blob: 7e81bfc751644b0cfb70c9c4291ab6e7fb644ea3 [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.
Paul Bolle395cf962011-08-15 02:02:26 +020063 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * 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>
Al Viro9f3acc32008-04-24 07:44:08 -0400119#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000128#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130#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
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400136static bool lease_breaking(struct file_lock *fl)
137{
J. Bruce Fields778fc542011-07-26 18:25:49 -0400138 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
139}
140
141static int target_leasetype(struct file_lock *fl)
142{
143 if (fl->fl_flags & FL_UNLOCK_PENDING)
144 return F_UNLCK;
145 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
146 return F_RDLCK;
147 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400148}
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150int leases_enable = 1;
151int lease_break_time = 45;
152
153#define for_each_lock(inode, lockp) \
154 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
155
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500156static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157static LIST_HEAD(blocked_list);
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200158static DEFINE_SPINLOCK(file_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200160/*
161 * Protects the two list heads above, plus the inode->i_flock list
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200162 */
163void lock_flocks(void)
164{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200165 spin_lock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200166}
167EXPORT_SYMBOL_GPL(lock_flocks);
168
169void unlock_flocks(void)
170{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200171 spin_unlock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200172}
173EXPORT_SYMBOL_GPL(unlock_flocks);
174
Christoph Lametere18b8902006-12-06 20:33:20 -0800175static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Miklos Szerediee19cc42011-07-07 13:06:09 +0200177static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200178{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200179 INIT_LIST_HEAD(&fl->fl_link);
180 INIT_LIST_HEAD(&fl->fl_block);
181 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200185struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200187 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200188
189 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200190 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200191
192 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200194EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500196void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500197{
198 if (fl->fl_ops) {
199 if (fl->fl_ops->fl_release_private)
200 fl->fl_ops->fl_release_private(fl);
201 fl->fl_ops = NULL;
202 }
J. Bruce Fields068535f2012-08-01 07:56:16 -0400203 fl->fl_lmops = NULL;
Trond Myklebust47831f32006-03-20 13:44:05 -0500204
205}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500206EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500207
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400209void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800211 BUG_ON(waitqueue_active(&fl->fl_wait));
212 BUG_ON(!list_empty(&fl->fl_block));
213 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214
Trond Myklebust47831f32006-03-20 13:44:05 -0500215 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 kmem_cache_free(filelock_cache, fl);
217}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400218EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220void locks_init_lock(struct file_lock *fl)
221{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200222 memset(fl, 0, sizeof(struct file_lock));
223 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226EXPORT_SYMBOL(locks_init_lock);
227
Trond Myklebust47831f32006-03-20 13:44:05 -0500228static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
229{
230 if (fl->fl_ops) {
231 if (fl->fl_ops->fl_copy_lock)
232 fl->fl_ops->fl_copy_lock(new, fl);
233 new->fl_ops = fl->fl_ops;
234 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400235 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500236 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500237}
238
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239/*
240 * Initialize a new lock from an existing file_lock structure.
241 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400242void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243{
244 new->fl_owner = fl->fl_owner;
245 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500246 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 new->fl_flags = fl->fl_flags;
248 new->fl_type = fl->fl_type;
249 new->fl_start = fl->fl_start;
250 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500251 new->fl_ops = NULL;
252 new->fl_lmops = NULL;
253}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700254EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500255
256void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
257{
258 locks_release_private(new);
259
260 __locks_copy_lock(new, fl);
261 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 new->fl_ops = fl->fl_ops;
263 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500264
265 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
268EXPORT_SYMBOL(locks_copy_lock);
269
270static inline int flock_translate_cmd(int cmd) {
271 if (cmd & LOCK_MAND)
272 return cmd & (LOCK_MAND | LOCK_RW);
273 switch (cmd) {
274 case LOCK_SH:
275 return F_RDLCK;
276 case LOCK_EX:
277 return F_WRLCK;
278 case LOCK_UN:
279 return F_UNLCK;
280 }
281 return -EINVAL;
282}
283
284/* Fill in a file_lock structure with an appropriate FLOCK lock. */
285static int flock_make_lock(struct file *filp, struct file_lock **lock,
286 unsigned int cmd)
287{
288 struct file_lock *fl;
289 int type = flock_translate_cmd(cmd);
290 if (type < 0)
291 return type;
292
293 fl = locks_alloc_lock();
294 if (fl == NULL)
295 return -ENOMEM;
296
297 fl->fl_file = filp;
298 fl->fl_pid = current->tgid;
299 fl->fl_flags = FL_FLOCK;
300 fl->fl_type = type;
301 fl->fl_end = OFFSET_MAX;
302
303 *lock = fl;
304 return 0;
305}
306
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400307static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
309 switch (type) {
310 case F_RDLCK:
311 case F_WRLCK:
312 case F_UNLCK:
313 fl->fl_type = type;
314 break;
315 default:
316 return -EINVAL;
317 }
318 return 0;
319}
320
321/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
322 * style lock.
323 */
324static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
325 struct flock *l)
326{
327 off_t start, end;
328
329 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700330 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 start = 0;
332 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700333 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 start = filp->f_pos;
335 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700336 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800337 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 break;
339 default:
340 return -EINVAL;
341 }
342
343 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
344 POSIX-2001 defines it. */
345 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 if (start < 0)
347 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700348 fl->fl_end = OFFSET_MAX;
349 if (l->l_len > 0) {
350 end = start + l->l_len - 1;
351 fl->fl_end = end;
352 } else if (l->l_len < 0) {
353 end = start - 1;
354 fl->fl_end = end;
355 start += l->l_len;
356 if (start < 0)
357 return -EINVAL;
358 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700360 if (fl->fl_end < fl->fl_start)
361 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 fl->fl_owner = current->files;
364 fl->fl_pid = current->tgid;
365 fl->fl_file = filp;
366 fl->fl_flags = FL_POSIX;
367 fl->fl_ops = NULL;
368 fl->fl_lmops = NULL;
369
370 return assign_type(fl, l->l_type);
371}
372
373#if BITS_PER_LONG == 32
374static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
375 struct flock64 *l)
376{
377 loff_t start;
378
379 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700380 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 start = 0;
382 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700383 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 start = filp->f_pos;
385 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700386 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800387 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 break;
389 default:
390 return -EINVAL;
391 }
392
Trond Myklebust4c780a42005-10-18 14:20:21 -0700393 start += l->l_start;
394 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700396 fl->fl_end = OFFSET_MAX;
397 if (l->l_len > 0) {
398 fl->fl_end = start + l->l_len - 1;
399 } else if (l->l_len < 0) {
400 fl->fl_end = start - 1;
401 start += l->l_len;
402 if (start < 0)
403 return -EINVAL;
404 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700406 if (fl->fl_end < fl->fl_start)
407 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409 fl->fl_owner = current->files;
410 fl->fl_pid = current->tgid;
411 fl->fl_file = filp;
412 fl->fl_flags = FL_POSIX;
413 fl->fl_ops = NULL;
414 fl->fl_lmops = NULL;
415
Namhyung Kimf32cb532011-01-17 15:45:59 +0900416 return assign_type(fl, l->l_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417}
418#endif
419
420/* default lease lock manager operations */
421static void lease_break_callback(struct file_lock *fl)
422{
423 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
424}
425
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700426static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400427 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400428 .lm_change = lease_modify,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429};
430
431/*
432 * Initialize a lease, use the default lock manager operations
433 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400434static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400436 if (assign_type(fl, type) != 0)
437 return -EINVAL;
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 fl->fl_owner = current->files;
440 fl->fl_pid = current->tgid;
441
442 fl->fl_file = filp;
443 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 fl->fl_start = 0;
445 fl->fl_end = OFFSET_MAX;
446 fl->fl_ops = NULL;
447 fl->fl_lmops = &lease_manager_ops;
448 return 0;
449}
450
451/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400452static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400455 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500458 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400461 if (error) {
462 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500463 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400464 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500465 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/* Check if two locks overlap each other.
