blob: 4f441e46cef47bc67b08a3e82b78f389dfbbf818 [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 }
203 if (fl->fl_lmops) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400204 if (fl->fl_lmops->lm_release_private)
205 fl->fl_lmops->lm_release_private(fl);
Trond Myklebust47831f32006-03-20 13:44:05 -0500206 fl->fl_lmops = NULL;
207 }
208
209}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500210EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500211
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400213void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800215 BUG_ON(waitqueue_active(&fl->fl_wait));
216 BUG_ON(!list_empty(&fl->fl_block));
217 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
Trond Myklebust47831f32006-03-20 13:44:05 -0500219 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 kmem_cache_free(filelock_cache, fl);
221}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400222EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
224void locks_init_lock(struct file_lock *fl)
225{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200226 memset(fl, 0, sizeof(struct file_lock));
227 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228}
229
230EXPORT_SYMBOL(locks_init_lock);
231
Trond Myklebust47831f32006-03-20 13:44:05 -0500232static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
233{
234 if (fl->fl_ops) {
235 if (fl->fl_ops->fl_copy_lock)
236 fl->fl_ops->fl_copy_lock(new, fl);
237 new->fl_ops = fl->fl_ops;
238 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400239 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500240 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500241}
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243/*
244 * Initialize a new lock from an existing file_lock structure.
245 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400246void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247{
248 new->fl_owner = fl->fl_owner;
249 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500250 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 new->fl_flags = fl->fl_flags;
252 new->fl_type = fl->fl_type;
253 new->fl_start = fl->fl_start;
254 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500255 new->fl_ops = NULL;
256 new->fl_lmops = NULL;
257}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700258EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500259
260void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
261{
262 locks_release_private(new);
263
264 __locks_copy_lock(new, fl);
265 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 new->fl_ops = fl->fl_ops;
267 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500268
269 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270}
271
272EXPORT_SYMBOL(locks_copy_lock);
273
274static inline int flock_translate_cmd(int cmd) {
275 if (cmd & LOCK_MAND)
276 return cmd & (LOCK_MAND | LOCK_RW);
277 switch (cmd) {
278 case LOCK_SH:
279 return F_RDLCK;
280 case LOCK_EX:
281 return F_WRLCK;
282 case LOCK_UN:
283 return F_UNLCK;
284 }
285 return -EINVAL;
286}
287
288/* Fill in a file_lock structure with an appropriate FLOCK lock. */
289static int flock_make_lock(struct file *filp, struct file_lock **lock,
290 unsigned int cmd)
291{
292 struct file_lock *fl;
293 int type = flock_translate_cmd(cmd);
294 if (type < 0)
295 return type;
296
297 fl = locks_alloc_lock();
298 if (fl == NULL)
299 return -ENOMEM;
300
301 fl->fl_file = filp;
302 fl->fl_pid = current->tgid;
303 fl->fl_flags = FL_FLOCK;
304 fl->fl_type = type;
305 fl->fl_end = OFFSET_MAX;
306
307 *lock = fl;
308 return 0;
309}
310
311static int assign_type(struct file_lock *fl, int type)
312{
313 switch (type) {
314 case F_RDLCK:
315 case F_WRLCK:
316 case F_UNLCK:
317 fl->fl_type = type;
318 break;
319 default:
320 return -EINVAL;
321 }
322 return 0;
323}
324
325/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
326 * style lock.
327 */
328static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
329 struct flock *l)
330{
331 off_t start, end;
332
333 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700334 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 start = 0;
336 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700337 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 start = filp->f_pos;
339 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700340 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800341 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 break;
343 default:
344 return -EINVAL;
345 }
346
347 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
348 POSIX-2001 defines it. */
349 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 if (start < 0)
351 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700352 fl->fl_end = OFFSET_MAX;
353 if (l->l_len > 0) {
354 end = start + l->l_len - 1;
355 fl->fl_end = end;
356 } else if (l->l_len < 0) {
357 end = start - 1;
358 fl->fl_end = end;
359 start += l->l_len;
360 if (start < 0)
361 return -EINVAL;
362 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700364 if (fl->fl_end < fl->fl_start)
365 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366
367 fl->fl_owner = current->files;
368 fl->fl_pid = current->tgid;
369 fl->fl_file = filp;
370 fl->fl_flags = FL_POSIX;
371 fl->fl_ops = NULL;
372 fl->fl_lmops = NULL;
373
374 return assign_type(fl, l->l_type);
375}
376
377#if BITS_PER_LONG == 32
378static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
379 struct flock64 *l)
380{
381 loff_t start;
382
383 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700384 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 start = 0;
386 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700387 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 start = filp->f_pos;
389 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700390 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800391 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 break;
393 default:
394 return -EINVAL;
395 }
396
Trond Myklebust4c780a42005-10-18 14:20:21 -0700397 start += l->l_start;
398 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700400 fl->fl_end = OFFSET_MAX;
401 if (l->l_len > 0) {
402 fl->fl_end = start + l->l_len - 1;
403 } else if (l->l_len < 0) {
404 fl->fl_end = start - 1;
405 start += l->l_len;
406 if (start < 0)
407 return -EINVAL;
408 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700410 if (fl->fl_end < fl->fl_start)
411 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 fl->fl_owner = current->files;
414 fl->fl_pid = current->tgid;
415 fl->fl_file = filp;
416 fl->fl_flags = FL_POSIX;
417 fl->fl_ops = NULL;
418 fl->fl_lmops = NULL;
419
Namhyung Kimf32cb532011-01-17 15:45:59 +0900420 return assign_type(fl, l->l_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421}
422#endif
423
424/* default lease lock manager operations */
425static void lease_break_callback(struct file_lock *fl)
426{
427 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
428}
429
430static void lease_release_private_callback(struct file_lock *fl)
431{
432 if (!fl->fl_file)
433 return;
434
435 f_delown(fl->fl_file);
436 fl->fl_file->f_owner.signum = 0;
437}
438
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700439static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400440 .lm_break = lease_break_callback,
441 .lm_release_private = lease_release_private_callback,
442 .lm_change = lease_modify,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443};
444
445/*
446 * Initialize a lease, use the default lock manager operations
447 */
448static int lease_init(struct file *filp, int type, struct file_lock *fl)
449 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400450 if (assign_type(fl, type) != 0)
451 return -EINVAL;
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 fl->fl_owner = current->files;
454 fl->fl_pid = current->tgid;
455
456 fl->fl_file = filp;
457 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 fl->fl_start = 0;
459 fl->fl_end = OFFSET_MAX;
460 fl->fl_ops = NULL;
461 fl->fl_lmops = &lease_manager_ops;
462 return 0;
463}
464
465/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500466static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467{
468 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400469 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
471 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500472 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400475 if (error) {
476 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500477 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400478 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500479 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
482/* Check if two locks overlap each other.
483 */
484static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
485{
486 return ((fl1->fl_end >= fl2->fl_start) &&
487 (fl2->fl_end >= fl1->fl_start));
488}
489
490/*
491 * Check whether two locks have the same owner.
