blob: 3e946cda98c65d87f8b1df46cf154b7f1c7a4b64 [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 */
513static void locks_delete_block(struct file_lock *waiter)
514{
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}
519
520/* Insert waiter into blocker's block list.
521 * We use a circular list so that processes can be easily woken up in
522 * the order they blocked. The documentation doesn't require this but
523 * it seems like the reasonable thing to do.
524 */
525static void locks_insert_block(struct file_lock *blocker,
526 struct file_lock *waiter)
527{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800528 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 list_add_tail(&waiter->fl_block, &blocker->fl_block);
530 waiter->fl_next = blocker;
531 if (IS_POSIX(blocker))
532 list_add(&waiter->fl_link, &blocked_list);
533}
534
535/* Wake up processes blocked waiting for blocker.
536 * If told to wait then schedule the processes until the block list
537 * is empty, otherwise empty the block list ourselves.
538 */
539static void locks_wake_up_blocks(struct file_lock *blocker)
540{
541 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400542 struct file_lock *waiter;
543
544 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 struct file_lock, fl_block);
546 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400547 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
548 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 else
550 wake_up(&waiter->fl_wait);
551 }
552}
553
554/* Insert file lock fl into an inode's lock list at the position indicated
555 * by pos. At the same time add the lock to the global file lock list.
556 */
557static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
558{
559 list_add(&fl->fl_link, &file_lock_list);
560
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000561 fl->fl_nspid = get_pid(task_tgid(current));
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 /* insert into file's list */
564 fl->fl_next = *pos;
565 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566}
567
568/*
569 * Delete a lock and then free it.
570 * Wake up processes that are blocked waiting for this lock,
571 * notify the FS that the lock has been cleared and
572 * finally free the lock.
573 */
574static void locks_delete_lock(struct file_lock **thisfl_p)
575{
576 struct file_lock *fl = *thisfl_p;
577
578 *thisfl_p = fl->fl_next;
579 fl->fl_next = NULL;
580 list_del_init(&fl->fl_link);
581
582 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
583 if (fl->fl_fasync != NULL) {
584 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
585 fl->fl_fasync = NULL;
586 }
587
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000588 if (fl->fl_nspid) {
589 put_pid(fl->fl_nspid);
590 fl->fl_nspid = NULL;
591 }
592
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 locks_wake_up_blocks(fl);
594 locks_free_lock(fl);
595}
596
597/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
598 * checks for shared/exclusive status of overlapping locks.
599 */
600static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
601{
602 if (sys_fl->fl_type == F_WRLCK)
603 return 1;
604 if (caller_fl->fl_type == F_WRLCK)
605 return 1;
606 return 0;
607}
608
609/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
610 * checking before calling the locks_conflict().
611 */
612static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
613{
614 /* POSIX locks owned by the same process do not conflict with
615 * each other.
616 */
617 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
618 return (0);
619
620 /* Check whether they overlap */
621 if (!locks_overlap(caller_fl, sys_fl))
622 return 0;
623
624 return (locks_conflict(caller_fl, sys_fl));
625}
626
627/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
628 * checking before calling the locks_conflict().
629 */
630static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
631{
632 /* FLOCK locks referring to the same filp do not conflict with
633 * each other.
634 */
635 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
636 return (0);
637 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
638 return 0;
639
640 return (locks_conflict(caller_fl, sys_fl));
641}
642
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400643void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500644posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
646 struct file_lock *cfl;
647
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200648 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800649 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 if (!IS_POSIX(cfl))
651 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400652 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 break;
654 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000655 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500656 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000657 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800658 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000659 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400660 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200661 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400662 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664EXPORT_SYMBOL(posix_test_lock);
665
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400666/*
667 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400669 * We attempt to detect deadlocks that are due purely to posix file
670 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400672 * We assume that a task can be waiting for at most one lock at a time.
673 * So for any acquired lock, the process holding that lock may be
674 * waiting on at most one other lock. That lock in turns may be held by
675 * someone waiting for at most one other lock. Given a requested lock
676 * caller_fl which is about to wait for a conflicting lock block_fl, we
677 * follow this chain of waiters to ensure we are not about to create a
678 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400679 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400680 * Since we do this before we ever put a process to sleep on a lock, we
681 * are ensured that there is never a cycle; that is what guarantees that
682 * the while() loop in posix_locks_deadlock() eventually completes.