469 */
470static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
471{
472 return ((fl1->fl_end >= fl2->fl_start) &&
473 (fl2->fl_end >= fl1->fl_start));
474}
475
476/*
477 * Check whether two locks have the same owner.
478 */
Matt Mackall33443c42006-01-08 01:05:22 -0800479static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400481 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400483 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 return fl1->fl_owner == fl2->fl_owner;
485}
486
487/* Remove waiter from blocker's block list.
488 * When blocker ends up pointing to itself then the list is empty.
489 */
Matt Mackall33443c42006-01-08 01:05:22 -0800490static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 list_del_init(&waiter->fl_block);
493 list_del_init(&waiter->fl_link);
494 waiter->fl_next = NULL;
495}
496
497/*
498 */
Pavel Shilovsky66189be2012-03-28 21:56:19 +0400499void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500{
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200501 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 __locks_delete_block(waiter);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200503 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504}
Pavel Shilovsky66189be2012-03-28 21:56:19 +0400505EXPORT_SYMBOL(locks_delete_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
507/* Insert waiter into blocker's block list.
508 * We use a circular list so that processes can be easily woken up in
509 * the order they blocked. The documentation doesn't require this but
510 * it seems like the reasonable thing to do.
511 */
512static void locks_insert_block(struct file_lock *blocker,
513 struct file_lock *waiter)
514{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800515 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 list_add_tail(&waiter->fl_block, &blocker->fl_block);
517 waiter->fl_next = blocker;
518 if (IS_POSIX(blocker))
519 list_add(&waiter->fl_link, &blocked_list);
520}
521
522/* Wake up processes blocked waiting for blocker.
523 * If told to wait then schedule the processes until the block list
524 * is empty, otherwise empty the block list ourselves.
525 */
526static void locks_wake_up_blocks(struct file_lock *blocker)
527{
528 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400529 struct file_lock *waiter;
530
531 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 struct file_lock, fl_block);
533 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400534 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
535 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 else
537 wake_up(&waiter->fl_wait);
538 }
539}
540
541/* Insert file lock fl into an inode's lock list at the position indicated
542 * by pos. At the same time add the lock to the global file lock list.
543 */
544static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
545{
546 list_add(&fl->fl_link, &file_lock_list);
547
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000548 fl->fl_nspid = get_pid(task_tgid(current));
549
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 /* insert into file's list */
551 fl->fl_next = *pos;
552 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
554
555/*
556 * Delete a lock and then free it.
557 * Wake up processes that are blocked waiting for this lock,
558 * notify the FS that the lock has been cleared and
559 * finally free the lock.
560 */
561static void locks_delete_lock(struct file_lock **thisfl_p)
562{
563 struct file_lock *fl = *thisfl_p;
564
565 *thisfl_p = fl->fl_next;
566 fl->fl_next = NULL;
567 list_del_init(&fl->fl_link);
568
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000569 if (fl->fl_nspid) {
570 put_pid(fl->fl_nspid);
571 fl->fl_nspid = NULL;
572 }
573
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 locks_wake_up_blocks(fl);
575 locks_free_lock(fl);
576}
577
578/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
579 * checks for shared/exclusive status of overlapping locks.
580 */
581static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
582{
583 if (sys_fl->fl_type == F_WRLCK)
584 return 1;
585 if (caller_fl->fl_type == F_WRLCK)
586 return 1;
587 return 0;
588}
589
590/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
591 * checking before calling the locks_conflict().
592 */
593static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
594{
595 /* POSIX locks owned by the same process do not conflict with
596 * each other.
597 */
598 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
599 return (0);
600
601 /* Check whether they overlap */
602 if (!locks_overlap(caller_fl, sys_fl))
603 return 0;
604
605 return (locks_conflict(caller_fl, sys_fl));
606}
607
608/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
609 * checking before calling the locks_conflict().
610 */
611static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
612{
613 /* FLOCK locks referring to the same filp do not conflict with
614 * each other.
615 */
616 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
617 return (0);
618 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
619 return 0;
620
621 return (locks_conflict(caller_fl, sys_fl));
622}
623
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400624void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500625posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626{
627 struct file_lock *cfl;
628
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200629 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800630 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 if (!IS_POSIX(cfl))
632 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400633 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 break;
635 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000636 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500637 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000638 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800639 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000640 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400641 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200642 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400643 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645EXPORT_SYMBOL(posix_test_lock);
646
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400647/*
648 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400650 * We attempt to detect deadlocks that are due purely to posix file
651 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400653 * We assume that a task can be waiting for at most one lock at a time.
654 * So for any acquired lock, the process holding that lock may be
655 * waiting on at most one other lock. That lock in turns may be held by
656 * someone waiting for at most one other lock. Given a requested lock
657 * caller_fl which is about to wait for a conflicting lock block_fl, we
658 * follow this chain of waiters to ensure we are not about to create a
659 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400660 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400661 * Since we do this before we ever put a process to sleep on a lock, we
662 * are ensured that there is never a cycle; that is what guarantees that
663 * the while() loop in posix_locks_deadlock() eventually completes.
664 *
665 * Note: the above assumption may not be true when handling lock
666 * requests from a broken NFS client. It may also fail in the presence
667 * of tasks (such as posix threads) sharing the same open file table.
668 *
669 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400671
672#define MAX_DEADLK_ITERATIONS 10
673
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400674/* Find a lock that the owner of the given block_fl is blocking on. */
675static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
676{
677 struct file_lock *fl;
678
679 list_for_each_entry(fl, &blocked_list, fl_link) {
680 if (posix_same_owner(fl, block_fl))
681 return fl->fl_next;
682 }
683 return NULL;
684}
685
Adrian Bunkb0904e12006-06-23 02:05:13 -0700686static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 struct file_lock *block_fl)
688{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400689 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400691 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
692 if (i++ > MAX_DEADLK_ITERATIONS)
693 return 0;
694 if (posix_same_owner(caller_fl, block_fl))
695 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
697 return 0;
698}
699
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400701 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400702 *
703 * Note that if called with an FL_EXISTS argument, the caller may determine
704 * whether or not a lock was successfully freed by testing the return
705 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800707static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800709 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800711 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 int error = 0;
713 int found = 0;
714
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200715 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
716 new_fl = locks_alloc_lock();
717 if (!new_fl)
718 return -ENOMEM;
719 }
720
721 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400722 if (request->fl_flags & FL_ACCESS)
723 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400724
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 for_each_lock(inode, before) {
726 struct file_lock *fl = *before;
727 if (IS_POSIX(fl))
728 break;
729 if (IS_LEASE(fl))
730 continue;
731 if (filp != fl->fl_file)
732 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800733 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 goto out;
735 found = 1;
736 locks_delete_lock(before);
737 break;
738 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739
Trond Myklebustf475ae92006-06-29 16:38:32 -0400740 if (request->fl_type == F_UNLCK) {
741 if ((request->fl_flags & FL_EXISTS) && !found)
742 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800743 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400744 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745
746 /*
747 * If a higher-priority process was blocked on the old file lock,
748 * give it the opportunity to lock the file.