492 */
Matt Mackall33443c42006-01-08 01:05:22 -0800493static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400495 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400497 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return fl1->fl_owner == fl2->fl_owner;
499}
500
501/* Remove waiter from blocker's block list.
502 * When blocker ends up pointing to itself then the list is empty.
503 */
Matt Mackall33443c42006-01-08 01:05:22 -0800504static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
506 list_del_init(&waiter->fl_block);
507 list_del_init(&waiter->fl_link);
508 waiter->fl_next = NULL;
509}
510
511/*
512 */
Pavel Shilovsky66189be2012-03-28 21:56:19 +0400513void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200515 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 __locks_delete_block(waiter);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200517 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518}
Pavel Shilovsky66189be2012-03-28 21:56:19 +0400519EXPORT_SYMBOL(locks_delete_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521/* Insert waiter into blocker's block list.
522 * We use a circular list so that processes can be easily woken up in
523 * the order they blocked. The documentation doesn't require this but
524 * it seems like the reasonable thing to do.
525 */
526static void locks_insert_block(struct file_lock *blocker,
527 struct file_lock *waiter)
528{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800529 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 list_add_tail(&waiter->fl_block, &blocker->fl_block);
531 waiter->fl_next = blocker;
532 if (IS_POSIX(blocker))
533 list_add(&waiter->fl_link, &blocked_list);
534}
535
536/* Wake up processes blocked waiting for blocker.
537 * If told to wait then schedule the processes until the block list
538 * is empty, otherwise empty the block list ourselves.
539 */
540static void locks_wake_up_blocks(struct file_lock *blocker)
541{
542 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400543 struct file_lock *waiter;
544
545 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 struct file_lock, fl_block);
547 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400548 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
549 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550 else
551 wake_up(&waiter->fl_wait);
552 }
553}
554
555/* Insert file lock fl into an inode's lock list at the position indicated
556 * by pos. At the same time add the lock to the global file lock list.
557 */
558static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
559{
560 list_add(&fl->fl_link, &file_lock_list);
561
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000562 fl->fl_nspid = get_pid(task_tgid(current));
563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 /* insert into file's list */
565 fl->fl_next = *pos;
566 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
568
569/*
570 * Delete a lock and then free it.
571 * Wake up processes that are blocked waiting for this lock,
572 * notify the FS that the lock has been cleared and
573 * finally free the lock.
574 */
575static void locks_delete_lock(struct file_lock **thisfl_p)
576{
577 struct file_lock *fl = *thisfl_p;
578
579 *thisfl_p = fl->fl_next;
580 fl->fl_next = NULL;
581 list_del_init(&fl->fl_link);
582
583 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
584 if (fl->fl_fasync != NULL) {
585 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
586 fl->fl_fasync = NULL;
587 }
588
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000589 if (fl->fl_nspid) {
590 put_pid(fl->fl_nspid);
591 fl->fl_nspid = NULL;
592 }
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 locks_wake_up_blocks(fl);
595 locks_free_lock(fl);
596}
597
598/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
599 * checks for shared/exclusive status of overlapping locks.
600 */
601static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
602{
603 if (sys_fl->fl_type == F_WRLCK)
604 return 1;
605 if (caller_fl->fl_type == F_WRLCK)
606 return 1;
607 return 0;
608}
609
610/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
611 * checking before calling the locks_conflict().
612 */
613static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
614{
615 /* POSIX locks owned by the same process do not conflict with
616 * each other.
617 */
618 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
619 return (0);
620
621 /* Check whether they overlap */
622 if (!locks_overlap(caller_fl, sys_fl))
623 return 0;
624
625 return (locks_conflict(caller_fl, sys_fl));
626}
627
628/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
629 * checking before calling the locks_conflict().
630 */
631static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
632{
633 /* FLOCK locks referring to the same filp do not conflict with
634 * each other.
635 */
636 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
637 return (0);
638 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
639 return 0;
640
641 return (locks_conflict(caller_fl, sys_fl));
642}
643
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400644void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500645posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
647 struct file_lock *cfl;
648
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200649 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800650 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (!IS_POSIX(cfl))
652 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400653 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 break;
655 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000656 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500657 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000658 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800659 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000660 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400661 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200662 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400663 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665EXPORT_SYMBOL(posix_test_lock);
666
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400667/*
668 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400670 * We attempt to detect deadlocks that are due purely to posix file
671 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400673 * We assume that a task can be waiting for at most one lock at a time.
674 * So for any acquired lock, the process holding that lock may be
675 * waiting on at most one other lock. That lock in turns may be held by
676 * someone waiting for at most one other lock. Given a requested lock
677 * caller_fl which is about to wait for a conflicting lock block_fl, we
678 * follow this chain of waiters to ensure we are not about to create a
679 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400680 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400681 * Since we do this before we ever put a process to sleep on a lock, we
682 * are ensured that there is never a cycle; that is what guarantees that
683 * the while() loop in posix_locks_deadlock() eventually completes.
684 *
685 * Note: the above assumption may not be true when handling lock
686 * requests from a broken NFS client. It may also fail in the presence
687 * of tasks (such as posix threads) sharing the same open file table.
688 *
689 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400691
692#define MAX_DEADLK_ITERATIONS 10
693
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400694/* Find a lock that the owner of the given block_fl is blocking on. */
695static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
696{
697 struct file_lock *fl;
698
699 list_for_each_entry(fl, &blocked_list, fl_link) {
700 if (posix_same_owner(fl, block_fl))
701 return fl->fl_next;
702 }
703 return NULL;
704}
705
Adrian Bunkb0904e12006-06-23 02:05:13 -0700706static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 struct file_lock *block_fl)
708{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400709 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400711 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
712 if (i++ > MAX_DEADLK_ITERATIONS)
713 return 0;
714 if (posix_same_owner(caller_fl, block_fl))
715 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 }
717 return 0;
718}
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400721 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400722 *
723 * Note that if called with an FL_EXISTS argument, the caller may determine
724 * whether or not a lock was successfully freed by testing the return
725 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800727static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800729 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800731 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 int error = 0;
733 int found = 0;
734
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200735 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
736 new_fl = locks_alloc_lock();
737 if (!new_fl)
738 return -ENOMEM;
739 }
740
741 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400742 if (request->fl_flags & FL_ACCESS)
743 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400744
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 for_each_lock(inode, before) {
746 struct file_lock *fl = *before;
747 if (IS_POSIX(fl))
748 break;
749 if (IS_LEASE(fl))
750 continue;
751 if (filp != fl->fl_file)
752 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800753 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 goto out;
755 found = 1;
756 locks_delete_lock(before);
757 break;
758 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
Trond Myklebustf475ae92006-06-29 16:38:32 -0400760 if (request->fl_type == F_UNLCK) {
761 if ((request->fl_flags & FL_EXISTS) && !found)
762 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800763 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
766 /*
767 * If a higher-priority process was blocked on the old file lock,
768 * give it the opportunity to lock the file.
