blob: 53b0cd15320204ee8cf470729a4dbc144b626c5c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
119#include <linux/fs.h>
120#include <linux/init.h>
121#include <linux/module.h>
122#include <linux/security.h>
123#include <linux/slab.h>
124#include <linux/smp_lock.h>
125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129#include <asm/semaphore.h>
130#include <asm/uaccess.h>
131
132#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
135
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
144
Christoph Lametere18b8902006-12-06 20:33:20 -0800145static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
147/* Allocate an empty lock structure. */
148static struct file_lock *locks_alloc_lock(void)
149{
Christoph Lametere94b1762006-12-06 20:33:17 -0800150 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Trond Myklebust47831f32006-03-20 13:44:05 -0500153static void locks_release_private(struct file_lock *fl)
154{
155 if (fl->fl_ops) {
156 if (fl->fl_ops->fl_release_private)
157 fl->fl_ops->fl_release_private(fl);
158 fl->fl_ops = NULL;
159 }
160 if (fl->fl_lmops) {
161 if (fl->fl_lmops->fl_release_private)
162 fl->fl_lmops->fl_release_private(fl);
163 fl->fl_lmops = NULL;
164 }
165
166}
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168/* Free a lock which is not in use. */
Matt Mackall33443c42006-01-08 01:05:22 -0800169static void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800171 BUG_ON(waitqueue_active(&fl->fl_wait));
172 BUG_ON(!list_empty(&fl->fl_block));
173 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Trond Myklebust47831f32006-03-20 13:44:05 -0500175 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 kmem_cache_free(filelock_cache, fl);
177}
178
179void locks_init_lock(struct file_lock *fl)
180{
181 INIT_LIST_HEAD(&fl->fl_link);
182 INIT_LIST_HEAD(&fl->fl_block);
183 init_waitqueue_head(&fl->fl_wait);
184 fl->fl_next = NULL;
185 fl->fl_fasync = NULL;
186 fl->fl_owner = NULL;
187 fl->fl_pid = 0;
188 fl->fl_file = NULL;
189 fl->fl_flags = 0;
190 fl->fl_type = 0;
191 fl->fl_start = fl->fl_end = 0;
192 fl->fl_ops = NULL;
193 fl->fl_lmops = NULL;
194}
195
196EXPORT_SYMBOL(locks_init_lock);
197
198/*
199 * Initialises the fields of the file lock which are invariant for
200 * free file_locks.
201 */
Christoph Lametere18b8902006-12-06 20:33:20 -0800202static void init_once(void *foo, struct kmem_cache *cache, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 struct file_lock *lock = (struct file_lock *) foo;
205
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:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800324 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 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:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800374 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 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
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500669posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
671 struct file_lock *cfl;
672
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500673 fl->fl_type = F_UNLCK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800675 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 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) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500682 __locks_copy_lock(fl, 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;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800737 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 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
Marc Eshel150b3932007-01-18 16:15:35 -0500804static int __posix_lock_file(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
Marc Eshel150b3932007-01-18 16:15:35 -05001010 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 *
1012 * Add a POSIX style lock to a file.
1013 * We merge adjacent & overlapping locks whenever possible.
1014 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001015 *
1016 * Note that if called with an FL_EXISTS argument, the caller may determine
1017 * whether or not a lock was successfully freed by testing the return
1018 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 */
Marc Eshel150b3932007-01-18 16:15:35 -05001020int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001021 struct file_lock *conflock)
1022{
Marc Eshel150b3932007-01-18 16:15:35 -05001023 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001024}
Marc Eshel150b3932007-01-18 16:15:35 -05001025EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026
1027/**
1028 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1029 * @filp: The file to apply the lock to
1030 * @fl: The lock to be applied
1031 *
1032 * Add a POSIX style lock to a file.
1033 * We merge adjacent & overlapping locks whenever possible.
1034 * POSIX locks are sorted by owner task, then by starting address
1035 */
1036int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1037{
1038 int error;
1039 might_sleep ();
1040 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001041 error = posix_lock_file(filp, fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1043 break;
1044 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1045 if (!error)
1046 continue;
1047
1048 locks_delete_block(fl);
1049 break;
1050 }
1051 return error;
1052}
1053EXPORT_SYMBOL(posix_lock_file_wait);
1054
1055/**
1056 * locks_mandatory_locked - Check for an active lock
1057 * @inode: the file to check
1058 *
1059 * Searches the inode's list of locks to find any POSIX locks which conflict.
1060 * This function is called from locks_verify_locked() only.
1061 */
1062int locks_mandatory_locked(struct inode *inode)
1063{
1064 fl_owner_t owner = current->files;
1065 struct file_lock *fl;
1066
1067 /*
1068 * Search the lock list for this inode for any POSIX locks.
1069 */
1070 lock_kernel();
1071 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1072 if (!IS_POSIX(fl))
1073 continue;
1074 if (fl->fl_owner != owner)
1075 break;
1076 }
1077 unlock_kernel();
1078 return fl ? -EAGAIN : 0;
1079}
1080
1081/**
1082 * locks_mandatory_area - Check for a conflicting lock
1083 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1084 * for shared
1085 * @inode: the file to check
1086 * @filp: how the file was opened (if it was)
1087 * @offset: start of area to check
1088 * @count: length of area to check
1089 *
1090 * Searches the inode's list of locks to find any POSIX locks which conflict.
