473 * fcntl_setlease - sets a lease on an open file
1474 * @fd: open file descriptor
1475 * @filp: file pointer
1476 * @arg: type of lease to obtain
1477 *
1478 * Call this fcntl to establish a lease on the file.
1479 * Note that you also need to call %F_SETSIG to
1480 * receive a signal when the lease is broken.
1481 */
1482int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1483{
1484 struct file_lock fl, *flp = &fl;
1485 struct dentry *dentry = filp->f_dentry;
1486 struct inode *inode = dentry->d_inode;
1487 int error;
1488
1489 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1490 return -EACCES;
1491 if (!S_ISREG(inode->i_mode))
1492 return -EINVAL;
1493 error = security_file_lock(filp, arg);
1494 if (error)
1495 return error;
1496
1497 locks_init_lock(&fl);
1498 error = lease_init(filp, arg, &fl);
1499 if (error)
1500 return error;
1501
1502 lock_kernel();
1503
1504 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001505 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 goto out_unlock;
1507
1508 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1509 if (error < 0) {
1510 /* remove lease just inserted by __setlease */
1511 flp->fl_type = F_UNLCK | F_INPROGRESS;
1512 flp->fl_break_time = jiffies- 10;
1513 time_out_leases(inode);
1514 goto out_unlock;
1515 }
1516
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001517 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518out_unlock:
1519 unlock_kernel();
1520 return error;
1521}
1522
1523/**
1524 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1525 * @filp: The file to apply the lock to
1526 * @fl: The lock to be applied
1527 *
1528 * Add a FLOCK style lock to a file.
1529 */
1530int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1531{
1532 int error;
1533 might_sleep();
1534 for (;;) {
1535 error = flock_lock_file(filp, fl);
1536 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1537 break;
1538 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1539 if (!error)
1540 continue;
1541
1542 locks_delete_block(fl);
1543 break;
1544 }
1545 return error;
1546}
1547
1548EXPORT_SYMBOL(flock_lock_file_wait);
1549
1550/**
1551 * sys_flock: - flock() system call.
1552 * @fd: the file descriptor to lock.
1553 * @cmd: the type of lock to apply.
1554 *
1555 * Apply a %FL_FLOCK style lock to an open file descriptor.
1556 * The @cmd can be one of
1557 *
1558 * %LOCK_SH -- a shared lock.
1559 *
1560 * %LOCK_EX -- an exclusive lock.
1561 *
1562 * %LOCK_UN -- remove an existing lock.
1563 *
1564 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1565 *
1566 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1567 * processes read and write access respectively.
1568 */
1569asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1570{
1571 struct file *filp;
1572 struct file_lock *lock;
1573 int can_sleep, unlock;
1574 int error;
1575
1576 error = -EBADF;
1577 filp = fget(fd);
1578 if (!filp)
1579 goto out;
1580
1581 can_sleep = !(cmd & LOCK_NB);
1582 cmd &= ~LOCK_NB;
1583 unlock = (cmd == LOCK_UN);
1584
1585 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1586 goto out_putf;
1587
1588 error = flock_make_lock(filp, &lock, cmd);
1589 if (error)
1590 goto out_putf;
1591 if (can_sleep)
1592 lock->fl_flags |= FL_SLEEP;
1593
1594 error = security_file_lock(filp, cmd);
1595 if (error)
1596 goto out_free;
1597
1598 if (filp->f_op && filp->f_op->flock)
1599 error = filp->f_op->flock(filp,
1600 (can_sleep) ? F_SETLKW : F_SETLK,
1601 lock);
1602 else
1603 error = flock_lock_file_wait(filp, lock);
1604
1605 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001606 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607
1608 out_putf:
1609 fput(filp);
1610 out:
1611 return error;
1612}
1613
1614/* Report the first existing lock that would conflict with l.
1615 * This implements the F_GETLK command of fcntl().
1616 */
1617int fcntl_getlk(struct file *filp, struct flock __user *l)
1618{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001619 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 struct flock flock;
1621 int error;
1622
1623 error = -EFAULT;
1624 if (copy_from_user(&flock, l, sizeof(flock)))
1625 goto out;
1626 error = -EINVAL;
1627 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1628 goto out;
1629
1630 error = flock_to_posix_lock(filp, &file_lock, &flock);
1631 if (error)
1632 goto out;
1633
1634 if (filp->f_op && filp->f_op->lock) {
1635 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001636 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1637 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 if (error < 0)
1639 goto out;
1640 else
1641 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1642 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001643 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 }
1645
1646 flock.l_type = F_UNLCK;
1647 if (fl != NULL) {
1648 flock.l_pid = fl->fl_pid;
1649#if BITS_PER_LONG == 32
1650 /*
1651 * Make sure we can represent the posix lock via
1652 * legacy 32bit flock.