683 *
684 * Note: the above assumption may not be true when handling lock
685 * requests from a broken NFS client. It may also fail in the presence
686 * of tasks (such as posix threads) sharing the same open file table.
687 *
688 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400690
691#define MAX_DEADLK_ITERATIONS 10
692
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400693/* Find a lock that the owner of the given block_fl is blocking on. */
694static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
695{
696 struct file_lock *fl;
697
698 list_for_each_entry(fl, &blocked_list, fl_link) {
699 if (posix_same_owner(fl, block_fl))
700 return fl->fl_next;
701 }
702 return NULL;
703}
704
Adrian Bunkb0904e12006-06-23 02:05:13 -0700705static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 struct file_lock *block_fl)
707{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400708 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400710 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
711 if (i++ > MAX_DEADLK_ITERATIONS)
712 return 0;
713 if (posix_same_owner(caller_fl, block_fl))
714 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 }
716 return 0;
717}
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400720 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400721 *
722 * Note that if called with an FL_EXISTS argument, the caller may determine
723 * whether or not a lock was successfully freed by testing the return
724 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800726static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800728 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800730 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 int error = 0;
732 int found = 0;
733
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200734 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
735 new_fl = locks_alloc_lock();
736 if (!new_fl)
737 return -ENOMEM;
738 }
739
740 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400741 if (request->fl_flags & FL_ACCESS)
742 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400743
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 for_each_lock(inode, before) {
745 struct file_lock *fl = *before;
746 if (IS_POSIX(fl))
747 break;
748 if (IS_LEASE(fl))
749 continue;
750 if (filp != fl->fl_file)
751 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800752 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 goto out;
754 found = 1;
755 locks_delete_lock(before);
756 break;
757 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Trond Myklebustf475ae92006-06-29 16:38:32 -0400759 if (request->fl_type == F_UNLCK) {
760 if ((request->fl_flags & FL_EXISTS) && !found)
761 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800762 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400763 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 /*
766 * If a higher-priority process was blocked on the old file lock,
767 * give it the opportunity to lock the file.
768 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200769 if (found) {
770 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200771 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200772 lock_flocks();
773 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400775find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 for_each_lock(inode, before) {
777 struct file_lock *fl = *before;
778 if (IS_POSIX(fl))
779 break;
780 if (IS_LEASE(fl))
781 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800782 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 continue;
784 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700785 if (!(request->fl_flags & FL_SLEEP))
786 goto out;
787 error = FILE_LOCK_DEFERRED;
788 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 goto out;
790 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400791 if (request->fl_flags & FL_ACCESS)
792 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800793 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700794 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800795 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400796 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200799 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800800 if (new_fl)
801 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 return error;
803}
804
Marc Eshel150b3932007-01-18 16:15:35 -0500805static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806{
807 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700808 struct file_lock *new_fl = NULL;
809 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 struct file_lock *left = NULL;
811 struct file_lock *right = NULL;
812 struct file_lock **before;
813 int error, added = 0;
814
815 /*
816 * We may need two file_lock structures for this operation,
817 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700818 *
819 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700821 if (!(request->fl_flags & FL_ACCESS) &&
822 (request->fl_type != F_UNLCK ||
823 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
824 new_fl = locks_alloc_lock();
825 new_fl2 = locks_alloc_lock();
826 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200828 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 if (request->fl_type != F_UNLCK) {
830 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500831 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 if (!IS_POSIX(fl))
833 continue;
834 if (!posix_locks_conflict(request, fl))
835 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800836 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400837 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 error = -EAGAIN;
839 if (!(request->fl_flags & FL_SLEEP))
840 goto out;
841 error = -EDEADLK;
842 if (posix_locks_deadlock(request, fl))
843 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700844 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 locks_insert_block(fl, request);
846 goto out;
847 }
848 }
849
850 /* If we're just looking for a conflict, we're done. */
851 error = 0;
852 if (request->fl_flags & FL_ACCESS)
853 goto out;
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 * Find the first old lock with the same owner as the new lock.