1091 * This function is called from rw_verify_area() and
1092 * locks_verify_truncate().
1093 */
1094int locks_mandatory_area(int read_write, struct inode *inode,
1095 struct file *filp, loff_t offset,
1096 size_t count)
1097{
1098 struct file_lock fl;
1099 int error;
1100
1101 locks_init_lock(&fl);
1102 fl.fl_owner = current->files;
1103 fl.fl_pid = current->tgid;
1104 fl.fl_file = filp;
1105 fl.fl_flags = FL_POSIX | FL_ACCESS;
1106 if (filp && !(filp->f_flags & O_NONBLOCK))
1107 fl.fl_flags |= FL_SLEEP;
1108 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1109 fl.fl_start = offset;
1110 fl.fl_end = offset + count - 1;
1111
1112 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001113 error = __posix_lock_file(inode, &fl, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 if (error != -EAGAIN)
1115 break;
1116 if (!(fl.fl_flags & FL_SLEEP))
1117 break;
1118 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1119 if (!error) {
1120 /*
1121 * If we've been sleeping someone might have
1122 * changed the permissions behind our back.
1123 */
1124 if ((inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID)
1125 continue;
1126 }
1127
1128 locks_delete_block(&fl);
1129 break;
1130 }
1131
1132 return error;
1133}
1134
1135EXPORT_SYMBOL(locks_mandatory_area);
1136
1137/* We already had a lease on this file; just change its type */
1138int lease_modify(struct file_lock **before, int arg)
1139{
1140 struct file_lock *fl = *before;
1141 int error = assign_type(fl, arg);
1142
1143 if (error)
1144 return error;
1145 locks_wake_up_blocks(fl);
1146 if (arg == F_UNLCK)
1147 locks_delete_lock(before);
1148 return 0;
1149}
1150
1151EXPORT_SYMBOL(lease_modify);
1152
1153static void time_out_leases(struct inode *inode)
1154{
1155 struct file_lock **before;
1156 struct file_lock *fl;
1157
1158 before = &inode->i_flock;
1159 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1160 if ((fl->fl_break_time == 0)
1161 || time_before(jiffies, fl->fl_break_time)) {
1162 before = &fl->fl_next;
1163 continue;
1164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1166 if (fl == *before) /* lease_modify may have freed fl */
1167 before = &fl->fl_next;
1168 }
1169}
1170
1171/**
1172 * __break_lease - revoke all outstanding leases on file
1173 * @inode: the inode of the file to return
1174 * @mode: the open mode (read or write)
1175 *
1176 * break_lease (inlined for speed) has checked there already
1177 * is a lease on this file. Leases are broken on a call to open()
1178 * or truncate(). This function can sleep unless you
1179 * specified %O_NONBLOCK to your open().
1180 */
1181int __break_lease(struct inode *inode, unsigned int mode)
1182{
1183 int error = 0, future;
1184 struct file_lock *new_fl, *flock;
1185 struct file_lock *fl;
1186 int alloc_err;
1187 unsigned long break_time;
1188 int i_have_this_lease = 0;
1189
1190 alloc_err = lease_alloc(NULL, mode & FMODE_WRITE ? F_WRLCK : F_RDLCK,
1191 &new_fl);
1192
1193 lock_kernel();
1194
1195 time_out_leases(inode);
1196
1197 flock = inode->i_flock;
1198 if ((flock == NULL) || !IS_LEASE(flock))
1199 goto out;
1200
1201 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1202 if (fl->fl_owner == current->files)
1203 i_have_this_lease = 1;
1204
1205 if (mode & FMODE_WRITE) {
1206 /* If we want write access, we have to revoke any lease. */
1207 future = F_UNLCK | F_INPROGRESS;
1208 } else if (flock->fl_type & F_INPROGRESS) {
1209 /* If the lease is already being broken, we just leave it */
1210 future = flock->fl_type;
1211 } else if (flock->fl_type & F_WRLCK) {
1212 /* Downgrade the exclusive lease to a read-only lease. */
1213 future = F_RDLCK | F_INPROGRESS;
1214 } else {
1215 /* the existing lease was read-only, so we can read too. */
1216 goto out;
1217 }
1218
1219 if (alloc_err && !i_have_this_lease && ((mode & O_NONBLOCK) == 0)) {
1220 error = alloc_err;
1221 goto out;
1222 }
1223
1224 break_time = 0;
1225 if (lease_break_time > 0) {
1226 break_time = jiffies + lease_break_time * HZ;
1227 if (break_time == 0)
1228 break_time++; /* so that 0 means no break time */
1229 }
1230
1231 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1232 if (fl->fl_type != future) {
1233 fl->fl_type = future;
1234 fl->fl_break_time = break_time;
1235 /* lease must have lmops break callback */
1236 fl->fl_lmops->fl_break(fl);
1237 }
1238 }
1239
1240 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1241 error = -EWOULDBLOCK;
1242 goto out;
1243 }
1244
1245restart:
1246 break_time = flock->fl_break_time;
1247 if (break_time != 0) {
1248 break_time -= jiffies;
1249 if (break_time == 0)
1250 break_time++;
1251 }
1252 error = locks_block_on_timeout(flock, new_fl, break_time);
1253 if (error >= 0) {
1254 if (error == 0)
1255 time_out_leases(inode);
1256 /* Wait for the next lease that has not been broken yet */
1257 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1258 flock = flock->fl_next) {
1259 if (flock->fl_type & F_INPROGRESS)
1260 goto restart;
1261 }
1262 error = 0;
1263 }
1264
1265out:
1266 unlock_kernel();
1267 if (!alloc_err)
1268 locks_free_lock(new_fl);
1269 return error;
1270}
1271
1272EXPORT_SYMBOL(__break_lease);
1273
1274/**
1275 * lease_get_mtime
1276 * @inode: the inode
1277 * @time: pointer to a timespec which will contain the last modified time
1278 *
1279 * This is to force NFS clients to flush their caches for files with
1280 * exclusive leases. The justification is that if someone has an
1281 * exclusive lease, then they could be modifiying it.