749 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200750 if (found) {
751 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200752 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200753 lock_flocks();
754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400756find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 for_each_lock(inode, before) {
758 struct file_lock *fl = *before;
759 if (IS_POSIX(fl))
760 break;
761 if (IS_LEASE(fl))
762 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800763 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 continue;
765 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700766 if (!(request->fl_flags & FL_SLEEP))
767 goto out;
768 error = FILE_LOCK_DEFERRED;
769 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 goto out;
771 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400772 if (request->fl_flags & FL_ACCESS)
773 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800774 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700775 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800776 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400777 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200780 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800781 if (new_fl)
782 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 return error;
784}
785
Marc Eshel150b3932007-01-18 16:15:35 -0500786static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787{
788 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700789 struct file_lock *new_fl = NULL;
790 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 struct file_lock *left = NULL;
792 struct file_lock *right = NULL;
793 struct file_lock **before;
794 int error, added = 0;
795
796 /*
797 * We may need two file_lock structures for this operation,
798 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700799 *
800 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700802 if (!(request->fl_flags & FL_ACCESS) &&
803 (request->fl_type != F_UNLCK ||
804 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
805 new_fl = locks_alloc_lock();
806 new_fl2 = locks_alloc_lock();
807 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200809 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 if (request->fl_type != F_UNLCK) {
811 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500812 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 if (!IS_POSIX(fl))
814 continue;
815 if (!posix_locks_conflict(request, fl))
816 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800817 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400818 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 error = -EAGAIN;
820 if (!(request->fl_flags & FL_SLEEP))
821 goto out;
822 error = -EDEADLK;
823 if (posix_locks_deadlock(request, fl))
824 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700825 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 locks_insert_block(fl, request);
827 goto out;
828 }
829 }
830
831 /* If we're just looking for a conflict, we're done. */
832 error = 0;
833 if (request->fl_flags & FL_ACCESS)
834 goto out;
835
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 * Find the first old lock with the same owner as the new lock.
838 */
839
840 before = &inode->i_flock;
841
842 /* First skip locks owned by other processes. */
843 while ((fl = *before) && (!IS_POSIX(fl) ||
844 !posix_same_owner(request, fl))) {
845 before = &fl->fl_next;
846 }
847
848 /* Process locks with this owner. */
849 while ((fl = *before) && posix_same_owner(request, fl)) {
850 /* Detect adjacent or overlapping regions (if same lock type)
851 */
852 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700853 /* In all comparisons of start vs end, use
854 * "start - 1" rather than "end + 1". If end
855 * is OFFSET_MAX, end + 1 will become negative.
856 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 if (fl->fl_end < request->fl_start - 1)
858 goto next_lock;
859 /* If the next lock in the list has entirely bigger
860 * addresses than the new one, insert the lock here.
861 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700862 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 break;
864
865 /* If we come here, the new and old lock are of the
866 * same type and adjacent or overlapping. Make one
867 * lock yielding from the lower start address of both
868 * locks to the higher end address.
869 */
870 if (fl->fl_start > request->fl_start)
871 fl->fl_start = request->fl_start;
872 else
873 request->fl_start = fl->fl_start;
874 if (fl->fl_end < request->fl_end)
875 fl->fl_end = request->fl_end;
876 else
877 request->fl_end = fl->fl_end;
878 if (added) {
879 locks_delete_lock(before);
880 continue;
881 }
882 request = fl;
883 added = 1;
884 }
885 else {
886 /* Processing for different lock types is a bit
887 * more complex.
888 */
889 if (fl->fl_end < request->fl_start)
890 goto next_lock;
891 if (fl->fl_start > request->fl_end)
892 break;
893 if (request->fl_type == F_UNLCK)
894 added = 1;
895 if (fl->fl_start < request->fl_start)
896 left = fl;
897 /* If the next lock in the list has a higher end
898 * address than the new one, insert the new one here.
899 */
900 if (fl->fl_end > request->fl_end) {
901 right = fl;
902 break;
903 }
904 if (fl->fl_start >= request->fl_start) {
905 /* The new lock completely replaces an old
906 * one (This may happen several times).
907 */
908 if (added) {
909 locks_delete_lock(before);
910 continue;
911 }
912 /* Replace the old lock with the new one.
913 * Wake up anybody waiting for the old one,
914 * as the change in lock type might satisfy
915 * their needs.
916 */
917 locks_wake_up_blocks(fl);
918 fl->fl_start = request->fl_start;
919 fl->fl_end = request->fl_end;
920 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500921 locks_release_private(fl);
922 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 request = fl;
924 added = 1;
925 }
926 }
927 /* Go on to next lock.
928 */
929 next_lock:
930 before = &fl->fl_next;
931 }
932
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700933 /*
934 * The above code only modifies existing locks in case of
935 * merging or replacing. If new lock(s) need to be inserted
936 * all modifications are done bellow this, so it's safe yet to
937 * bail out.
938 */
939 error = -ENOLCK; /* "no luck" */
940 if (right && left == right && !new_fl2)
941 goto out;
942
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 error = 0;
944 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400945 if (request->fl_type == F_UNLCK) {
946 if (request->fl_flags & FL_EXISTS)
947 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400949 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700950
951 if (!new_fl) {
952 error = -ENOLCK;
953 goto out;
954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 locks_copy_lock(new_fl, request);
956 locks_insert_lock(before, new_fl);
957 new_fl = NULL;
958 }
959 if (right) {
960 if (left == right) {
961 /* The new lock breaks the old one in two pieces,
962 * so we have to use the second new lock.
963 */
964 left = new_fl2;
965 new_fl2 = NULL;
966 locks_copy_lock(left, right);
967 locks_insert_lock(before, left);
968 }
969 right->fl_start = request->fl_end + 1;
970 locks_wake_up_blocks(right);
971 }
972 if (left) {
973 left->fl_end = request->fl_start - 1;
974 locks_wake_up_blocks(left);
975 }
976 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200977 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 /*
979 * Free any unused locks.
980 */
981 if (new_fl)
982 locks_free_lock(new_fl);
983 if (new_fl2)
984 locks_free_lock(new_fl2);
985 return error;
986}
987
988/**
989 * posix_lock_file - Apply a POSIX-style lock to a file
990 * @filp: The file to apply the lock to
991 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -0500992 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 *
994 * Add a POSIX style lock to a file.