769 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200770 if (found) {
771 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200772 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200773 lock_flocks();
774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400776find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 for_each_lock(inode, before) {
778 struct file_lock *fl = *before;
779 if (IS_POSIX(fl))
780 break;
781 if (IS_LEASE(fl))
782 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800783 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 continue;
785 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700786 if (!(request->fl_flags & FL_SLEEP))
787 goto out;
788 error = FILE_LOCK_DEFERRED;
789 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 goto out;
791 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400792 if (request->fl_flags & FL_ACCESS)
793 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800794 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700795 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800796 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400797 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
799out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200800 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800801 if (new_fl)
802 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 return error;
804}
805
Marc Eshel150b3932007-01-18 16:15:35 -0500806static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807{
808 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700809 struct file_lock *new_fl = NULL;
810 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 struct file_lock *left = NULL;
812 struct file_lock *right = NULL;
813 struct file_lock **before;
814 int error, added = 0;
815
816 /*
817 * We may need two file_lock structures for this operation,
818 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700819 *
820 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700822 if (!(request->fl_flags & FL_ACCESS) &&
823 (request->fl_type != F_UNLCK ||
824 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
825 new_fl = locks_alloc_lock();
826 new_fl2 = locks_alloc_lock();
827 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200829 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (request->fl_type != F_UNLCK) {
831 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500832 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 if (!IS_POSIX(fl))
834 continue;
835 if (!posix_locks_conflict(request, fl))
836 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800837 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400838 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 error = -EAGAIN;
840 if (!(request->fl_flags & FL_SLEEP))
841 goto out;
842 error = -EDEADLK;
843 if (posix_locks_deadlock(request, fl))
844 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700845 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 locks_insert_block(fl, request);
847 goto out;
848 }
849 }
850
851 /* If we're just looking for a conflict, we're done. */
852 error = 0;
853 if (request->fl_flags & FL_ACCESS)
854 goto out;
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 * Find the first old lock with the same owner as the new lock.
858 */
859
860 before = &inode->i_flock;
861
862 /* First skip locks owned by other processes. */
863 while ((fl = *before) && (!IS_POSIX(fl) ||
864 !posix_same_owner(request, fl))) {
865 before = &fl->fl_next;
866 }
867
868 /* Process locks with this owner. */
869 while ((fl = *before) && posix_same_owner(request, fl)) {
870 /* Detect adjacent or overlapping regions (if same lock type)
871 */
872 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700873 /* In all comparisons of start vs end, use
874 * "start - 1" rather than "end + 1". If end
875 * is OFFSET_MAX, end + 1 will become negative.
876 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 if (fl->fl_end < request->fl_start - 1)
878 goto next_lock;
879 /* If the next lock in the list has entirely bigger
880 * addresses than the new one, insert the lock here.
881 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700882 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 break;
884
885 /* If we come here, the new and old lock are of the
886 * same type and adjacent or overlapping. Make one
887 * lock yielding from the lower start address of both
888 * locks to the higher end address.
889 */
890 if (fl->fl_start > request->fl_start)
891 fl->fl_start = request->fl_start;
892 else
893 request->fl_start = fl->fl_start;
894 if (fl->fl_end < request->fl_end)
895 fl->fl_end = request->fl_end;
896 else
897 request->fl_end = fl->fl_end;
898 if (added) {
899 locks_delete_lock(before);
900 continue;
901 }
902 request = fl;
903 added = 1;
904 }
905 else {
906 /* Processing for different lock types is a bit
907 * more complex.
908 */
909 if (fl->fl_end < request->fl_start)
910 goto next_lock;
911 if (fl->fl_start > request->fl_end)
912 break;
913 if (request->fl_type == F_UNLCK)
914 added = 1;
915 if (fl->fl_start < request->fl_start)
916 left = fl;
917 /* If the next lock in the list has a higher end
918 * address than the new one, insert the new one here.
919 */
920 if (fl->fl_end > request->fl_end) {
921 right = fl;
922 break;
923 }
924 if (fl->fl_start >= request->fl_start) {
925 /* The new lock completely replaces an old
926 * one (This may happen several times).
927 */
928 if (added) {
929 locks_delete_lock(before);
930 continue;
931 }
932 /* Replace the old lock with the new one.
933 * Wake up anybody waiting for the old one,
934 * as the change in lock type might satisfy
935 * their needs.
936 */
937 locks_wake_up_blocks(fl);
938 fl->fl_start = request->fl_start;
939 fl->fl_end = request->fl_end;
940 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500941 locks_release_private(fl);
942 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 request = fl;
944 added = 1;
945 }
946 }
947 /* Go on to next lock.
948 */
949 next_lock:
950 before = &fl->fl_next;
951 }
952
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700953 /*
954 * The above code only modifies existing locks in case of
955 * merging or replacing. If new lock(s) need to be inserted
956 * all modifications are done bellow this, so it's safe yet to
957 * bail out.
958 */
959 error = -ENOLCK; /* "no luck" */
960 if (right && left == right && !new_fl2)
961 goto out;
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 error = 0;
964 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400965 if (request->fl_type == F_UNLCK) {
966 if (request->fl_flags & FL_EXISTS)
967 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400969 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700970
971 if (!new_fl) {
972 error = -ENOLCK;
973 goto out;
974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 locks_copy_lock(new_fl, request);
976 locks_insert_lock(before, new_fl);
977 new_fl = NULL;
978 }
979 if (right) {
980 if (left == right) {
981 /* The new lock breaks the old one in two pieces,
982 * so we have to use the second new lock.
983 */
984 left = new_fl2;
985 new_fl2 = NULL;
986 locks_copy_lock(left, right);
987 locks_insert_lock(before, left);
988 }
989 right->fl_start = request->fl_end + 1;
990 locks_wake_up_blocks(right);
991 }
992 if (left) {
993 left->fl_end = request->fl_start - 1;
994 locks_wake_up_blocks(left);
995 }
996 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200997 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 /*
999 * Free any unused locks.
1000 */
1001 if (new_fl)
1002 locks_free_lock(new_fl);
1003 if (new_fl2)
1004 locks_free_lock(new_fl2);
1005 return error;
1006}
1007
1008/**
1009 * posix_lock_file - Apply a POSIX-style lock to a file
1010 * @filp: The file to apply the lock to
1011 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001012 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 *
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
Trond Myklebustf475ae92006-06-29 16:38:32 -04001017 *
1018 * Note that if called with an FL_EXISTS argument, the caller may determine
1019 * whether or not a lock was successfully freed by testing the return
1020 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 */
Marc Eshel150b3932007-01-18 16:15:35 -05001022int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001023 struct file_lock *conflock)
1024{
Marc Eshel150b3932007-01-18 16:15:35 -05001025 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001026}
Marc Eshel150b3932007-01-18 16:15:35 -05001027EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029/**
1030 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1031 * @filp: The file to apply the lock to
1032 * @fl: The lock to be applied
1033 *
1034 * Add a POSIX style lock to a file.