1282 */
1283void lease_get_mtime(struct inode *inode, struct timespec *time)
1284{
1285 struct file_lock *flock = inode->i_flock;
1286 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1287 *time = current_fs_time(inode->i_sb);
1288 else
1289 *time = inode->i_mtime;
1290}
1291
1292EXPORT_SYMBOL(lease_get_mtime);
1293
1294/**
1295 * fcntl_getlease - Enquire what lease is currently active
1296 * @filp: the file
1297 *
1298 * The value returned by this function will be one of
1299 * (if no lease break is pending):
1300 *
1301 * %F_RDLCK to indicate a shared lease is held.
1302 *
1303 * %F_WRLCK to indicate an exclusive lease is held.
1304 *
1305 * %F_UNLCK to indicate no lease is held.
1306 *
1307 * (if a lease break is pending):
1308 *
1309 * %F_RDLCK to indicate an exclusive lease needs to be
1310 * changed to a shared lease (or removed).
1311 *
1312 * %F_UNLCK to indicate the lease needs to be removed.
1313 *
1314 * XXX: sfr & willy disagree over whether F_INPROGRESS
1315 * should be returned to userspace.
1316 */
1317int fcntl_getlease(struct file *filp)
1318{
1319 struct file_lock *fl;
1320 int type = F_UNLCK;
1321
1322 lock_kernel();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001323 time_out_leases(filp->f_path.dentry->d_inode);
1324 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 fl = fl->fl_next) {
1326 if (fl->fl_file == filp) {
1327 type = fl->fl_type & ~F_INPROGRESS;
1328 break;
1329 }
1330 }
1331 unlock_kernel();
1332 return type;
1333}
1334
1335/**
1336 * __setlease - sets a lease on an open file
1337 * @filp: file pointer
1338 * @arg: type of lease to obtain
1339 * @flp: input - file_lock to use, output - file_lock inserted
1340 *
1341 * The (input) flp->fl_lmops->fl_break function is required
1342 * by break_lease().
1343 *
1344 * Called with kernel lock held.
1345 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07001346static int __setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001348 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001349 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 struct inode *inode = dentry->d_inode;
1351 int error, rdlease_count = 0, wrlease_count = 0;
1352
1353 time_out_leases(inode);
1354
1355 error = -EINVAL;
1356 if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
1357 goto out;
1358
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001359 lease = *flp;
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 error = -EAGAIN;
1362 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1363 goto out;
1364 if ((arg == F_WRLCK)
1365 && ((atomic_read(&dentry->d_count) > 1)
1366 || (atomic_read(&inode->i_count) > 1)))
1367 goto out;
1368
1369 /*
1370 * At this point, we know that if there is an exclusive
1371 * lease on this file, then we hold it on this filp
1372 * (otherwise our open of this file would have blocked).
1373 * And if we are trying to acquire an exclusive lease,
1374 * then the file is not open by anyone (including us)
1375 * except for this filp.
1376 */
1377 for (before = &inode->i_flock;
1378 ((fl = *before) != NULL) && IS_LEASE(fl);
1379 before = &fl->fl_next) {
1380 if (lease->fl_lmops->fl_mylease(fl, lease))
1381 my_before = before;
1382 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1383 /*
1384 * Someone is in the process of opening this
1385 * file for writing so we may not take an
1386 * exclusive lease on it.
1387 */
1388 wrlease_count++;
1389 else
1390 rdlease_count++;
1391 }
1392
1393 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1394 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1395 goto out;
1396
1397 if (my_before != NULL) {
Trond Myklebust75dff552006-05-07 23:02:42 -04001398 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 error = lease->fl_lmops->fl_change(my_before, arg);
1400 goto out;
1401 }
1402
1403 error = 0;
1404 if (arg == F_UNLCK)
1405 goto out;
1406
1407 error = -EINVAL;
1408 if (!leases_enable)
1409 goto out;
1410
Trond Myklebust74361cb2006-08-14 08:54:48 -07001411 error = -ENOMEM;
1412 fl = locks_alloc_lock();
1413 if (fl == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 goto out;
1415
1416 locks_copy_lock(fl, lease);
1417
1418 locks_insert_lock(before, fl);
1419
1420 *flp = fl;
Trond Myklebust74361cb2006-08-14 08:54:48 -07001421 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422out:
1423 return error;
1424}
1425
1426 /**
1427 * setlease - sets a lease on an open file
1428 * @filp: file pointer
1429 * @arg: type of lease to obtain
1430 * @lease: file_lock to use
1431 *
1432 * Call this to establish a lease on the file.