995 * We merge adjacent & overlapping locks whenever possible.
996 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -0400997 *
998 * Note that if called with an FL_EXISTS argument, the caller may determine
999 * whether or not a lock was successfully freed by testing the return
1000 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 */
Marc Eshel150b3932007-01-18 16:15:35 -05001002int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001003 struct file_lock *conflock)
1004{
Marc Eshel150b3932007-01-18 16:15:35 -05001005 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001006}
Marc Eshel150b3932007-01-18 16:15:35 -05001007EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
1009/**
1010 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1011 * @filp: The file to apply the lock to
1012 * @fl: The lock to be applied
1013 *
1014 * Add a POSIX style lock to a file.
1015 * We merge adjacent & overlapping locks whenever possible.
1016 * POSIX locks are sorted by owner task, then by starting address
1017 */
1018int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1019{
1020 int error;
1021 might_sleep ();
1022 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001023 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001024 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 break;
1026 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1027 if (!error)
1028 continue;
1029
1030 locks_delete_block(fl);
1031 break;
1032 }
1033 return error;
1034}
1035EXPORT_SYMBOL(posix_lock_file_wait);
1036
1037/**
1038 * locks_mandatory_locked - Check for an active lock
1039 * @inode: the file to check
1040 *
1041 * Searches the inode's list of locks to find any POSIX locks which conflict.
1042 * This function is called from locks_verify_locked() only.
1043 */
1044int locks_mandatory_locked(struct inode *inode)
1045{
1046 fl_owner_t owner = current->files;
1047 struct file_lock *fl;
1048
1049 /*
1050 * Search the lock list for this inode for any POSIX locks.
1051 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001052 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1054 if (!IS_POSIX(fl))
1055 continue;
1056 if (fl->fl_owner != owner)
1057 break;
1058 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001059 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 return fl ? -EAGAIN : 0;
1061}
1062
1063/**
1064 * locks_mandatory_area - Check for a conflicting lock
1065 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1066 * for shared
1067 * @inode: the file to check
1068 * @filp: how the file was opened (if it was)
1069 * @offset: start of area to check
1070 * @count: length of area to check
1071 *
1072 * Searches the inode's list of locks to find any POSIX locks which conflict.
1073 * This function is called from rw_verify_area() and
1074 * locks_verify_truncate().
1075 */
1076int locks_mandatory_area(int read_write, struct inode *inode,
1077 struct file *filp, loff_t offset,
1078 size_t count)
1079{
1080 struct file_lock fl;
1081 int error;
1082
1083 locks_init_lock(&fl);
1084 fl.fl_owner = current->files;
1085 fl.fl_pid = current->tgid;
1086 fl.fl_file = filp;
1087 fl.fl_flags = FL_POSIX | FL_ACCESS;
1088 if (filp && !(filp->f_flags & O_NONBLOCK))
1089 fl.fl_flags |= FL_SLEEP;
1090 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1091 fl.fl_start = offset;
1092 fl.fl_end = offset + count - 1;
1093
1094 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001095 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001096 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 break;
1098 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1099 if (!error) {
1100 /*
1101 * If we've been sleeping someone might have
1102 * changed the permissions behind our back.
1103 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001104 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 continue;
1106 }
1107
1108 locks_delete_block(&fl);
1109 break;
1110 }
1111
1112 return error;
1113}
1114
1115EXPORT_SYMBOL(locks_mandatory_area);
1116
J. Bruce Fields778fc542011-07-26 18:25:49 -04001117static void lease_clear_pending(struct file_lock *fl, int arg)
1118{
1119 switch (arg) {
1120 case F_UNLCK:
1121 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1122 /* fall through: */
1123 case F_RDLCK:
1124 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1125 }
1126}
1127
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128/* We already had a lease on this file; just change its type */
1129int lease_modify(struct file_lock **before, int arg)
1130{
1131 struct file_lock *fl = *before;
1132 int error = assign_type(fl, arg);
1133
1134 if (error)
1135 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001136 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001138 if (arg == F_UNLCK) {
1139 struct file *filp = fl->fl_file;
1140
1141 f_delown(filp);
1142 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001143 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1144 if (fl->fl_fasync != NULL) {
1145 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1146 fl->fl_fasync = NULL;
1147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 locks_delete_lock(before);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 return 0;
1151}
1152
1153EXPORT_SYMBOL(lease_modify);
1154
J. Bruce Fields778fc542011-07-26 18:25:49 -04001155static bool past_time(unsigned long then)
1156{
1157 if (!then)
1158 /* 0 is a special value meaning "this never expires": */
1159 return false;
1160 return time_after(jiffies, then);
1161}
1162
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163static void time_out_leases(struct inode *inode)
1164{
1165 struct file_lock **before;
1166 struct file_lock *fl;
1167
1168 before = &inode->i_flock;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04001169 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001170 if (past_time(fl->fl_downgrade_time))
1171 lease_modify(before, F_RDLCK);
1172 if (past_time(fl->fl_break_time))
1173 lease_modify(before, F_UNLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 if (fl == *before) /* lease_modify may have freed fl */
1175 before = &fl->fl_next;
1176 }
1177}
1178
1179/**
1180 * __break_lease - revoke all outstanding leases on file
1181 * @inode: the inode of the file to return
1182 * @mode: the open mode (read or write)
1183 *
david m. richter87250dd2007-05-09 16:10:27 -04001184 * break_lease (inlined for speed) has checked there already is at least
1185 * some kind of lock (maybe a lease) on this file. Leases are broken on
1186 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 * specified %O_NONBLOCK to your open().
1188 */
1189int __break_lease(struct inode *inode, unsigned int mode)
1190{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001191 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 struct file_lock *new_fl, *flock;
1193 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 unsigned long break_time;
1195 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001196 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197
Al Viro8737c932009-12-24 06:47:55 -05001198 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001199 if (IS_ERR(new_fl))
1200 return PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001202 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 time_out_leases(inode);
1205
1206 flock = inode->i_flock;
1207 if ((flock == NULL) || !IS_LEASE(flock))
1208 goto out;
1209
J. Bruce Fields778fc542011-07-26 18:25:49 -04001210 if (!locks_conflict(flock, new_fl))
1211 goto out;
1212
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1214 if (fl->fl_owner == current->files)
1215 i_have_this_lease = 1;
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 break_time = 0;
1218 if (lease_break_time > 0) {
1219 break_time = jiffies + lease_break_time * HZ;
1220 if (break_time == 0)
1221 break_time++; /* so that 0 means no break time */
1222 }
1223
1224 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001225 if (want_write) {
1226 if (fl->fl_flags & FL_UNLOCK_PENDING)
1227 continue;
1228 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001230 } else {
1231 if (lease_breaking(flock))
1232 continue;
1233 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1234 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 }
J. Bruce Fields778fc542011-07-26 18:25:49 -04001236 fl->fl_lmops->lm_break(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 }
1238
1239 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1240 error = -EWOULDBLOCK;
1241 goto out;
1242 }
1243
1244restart:
1245 break_time = flock->fl_break_time;
1246 if (break_time != 0) {
1247 break_time -= jiffies;
1248 if (break_time == 0)
1249 break_time++;
1250 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001251 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001252 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001253 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1254 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001255 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001256 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 if (error >= 0) {
1258 if (error == 0)
1259 time_out_leases(inode);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001260 /*
1261 * Wait for the next conflicting lease that has not been
1262 * broken yet
1263 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1265 flock = flock->fl_next) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001266 if (locks_conflict(new_fl, flock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 goto restart;
1268 }
1269 error = 0;
1270 }
1271
1272out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001273 unlock_flocks();
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001274 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 return error;
1276}
1277
1278EXPORT_SYMBOL(__break_lease);
1279
1280/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001281 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 * @inode: the inode
1283 * @time: pointer to a timespec which will contain the last modified time
1284 *
1285 * This is to force NFS clients to flush their caches for files with
1286 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001287 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 */
1289void lease_get_mtime(struct inode *inode, struct timespec *time)
1290{
1291 struct file_lock *flock = inode->i_flock;
1292 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1293 *time = current_fs_time(inode->i_sb);
1294 else
1295 *time = inode->i_mtime;
1296}
1297
1298EXPORT_SYMBOL(lease_get_mtime);
1299
1300/**
1301 * fcntl_getlease - Enquire what lease is currently active
1302 * @filp: the file
1303 *
1304 * The value returned by this function will be one of
1305 * (if no lease break is pending):
1306 *
1307 * %F_RDLCK to indicate a shared lease is held.