1035 * We merge adjacent & overlapping locks whenever possible.
1036 * POSIX locks are sorted by owner task, then by starting address
1037 */
1038int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1039{
1040 int error;
1041 might_sleep ();
1042 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001043 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001044 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 break;
1046 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1047 if (!error)
1048 continue;
1049
1050 locks_delete_block(fl);
1051 break;
1052 }
1053 return error;
1054}
1055EXPORT_SYMBOL(posix_lock_file_wait);
1056
1057/**
1058 * locks_mandatory_locked - Check for an active lock
1059 * @inode: the file to check
1060 *
1061 * Searches the inode's list of locks to find any POSIX locks which conflict.
1062 * This function is called from locks_verify_locked() only.
1063 */
1064int locks_mandatory_locked(struct inode *inode)
1065{
1066 fl_owner_t owner = current->files;
1067 struct file_lock *fl;
1068
1069 /*
1070 * Search the lock list for this inode for any POSIX locks.
1071 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001072 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1074 if (!IS_POSIX(fl))
1075 continue;
1076 if (fl->fl_owner != owner)
1077 break;
1078 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001079 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 return fl ? -EAGAIN : 0;
1081}
1082
1083/**
1084 * locks_mandatory_area - Check for a conflicting lock
1085 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1086 * for shared
1087 * @inode: the file to check
1088 * @filp: how the file was opened (if it was)
1089 * @offset: start of area to check
1090 * @count: length of area to check
1091 *
1092 * Searches the inode's list of locks to find any POSIX locks which conflict.
1093 * This function is called from rw_verify_area() and
1094 * locks_verify_truncate().
1095 */
1096int locks_mandatory_area(int read_write, struct inode *inode,
1097 struct file *filp, loff_t offset,
1098 size_t count)
1099{
1100 struct file_lock fl;
1101 int error;
1102
1103 locks_init_lock(&fl);
1104 fl.fl_owner = current->files;
1105 fl.fl_pid = current->tgid;
1106 fl.fl_file = filp;
1107 fl.fl_flags = FL_POSIX | FL_ACCESS;
1108 if (filp && !(filp->f_flags & O_NONBLOCK))
1109 fl.fl_flags |= FL_SLEEP;
1110 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1111 fl.fl_start = offset;
1112 fl.fl_end = offset + count - 1;
1113
1114 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001115 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001116 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 break;
1118 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1119 if (!error) {
1120 /*
1121 * If we've been sleeping someone might have
1122 * changed the permissions behind our back.
1123 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001124 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 continue;
1126 }
1127
1128 locks_delete_block(&fl);
1129 break;
1130 }
1131
1132 return error;
1133}
1134
1135EXPORT_SYMBOL(locks_mandatory_area);
1136
J. Bruce Fields778fc542011-07-26 18:25:49 -04001137static void lease_clear_pending(struct file_lock *fl, int arg)
1138{
1139 switch (arg) {
1140 case F_UNLCK:
1141 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1142 /* fall through: */
1143 case F_RDLCK:
1144 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1145 }
1146}
1147
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148/* We already had a lease on this file; just change its type */
1149int lease_modify(struct file_lock **before, int arg)
1150{
1151 struct file_lock *fl = *before;
1152 int error = assign_type(fl, arg);
1153
1154 if (error)
1155 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001156 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 locks_wake_up_blocks(fl);
1158 if (arg == F_UNLCK)
1159 locks_delete_lock(before);
1160 return 0;
1161}
1162
1163EXPORT_SYMBOL(lease_modify);
1164
J. Bruce Fields778fc542011-07-26 18:25:49 -04001165static bool past_time(unsigned long then)
1166{
1167 if (!then)
1168 /* 0 is a special value meaning "this never expires": */
1169 return false;
1170 return time_after(jiffies, then);
1171}
1172
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173static void time_out_leases(struct inode *inode)
1174{
1175 struct file_lock **before;
1176 struct file_lock *fl;
1177
1178 before = &inode->i_flock;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04001179 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001180 if (past_time(fl->fl_downgrade_time))
1181 lease_modify(before, F_RDLCK);
1182 if (past_time(fl->fl_break_time))
1183 lease_modify(before, F_UNLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (fl == *before) /* lease_modify may have freed fl */
1185 before = &fl->fl_next;
1186 }
1187}
1188
1189/**
1190 * __break_lease - revoke all outstanding leases on file
1191 * @inode: the inode of the file to return
1192 * @mode: the open mode (read or write)
1193 *
david m. richter87250dd2007-05-09 16:10:27 -04001194 * break_lease (inlined for speed) has checked there already is at least
1195 * some kind of lock (maybe a lease) on this file. Leases are broken on
1196 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 * specified %O_NONBLOCK to your open().
1198 */
1199int __break_lease(struct inode *inode, unsigned int mode)
1200{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001201 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 struct file_lock *new_fl, *flock;
1203 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 unsigned long break_time;
1205 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001206 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
Al Viro8737c932009-12-24 06:47:55 -05001208 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001209 if (IS_ERR(new_fl))
1210 return PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001212 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
1214 time_out_leases(inode);
1215
1216 flock = inode->i_flock;
1217 if ((flock == NULL) || !IS_LEASE(flock))
1218 goto out;
1219
J. Bruce Fields778fc542011-07-26 18:25:49 -04001220 if (!locks_conflict(flock, new_fl))
1221 goto out;
1222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1224 if (fl->fl_owner == current->files)
1225 i_have_this_lease = 1;
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 break_time = 0;
1228 if (lease_break_time > 0) {
1229 break_time = jiffies + lease_break_time * HZ;
1230 if (break_time == 0)
1231 break_time++; /* so that 0 means no break time */
1232 }
1233
1234 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001235 if (want_write) {
1236 if (fl->fl_flags & FL_UNLOCK_PENDING)
1237 continue;
1238 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001240 } else {
1241 if (lease_breaking(flock))
1242 continue;
1243 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1244 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245 }
J. Bruce Fields778fc542011-07-26 18:25:49 -04001246 fl->fl_lmops->lm_break(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 }
1248
1249 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1250 error = -EWOULDBLOCK;
1251 goto out;
1252 }
1253
1254restart:
1255 break_time = flock->fl_break_time;
1256 if (break_time != 0) {
1257 break_time -= jiffies;
1258 if (break_time == 0)
1259 break_time++;
1260 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001261 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001262 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001263 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1264 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001265 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001266 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 if (error >= 0) {
1268 if (error == 0)
1269 time_out_leases(inode);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001270 /*
1271 * Wait for the next conflicting lease that has not been
1272 * broken yet
1273 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1275 flock = flock->fl_next) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001276 if (locks_conflict(new_fl, flock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 goto restart;
1278 }
1279 error = 0;
1280 }
1281
1282out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001283 unlock_flocks();
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001284 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 return error;
1286}
1287
1288EXPORT_SYMBOL(__break_lease);
1289
1290/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001291 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 * @inode: the inode
1293 * @time: pointer to a timespec which will contain the last modified time
1294 *
1295 * This is to force NFS clients to flush their caches for files with
1296 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001297 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 */
1299void lease_get_mtime(struct inode *inode, struct timespec *time)
1300{
1301 struct file_lock *flock = inode->i_flock;
1302 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1303 *time = current_fs_time(inode->i_sb);
1304 else
1305 *time = inode->i_mtime;
1306}
1307
1308EXPORT_SYMBOL(lease_get_mtime);
1309
1310/**
1311 * fcntl_getlease - Enquire what lease is currently active
1312 * @filp: the file
1313 *
1314 * The value returned by this function will be one of
1315 * (if no lease break is pending):
1316 *
1317 * %F_RDLCK to indicate a shared lease is held.