1653 */
1654 error = -EOVERFLOW;
1655 if (fl->fl_start > OFFT_OFFSET_MAX)
1656 goto out;
1657 if ((fl->fl_end != OFFSET_MAX)
1658 && (fl->fl_end > OFFT_OFFSET_MAX))
1659 goto out;
1660#endif
1661 flock.l_start = fl->fl_start;
1662 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1663 fl->fl_end - fl->fl_start + 1;
1664 flock.l_whence = 0;
1665 flock.l_type = fl->fl_type;
1666 }
1667 error = -EFAULT;
1668 if (!copy_to_user(l, &flock, sizeof(flock)))
1669 error = 0;
1670out:
1671 return error;
1672}
1673
1674/* Apply the lock described by l to an open file descriptor.
1675 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1676 */
Peter Staubachc2936212005-07-27 11:45:09 -07001677int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1678 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679{
1680 struct file_lock *file_lock = locks_alloc_lock();
1681 struct flock flock;
1682 struct inode *inode;
1683 int error;
1684
1685 if (file_lock == NULL)
1686 return -ENOLCK;
1687
1688 /*
1689 * This might block, so we do it before checking the inode.
1690 */
1691 error = -EFAULT;
1692 if (copy_from_user(&flock, l, sizeof(flock)))
1693 goto out;
1694
1695 inode = filp->f_dentry->d_inode;
1696
1697 /* Don't allow mandatory locks on files that may be memory mapped
1698 * and shared.
1699 */
1700 if (IS_MANDLOCK(inode) &&
1701 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1702 mapping_writably_mapped(filp->f_mapping)) {
1703 error = -EAGAIN;
1704 goto out;
1705 }
1706
Peter Staubachc2936212005-07-27 11:45:09 -07001707again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 error = flock_to_posix_lock(filp, file_lock, &flock);
1709 if (error)
1710 goto out;
1711 if (cmd == F_SETLKW) {
1712 file_lock->fl_flags |= FL_SLEEP;
1713 }
1714
1715 error = -EBADF;
1716 switch (flock.l_type) {
1717 case F_RDLCK:
1718 if (!(filp->f_mode & FMODE_READ))
1719 goto out;
1720 break;
1721 case F_WRLCK:
1722 if (!(filp->f_mode & FMODE_WRITE))
1723 goto out;
1724 break;
1725 case F_UNLCK:
1726 break;
1727 default:
1728 error = -EINVAL;
1729 goto out;
1730 }
1731
1732 error = security_file_lock(filp, file_lock->fl_type);
1733 if (error)
1734 goto out;
1735
Peter Staubachc2936212005-07-27 11:45:09 -07001736 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001738 else {
1739 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001740 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001741 if ((error != -EAGAIN) || (cmd == F_SETLK))
1742 break;
1743 error = wait_event_interruptible(file_lock->fl_wait,
1744 !file_lock->fl_next);
1745 if (!error)
1746 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Peter Staubachc2936212005-07-27 11:45:09 -07001748 locks_delete_block(file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 break;
Peter Staubachc2936212005-07-27 11:45:09 -07001750 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 }
1752
Peter Staubachc2936212005-07-27 11:45:09 -07001753 /*
1754 * Attempt to detect a close/fcntl race and recover by
1755 * releasing the lock that was just acquired.
1756 */
1757 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1758 flock.l_type = F_UNLCK;
1759 goto again;
1760 }
1761
1762out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 locks_free_lock(file_lock);
1764 return error;
1765}
1766
1767#if BITS_PER_LONG == 32
1768/* Report the first existing lock that would conflict with l.
1769 * This implements the F_GETLK command of fcntl().