857 */
858
859 before = &inode->i_flock;
860
861 /* First skip locks owned by other processes. */
862 while ((fl = *before) && (!IS_POSIX(fl) ||
863 !posix_same_owner(request, fl))) {
864 before = &fl->fl_next;
865 }
866
867 /* Process locks with this owner. */
868 while ((fl = *before) && posix_same_owner(request, fl)) {
869 /* Detect adjacent or overlapping regions (if same lock type)
870 */
871 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700872 /* In all comparisons of start vs end, use
873 * "start - 1" rather than "end + 1". If end
874 * is OFFSET_MAX, end + 1 will become negative.
875 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 if (fl->fl_end < request->fl_start - 1)
877 goto next_lock;
878 /* If the next lock in the list has entirely bigger
879 * addresses than the new one, insert the lock here.
880 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700881 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 break;
883
884 /* If we come here, the new and old lock are of the
885 * same type and adjacent or overlapping. Make one
886 * lock yielding from the lower start address of both
887 * locks to the higher end address.
888 */
889 if (fl->fl_start > request->fl_start)
890 fl->fl_start = request->fl_start;
891 else
892 request->fl_start = fl->fl_start;
893 if (fl->fl_end < request->fl_end)
894 fl->fl_end = request->fl_end;
895 else
896 request->fl_end = fl->fl_end;
897 if (added) {
898 locks_delete_lock(before);
899 continue;
900 }
901 request = fl;
902 added = 1;
903 }
904 else {
905 /* Processing for different lock types is a bit
906 * more complex.
907 */
908 if (fl->fl_end < request->fl_start)
909 goto next_lock;
910 if (fl->fl_start > request->fl_end)
911 break;
912 if (request->fl_type == F_UNLCK)
913 added = 1;
914 if (fl->fl_start < request->fl_start)
915 left = fl;
916 /* If the next lock in the list has a higher end
917 * address than the new one, insert the new one here.
918 */
919 if (fl->fl_end > request->fl_end) {
920 right = fl;
921 break;
922 }
923 if (fl->fl_start >= request->fl_start) {
924 /* The new lock completely replaces an old
925 * one (This may happen several times).
926 */
927 if (added) {
928 locks_delete_lock(before);
929 continue;
930 }
931 /* Replace the old lock with the new one.
932 * Wake up anybody waiting for the old one,
933 * as the change in lock type might satisfy
934 * their needs.
935 */
936 locks_wake_up_blocks(fl);
937 fl->fl_start = request->fl_start;
938 fl->fl_end = request->fl_end;
939 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500940 locks_release_private(fl);
941 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 request = fl;
943 added = 1;
944 }
945 }
946 /* Go on to next lock.
947 */
948 next_lock:
949 before = &fl->fl_next;
950 }
951
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700952 /*
953 * The above code only modifies existing locks in case of
954 * merging or replacing. If new lock(s) need to be inserted
955 * all modifications are done bellow this, so it's safe yet to
956 * bail out.
957 */
958 error = -ENOLCK; /* "no luck" */
959 if (right && left == right && !new_fl2)
960 goto out;
961
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 error = 0;
963 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400964 if (request->fl_type == F_UNLCK) {
965 if (request->fl_flags & FL_EXISTS)
966 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400968 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700969
970 if (!new_fl) {
971 error = -ENOLCK;
972 goto out;
973 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 locks_copy_lock(new_fl, request);
975 locks_insert_lock(before, new_fl);
976 new_fl = NULL;
977 }
978 if (right) {
979 if (left == right) {
980 /* The new lock breaks the old one in two pieces,
981 * so we have to use the second new lock.
982 */
983 left = new_fl2;
984 new_fl2 = NULL;
985 locks_copy_lock(left, right);
986 locks_insert_lock(before, left);
987 }
988 right->fl_start = request->fl_end + 1;
989 locks_wake_up_blocks(right);
990 }
991 if (left) {
992 left->fl_end = request->fl_start - 1;
993 locks_wake_up_blocks(left);
994 }
995 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200996 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 /*
998 * Free any unused locks.
999 */
1000 if (new_fl)
1001 locks_free_lock(new_fl);
1002 if (new_fl2)
1003 locks_free_lock(new_fl2);
1004 return error;
1005}
1006
1007/**
1008 * posix_lock_file - Apply a POSIX-style lock to a file
1009 * @filp: The file to apply the lock to
1010 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001011 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 *
1013 * Add a POSIX style lock to a file.
1014 * We merge adjacent & overlapping locks whenever possible.