1318 *
1319 * %F_WRLCK to indicate an exclusive lease is held.
1320 *
1321 * %F_UNLCK to indicate no lease is held.
1322 *
1323 * (if a lease break is pending):
1324 *
1325 * %F_RDLCK to indicate an exclusive lease needs to be
1326 * changed to a shared lease (or removed).
1327 *
1328 * %F_UNLCK to indicate the lease needs to be removed.
1329 *
1330 * XXX: sfr & willy disagree over whether F_INPROGRESS
1331 * should be returned to userspace.
1332 */
1333int fcntl_getlease(struct file *filp)
1334{
1335 struct file_lock *fl;
1336 int type = F_UNLCK;
1337
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001338 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001339 time_out_leases(filp->f_path.dentry->d_inode);
1340 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 fl = fl->fl_next) {
1342 if (fl->fl_file == filp) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001343 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 break;
1345 }
1346 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001347 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 return type;
1349}
1350
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001351int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001353 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001354 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 struct inode *inode = dentry->d_inode;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001356 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
J. Bruce Fields096657b2010-10-30 17:31:14 -04001358 lease = *flp;
1359
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001360 error = -EAGAIN;
1361 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001362 goto out;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001363 if ((arg == F_WRLCK)
1364 && ((dentry->d_count > 1)
1365 || (atomic_read(&inode->i_count) > 1)))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001366 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001367
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 /*
1369 * At this point, we know that if there is an exclusive
1370 * lease on this file, then we hold it on this filp
1371 * (otherwise our open of this file would have blocked).
1372 * And if we are trying to acquire an exclusive lease,
1373 * then the file is not open by anyone (including us)
1374 * except for this filp.
1375 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001376 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 for (before = &inode->i_flock;
1378 ((fl = *before) != NULL) && IS_LEASE(fl);
1379 before = &fl->fl_next) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001380 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 my_before = before;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001382 continue;
1383 }
1384 /*
1385 * No exclusive leases if someone else has a lease on
1386 * this file:
1387 */
1388 if (arg == F_WRLCK)
1389 goto out;
1390 /*
1391 * Modifying our existing lease is OK, but no getting a
1392 * new lease if someone else is opening for write:
1393 */
1394 if (fl->fl_flags & FL_UNLOCK_PENDING)
1395 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 }
1397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 if (my_before != NULL) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001399 error = lease->fl_lmops->lm_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001400 if (!error)
1401 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 goto out;
1403 }
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 error = -EINVAL;
1406 if (!leases_enable)
1407 goto out;
1408
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001409 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001410 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412out:
1413 return error;
1414}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001415
1416int generic_delete_lease(struct file *filp, struct file_lock **flp)
1417{
1418 struct file_lock *fl, **before;
1419 struct dentry *dentry = filp->f_path.dentry;
1420 struct inode *inode = dentry->d_inode;
1421
1422 for (before = &inode->i_flock;
1423 ((fl = *before) != NULL) && IS_LEASE(fl);
1424 before = &fl->fl_next) {
1425 if (fl->fl_file != filp)
1426 continue;
1427 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1428 }
1429 return -EAGAIN;
1430}
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432/**
1433 * generic_setlease - sets a lease on an open file
1434 * @filp: file pointer
1435 * @arg: type of lease to obtain
1436 * @flp: input - file_lock to use, output - file_lock inserted
1437 *
1438 * The (input) flp->fl_lmops->lm_break function is required
1439 * by break_lease().
1440 *
1441 * Called with file_lock_lock held.
1442 */
1443int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1444{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 struct dentry *dentry = filp->f_path.dentry;
1446 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001447 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001449 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001450 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001452 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 error = security_file_lock(filp, arg);
1454 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001455 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456
1457 time_out_leases(inode);
1458
1459 BUG_ON(!(*flp)->fl_lmops->lm_break);
1460
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001461 switch (arg) {
1462 case F_UNLCK:
1463 return generic_delete_lease(filp, flp);
1464 case F_RDLCK:
1465 case F_WRLCK:
1466 return generic_add_lease(filp, arg, flp);
1467 default:
1468 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001471EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001473static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1474{
1475 if (filp->f_op && filp->f_op->setlease)
1476 return filp->f_op->setlease(filp, arg, lease);
1477 else
1478 return generic_setlease(filp, arg, lease);
1479}
1480
1481/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001482 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 * @filp: file pointer
1484 * @arg: type of lease to obtain
1485 * @lease: file_lock to use
1486 *
1487 * Call this to establish a lease on the file.
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001488 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001489 * break_lease will oops!
1490 *
1491 * This will call the filesystem's setlease file method, if
1492 * defined. Note that there is no getlease method; instead, the
1493 * filesystem setlease method should call back to setlease() to
1494 * add a lease to the inode's lease list, where fcntl_getlease() can
1495 * find it. Since fcntl_getlease() only reports whether the current
1496 * task holds a lease, a cluster filesystem need only do this for
1497 * leases held by processes on this node.
1498 *
1499 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001500 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001501 * filesystem's open, create, and (on truncate) setattr methods.
1502 *
1503 * Warning: the only current setlease methods exist only to disable
1504 * leases in certain cases. More vfs changes may be required to
1505 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 */
1507
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001508int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 int error;
1511
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001512 lock_flocks();
1513 error = __vfs_setlease(filp, arg, lease);
1514 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515
1516 return error;
1517}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001518EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001520static int do_fcntl_delete_lease(struct file *filp)
1521{
1522 struct file_lock fl, *flp = &fl;
1523
1524 lease_init(filp, F_UNLCK, flp);
1525
1526 return vfs_setlease(filp, F_UNLCK, &flp);
1527}
1528
1529static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001531 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001532 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 int error;
1534
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001535 fl = lease_alloc(filp, arg);
1536 if (IS_ERR(fl))
1537 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001539 new = fasync_alloc();
1540 if (!new) {
1541 locks_free_lock(fl);
1542 return -ENOMEM;
1543 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001544 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001545 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001546 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001547 if (error) {
1548 unlock_flocks();
1549 locks_free_lock(fl);
1550 goto out_free_fasync;
1551 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001552 if (ret != fl)
1553 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001555 /*
1556 * fasync_insert_entry() returns the old entry if any.