1770 */
1771int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1772{
Andy Adamson8dc7c312006-03-20 13:44:26 -05001773 struct file_lock *fl, cfl, file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 struct flock64 flock;
1775 int error;
1776
1777 error = -EFAULT;
1778 if (copy_from_user(&flock, l, sizeof(flock)))
1779 goto out;
1780 error = -EINVAL;
1781 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1782 goto out;
1783
1784 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1785 if (error)
1786 goto out;
1787
1788 if (filp->f_op && filp->f_op->lock) {
1789 error = filp->f_op->lock(filp, F_GETLK, &file_lock);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001790 if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
1791 file_lock.fl_ops->fl_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 if (error < 0)
1793 goto out;
1794 else
1795 fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
1796 } else {
Andy Adamson8dc7c312006-03-20 13:44:26 -05001797 fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 }
1799
1800 flock.l_type = F_UNLCK;
1801 if (fl != NULL) {
1802 flock.l_pid = fl->fl_pid;
1803 flock.l_start = fl->fl_start;
1804 flock.l_len = fl->fl_end == OFFSET_MAX ? 0 :
1805 fl->fl_end - fl->fl_start + 1;
1806 flock.l_whence = 0;
1807 flock.l_type = fl->fl_type;
1808 }
1809 error = -EFAULT;
1810 if (!copy_to_user(l, &flock, sizeof(flock)))
1811 error = 0;
1812
1813out:
1814 return error;
1815}
1816
1817/* Apply the lock described by l to an open file descriptor.
1818 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1819 */
Peter Staubachc2936212005-07-27 11:45:09 -07001820int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1821 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822{
1823 struct file_lock *file_lock = locks_alloc_lock();
1824 struct flock64 flock;
1825 struct inode *inode;
1826 int error;
1827
1828 if (file_lock == NULL)
1829 return -ENOLCK;
1830
1831 /*
1832 * This might block, so we do it before checking the inode.
1833 */
1834 error = -EFAULT;
1835 if (copy_from_user(&flock, l, sizeof(flock)))
1836 goto out;
1837
1838 inode = filp->f_dentry->d_inode;
1839
1840 /* Don't allow mandatory locks on files that may be memory mapped
1841 * and shared.
1842 */
1843 if (IS_MANDLOCK(inode) &&
1844 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1845 mapping_writably_mapped(filp->f_mapping)) {
1846 error = -EAGAIN;
1847 goto out;
1848 }
1849
Peter Staubachc2936212005-07-27 11:45:09 -07001850again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 error = flock64_to_posix_lock(filp, file_lock, &flock);
1852 if (error)
1853 goto out;
1854 if (cmd == F_SETLKW64) {
1855 file_lock->fl_flags |= FL_SLEEP;
1856 }
1857
1858 error = -EBADF;
1859 switch (flock.l_type) {
1860 case F_RDLCK:
1861 if (!(filp->f_mode & FMODE_READ))
1862 goto out;
1863 break;
1864 case F_WRLCK:
1865 if (!(filp->f_mode & FMODE_WRITE))
1866 goto out;
1867 break;
1868 case F_UNLCK:
1869 break;
1870 default:
1871 error = -EINVAL;
1872 goto out;
1873 }
1874
1875 error = security_file_lock(filp, file_lock->fl_type);
1876 if (error)
1877 goto out;
1878
Peter Staubachc2936212005-07-27 11:45:09 -07001879 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 error = filp->f_op->lock(filp, cmd, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001881 else {
1882 for (;;) {
Andy Adamson5842add2006-03-26 01:37:26 -08001883 error = posix_lock_file(filp, file_lock);
Peter Staubachc2936212005-07-27 11:45:09 -07001884 if ((error != -EAGAIN) || (cmd == F_SETLK64))
1885 break;
1886 error = wait_event_interruptible(file_lock->fl_wait,
1887 !file_lock->fl_next);
1888 if (!error)
1889 continue;
1890
1891 locks_delete_block(file_lock);
1892 break;
1893 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
1895
Peter Staubachc2936212005-07-27 11:45:09 -07001896 /*
1897 * Attempt to detect a close/fcntl race and recover by
1898 * releasing the lock that was just acquired.
1899 */
1900 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1901 flock.l_type = F_UNLCK;
1902 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 }
1904
1905out:
1906 locks_free_lock(file_lock);
1907 return error;
1908}
1909#endif /* BITS_PER_LONG == 32 */
1910
1911/*
1912 * This function is called when the file is being removed
1913 * from the task's fd array. POSIX locks belonging to this task
1914 * are deleted at this time.
1915 */
1916void locks_remove_posix(struct file *filp, fl_owner_t owner)
1917{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001918 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920 /*
1921 * If there are no locks held on this file, we don't need to call
1922 * posix_lock_file(). Another process could be setting a lock on this
1923 * file at the same time, but we wouldn't remove that lock anyway.