1308 *
1309 * %F_WRLCK to indicate an exclusive lease is held.
1310 *
1311 * %F_UNLCK to indicate no lease is held.
1312 *
1313 * (if a lease break is pending):
1314 *
1315 * %F_RDLCK to indicate an exclusive lease needs to be
1316 * changed to a shared lease (or removed).
1317 *
1318 * %F_UNLCK to indicate the lease needs to be removed.
1319 *
1320 * XXX: sfr & willy disagree over whether F_INPROGRESS
1321 * should be returned to userspace.
1322 */
1323int fcntl_getlease(struct file *filp)
1324{
1325 struct file_lock *fl;
1326 int type = F_UNLCK;
1327
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001328 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001329 time_out_leases(filp->f_path.dentry->d_inode);
1330 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 fl = fl->fl_next) {
1332 if (fl->fl_file == filp) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001333 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 break;
1335 }
1336 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001337 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 return type;
1339}
1340
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001341int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001343 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001344 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 struct inode *inode = dentry->d_inode;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001346 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347
J. Bruce Fields096657b2010-10-30 17:31:14 -04001348 lease = *flp;
1349
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001350 error = -EAGAIN;
1351 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001352 goto out;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001353 if ((arg == F_WRLCK)
1354 && ((dentry->d_count > 1)
1355 || (atomic_read(&inode->i_count) > 1)))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001356 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 /*
1359 * At this point, we know that if there is an exclusive
1360 * lease on this file, then we hold it on this filp
1361 * (otherwise our open of this file would have blocked).
1362 * And if we are trying to acquire an exclusive lease,
1363 * then the file is not open by anyone (including us)
1364 * except for this filp.
1365 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001366 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 for (before = &inode->i_flock;
1368 ((fl = *before) != NULL) && IS_LEASE(fl);
1369 before = &fl->fl_next) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001370 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 my_before = before;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001372 continue;
1373 }
1374 /*
1375 * No exclusive leases if someone else has a lease on
1376 * this file:
1377 */
1378 if (arg == F_WRLCK)
1379 goto out;
1380 /*
1381 * Modifying our existing lease is OK, but no getting a
1382 * new lease if someone else is opening for write:
1383 */
1384 if (fl->fl_flags & FL_UNLOCK_PENDING)
1385 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 }
1387
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (my_before != NULL) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001389 error = lease->fl_lmops->lm_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001390 if (!error)
1391 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 goto out;
1393 }
1394
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 error = -EINVAL;
1396 if (!leases_enable)
1397 goto out;
1398
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001399 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001400 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402out:
1403 return error;
1404}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001405
1406int generic_delete_lease(struct file *filp, struct file_lock **flp)
1407{
1408 struct file_lock *fl, **before;
1409 struct dentry *dentry = filp->f_path.dentry;
1410 struct inode *inode = dentry->d_inode;
1411
1412 for (before = &inode->i_flock;
1413 ((fl = *before) != NULL) && IS_LEASE(fl);
1414 before = &fl->fl_next) {
1415 if (fl->fl_file != filp)
1416 continue;
1417 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1418 }
1419 return -EAGAIN;
1420}
1421
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422/**
1423 * generic_setlease - sets a lease on an open file
1424 * @filp: file pointer
1425 * @arg: type of lease to obtain
1426 * @flp: input - file_lock to use, output - file_lock inserted
1427 *
1428 * The (input) flp->fl_lmops->lm_break function is required
1429 * by break_lease().
1430 *
1431 * Called with file_lock_lock held.
1432 */
1433int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1434{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 struct dentry *dentry = filp->f_path.dentry;
1436 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001437 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001439 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001440 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001442 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 error = security_file_lock(filp, arg);
1444 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001445 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
1447 time_out_leases(inode);
1448
1449 BUG_ON(!(*flp)->fl_lmops->lm_break);
1450
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001451 switch (arg) {
1452 case F_UNLCK:
1453 return generic_delete_lease(filp, flp);
1454 case F_RDLCK:
1455 case F_WRLCK:
1456 return generic_add_lease(filp, arg, flp);
1457 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001458 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001461EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001463static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1464{
1465 if (filp->f_op && filp->f_op->setlease)
1466 return filp->f_op->setlease(filp, arg, lease);
1467 else
1468 return generic_setlease(filp, arg, lease);
1469}
1470
1471/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001472 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 * @filp: file pointer
1474 * @arg: type of lease to obtain
1475 * @lease: file_lock to use
1476 *
1477 * Call this to establish a lease on the file.
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001478 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001479 * break_lease will oops!
1480 *
1481 * This will call the filesystem's setlease file method, if
1482 * defined. Note that there is no getlease method; instead, the
1483 * filesystem setlease method should call back to setlease() to
1484 * add a lease to the inode's lease list, where fcntl_getlease() can
1485 * find it. Since fcntl_getlease() only reports whether the current
1486 * task holds a lease, a cluster filesystem need only do this for
1487 * leases held by processes on this node.
1488 *
1489 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001490 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001491 * filesystem's open, create, and (on truncate) setattr methods.
1492 *
1493 * Warning: the only current setlease methods exist only to disable
1494 * leases in certain cases. More vfs changes may be required to
1495 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496 */
1497
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001498int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 int error;
1501
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001502 lock_flocks();
1503 error = __vfs_setlease(filp, arg, lease);
1504 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 return error;
1507}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001508EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001510static int do_fcntl_delete_lease(struct file *filp)
1511{
1512 struct file_lock fl, *flp = &fl;
1513
1514 lease_init(filp, F_UNLCK, flp);
1515
1516 return vfs_setlease(filp, F_UNLCK, &flp);
1517}
1518
1519static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001521 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001522 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 int error;
1524
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001525 fl = lease_alloc(filp, arg);
1526 if (IS_ERR(fl))
1527 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001529 new = fasync_alloc();
1530 if (!new) {
1531 locks_free_lock(fl);
1532 return -ENOMEM;
1533 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001534 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001535 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001536 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001537 if (error) {
1538 unlock_flocks();
1539 locks_free_lock(fl);
1540 goto out_free_fasync;
1541 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001542 if (ret != fl)
1543 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001545 /*
1546 * fasync_insert_entry() returns the old entry if any.