1557 * If there was no old entry, then it used 'new' and
1558 * inserted it into the fasync list. Clear new so that
1559 * we don't release it here.
1560 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001561 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001562 new = NULL;
1563
J. Bruce Fields8896b932010-11-03 18:09:18 -04001564 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001565 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001566
1567out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001568 if (new)
1569 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return error;
1571}
1572
1573/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001574 * fcntl_setlease - sets a lease on an open file
1575 * @fd: open file descriptor
1576 * @filp: file pointer
1577 * @arg: type of lease to obtain
1578 *
1579 * Call this fcntl to establish a lease on the file.
1580 * Note that you also need to call %F_SETSIG to
1581 * receive a signal when the lease is broken.
1582 */
1583int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1584{
1585 if (arg == F_UNLCK)
1586 return do_fcntl_delete_lease(filp);
1587 return do_fcntl_add_lease(fd, filp, arg);
1588}
1589
1590/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1592 * @filp: The file to apply the lock to
1593 * @fl: The lock to be applied
1594 *
1595 * Add a FLOCK style lock to a file.
1596 */
1597int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1598{
1599 int error;
1600 might_sleep();
1601 for (;;) {
1602 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001603 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 break;
1605 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1606 if (!error)
1607 continue;
1608
1609 locks_delete_block(fl);
1610 break;
1611 }
1612 return error;
1613}
1614
1615EXPORT_SYMBOL(flock_lock_file_wait);
1616
1617/**
1618 * sys_flock: - flock() system call.
1619 * @fd: the file descriptor to lock.
1620 * @cmd: the type of lock to apply.
1621 *
1622 * Apply a %FL_FLOCK style lock to an open file descriptor.
1623 * The @cmd can be one of
1624 *
1625 * %LOCK_SH -- a shared lock.
1626 *
1627 * %LOCK_EX -- an exclusive lock.
1628 *
1629 * %LOCK_UN -- remove an existing lock.
1630 *
1631 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1632 *
1633 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1634 * processes read and write access respectively.
1635 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001636SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
1638 struct file *filp;
1639 struct file_lock *lock;
1640 int can_sleep, unlock;
1641 int error;
1642
1643 error = -EBADF;
1644 filp = fget(fd);
1645 if (!filp)
1646 goto out;
1647
1648 can_sleep = !(cmd & LOCK_NB);
1649 cmd &= ~LOCK_NB;
1650 unlock = (cmd == LOCK_UN);
1651
Al Viroaeb5d722008-09-02 15:28:45 -04001652 if (!unlock && !(cmd & LOCK_MAND) &&
1653 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 goto out_putf;
1655
1656 error = flock_make_lock(filp, &lock, cmd);
1657 if (error)
1658 goto out_putf;
1659 if (can_sleep)
1660 lock->fl_flags |= FL_SLEEP;
1661
Sten Spans713c0ec2009-07-16 09:41:39 +02001662 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (error)
1664 goto out_free;
1665
1666 if (filp->f_op && filp->f_op->flock)
1667 error = filp->f_op->flock(filp,
1668 (can_sleep) ? F_SETLKW : F_SETLK,
1669 lock);
1670 else
1671 error = flock_lock_file_wait(filp, lock);
1672
1673 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001674 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675
1676 out_putf:
1677 fput(filp);
1678 out:
1679 return error;
1680}
1681
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001682/**
1683 * vfs_test_lock - test file byte range lock
1684 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001685 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001686 *
1687 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1688 * setting conf->fl_type to something other than F_UNLCK.
1689 */
1690int vfs_test_lock(struct file *filp, struct file_lock *fl)
1691{
1692 if (filp->f_op && filp->f_op->lock)
1693 return filp->f_op->lock(filp, F_GETLK, fl);
1694 posix_test_lock(filp, fl);
1695 return 0;
1696}
1697EXPORT_SYMBOL_GPL(vfs_test_lock);
1698
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001699static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1700{
1701 flock->l_pid = fl->fl_pid;
1702#if BITS_PER_LONG == 32
1703 /*
1704 * Make sure we can represent the posix lock via
1705 * legacy 32bit flock.
1706 */
1707 if (fl->fl_start > OFFT_OFFSET_MAX)
1708 return -EOVERFLOW;
1709 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1710 return -EOVERFLOW;
1711#endif
1712 flock->l_start = fl->fl_start;
1713 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1714 fl->fl_end - fl->fl_start + 1;
1715 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001716 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001717 return 0;
1718}
1719
1720#if BITS_PER_LONG == 32
1721static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1722{
1723 flock->l_pid = fl->fl_pid;
1724 flock->l_start = fl->fl_start;
1725 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1726 fl->fl_end - fl->fl_start + 1;
1727 flock->l_whence = 0;
1728 flock->l_type = fl->fl_type;
1729}
1730#endif
1731
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732/* Report the first existing lock that would conflict with l.
1733 * This implements the F_GETLK command of fcntl().
1734 */
1735int fcntl_getlk(struct file *filp, struct flock __user *l)
1736{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001737 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738 struct flock flock;
1739 int error;
1740
1741 error = -EFAULT;
1742 if (copy_from_user(&flock, l, sizeof(flock)))
1743 goto out;
1744 error = -EINVAL;
1745 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1746 goto out;
1747
1748 error = flock_to_posix_lock(filp, &file_lock, &flock);
1749 if (error)
1750 goto out;
1751
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001752 error = vfs_test_lock(filp, &file_lock);
1753 if (error)
1754 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001756 flock.l_type = file_lock.fl_type;
1757 if (file_lock.fl_type != F_UNLCK) {
1758 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001759 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 }
1762 error = -EFAULT;
1763 if (!copy_to_user(l, &flock, sizeof(flock)))
1764 error = 0;
1765out:
1766 return error;
1767}
1768
Marc Eshel7723ec92007-01-18 15:08:55 -05001769/**
1770 * vfs_lock_file - file byte range lock
1771 * @filp: The file to apply the lock to
1772 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1773 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001774 * @conf: Place to return a copy of the conflicting lock, if found.
1775 *
1776 * A caller that doesn't care about the conflicting lock may pass NULL
1777 * as the final argument.
1778 *
1779 * If the filesystem defines a private ->lock() method, then @conf will
1780 * be left unchanged; so a caller that cares should initialize it to
1781 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001782 *
1783 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1784 * locks, the ->lock() interface may return asynchronously, before the lock has
1785 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001786 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05001787 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1788 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001789 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001790 * request completes.
1791 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001792 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1793 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001794 * nonzero return code and the file system should release the lock. The file
1795 * system is also responsible to keep a corresponding posix lock when it
1796 * grants a lock so the VFS can find out which locks are locally held and do
1797 * the correct lock cleanup when required.