1924 */
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001925 if (!filp->f_dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 return;
1927
1928 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001929 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 lock.fl_start = 0;
1931 lock.fl_end = OFFSET_MAX;
1932 lock.fl_owner = owner;
1933 lock.fl_pid = current->tgid;
1934 lock.fl_file = filp;
1935 lock.fl_ops = NULL;
1936 lock.fl_lmops = NULL;
1937
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001938 if (filp->f_op && filp->f_op->lock != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 filp->f_op->lock(filp, F_SETLK, &lock);
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001940 else
1941 posix_lock_file(filp, &lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1944 lock.fl_ops->fl_release_private(&lock);
1945}
1946
1947EXPORT_SYMBOL(locks_remove_posix);
1948
1949/*
1950 * This function is called on the last close of an open file.
1951 */
1952void locks_remove_flock(struct file *filp)
1953{
1954 struct inode * inode = filp->f_dentry->d_inode;
1955 struct file_lock *fl;
1956 struct file_lock **before;
1957
1958 if (!inode->i_flock)
1959 return;
1960
1961 if (filp->f_op && filp->f_op->flock) {
1962 struct file_lock fl = {
1963 .fl_pid = current->tgid,
1964 .fl_file = filp,
1965 .fl_flags = FL_FLOCK,
1966 .fl_type = F_UNLCK,
1967 .fl_end = OFFSET_MAX,
1968 };
1969 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00001970 if (fl.fl_ops && fl.fl_ops->fl_release_private)
1971 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 }
1973
1974 lock_kernel();
1975 before = &inode->i_flock;
1976
1977 while ((fl = *before) != NULL) {
1978 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07001979 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 locks_delete_lock(before);
1981 continue;
1982 }
1983 if (IS_LEASE(fl)) {
1984 lease_modify(before, F_UNLCK);
1985 continue;
1986 }
1987 /* What? */
1988 BUG();
1989 }
1990 before = &fl->fl_next;
1991 }
1992 unlock_kernel();
1993}
1994
1995/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 * posix_unblock_lock - stop waiting for a file lock
1997 * @filp: how the file was opened
1998 * @waiter: the lock which was waiting
1999 *
2000 * lockd needs to block waiting for locks.
2001 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002002int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2004{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002005 int status = 0;
2006
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002008 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002010 else
2011 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002012 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002013 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014}
2015
2016EXPORT_SYMBOL(posix_unblock_lock);
2017
2018static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2019{
2020 struct inode *inode = NULL;
2021
2022 if (fl->fl_file != NULL)
2023 inode = fl->fl_file->f_dentry->d_inode;
2024
2025 out += sprintf(out, "%d:%s ", id, pfx);
2026 if (IS_POSIX(fl)) {
2027 out += sprintf(out, "%6s %s ",
2028 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2029 (inode == NULL) ? "*NOINODE*" :
2030 (IS_MANDLOCK(inode) &&
2031 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2032 "MANDATORY" : "ADVISORY ");
2033 } else if (IS_FLOCK(fl)) {
2034 if (fl->fl_type & LOCK_MAND) {
2035 out += sprintf(out, "FLOCK MSNFS ");
2036 } else {
2037 out += sprintf(out, "FLOCK ADVISORY ");
2038 }
2039 } else if (IS_LEASE(fl)) {
2040 out += sprintf(out, "LEASE ");
2041 if (fl->fl_type & F_INPROGRESS)
2042 out += sprintf(out, "BREAKING ");
2043 else if (fl->fl_file)
2044 out += sprintf(out, "ACTIVE ");
2045 else
2046 out += sprintf(out, "BREAKER ");
2047 } else {
2048 out += sprintf(out, "UNKNOWN UNKNOWN ");
2049 }
2050 if (fl->fl_type & LOCK_MAND) {
2051 out += sprintf(out, "%s ",
2052 (fl->fl_type & LOCK_READ)
2053 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2054 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2055 } else {
2056 out += sprintf(out, "%s ",
2057 (fl->fl_type & F_INPROGRESS)
2058 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2059 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2060 }
2061 if (inode) {
2062#ifdef WE_CAN_BREAK_LSLK_NOW
2063 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2064 inode->i_sb->s_id, inode->i_ino);
2065#else
2066 /* userspace relies on this representation of dev_t ;-( */
2067 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2068 MAJOR(inode->i_sb->s_dev),
2069 MINOR(inode->i_sb->s_dev), inode->i_ino);
2070#endif
2071 } else {
2072 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2073 }
2074 if (IS_POSIX(fl)) {
2075 if (fl->fl_end == OFFSET_MAX)
2076 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2077 else
2078 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2079 fl->fl_end);
2080 } else {
2081 out += sprintf(out, "0 EOF\n");
2082 }
2083}
2084
2085static void move_lock_status(char **p, off_t* pos, off_t offset)
2086{
2087 int len;
2088 len = strlen(*p);
2089 if(*pos >= offset) {
2090 /* the complete line is valid */
2091 *p += len;
2092 *pos += len;
2093 return;
2094 }
2095 if(*pos+len > offset) {
2096 /* use the second part of the line */
2097 int i = offset-*pos;
2098 memmove(*p,*p+i,len-i);
2099 *p += len-i;
2100 *pos += len;
2101 return;
2102 }
2103 /* discard the complete line */
2104 *pos += len;
2105}
2106
2107/**
2108 * get_locks_status - reports lock usage in /proc/locks
2109 * @buffer: address in userspace to write into
2110 * @start: ?