1433 * The fl_lmops fl_break function is required by break_lease
1434 */
1435
1436int setlease(struct file *filp, long arg, struct file_lock **lease)
1437{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001438 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 struct inode *inode = dentry->d_inode;
1440 int error;
1441
1442 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1443 return -EACCES;
1444 if (!S_ISREG(inode->i_mode))
1445 return -EINVAL;
1446 error = security_file_lock(filp, arg);
1447 if (error)
1448 return error;
1449
1450 lock_kernel();
1451 error = __setlease(filp, arg, lease);
1452 unlock_kernel();
1453
1454 return error;
1455}
1456
1457EXPORT_SYMBOL(setlease);
1458
1459/**
1460 * fcntl_setlease - sets a lease on an open file
1461 * @fd: open file descriptor
1462 * @filp: file pointer
1463 * @arg: type of lease to obtain
1464 *
1465 * Call this fcntl to establish a lease on the file.
1466 * Note that you also need to call %F_SETSIG to
1467 * receive a signal when the lease is broken.
1468 */
1469int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1470{
1471 struct file_lock fl, *flp = &fl;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001472 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 struct inode *inode = dentry->d_inode;
1474 int error;
1475
1476 if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
1477 return -EACCES;
1478 if (!S_ISREG(inode->i_mode))
1479 return -EINVAL;
1480 error = security_file_lock(filp, arg);
1481 if (error)
1482 return error;
1483
1484 locks_init_lock(&fl);
1485 error = lease_init(filp, arg, &fl);
1486 if (error)
1487 return error;
1488
1489 lock_kernel();
1490
1491 error = __setlease(filp, arg, &flp);
J. Bruce Fieldsdc15ae12005-11-10 19:08:00 -05001492 if (error || arg == F_UNLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 goto out_unlock;
1494
1495 error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
1496 if (error < 0) {
1497 /* remove lease just inserted by __setlease */
1498 flp->fl_type = F_UNLCK | F_INPROGRESS;
1499 flp->fl_break_time = jiffies- 10;
1500 time_out_leases(inode);
1501 goto out_unlock;
1502 }
1503
Eric W. Biederman609d7fa2006-10-02 02:17:15 -07001504 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505out_unlock:
1506 unlock_kernel();
1507 return error;
1508}
1509
1510/**
1511 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1512 * @filp: The file to apply the lock to
1513 * @fl: The lock to be applied
1514 *
1515 * Add a FLOCK style lock to a file.
1516 */
1517int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1518{
1519 int error;
1520 might_sleep();
1521 for (;;) {
1522 error = flock_lock_file(filp, fl);
1523 if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
1524 break;
1525 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1526 if (!error)
1527 continue;
1528
1529 locks_delete_block(fl);
1530 break;
1531 }
1532 return error;
1533}
1534
1535EXPORT_SYMBOL(flock_lock_file_wait);
1536
1537/**
1538 * sys_flock: - flock() system call.
1539 * @fd: the file descriptor to lock.
1540 * @cmd: the type of lock to apply.
1541 *
1542 * Apply a %FL_FLOCK style lock to an open file descriptor.
1543 * The @cmd can be one of
1544 *
1545 * %LOCK_SH -- a shared lock.
1546 *
1547 * %LOCK_EX -- an exclusive lock.
1548 *
1549 * %LOCK_UN -- remove an existing lock.
1550 *
1551 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1552 *
1553 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1554 * processes read and write access respectively.
1555 */
1556asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
1557{
1558 struct file *filp;
1559 struct file_lock *lock;
1560 int can_sleep, unlock;
1561 int error;
1562
1563 error = -EBADF;
1564 filp = fget(fd);
1565 if (!filp)
1566 goto out;
1567
1568 can_sleep = !(cmd & LOCK_NB);
1569 cmd &= ~LOCK_NB;
1570 unlock = (cmd == LOCK_UN);
1571
1572 if (!unlock && !(cmd & LOCK_MAND) && !(filp->f_mode & 3))
1573 goto out_putf;
1574
1575 error = flock_make_lock(filp, &lock, cmd);
1576 if (error)
1577 goto out_putf;
1578 if (can_sleep)
1579 lock->fl_flags |= FL_SLEEP;
1580
1581 error = security_file_lock(filp, cmd);
1582 if (error)
1583 goto out_free;
1584
1585 if (filp->f_op && filp->f_op->flock)
1586 error = filp->f_op->flock(filp,
1587 (can_sleep) ? F_SETLKW : F_SETLK,
1588 lock);
1589 else
1590 error = flock_lock_file_wait(filp, lock);
1591
1592 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001593 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595 out_putf:
1596 fput(filp);
1597 out:
1598 return error;
1599}
1600
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001601/**
1602 * vfs_test_lock - test file byte range lock
1603 * @filp: The file to test lock for
1604 * @fl: The lock to test
1605 * @conf: Place to return a copy of the conflicting lock, if found
1606 *
1607 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1608 * setting conf->fl_type to something other than F_UNLCK.