1547 * If there was no old entry, then it used 'new' and
1548 * inserted it into the fasync list. Clear new so that
1549 * we don't release it here.
1550 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001551 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001552 new = NULL;
1553
J. Bruce Fields8896b932010-11-03 18:09:18 -04001554 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001555 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001556
1557out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001558 if (new)
1559 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 return error;
1561}
1562
1563/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001564 * fcntl_setlease - sets a lease on an open file
1565 * @fd: open file descriptor
1566 * @filp: file pointer
1567 * @arg: type of lease to obtain
1568 *
1569 * Call this fcntl to establish a lease on the file.
1570 * Note that you also need to call %F_SETSIG to
1571 * receive a signal when the lease is broken.
1572 */
1573int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1574{
1575 if (arg == F_UNLCK)
1576 return do_fcntl_delete_lease(filp);
1577 return do_fcntl_add_lease(fd, filp, arg);
1578}
1579
1580/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1582 * @filp: The file to apply the lock to
1583 * @fl: The lock to be applied
1584 *
1585 * Add a FLOCK style lock to a file.
1586 */
1587int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1588{
1589 int error;
1590 might_sleep();
1591 for (;;) {
1592 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001593 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 break;
1595 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1596 if (!error)
1597 continue;
1598
1599 locks_delete_block(fl);
1600 break;
1601 }
1602 return error;
1603}
1604
1605EXPORT_SYMBOL(flock_lock_file_wait);
1606
1607/**
1608 * sys_flock: - flock() system call.
1609 * @fd: the file descriptor to lock.
1610 * @cmd: the type of lock to apply.
1611 *
1612 * Apply a %FL_FLOCK style lock to an open file descriptor.
1613 * The @cmd can be one of
1614 *
1615 * %LOCK_SH -- a shared lock.
1616 *
1617 * %LOCK_EX -- an exclusive lock.
1618 *
1619 * %LOCK_UN -- remove an existing lock.
1620 *
1621 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1622 *
1623 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1624 * processes read and write access respectively.
1625 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001626SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627{
1628 struct file *filp;
Al Virobdc68952012-04-21 18:46:53 -04001629 int fput_needed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 struct file_lock *lock;
1631 int can_sleep, unlock;
1632 int error;
1633
1634 error = -EBADF;
Al Virobdc68952012-04-21 18:46:53 -04001635 filp = fget_light(fd, &fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (!filp)
1637 goto out;
1638
1639 can_sleep = !(cmd & LOCK_NB);
1640 cmd &= ~LOCK_NB;
1641 unlock = (cmd == LOCK_UN);
1642
Al Viroaeb5d722008-09-02 15:28:45 -04001643 if (!unlock && !(cmd & LOCK_MAND) &&
1644 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 goto out_putf;
1646
1647 error = flock_make_lock(filp, &lock, cmd);
1648 if (error)
1649 goto out_putf;
1650 if (can_sleep)
1651 lock->fl_flags |= FL_SLEEP;
1652
Sten Spans713c0ec2009-07-16 09:41:39 +02001653 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 if (error)
1655 goto out_free;
1656
1657 if (filp->f_op && filp->f_op->flock)
1658 error = filp->f_op->flock(filp,
1659 (can_sleep) ? F_SETLKW : F_SETLK,
1660 lock);
1661 else
1662 error = flock_lock_file_wait(filp, lock);
1663
1664 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001665 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
1667 out_putf:
Al Virobdc68952012-04-21 18:46:53 -04001668 fput_light(filp, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 out:
1670 return error;
1671}
1672
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001673/**
1674 * vfs_test_lock - test file byte range lock
1675 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001676 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001677 *
1678 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1679 * setting conf->fl_type to something other than F_UNLCK.
1680 */
1681int vfs_test_lock(struct file *filp, struct file_lock *fl)
1682{
1683 if (filp->f_op && filp->f_op->lock)
1684 return filp->f_op->lock(filp, F_GETLK, fl);
1685 posix_test_lock(filp, fl);
1686 return 0;
1687}
1688EXPORT_SYMBOL_GPL(vfs_test_lock);
1689
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001690static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1691{
1692 flock->l_pid = fl->fl_pid;
1693#if BITS_PER_LONG == 32
1694 /*
1695 * Make sure we can represent the posix lock via
1696 * legacy 32bit flock.
1697 */
1698 if (fl->fl_start > OFFT_OFFSET_MAX)
1699 return -EOVERFLOW;
1700 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1701 return -EOVERFLOW;
1702#endif
1703 flock->l_start = fl->fl_start;
1704 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1705 fl->fl_end - fl->fl_start + 1;
1706 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001707 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001708 return 0;
1709}
1710
1711#if BITS_PER_LONG == 32
1712static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1713{
1714 flock->l_pid = fl->fl_pid;
1715 flock->l_start = fl->fl_start;
1716 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1717 fl->fl_end - fl->fl_start + 1;
1718 flock->l_whence = 0;
1719 flock->l_type = fl->fl_type;
1720}
1721#endif
1722
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723/* Report the first existing lock that would conflict with l.
1724 * This implements the F_GETLK command of fcntl().
1725 */
1726int fcntl_getlk(struct file *filp, struct flock __user *l)
1727{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001728 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 struct flock flock;
1730 int error;
1731
1732 error = -EFAULT;
1733 if (copy_from_user(&flock, l, sizeof(flock)))
1734 goto out;
1735 error = -EINVAL;
1736 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1737 goto out;
1738
1739 error = flock_to_posix_lock(filp, &file_lock, &flock);
1740 if (error)
1741 goto out;
1742
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001743 error = vfs_test_lock(filp, &file_lock);
1744 if (error)
1745 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001747 flock.l_type = file_lock.fl_type;
1748 if (file_lock.fl_type != F_UNLCK) {
1749 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001750 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 }
1753 error = -EFAULT;
1754 if (!copy_to_user(l, &flock, sizeof(flock)))
1755 error = 0;
1756out:
1757 return error;
1758}
1759
Marc Eshel7723ec92007-01-18 15:08:55 -05001760/**
1761 * vfs_lock_file - file byte range lock
1762 * @filp: The file to apply the lock to
1763 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1764 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001765 * @conf: Place to return a copy of the conflicting lock, if found.
1766 *
1767 * A caller that doesn't care about the conflicting lock may pass NULL
1768 * as the final argument.
1769 *
1770 * If the filesystem defines a private ->lock() method, then @conf will
1771 * be left unchanged; so a caller that cares should initialize it to
1772 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001773 *
1774 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1775 * locks, the ->lock() interface may return asynchronously, before the lock has
1776 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001777 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05001778 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1779 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001780 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001781 * request completes.