2111 * @offset: how far we are through the buffer
2112 * @length: how much to read
2113 */
2114
2115int get_locks_status(char *buffer, char **start, off_t offset, int length)
2116{
2117 struct list_head *tmp;
2118 char *q = buffer;
2119 off_t pos = 0;
2120 int i = 0;
2121
2122 lock_kernel();
2123 list_for_each(tmp, &file_lock_list) {
2124 struct list_head *btmp;
2125 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2126 lock_get_status(q, fl, ++i, "");
2127 move_lock_status(&q, &pos, offset);
2128
2129 if(pos >= offset+length)
2130 goto done;
2131
2132 list_for_each(btmp, &fl->fl_block) {
2133 struct file_lock *bfl = list_entry(btmp,
2134 struct file_lock, fl_block);
2135 lock_get_status(q, bfl, i, " ->");
2136 move_lock_status(&q, &pos, offset);
2137
2138 if(pos >= offset+length)
2139 goto done;
2140 }
2141 }
2142done:
2143 unlock_kernel();
2144 *start = buffer;
2145 if(q-buffer < length)
2146 return (q-buffer);
2147 return length;
2148}
2149
2150/**
2151 * lock_may_read - checks that the region is free of locks
2152 * @inode: the inode that is being read
2153 * @start: the first byte to read
2154 * @len: the number of bytes to read
2155 *
2156 * Emulates Windows locking requirements. Whole-file
2157 * mandatory locks (share modes) can prohibit a read and
2158 * byte-range POSIX locks can prohibit a read if they overlap.
2159 *
2160 * N.B. this function is only ever called
2161 * from knfsd and ownership of locks is never checked.
2162 */
2163int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2164{
2165 struct file_lock *fl;
2166 int result = 1;
2167 lock_kernel();
2168 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2169 if (IS_POSIX(fl)) {
2170 if (fl->fl_type == F_RDLCK)
2171 continue;
2172 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2173 continue;
2174 } else if (IS_FLOCK(fl)) {
2175 if (!(fl->fl_type & LOCK_MAND))
2176 continue;
2177 if (fl->fl_type & LOCK_READ)
2178 continue;
2179 } else
2180 continue;
2181 result = 0;
2182 break;
2183 }
2184 unlock_kernel();
2185 return result;
2186}
2187
2188EXPORT_SYMBOL(lock_may_read);
2189
2190/**
2191 * lock_may_write - checks that the region is free of locks
2192 * @inode: the inode that is being written
2193 * @start: the first byte to write
2194 * @len: the number of bytes to write
2195 *
2196 * Emulates Windows locking requirements. Whole-file
2197 * mandatory locks (share modes) can prohibit a write and
2198 * byte-range POSIX locks can prohibit a write if they overlap.
2199 *
2200 * N.B. this function is only ever called
2201 * from knfsd and ownership of locks is never checked.
2202 */
2203int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2204{
2205 struct file_lock *fl;
2206 int result = 1;
2207 lock_kernel();
2208 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2209 if (IS_POSIX(fl)) {
2210 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2211 continue;
2212 } else if (IS_FLOCK(fl)) {
2213 if (!(fl->fl_type & LOCK_MAND))
2214 continue;
2215 if (fl->fl_type & LOCK_WRITE)
2216 continue;
2217 } else
2218 continue;
2219 result = 0;
2220 break;
2221 }
2222 unlock_kernel();
2223 return result;
2224}
2225
2226EXPORT_SYMBOL(lock_may_write);
2227
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228static int __init filelock_init(void)
2229{
2230 filelock_cache = kmem_cache_create("file_lock_cache",
2231 sizeof(struct file_lock), 0, SLAB_PANIC,
2232 init_once, NULL);
2233 return 0;
2234}
2235
2236core_initcall(filelock_init);