1609 */
1610int vfs_test_lock(struct file *filp, struct file_lock *fl)
1611{
1612 if (filp->f_op && filp->f_op->lock)
1613 return filp->f_op->lock(filp, F_GETLK, fl);
1614 posix_test_lock(filp, fl);
1615 return 0;
1616}
1617EXPORT_SYMBOL_GPL(vfs_test_lock);
1618
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001619static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1620{
1621 flock->l_pid = fl->fl_pid;
1622#if BITS_PER_LONG == 32
1623 /*
1624 * Make sure we can represent the posix lock via
1625 * legacy 32bit flock.
1626 */
1627 if (fl->fl_start > OFFT_OFFSET_MAX)
1628 return -EOVERFLOW;
1629 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1630 return -EOVERFLOW;
1631#endif
1632 flock->l_start = fl->fl_start;
1633 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1634 fl->fl_end - fl->fl_start + 1;
1635 flock->l_whence = 0;
1636 return 0;
1637}
1638
1639#if BITS_PER_LONG == 32
1640static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1641{
1642 flock->l_pid = fl->fl_pid;
1643 flock->l_start = fl->fl_start;
1644 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1645 fl->fl_end - fl->fl_start + 1;
1646 flock->l_whence = 0;
1647 flock->l_type = fl->fl_type;
1648}
1649#endif
1650
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651/* Report the first existing lock that would conflict with l.
1652 * This implements the F_GETLK command of fcntl().
1653 */
1654int fcntl_getlk(struct file *filp, struct flock __user *l)
1655{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001656 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 struct flock flock;
1658 int error;
1659
1660 error = -EFAULT;
1661 if (copy_from_user(&flock, l, sizeof(flock)))
1662 goto out;
1663 error = -EINVAL;
1664 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1665 goto out;
1666
1667 error = flock_to_posix_lock(filp, &file_lock, &flock);
1668 if (error)
1669 goto out;
1670
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001671 error = vfs_test_lock(filp, &file_lock);
1672 if (error)
1673 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001675 flock.l_type = file_lock.fl_type;
1676 if (file_lock.fl_type != F_UNLCK) {
1677 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001678 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
1681 error = -EFAULT;
1682 if (!copy_to_user(l, &flock, sizeof(flock)))
1683 error = 0;
1684out:
1685 return error;
1686}
1687
Marc Eshel7723ec92007-01-18 15:08:55 -05001688/**
1689 * vfs_lock_file - file byte range lock
1690 * @filp: The file to apply the lock to
1691 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1692 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001693 * @conf: Place to return a copy of the conflicting lock, if found.
1694 *
1695 * A caller that doesn't care about the conflicting lock may pass NULL
1696 * as the final argument.
1697 *
1698 * If the filesystem defines a private ->lock() method, then @conf will
1699 * be left unchanged; so a caller that cares should initialize it to
1700 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001701 *
1702 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1703 * locks, the ->lock() interface may return asynchronously, before the lock has
1704 * been granted or denied by the underlying filesystem, if (and only if)
1705 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1706 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1707 * the request is for a blocking lock. When ->lock() does return asynchronously,
1708 * it must return -EINPROGRESS, and call ->fl_grant() when the lock
1709 * request completes.
1710 * If the request is for non-blocking lock the file system should return
1711 * -EINPROGRESS then try to get the lock and call the callback routine with
1712 * the result. If the request timed out the callback routine will return a
1713 * nonzero return code and the file system should release the lock. The file
1714 * system is also responsible to keep a corresponding posix lock when it
1715 * grants a lock so the VFS can find out which locks are locally held and do
1716 * the correct lock cleanup when required.
1717 * The underlying filesystem must not drop the kernel lock or call
1718 * ->fl_grant() before returning to the caller with a -EINPROGRESS
1719 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001720 */
Marc Eshel150b3932007-01-18 16:15:35 -05001721int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001722{
1723 if (filp->f_op && filp->f_op->lock)
1724 return filp->f_op->lock(filp, cmd, fl);
1725 else
Marc Eshel150b3932007-01-18 16:15:35 -05001726 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001727}
1728EXPORT_SYMBOL_GPL(vfs_lock_file);
1729
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730/* Apply the lock described by l to an open file descriptor.
1731 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1732 */
Peter Staubachc2936212005-07-27 11:45:09 -07001733int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1734 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735{
1736 struct file_lock *file_lock = locks_alloc_lock();
1737 struct flock flock;
1738 struct inode *inode;
1739 int error;
1740
1741 if (file_lock == NULL)
1742 return -ENOLCK;
1743
1744 /*
1745 * This might block, so we do it before checking the inode.
1746 */
1747 error = -EFAULT;
1748 if (copy_from_user(&flock, l, sizeof(flock)))
1749 goto out;
1750
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001751 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
1753 /* Don't allow mandatory locks on files that may be memory mapped
1754 * and shared.