1015 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001016 *
1017 * Note that if called with an FL_EXISTS argument, the caller may determine
1018 * whether or not a lock was successfully freed by testing the return
1019 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 */
Marc Eshel150b3932007-01-18 16:15:35 -05001021int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001022 struct file_lock *conflock)
1023{
Marc Eshel150b3932007-01-18 16:15:35 -05001024 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001025}
Marc Eshel150b3932007-01-18 16:15:35 -05001026EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
1028/**
1029 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1030 * @filp: The file to apply the lock to
1031 * @fl: The lock to be applied
1032 *
1033 * Add a POSIX style lock to a file.
1034 * We merge adjacent & overlapping locks whenever possible.
1035 * POSIX locks are sorted by owner task, then by starting address
1036 */
1037int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1038{
1039 int error;
1040 might_sleep ();
1041 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001042 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001043 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 break;
1045 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1046 if (!error)
1047 continue;
1048
1049 locks_delete_block(fl);
1050 break;
1051 }
1052 return error;
1053}
1054EXPORT_SYMBOL(posix_lock_file_wait);
1055
1056/**
1057 * locks_mandatory_locked - Check for an active lock
1058 * @inode: the file to check
1059 *
1060 * Searches the inode's list of locks to find any POSIX locks which conflict.
1061 * This function is called from locks_verify_locked() only.
1062 */
1063int locks_mandatory_locked(struct inode *inode)
1064{
1065 fl_owner_t owner = current->files;
1066 struct file_lock *fl;
1067
1068 /*
1069 * Search the lock list for this inode for any POSIX locks.
1070 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001071 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1073 if (!IS_POSIX(fl))
1074 continue;
1075 if (fl->fl_owner != owner)
1076 break;
1077 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001078 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 return fl ? -EAGAIN : 0;
1080}
1081
1082/**
1083 * locks_mandatory_area - Check for a conflicting lock
1084 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1085 * for shared
1086 * @inode: the file to check
1087 * @filp: how the file was opened (if it was)
1088 * @offset: start of area to check
1089 * @count: length of area to check
1090 *
1091 * Searches the inode's list of locks to find any POSIX locks which conflict.
1092 * This function is called from rw_verify_area() and
1093 * locks_verify_truncate().
1094 */
1095int locks_mandatory_area(int read_write, struct inode *inode,
1096 struct file *filp, loff_t offset,
1097 size_t count)
1098{
1099 struct file_lock fl;
1100 int error;
1101
1102 locks_init_lock(&fl);
1103 fl.fl_owner = current->files;
1104 fl.fl_pid = current->tgid;
1105 fl.fl_file = filp;
1106 fl.fl_flags = FL_POSIX | FL_ACCESS;
1107 if (filp && !(filp->f_flags & O_NONBLOCK))
1108 fl.fl_flags |= FL_SLEEP;
1109 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1110 fl.fl_start = offset;
1111 fl.fl_end = offset + count - 1;
1112
1113 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001114 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001115 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 break;
1117 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1118 if (!error) {
1119 /*
1120 * If we've been sleeping someone might have
1121 * changed the permissions behind our back.
1122 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001123 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 continue;
1125 }
1126
1127 locks_delete_block(&fl);
1128 break;
1129 }
1130
1131 return error;
1132}
1133
1134EXPORT_SYMBOL(locks_mandatory_area);
1135
J. Bruce Fields778fc542011-07-26 18:25:49 -04001136static void lease_clear_pending(struct file_lock *fl, int arg)
1137{
1138 switch (arg) {
1139 case F_UNLCK:
1140 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1141 /* fall through: */
1142 case F_RDLCK:
1143 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1144 }
1145}
1146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147/* We already had a lease on this file; just change its type */
1148int lease_modify(struct file_lock **before, int arg)
1149{
1150 struct file_lock *fl = *before;
1151 int error = assign_type(fl, arg);
1152
1153 if (error)
1154 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001155 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 locks_wake_up_blocks(fl);
1157 if (arg == F_UNLCK)
1158 locks_delete_lock(before);
1159 return 0;
1160}
1161
1162EXPORT_SYMBOL(lease_modify);
1163
J. Bruce Fields778fc542011-07-26 18:25:49 -04001164static bool past_time(unsigned long then)
1165{
1166 if (!then)
1167 /* 0 is a special value meaning "this never expires": */
1168 return false;
1169 return time_after(jiffies, then);
1170}
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172static void time_out_leases(struct inode *inode)
1173{
1174 struct file_lock **before;
1175 struct file_lock *fl;
1176
1177 before = &inode->i_flock;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04001178 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001179 if (past_time(fl->fl_downgrade_time))
1180 lease_modify(before, F_RDLCK);
1181 if (past_time(fl->fl_break_time))
1182 lease_modify(before, F_UNLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 if (fl == *before) /* lease_modify may have freed fl */
1184 before = &fl->fl_next;
1185 }
1186}
1187
1188/**
1189 * __break_lease - revoke all outstanding leases on file
1190 * @inode: the inode of the file to return
1191 * @mode: the open mode (read or write)
1192 *
david m. richter87250dd2007-05-09 16:10:27 -04001193 * break_lease (inlined for speed) has checked there already is at least
1194 * some kind of lock (maybe a lease) on this file. Leases are broken on
1195 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 * specified %O_NONBLOCK to your open().