1782 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001783 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1784 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001785 * nonzero return code and the file system should release the lock. The file
1786 * system is also responsible to keep a corresponding posix lock when it
1787 * grants a lock so the VFS can find out which locks are locally held and do
1788 * the correct lock cleanup when required.
1789 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001790 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001791 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001792 */
Marc Eshel150b3932007-01-18 16:15:35 -05001793int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001794{
1795 if (filp->f_op && filp->f_op->lock)
1796 return filp->f_op->lock(filp, cmd, fl);
1797 else
Marc Eshel150b3932007-01-18 16:15:35 -05001798 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001799}
1800EXPORT_SYMBOL_GPL(vfs_lock_file);
1801
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001802static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1803 struct file_lock *fl)
1804{
1805 int error;
1806
1807 error = security_file_lock(filp, fl->fl_type);
1808 if (error)
1809 return error;
1810
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001811 for (;;) {
1812 error = vfs_lock_file(filp, cmd, fl, NULL);
1813 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001814 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001815 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1816 if (!error)
1817 continue;
1818
1819 locks_delete_block(fl);
1820 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001821 }
1822
1823 return error;
1824}
1825
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826/* Apply the lock described by l to an open file descriptor.
1827 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1828 */
Peter Staubachc2936212005-07-27 11:45:09 -07001829int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1830 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831{
1832 struct file_lock *file_lock = locks_alloc_lock();
1833 struct flock flock;
1834 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001835 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 int error;
1837
1838 if (file_lock == NULL)
1839 return -ENOLCK;
1840
1841 /*
1842 * This might block, so we do it before checking the inode.
1843 */
1844 error = -EFAULT;
1845 if (copy_from_user(&flock, l, sizeof(flock)))
1846 goto out;
1847
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001848 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849
1850 /* Don't allow mandatory locks on files that may be memory mapped
1851 * and shared.
1852 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001853 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 error = -EAGAIN;
1855 goto out;
1856 }
1857
Peter Staubachc2936212005-07-27 11:45:09 -07001858again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 error = flock_to_posix_lock(filp, file_lock, &flock);
1860 if (error)
1861 goto out;
1862 if (cmd == F_SETLKW) {
1863 file_lock->fl_flags |= FL_SLEEP;
1864 }
1865
1866 error = -EBADF;
1867 switch (flock.l_type) {
1868 case F_RDLCK:
1869 if (!(filp->f_mode & FMODE_READ))
1870 goto out;
1871 break;
1872 case F_WRLCK:
1873 if (!(filp->f_mode & FMODE_WRITE))
1874 goto out;
1875 break;
1876 case F_UNLCK:
1877 break;
1878 default:
1879 error = -EINVAL;
1880 goto out;
1881 }
1882
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001883 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884
Peter Staubachc2936212005-07-27 11:45:09 -07001885 /*
1886 * Attempt to detect a close/fcntl race and recover by
1887 * releasing the lock that was just acquired.
1888 */
Al Viro0b2bac22008-05-06 13:58:34 -04001889 /*
1890 * we need that spin_lock here - it prevents reordering between
1891 * update of inode->i_flock and check for it done in close().
1892 * rcu_read_lock() wouldn't do.
1893 */
1894 spin_lock(&current->files->file_lock);
1895 f = fcheck(fd);
1896 spin_unlock(&current->files->file_lock);
1897 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001898 flock.l_type = F_UNLCK;
1899 goto again;
1900 }
1901
1902out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 locks_free_lock(file_lock);
1904 return error;
1905}
1906
1907#if BITS_PER_LONG == 32
1908/* Report the first existing lock that would conflict with l.
1909 * This implements the F_GETLK command of fcntl().
1910 */
1911int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1912{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001913 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 struct flock64 flock;
1915 int error;
1916
1917 error = -EFAULT;
1918 if (copy_from_user(&flock, l, sizeof(flock)))
1919 goto out;
1920 error = -EINVAL;
1921 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1922 goto out;
1923
1924 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1925 if (error)
1926 goto out;
1927
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001928 error = vfs_test_lock(filp, &file_lock);
1929 if (error)
1930 goto out;
1931
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001932 flock.l_type = file_lock.fl_type;
1933 if (file_lock.fl_type != F_UNLCK)
1934 posix_lock_to_flock64(&flock, &file_lock);
1935
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 error = -EFAULT;
1937 if (!copy_to_user(l, &flock, sizeof(flock)))
1938 error = 0;
1939
1940out:
1941 return error;
1942}
1943
1944/* Apply the lock described by l to an open file descriptor.
1945 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1946 */
Peter Staubachc2936212005-07-27 11:45:09 -07001947int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1948 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949{
1950 struct file_lock *file_lock = locks_alloc_lock();
1951 struct flock64 flock;
1952 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001953 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 int error;
1955
1956 if (file_lock == NULL)
1957 return -ENOLCK;
1958
1959 /*
1960 * This might block, so we do it before checking the inode.
1961 */
1962 error = -EFAULT;
1963 if (copy_from_user(&flock, l, sizeof(flock)))
1964 goto out;
1965
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001966 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
1968 /* Don't allow mandatory locks on files that may be memory mapped
1969 * and shared.
1970 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001971 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 error = -EAGAIN;
1973 goto out;
1974 }
1975
Peter Staubachc2936212005-07-27 11:45:09 -07001976again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 error = flock64_to_posix_lock(filp, file_lock, &flock);
1978 if (error)
1979 goto out;
1980 if (cmd == F_SETLKW64) {
1981 file_lock->fl_flags |= FL_SLEEP;
1982 }
1983
1984 error = -EBADF;
1985 switch (flock.l_type) {
1986 case F_RDLCK:
1987 if (!(filp->f_mode & FMODE_READ))
1988 goto out;
1989 break;
1990 case F_WRLCK:
1991 if (!(filp->f_mode & FMODE_WRITE))
1992 goto out;
1993 break;
1994 case F_UNLCK:
1995 break;
1996 default:
1997 error = -EINVAL;
1998 goto out;
1999 }
2000
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002001 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
Peter Staubachc2936212005-07-27 11:45:09 -07002003 /*
2004 * Attempt to detect a close/fcntl race and recover by
2005 * releasing the lock that was just acquired.
2006 */
Al Viro0b2bac22008-05-06 13:58:34 -04002007 spin_lock(&current->files->file_lock);
2008 f = fcheck(fd);
2009 spin_unlock(&current->files->file_lock);
2010 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002011 flock.l_type = F_UNLCK;
2012 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 }
2014
2015out:
2016 locks_free_lock(file_lock);
2017 return error;
2018}
2019#endif /* BITS_PER_LONG == 32 */
2020
2021/*
2022 * This function is called when the file is being removed
2023 * from the task's fd array. POSIX locks belonging to this task
2024 * are deleted at this time.
2025 */
2026void locks_remove_posix(struct file *filp, fl_owner_t owner)
2027{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002028 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
2030 /*
2031 * If there are no locks held on this file, we don't need to call
2032 * posix_lock_file(). Another process could be setting a lock on this
2033 * file at the same time, but we wouldn't remove that lock anyway.