1755 */
1756 if (IS_MANDLOCK(inode) &&
1757 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1758 mapping_writably_mapped(filp->f_mapping)) {
1759 error = -EAGAIN;
1760 goto out;
1761 }
1762
Peter Staubachc2936212005-07-27 11:45:09 -07001763again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 error = flock_to_posix_lock(filp, file_lock, &flock);
1765 if (error)
1766 goto out;
1767 if (cmd == F_SETLKW) {
1768 file_lock->fl_flags |= FL_SLEEP;
1769 }
1770
1771 error = -EBADF;
1772 switch (flock.l_type) {
1773 case F_RDLCK:
1774 if (!(filp->f_mode & FMODE_READ))
1775 goto out;
1776 break;
1777 case F_WRLCK:
1778 if (!(filp->f_mode & FMODE_WRITE))
1779 goto out;
1780 break;
1781 case F_UNLCK:
1782 break;
1783 default:
1784 error = -EINVAL;
1785 goto out;
1786 }
1787
1788 error = security_file_lock(filp, file_lock->fl_type);
1789 if (error)
1790 goto out;
1791
Marc Eshel7723ec92007-01-18 15:08:55 -05001792 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001793 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001794 if (error != -EAGAIN || cmd == F_SETLK)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001796 error = wait_event_interruptible(file_lock->fl_wait,
1797 !file_lock->fl_next);
1798 if (!error)
1799 continue;
1800
1801 locks_delete_block(file_lock);
1802 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 }
1804
Peter Staubachc2936212005-07-27 11:45:09 -07001805 /*
1806 * Attempt to detect a close/fcntl race and recover by
1807 * releasing the lock that was just acquired.
1808 */
1809 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1810 flock.l_type = F_UNLCK;
1811 goto again;
1812 }
1813
1814out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 locks_free_lock(file_lock);
1816 return error;
1817}
1818
1819#if BITS_PER_LONG == 32
1820/* Report the first existing lock that would conflict with l.
1821 * This implements the F_GETLK command of fcntl().
1822 */
1823int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1824{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001825 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 struct flock64 flock;
1827 int error;
1828
1829 error = -EFAULT;
1830 if (copy_from_user(&flock, l, sizeof(flock)))
1831 goto out;
1832 error = -EINVAL;
1833 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1834 goto out;
1835
1836 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1837 if (error)
1838 goto out;
1839
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001840 error = vfs_test_lock(filp, &file_lock);
1841 if (error)
1842 goto out;
1843
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001844 flock.l_type = file_lock.fl_type;
1845 if (file_lock.fl_type != F_UNLCK)
1846 posix_lock_to_flock64(&flock, &file_lock);
1847
Linus Torvalds1da177e2005-04-16 15:20:36 -07001848 error = -EFAULT;
1849 if (!copy_to_user(l, &flock, sizeof(flock)))
1850 error = 0;
1851
1852out:
1853 return error;
1854}
1855
1856/* Apply the lock described by l to an open file descriptor.
1857 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1858 */
Peter Staubachc2936212005-07-27 11:45:09 -07001859int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1860 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861{
1862 struct file_lock *file_lock = locks_alloc_lock();
1863 struct flock64 flock;
1864 struct inode *inode;
1865 int error;
1866
1867 if (file_lock == NULL)
1868 return -ENOLCK;
1869
1870 /*
1871 * This might block, so we do it before checking the inode.
1872 */
1873 error = -EFAULT;
1874 if (copy_from_user(&flock, l, sizeof(flock)))
1875 goto out;
1876
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001877 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878
1879 /* Don't allow mandatory locks on files that may be memory mapped
1880 * and shared.
1881 */
1882 if (IS_MANDLOCK(inode) &&
1883 (inode->i_mode & (S_ISGID | S_IXGRP)) == S_ISGID &&
1884 mapping_writably_mapped(filp->f_mapping)) {
1885 error = -EAGAIN;
1886 goto out;
1887 }
1888
Peter Staubachc2936212005-07-27 11:45:09 -07001889again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 error = flock64_to_posix_lock(filp, file_lock, &flock);
1891 if (error)
1892 goto out;
1893 if (cmd == F_SETLKW64) {
1894 file_lock->fl_flags |= FL_SLEEP;
1895 }
1896
1897 error = -EBADF;
1898 switch (flock.l_type) {
1899 case F_RDLCK:
1900 if (!(filp->f_mode & FMODE_READ))
1901 goto out;
1902 break;
1903 case F_WRLCK:
1904 if (!(filp->f_mode & FMODE_WRITE))
1905 goto out;
1906 break;
1907 case F_UNLCK:
1908 break;
1909 default:
1910 error = -EINVAL;
1911 goto out;
1912 }
1913
1914 error = security_file_lock(filp, file_lock->fl_type);
1915 if (error)
1916 goto out;
1917
Marc Eshel7723ec92007-01-18 15:08:55 -05001918 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001919 error = vfs_lock_file(filp, cmd, file_lock, NULL);
Marc Eshel7723ec92007-01-18 15:08:55 -05001920 if (error != -EAGAIN || cmd == F_SETLK64)
Peter Staubachc2936212005-07-27 11:45:09 -07001921 break;
Marc Eshel7723ec92007-01-18 15:08:55 -05001922 error = wait_event_interruptible(file_lock->fl_wait,
1923 !file_lock->fl_next);
1924 if (!error)
1925 continue;
1926
1927 locks_delete_block(file_lock);
1928 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 }
1930
Peter Staubachc2936212005-07-27 11:45:09 -07001931 /*
1932 * Attempt to detect a close/fcntl race and recover by
1933 * releasing the lock that was just acquired.
1934 */
1935 if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
1936 flock.l_type = F_UNLCK;
1937 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 }
1939
1940out:
1941 locks_free_lock(file_lock);
1942 return error;
1943}
1944#endif /* BITS_PER_LONG == 32 */
1945
1946/*
1947 * This function is called when the file is being removed
1948 * from the task's fd array. POSIX locks belonging to this task
1949 * are deleted at this time.