1197 */
1198int __break_lease(struct inode *inode, unsigned int mode)
1199{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001200 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 struct file_lock *new_fl, *flock;
1202 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 unsigned long break_time;
1204 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001205 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
Al Viro8737c932009-12-24 06:47:55 -05001207 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001208 if (IS_ERR(new_fl))
1209 return PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001211 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213 time_out_leases(inode);
1214
1215 flock = inode->i_flock;
1216 if ((flock == NULL) || !IS_LEASE(flock))
1217 goto out;
1218
J. Bruce Fields778fc542011-07-26 18:25:49 -04001219 if (!locks_conflict(flock, new_fl))
1220 goto out;
1221
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1223 if (fl->fl_owner == current->files)
1224 i_have_this_lease = 1;
1225
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 break_time = 0;
1227 if (lease_break_time > 0) {
1228 break_time = jiffies + lease_break_time * HZ;
1229 if (break_time == 0)
1230 break_time++; /* so that 0 means no break time */
1231 }
1232
1233 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001234 if (want_write) {
1235 if (fl->fl_flags & FL_UNLOCK_PENDING)
1236 continue;
1237 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001239 } else {
1240 if (lease_breaking(flock))
1241 continue;
1242 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1243 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 }
J. Bruce Fields778fc542011-07-26 18:25:49 -04001245 fl->fl_lmops->lm_break(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 }
1247
1248 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1249 error = -EWOULDBLOCK;
1250 goto out;
1251 }
1252
1253restart:
1254 break_time = flock->fl_break_time;
1255 if (break_time != 0) {
1256 break_time -= jiffies;
1257 if (break_time == 0)
1258 break_time++;
1259 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001260 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001261 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001262 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1263 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001264 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001265 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 if (error >= 0) {
1267 if (error == 0)
1268 time_out_leases(inode);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001269 /*
1270 * Wait for the next conflicting lease that has not been
1271 * broken yet
1272 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1274 flock = flock->fl_next) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001275 if (locks_conflict(new_fl, flock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 goto restart;
1277 }
1278 error = 0;
1279 }
1280
1281out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001282 unlock_flocks();
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001283 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 return error;
1285}
1286
1287EXPORT_SYMBOL(__break_lease);
1288
1289/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001290 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 * @inode: the inode
1292 * @time: pointer to a timespec which will contain the last modified time
1293 *
1294 * This is to force NFS clients to flush their caches for files with
1295 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001296 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 */
1298void lease_get_mtime(struct inode *inode, struct timespec *time)
1299{
1300 struct file_lock *flock = inode->i_flock;
1301 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1302 *time = current_fs_time(inode->i_sb);
1303 else
1304 *time = inode->i_mtime;
1305}
1306
1307EXPORT_SYMBOL(lease_get_mtime);
1308
1309/**
1310 * fcntl_getlease - Enquire what lease is currently active
1311 * @filp: the file
1312 *
1313 * The value returned by this function will be one of
1314 * (if no lease break is pending):
1315 *
1316 * %F_RDLCK to indicate a shared lease is held.
1317 *
1318 * %F_WRLCK to indicate an exclusive lease is held.
1319 *
1320 * %F_UNLCK to indicate no lease is held.
1321 *
1322 * (if a lease break is pending):
1323 *
1324 * %F_RDLCK to indicate an exclusive lease needs to be
1325 * changed to a shared lease (or removed).