1798 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001799 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001800 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001801 */
Marc Eshel150b3932007-01-18 16:15:35 -05001802int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001803{
1804 if (filp->f_op && filp->f_op->lock)
1805 return filp->f_op->lock(filp, cmd, fl);
1806 else
Marc Eshel150b3932007-01-18 16:15:35 -05001807 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001808}
1809EXPORT_SYMBOL_GPL(vfs_lock_file);
1810
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001811static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1812 struct file_lock *fl)
1813{
1814 int error;
1815
1816 error = security_file_lock(filp, fl->fl_type);
1817 if (error)
1818 return error;
1819
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001820 for (;;) {
1821 error = vfs_lock_file(filp, cmd, fl, NULL);
1822 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001823 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001824 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1825 if (!error)
1826 continue;
1827
1828 locks_delete_block(fl);
1829 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001830 }
1831
1832 return error;
1833}
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835/* Apply the lock described by l to an open file descriptor.
1836 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1837 */
Peter Staubachc2936212005-07-27 11:45:09 -07001838int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1839 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840{
1841 struct file_lock *file_lock = locks_alloc_lock();
1842 struct flock flock;
1843 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001844 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 int error;
1846
1847 if (file_lock == NULL)
1848 return -ENOLCK;
1849
1850 /*
1851 * This might block, so we do it before checking the inode.
1852 */
1853 error = -EFAULT;
1854 if (copy_from_user(&flock, l, sizeof(flock)))
1855 goto out;
1856
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001857 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 /* Don't allow mandatory locks on files that may be memory mapped
1860 * and shared.
1861 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001862 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863 error = -EAGAIN;
1864 goto out;
1865 }
1866
Peter Staubachc2936212005-07-27 11:45:09 -07001867again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 error = flock_to_posix_lock(filp, file_lock, &flock);
1869 if (error)
1870 goto out;
1871 if (cmd == F_SETLKW) {
1872 file_lock->fl_flags |= FL_SLEEP;
1873 }
1874
1875 error = -EBADF;
1876 switch (flock.l_type) {
1877 case F_RDLCK:
1878 if (!(filp->f_mode & FMODE_READ))
1879 goto out;
1880 break;
1881 case F_WRLCK:
1882 if (!(filp->f_mode & FMODE_WRITE))
1883 goto out;
1884 break;
1885 case F_UNLCK:
1886 break;
1887 default:
1888 error = -EINVAL;
1889 goto out;
1890 }
1891
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001892 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893
Peter Staubachc2936212005-07-27 11:45:09 -07001894 /*
1895 * Attempt to detect a close/fcntl race and recover by
1896 * releasing the lock that was just acquired.
1897 */
Al Viro0b2bac22008-05-06 13:58:34 -04001898 /*
1899 * we need that spin_lock here - it prevents reordering between
1900 * update of inode->i_flock and check for it done in close().
1901 * rcu_read_lock() wouldn't do.
1902 */
1903 spin_lock(&current->files->file_lock);
1904 f = fcheck(fd);
1905 spin_unlock(&current->files->file_lock);
1906 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001907 flock.l_type = F_UNLCK;
1908 goto again;
1909 }
1910
1911out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 locks_free_lock(file_lock);
1913 return error;
1914}
1915
1916#if BITS_PER_LONG == 32
1917/* Report the first existing lock that would conflict with l.
1918 * This implements the F_GETLK command of fcntl().
1919 */
1920int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1921{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001922 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 struct flock64 flock;
1924 int error;
1925
1926 error = -EFAULT;
1927 if (copy_from_user(&flock, l, sizeof(flock)))
1928 goto out;
1929 error = -EINVAL;
1930 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1931 goto out;
1932
1933 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1934 if (error)
1935 goto out;
1936
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001937 error = vfs_test_lock(filp, &file_lock);
1938 if (error)
1939 goto out;
1940
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001941 flock.l_type = file_lock.fl_type;
1942 if (file_lock.fl_type != F_UNLCK)
1943 posix_lock_to_flock64(&flock, &file_lock);
1944
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 error = -EFAULT;
1946 if (!copy_to_user(l, &flock, sizeof(flock)))
1947 error = 0;
1948
1949out:
1950 return error;
1951}
1952
1953/* Apply the lock described by l to an open file descriptor.
1954 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1955 */
Peter Staubachc2936212005-07-27 11:45:09 -07001956int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1957 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958{
1959 struct file_lock *file_lock = locks_alloc_lock();
1960 struct flock64 flock;
1961 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001962 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 int error;
1964
1965 if (file_lock == NULL)
1966 return -ENOLCK;
1967
1968 /*
1969 * This might block, so we do it before checking the inode.
1970 */
1971 error = -EFAULT;
1972 if (copy_from_user(&flock, l, sizeof(flock)))
1973 goto out;
1974
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001975 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
1977 /* Don't allow mandatory locks on files that may be memory mapped
1978 * and shared.
1979 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001980 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 error = -EAGAIN;
1982 goto out;
1983 }
1984
Peter Staubachc2936212005-07-27 11:45:09 -07001985again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 error = flock64_to_posix_lock(filp, file_lock, &flock);
1987 if (error)
1988 goto out;
1989 if (cmd == F_SETLKW64) {
1990 file_lock->fl_flags |= FL_SLEEP;
1991 }
1992
1993 error = -EBADF;
1994 switch (flock.l_type) {
1995 case F_RDLCK:
1996 if (!(filp->f_mode & FMODE_READ))
1997 goto out;
1998 break;
1999 case F_WRLCK:
2000 if (!(filp->f_mode & FMODE_WRITE))
2001 goto out;
2002 break;
2003 case F_UNLCK:
2004 break;
2005 default:
2006 error = -EINVAL;
2007 goto out;
2008 }
2009
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002010 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Peter Staubachc2936212005-07-27 11:45:09 -07002012 /*
2013 * Attempt to detect a close/fcntl race and recover by
2014 * releasing the lock that was just acquired.
2015 */
Al Viro0b2bac22008-05-06 13:58:34 -04002016 spin_lock(&current->files->file_lock);
2017 f = fcheck(fd);
2018 spin_unlock(&current->files->file_lock);
2019 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002020 flock.l_type = F_UNLCK;
2021 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 }
2023
2024out:
2025 locks_free_lock(file_lock);
2026 return error;
2027}
2028#endif /* BITS_PER_LONG == 32 */
2029
2030/*
2031 * This function is called when the file is being removed
2032 * from the task's fd array. POSIX locks belonging to this task
2033 * are deleted at this time.
2034 */
2035void locks_remove_posix(struct file *filp, fl_owner_t owner)
2036{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002037 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038
2039 /*
2040 * If there are no locks held on this file, we don't need to call
2041 * posix_lock_file(). Another process could be setting a lock on this
2042 * file at the same time, but we wouldn't remove that lock anyway.