1950 */
1951void locks_remove_posix(struct file *filp, fl_owner_t owner)
1952{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001953 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954
1955 /*
1956 * If there are no locks held on this file, we don't need to call
1957 * posix_lock_file(). Another process could be setting a lock on this
1958 * file at the same time, but we wouldn't remove that lock anyway.
1959 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001960 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 return;
1962
1963 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001964 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 lock.fl_start = 0;
1966 lock.fl_end = OFFSET_MAX;
1967 lock.fl_owner = owner;
1968 lock.fl_pid = current->tgid;
1969 lock.fl_file = filp;
1970 lock.fl_ops = NULL;
1971 lock.fl_lmops = NULL;
1972
Marc Eshel150b3932007-01-18 16:15:35 -05001973 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 if (lock.fl_ops && lock.fl_ops->fl_release_private)
1976 lock.fl_ops->fl_release_private(&lock);
1977}
1978
1979EXPORT_SYMBOL(locks_remove_posix);
1980
1981/*
1982 * This function is called on the last close of an open file.
1983 */
1984void locks_remove_flock(struct file *filp)
1985{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001986 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 struct file_lock *fl;
1988 struct file_lock **before;
1989
1990 if (!inode->i_flock)
1991 return;
1992
1993 if (filp->f_op && filp->f_op->flock) {
1994 struct file_lock fl = {
1995 .fl_pid = current->tgid,
1996 .fl_file = filp,
1997 .fl_flags = FL_FLOCK,
1998 .fl_type = F_UNLCK,
1999 .fl_end = OFFSET_MAX,
2000 };
2001 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002002 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2003 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 }
2005
2006 lock_kernel();
2007 before = &inode->i_flock;
2008
2009 while ((fl = *before) != NULL) {
2010 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002011 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 locks_delete_lock(before);
2013 continue;
2014 }
2015 if (IS_LEASE(fl)) {
2016 lease_modify(before, F_UNLCK);
2017 continue;
2018 }
2019 /* What? */
2020 BUG();
2021 }
2022 before = &fl->fl_next;
2023 }
2024 unlock_kernel();
2025}
2026
2027/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 * posix_unblock_lock - stop waiting for a file lock
2029 * @filp: how the file was opened
2030 * @waiter: the lock which was waiting
2031 *
2032 * lockd needs to block waiting for locks.
2033 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002034int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2036{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002037 int status = 0;
2038
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 lock_kernel();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002040 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002042 else
2043 status = -ENOENT;
J. Bruce Fields5996a292006-01-03 09:55:44 +01002044 unlock_kernel();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002045 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
2048EXPORT_SYMBOL(posix_unblock_lock);
2049
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002050/**
2051 * vfs_cancel_lock - file byte range unblock lock
2052 * @filp: The file to apply the unblock to
2053 * @fl: The lock to be unblocked
2054 *
2055 * Used by lock managers to cancel blocked requests
2056 */
2057int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2058{
2059 if (filp->f_op && filp->f_op->lock)
2060 return filp->f_op->lock(filp, F_CANCELLK, fl);
2061 return 0;
2062}
2063
2064EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2065
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066static void lock_get_status(char* out, struct file_lock *fl, int id, char *pfx)
2067{
2068 struct inode *inode = NULL;
2069
2070 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002071 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072
2073 out += sprintf(out, "%d:%s ", id, pfx);
2074 if (IS_POSIX(fl)) {
2075 out += sprintf(out, "%6s %s ",
2076 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2077 (inode == NULL) ? "*NOINODE*" :
2078 (IS_MANDLOCK(inode) &&
2079 (inode->i_mode & (S_IXGRP | S_ISGID)) == S_ISGID) ?