2034 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002035 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 return;
2037
2038 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002039 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 lock.fl_start = 0;
2041 lock.fl_end = OFFSET_MAX;
2042 lock.fl_owner = owner;
2043 lock.fl_pid = current->tgid;
2044 lock.fl_file = filp;
2045 lock.fl_ops = NULL;
2046 lock.fl_lmops = NULL;
2047
Marc Eshel150b3932007-01-18 16:15:35 -05002048 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2051 lock.fl_ops->fl_release_private(&lock);
2052}
2053
2054EXPORT_SYMBOL(locks_remove_posix);
2055
2056/*
2057 * This function is called on the last close of an open file.
2058 */
2059void locks_remove_flock(struct file *filp)
2060{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002061 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 struct file_lock *fl;
2063 struct file_lock **before;
2064
2065 if (!inode->i_flock)
2066 return;
2067
2068 if (filp->f_op && filp->f_op->flock) {
2069 struct file_lock fl = {
2070 .fl_pid = current->tgid,
2071 .fl_file = filp,
2072 .fl_flags = FL_FLOCK,
2073 .fl_type = F_UNLCK,
2074 .fl_end = OFFSET_MAX,
2075 };
2076 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002077 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2078 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 }
2080
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002081 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 before = &inode->i_flock;
2083
2084 while ((fl = *before) != NULL) {
2085 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002086 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 locks_delete_lock(before);
2088 continue;
2089 }
2090 if (IS_LEASE(fl)) {
2091 lease_modify(before, F_UNLCK);
2092 continue;
2093 }
2094 /* What? */
2095 BUG();
2096 }
2097 before = &fl->fl_next;
2098 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002099 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100}
2101
2102/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 * posix_unblock_lock - stop waiting for a file lock
2104 * @filp: how the file was opened
2105 * @waiter: the lock which was waiting
2106 *
2107 * lockd needs to block waiting for locks.
2108 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002109int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2111{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002112 int status = 0;
2113
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002114 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002115 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002117 else
2118 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002119 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002120 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121}
2122
2123EXPORT_SYMBOL(posix_unblock_lock);
2124
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002125/**
2126 * vfs_cancel_lock - file byte range unblock lock
2127 * @filp: The file to apply the unblock to
2128 * @fl: The lock to be unblocked
2129 *
2130 * Used by lock managers to cancel blocked requests
2131 */
2132int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2133{
2134 if (filp->f_op && filp->f_op->lock)
2135 return filp->f_op->lock(filp, F_CANCELLK, fl);
2136 return 0;
2137}
2138
2139EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2140
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002141#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002142#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002143#include <linux/seq_file.h>
2144
2145static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002146 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147{
2148 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002149 unsigned int fl_pid;
2150
2151 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002152 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002153 else
2154 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155
2156 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002157 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
Jerome Marchand99dc8292010-10-26 14:22:33 -07002159 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002161 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2163 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002164 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 } else if (IS_FLOCK(fl)) {
2166 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002167 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002169 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 }
2171 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002172 seq_printf(f, "LEASE ");
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002173 if (lease_breaking(fl))
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002174 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002176 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002178 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002180 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 }
2182 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002183 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 (fl->fl_type & LOCK_READ)
2185 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2186 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2187 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002188 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002189 (lease_breaking(fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2191 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2192 }
2193 if (inode) {
2194#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002195 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 inode->i_sb->s_id, inode->i_ino);
2197#else
2198 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002199 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 MAJOR(inode->i_sb->s_dev),
2201 MINOR(inode->i_sb->s_dev), inode->i_ino);
2202#endif
2203 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002204 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 }
2206 if (IS_POSIX(fl)) {
2207 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002208 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002210 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002212 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 }
2214}
2215
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002216static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002218 struct file_lock *fl, *bfl;
2219
2220 fl = list_entry(v, struct file_lock, fl_link);
2221
Jerome Marchand99dc8292010-10-26 14:22:33 -07002222 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002223
2224 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002225 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002226
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002227 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228}
2229
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002230static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002232 loff_t *p = f->private;
2233
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002234 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002235 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002236 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237}
2238
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002239static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2240{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002241 loff_t *p = f->private;
2242 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002243 return seq_list_next(v, &file_lock_list, pos);
2244}
2245
2246static void locks_stop(struct seq_file *f, void *v)
2247{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002248 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002249}
2250
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002251static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002252 .start = locks_start,
2253 .next = locks_next,
2254 .stop = locks_stop,
2255 .show = locks_show,
2256};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002257
2258static int locks_open(struct inode *inode, struct file *filp)
2259{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002260 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002261}
2262
2263static const struct file_operations proc_locks_operations = {
2264 .open = locks_open,
2265 .read = seq_read,
2266 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002267 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002268};
2269
2270static int __init proc_locks_init(void)
2271{
2272 proc_create("locks", 0, NULL, &proc_locks_operations);
2273 return 0;
2274}
2275module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002276#endif
2277
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278/**
2279 * lock_may_read - checks that the region is free of locks
2280 * @inode: the inode that is being read
2281 * @start: the first byte to read
2282 * @len: the number of bytes to read
2283 *
2284 * Emulates Windows locking requirements. Whole-file
2285 * mandatory locks (share modes) can prohibit a read and
2286 * byte-range POSIX locks can prohibit a read if they overlap.
2287 *
2288 * N.B. this function is only ever called
2289 * from knfsd and ownership of locks is never checked.
2290 */
2291int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2292{
2293 struct file_lock *fl;
2294 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002295 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2297 if (IS_POSIX(fl)) {
2298 if (fl->fl_type == F_RDLCK)
2299 continue;
2300 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2301 continue;
2302 } else if (IS_FLOCK(fl)) {
2303 if (!(fl->fl_type & LOCK_MAND))
2304 continue;
2305 if (fl->fl_type & LOCK_READ)
2306 continue;
2307 } else
2308 continue;
2309 result = 0;
2310 break;
2311 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002312 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 return result;
2314}
2315
2316EXPORT_SYMBOL(lock_may_read);
2317
2318/**
2319 * lock_may_write - checks that the region is free of locks
2320 * @inode: the inode that is being written
2321 * @start: the first byte to write
2322 * @len: the number of bytes to write
2323 *
2324 * Emulates Windows locking requirements. Whole-file
2325 * mandatory locks (share modes) can prohibit a write and
2326 * byte-range POSIX locks can prohibit a write if they overlap.
2327 *
2328 * N.B. this function is only ever called
2329 * from knfsd and ownership of locks is never checked.
2330 */
2331int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2332{
2333 struct file_lock *fl;
2334 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002335 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2337 if (IS_POSIX(fl)) {
2338 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2339 continue;
2340 } else if (IS_FLOCK(fl)) {
2341 if (!(fl->fl_type & LOCK_MAND))
2342 continue;
2343 if (fl->fl_type & LOCK_WRITE)
2344 continue;
2345 } else
2346 continue;
2347 result = 0;
2348 break;
2349 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002350 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 return result;
2352}
2353
2354EXPORT_SYMBOL(lock_may_write);
2355
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356static int __init filelock_init(void)
2357{
2358 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002359 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2360
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 return 0;
2362}
2363
2364core_initcall(filelock_init);