1326 *
1327 * %F_UNLCK to indicate the lease needs to be removed.
1328 *
1329 * XXX: sfr & willy disagree over whether F_INPROGRESS
1330 * should be returned to userspace.
1331 */
1332int fcntl_getlease(struct file *filp)
1333{
1334 struct file_lock *fl;
1335 int type = F_UNLCK;
1336
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001337 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001338 time_out_leases(filp->f_path.dentry->d_inode);
1339 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 fl = fl->fl_next) {
1341 if (fl->fl_file == filp) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001342 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 break;
1344 }
1345 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001346 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 return type;
1348}
1349
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001350int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001352 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001353 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 struct inode *inode = dentry->d_inode;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001355 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
J. Bruce Fields096657b2010-10-30 17:31:14 -04001357 lease = *flp;
1358
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001359 error = -EAGAIN;
1360 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001361 goto out;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001362 if ((arg == F_WRLCK)
1363 && ((dentry->d_count > 1)
1364 || (atomic_read(&inode->i_count) > 1)))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001365 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 /*
1368 * At this point, we know that if there is an exclusive
1369 * lease on this file, then we hold it on this filp
1370 * (otherwise our open of this file would have blocked).
1371 * And if we are trying to acquire an exclusive lease,
1372 * then the file is not open by anyone (including us)
1373 * except for this filp.
1374 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001375 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 for (before = &inode->i_flock;
1377 ((fl = *before) != NULL) && IS_LEASE(fl);
1378 before = &fl->fl_next) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001379 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 my_before = before;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001381 continue;
1382 }
1383 /*
1384 * No exclusive leases if someone else has a lease on
1385 * this file:
1386 */
1387 if (arg == F_WRLCK)
1388 goto out;
1389 /*
1390 * Modifying our existing lease is OK, but no getting a
1391 * new lease if someone else is opening for write:
1392 */
1393 if (fl->fl_flags & FL_UNLOCK_PENDING)
1394 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 }
1396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 if (my_before != NULL) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001398 error = lease->fl_lmops->lm_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001399 if (!error)
1400 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 goto out;
1402 }
1403
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 error = -EINVAL;
1405 if (!leases_enable)
1406 goto out;
1407
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001408 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001409 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411out:
1412 return error;
1413}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001414
1415int generic_delete_lease(struct file *filp, struct file_lock **flp)
1416{
1417 struct file_lock *fl, **before;
1418 struct dentry *dentry = filp->f_path.dentry;
1419 struct inode *inode = dentry->d_inode;
1420
1421 for (before = &inode->i_flock;
1422 ((fl = *before) != NULL) && IS_LEASE(fl);
1423 before = &fl->fl_next) {
1424 if (fl->fl_file != filp)
1425 continue;
1426 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1427 }
1428 return -EAGAIN;
1429}
1430
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431/**
1432 * generic_setlease - sets a lease on an open file
1433 * @filp: file pointer
1434 * @arg: type of lease to obtain
1435 * @flp: input - file_lock to use, output - file_lock inserted
1436 *
1437 * The (input) flp->fl_lmops->lm_break function is required
1438 * by break_lease().
1439 *
1440 * Called with file_lock_lock held.
1441 */
1442int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1443{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444 struct dentry *dentry = filp->f_path.dentry;
1445 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001446 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001448 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001449 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001451 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 error = security_file_lock(filp, arg);
1453 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001454 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
1456 time_out_leases(inode);
1457
1458 BUG_ON(!(*flp)->fl_lmops->lm_break);
1459
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001460 switch (arg) {
1461 case F_UNLCK:
1462 return generic_delete_lease(filp, flp);
1463 case F_RDLCK:
1464 case F_WRLCK:
1465 return generic_add_lease(filp, arg, flp);
1466 default:
1467 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001470EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001472static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1473{
1474 if (filp->f_op && filp->f_op->setlease)
1475 return filp->f_op->setlease(filp, arg, lease);
1476 else
1477 return generic_setlease(filp, arg, lease);
1478}
1479
1480/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001481 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 * @filp: file pointer
1483 * @arg: type of lease to obtain
1484 * @lease: file_lock to use
1485 *
1486 * Call this to establish a lease on the file.
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001487 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001488 * break_lease will oops!