2080 "MANDATORY" : "ADVISORY ");
2081 } else if (IS_FLOCK(fl)) {
2082 if (fl->fl_type & LOCK_MAND) {
2083 out += sprintf(out, "FLOCK MSNFS ");
2084 } else {
2085 out += sprintf(out, "FLOCK ADVISORY ");
2086 }
2087 } else if (IS_LEASE(fl)) {
2088 out += sprintf(out, "LEASE ");
2089 if (fl->fl_type & F_INPROGRESS)
2090 out += sprintf(out, "BREAKING ");
2091 else if (fl->fl_file)
2092 out += sprintf(out, "ACTIVE ");
2093 else
2094 out += sprintf(out, "BREAKER ");
2095 } else {
2096 out += sprintf(out, "UNKNOWN UNKNOWN ");
2097 }
2098 if (fl->fl_type & LOCK_MAND) {
2099 out += sprintf(out, "%s ",
2100 (fl->fl_type & LOCK_READ)
2101 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2102 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2103 } else {
2104 out += sprintf(out, "%s ",
2105 (fl->fl_type & F_INPROGRESS)
2106 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2107 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2108 }
2109 if (inode) {
2110#ifdef WE_CAN_BREAK_LSLK_NOW
2111 out += sprintf(out, "%d %s:%ld ", fl->fl_pid,
2112 inode->i_sb->s_id, inode->i_ino);
2113#else
2114 /* userspace relies on this representation of dev_t ;-( */
2115 out += sprintf(out, "%d %02x:%02x:%ld ", fl->fl_pid,
2116 MAJOR(inode->i_sb->s_dev),
2117 MINOR(inode->i_sb->s_dev), inode->i_ino);
2118#endif
2119 } else {
2120 out += sprintf(out, "%d <none>:0 ", fl->fl_pid);
2121 }
2122 if (IS_POSIX(fl)) {
2123 if (fl->fl_end == OFFSET_MAX)
2124 out += sprintf(out, "%Ld EOF\n", fl->fl_start);
2125 else
2126 out += sprintf(out, "%Ld %Ld\n", fl->fl_start,
2127 fl->fl_end);
2128 } else {
2129 out += sprintf(out, "0 EOF\n");
2130 }
2131}
2132
2133static void move_lock_status(char **p, off_t* pos, off_t offset)
2134{
2135 int len;
2136 len = strlen(*p);
2137 if(*pos >= offset) {
2138 /* the complete line is valid */
2139 *p += len;
2140 *pos += len;
2141 return;
2142 }
2143 if(*pos+len > offset) {
2144 /* use the second part of the line */
2145 int i = offset-*pos;
2146 memmove(*p,*p+i,len-i);
2147 *p += len-i;
2148 *pos += len;
2149 return;
2150 }
2151 /* discard the complete line */
2152 *pos += len;
2153}
2154
2155/**
2156 * get_locks_status - reports lock usage in /proc/locks
2157 * @buffer: address in userspace to write into
2158 * @start: ?
2159 * @offset: how far we are through the buffer
2160 * @length: how much to read
2161 */
2162
2163int get_locks_status(char *buffer, char **start, off_t offset, int length)
2164{
2165 struct list_head *tmp;
2166 char *q = buffer;
2167 off_t pos = 0;
2168 int i = 0;
2169
2170 lock_kernel();
2171 list_for_each(tmp, &file_lock_list) {
2172 struct list_head *btmp;
2173 struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
2174 lock_get_status(q, fl, ++i, "");
2175 move_lock_status(&q, &pos, offset);
2176
2177 if(pos >= offset+length)
2178 goto done;
2179
2180 list_for_each(btmp, &fl->fl_block) {
2181 struct file_lock *bfl = list_entry(btmp,
2182 struct file_lock, fl_block);
2183 lock_get_status(q, bfl, i, " ->");
2184 move_lock_status(&q, &pos, offset);
2185
2186 if(pos >= offset+length)
2187 goto done;
2188 }
2189 }
2190done:
2191 unlock_kernel();
2192 *start = buffer;
2193 if(q-buffer < length)
2194 return (q-buffer);
2195 return length;
2196}
2197
2198/**
2199 * lock_may_read - checks that the region is free of locks
2200 * @inode: the inode that is being read
2201 * @start: the first byte to read
2202 * @len: the number of bytes to read
2203 *
2204 * Emulates Windows locking requirements. Whole-file
2205 * mandatory locks (share modes) can prohibit a read and
2206 * byte-range POSIX locks can prohibit a read if they overlap.
2207 *
2208 * N.B. this function is only ever called
2209 * from knfsd and ownership of locks is never checked.
2210 */
2211int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2212{
2213 struct file_lock *fl;
2214 int result = 1;
2215 lock_kernel();
2216 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2217 if (IS_POSIX(fl)) {
2218 if (fl->fl_type == F_RDLCK)
2219 continue;
2220 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2221 continue;
2222 } else if (IS_FLOCK(fl)) {
2223 if (!(fl->fl_type & LOCK_MAND))
2224 continue;
2225 if (fl->fl_type & LOCK_READ)
2226 continue;
2227 } else
2228 continue;
2229 result = 0;
2230 break;
2231 }
2232 unlock_kernel();
2233 return result;
2234}
2235
2236EXPORT_SYMBOL(lock_may_read);
2237
2238/**
2239 * lock_may_write - checks that the region is free of locks
2240 * @inode: the inode that is being written
2241 * @start: the first byte to write
2242 * @len: the number of bytes to write
2243 *
2244 * Emulates Windows locking requirements. Whole-file
2245 * mandatory locks (share modes) can prohibit a write and
2246 * byte-range POSIX locks can prohibit a write if they overlap.
2247 *
2248 * N.B. this function is only ever called
2249 * from knfsd and ownership of locks is never checked.
2250 */
2251int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2252{
2253 struct file_lock *fl;
2254 int result = 1;
2255 lock_kernel();
2256 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2257 if (IS_POSIX(fl)) {
2258 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2259 continue;
2260 } else if (IS_FLOCK(fl)) {
2261 if (!(fl->fl_type & LOCK_MAND))
2262 continue;
2263 if (fl->fl_type & LOCK_WRITE)
2264 continue;
2265 } else
2266 continue;
2267 result = 0;
2268 break;
2269 }
2270 unlock_kernel();
2271 return result;
2272}
2273
2274EXPORT_SYMBOL(lock_may_write);
2275
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276static int __init filelock_init(void)
2277{
2278 filelock_cache = kmem_cache_create("file_lock_cache",
2279 sizeof(struct file_lock), 0, SLAB_PANIC,
2280 init_once, NULL);
2281 return 0;
2282}
2283
2284core_initcall(filelock_init);