2043 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002044 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 return;
2046
2047 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002048 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 lock.fl_start = 0;
2050 lock.fl_end = OFFSET_MAX;
2051 lock.fl_owner = owner;
2052 lock.fl_pid = current->tgid;
2053 lock.fl_file = filp;
2054 lock.fl_ops = NULL;
2055 lock.fl_lmops = NULL;
2056
Marc Eshel150b3932007-01-18 16:15:35 -05002057 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2060 lock.fl_ops->fl_release_private(&lock);
2061}
2062
2063EXPORT_SYMBOL(locks_remove_posix);
2064
2065/*
2066 * This function is called on the last close of an open file.
2067 */
2068void locks_remove_flock(struct file *filp)
2069{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002070 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 struct file_lock *fl;
2072 struct file_lock **before;
2073
2074 if (!inode->i_flock)
2075 return;
2076
2077 if (filp->f_op && filp->f_op->flock) {
2078 struct file_lock fl = {
2079 .fl_pid = current->tgid,
2080 .fl_file = filp,
2081 .fl_flags = FL_FLOCK,
2082 .fl_type = F_UNLCK,
2083 .fl_end = OFFSET_MAX,
2084 };
2085 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002086 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2087 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 }
2089
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002090 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 before = &inode->i_flock;
2092
2093 while ((fl = *before) != NULL) {
2094 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002095 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 locks_delete_lock(before);
2097 continue;
2098 }
2099 if (IS_LEASE(fl)) {
2100 lease_modify(before, F_UNLCK);
2101 continue;
2102 }
2103 /* What? */
2104 BUG();
2105 }
2106 before = &fl->fl_next;
2107 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002108 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109}
2110
2111/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 * posix_unblock_lock - stop waiting for a file lock
2113 * @filp: how the file was opened
2114 * @waiter: the lock which was waiting
2115 *
2116 * lockd needs to block waiting for locks.
2117 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002118int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2120{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002121 int status = 0;
2122
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002123 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002124 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002126 else
2127 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002128 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002129 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130}
2131
2132EXPORT_SYMBOL(posix_unblock_lock);
2133
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002134/**
2135 * vfs_cancel_lock - file byte range unblock lock
2136 * @filp: The file to apply the unblock to
2137 * @fl: The lock to be unblocked
2138 *
2139 * Used by lock managers to cancel blocked requests
2140 */
2141int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2142{
2143 if (filp->f_op && filp->f_op->lock)
2144 return filp->f_op->lock(filp, F_CANCELLK, fl);
2145 return 0;
2146}
2147
2148EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2149
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002150#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002151#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002152#include <linux/seq_file.h>
2153
2154static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002155 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156{
2157 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002158 unsigned int fl_pid;
2159
2160 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002161 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002162 else
2163 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
2165 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002166 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167
Jerome Marchand99dc8292010-10-26 14:22:33 -07002168 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002170 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2172 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002173 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 } else if (IS_FLOCK(fl)) {
2175 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002176 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002178 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 }
2180 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002181 seq_printf(f, "LEASE ");
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002182 if (lease_breaking(fl))
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002183 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002185 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002187 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002189 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 }
2191 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002192 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 (fl->fl_type & LOCK_READ)
2194 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2195 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2196 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002197 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002198 (lease_breaking(fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2200 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2201 }
2202 if (inode) {
2203#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002204 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 inode->i_sb->s_id, inode->i_ino);
2206#else
2207 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002208 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 MAJOR(inode->i_sb->s_dev),
2210 MINOR(inode->i_sb->s_dev), inode->i_ino);
2211#endif
2212 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002213 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 }
2215 if (IS_POSIX(fl)) {
2216 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002217 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002219 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002220 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002221 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 }
2223}
2224
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002225static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002227 struct file_lock *fl, *bfl;
2228
2229 fl = list_entry(v, struct file_lock, fl_link);
2230
Jerome Marchand99dc8292010-10-26 14:22:33 -07002231 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002232
2233 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002234 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002235
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002236 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237}
2238
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002239static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002241 loff_t *p = f->private;
2242
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002243 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002244 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002245 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246}
2247
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002248static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2249{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002250 loff_t *p = f->private;
2251 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002252 return seq_list_next(v, &file_lock_list, pos);
2253}
2254
2255static void locks_stop(struct seq_file *f, void *v)
2256{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002257 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002258}
2259
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002260static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002261 .start = locks_start,
2262 .next = locks_next,
2263 .stop = locks_stop,
2264 .show = locks_show,
2265};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002266
2267static int locks_open(struct inode *inode, struct file *filp)
2268{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002269 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002270}
2271
2272static const struct file_operations proc_locks_operations = {
2273 .open = locks_open,
2274 .read = seq_read,
2275 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002276 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002277};
2278
2279static int __init proc_locks_init(void)
2280{
2281 proc_create("locks", 0, NULL, &proc_locks_operations);
2282 return 0;
2283}
2284module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002285#endif
2286
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287/**
2288 * lock_may_read - checks that the region is free of locks
2289 * @inode: the inode that is being read
2290 * @start: the first byte to read
2291 * @len: the number of bytes to read
2292 *
2293 * Emulates Windows locking requirements. Whole-file
2294 * mandatory locks (share modes) can prohibit a read and
2295 * byte-range POSIX locks can prohibit a read if they overlap.
2296 *
2297 * N.B. this function is only ever called
2298 * from knfsd and ownership of locks is never checked.
2299 */
2300int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2301{
2302 struct file_lock *fl;
2303 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002304 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2306 if (IS_POSIX(fl)) {
2307 if (fl->fl_type == F_RDLCK)
2308 continue;
2309 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2310 continue;
2311 } else if (IS_FLOCK(fl)) {
2312 if (!(fl->fl_type & LOCK_MAND))
2313 continue;
2314 if (fl->fl_type & LOCK_READ)
2315 continue;
2316 } else
2317 continue;
2318 result = 0;
2319 break;
2320 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002321 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 return result;
2323}
2324
2325EXPORT_SYMBOL(lock_may_read);
2326
2327/**
2328 * lock_may_write - checks that the region is free of locks
2329 * @inode: the inode that is being written
2330 * @start: the first byte to write
2331 * @len: the number of bytes to write
2332 *
2333 * Emulates Windows locking requirements. Whole-file
2334 * mandatory locks (share modes) can prohibit a write and
2335 * byte-range POSIX locks can prohibit a write if they overlap.
2336 *
2337 * N.B. this function is only ever called
2338 * from knfsd and ownership of locks is never checked.
2339 */
2340int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2341{
2342 struct file_lock *fl;
2343 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002344 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2346 if (IS_POSIX(fl)) {
2347 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2348 continue;
2349 } else if (IS_FLOCK(fl)) {
2350 if (!(fl->fl_type & LOCK_MAND))
2351 continue;
2352 if (fl->fl_type & LOCK_WRITE)
2353 continue;
2354 } else
2355 continue;
2356 result = 0;
2357 break;
2358 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002359 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 return result;
2361}
2362
2363EXPORT_SYMBOL(lock_may_write);
2364
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365static int __init filelock_init(void)
2366{
2367 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002368 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2369
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 return 0;
2371}
2372
2373core_initcall(filelock_init);