1489 *
1490 * This will call the filesystem's setlease file method, if
1491 * defined. Note that there is no getlease method; instead, the
1492 * filesystem setlease method should call back to setlease() to
1493 * add a lease to the inode's lease list, where fcntl_getlease() can
1494 * find it. Since fcntl_getlease() only reports whether the current
1495 * task holds a lease, a cluster filesystem need only do this for
1496 * leases held by processes on this node.
1497 *
1498 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001499 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001500 * filesystem's open, create, and (on truncate) setattr methods.
1501 *
1502 * Warning: the only current setlease methods exist only to disable
1503 * leases in certain cases. More vfs changes may be required to
1504 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 */
1506
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001507int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 int error;
1510
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001511 lock_flocks();
1512 error = __vfs_setlease(filp, arg, lease);
1513 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
1515 return error;
1516}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001517EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001519static int do_fcntl_delete_lease(struct file *filp)
1520{
1521 struct file_lock fl, *flp = &fl;
1522
1523 lease_init(filp, F_UNLCK, flp);
1524
1525 return vfs_setlease(filp, F_UNLCK, &flp);
1526}
1527
1528static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001530 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001531 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 int error;
1533
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001534 fl = lease_alloc(filp, arg);
1535 if (IS_ERR(fl))
1536 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001538 new = fasync_alloc();
1539 if (!new) {
1540 locks_free_lock(fl);
1541 return -ENOMEM;
1542 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001543 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001544 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001545 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001546 if (error) {
1547 unlock_flocks();
1548 locks_free_lock(fl);
1549 goto out_free_fasync;
1550 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001551 if (ret != fl)
1552 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001554 /*
1555 * fasync_insert_entry() returns the old entry if any.
1556 * If there was no old entry, then it used 'new' and
1557 * inserted it into the fasync list. Clear new so that
1558 * we don't release it here.
1559 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001560 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001561 new = NULL;
1562
J. Bruce Fields8896b932010-11-03 18:09:18 -04001563 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001564 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001565
1566out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001567 if (new)
1568 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 return error;
1570}
1571
1572/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001573 * fcntl_setlease - sets a lease on an open file
1574 * @fd: open file descriptor
1575 * @filp: file pointer
1576 * @arg: type of lease to obtain
1577 *
1578 * Call this fcntl to establish a lease on the file.
1579 * Note that you also need to call %F_SETSIG to
1580 * receive a signal when the lease is broken.
1581 */
1582int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1583{
1584 if (arg == F_UNLCK)
1585 return do_fcntl_delete_lease(filp);
1586 return do_fcntl_add_lease(fd, filp, arg);
1587}
1588
1589/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1591 * @filp: The file to apply the lock to
1592 * @fl: The lock to be applied
1593 *
1594 * Add a FLOCK style lock to a file.
1595 */
1596int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1597{
1598 int error;
1599 might_sleep();
1600 for (;;) {
1601 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001602 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 break;
1604 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1605 if (!error)
1606 continue;
1607
1608 locks_delete_block(fl);
1609 break;
1610 }
1611 return error;
1612}
1613
1614EXPORT_SYMBOL(flock_lock_file_wait);
1615
1616/**
1617 * sys_flock: - flock() system call.
1618 * @fd: the file descriptor to lock.
1619 * @cmd: the type of lock to apply.
1620 *
1621 * Apply a %FL_FLOCK style lock to an open file descriptor.
1622 * The @cmd can be one of
1623 *
1624 * %LOCK_SH -- a shared lock.
1625 *
1626 * %LOCK_EX -- an exclusive lock.
1627 *
1628 * %LOCK_UN -- remove an existing lock.
1629 *
1630 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1631 *
1632 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1633 * processes read and write access respectively.
1634 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001635SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636{
1637 struct file *filp;
1638 struct file_lock *lock;
1639 int can_sleep, unlock;
1640 int error;
1641
1642 error = -EBADF;
1643 filp = fget(fd);
1644 if (!filp)
1645 goto out;
1646
1647 can_sleep = !(cmd & LOCK_NB);
1648 cmd &= ~LOCK_NB;
1649 unlock = (cmd == LOCK_UN);
1650
Al Viroaeb5d722008-09-02 15:28:45 -04001651 if (!unlock && !(cmd & LOCK_MAND) &&
1652 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 goto out_putf;
1654
1655 error = flock_make_lock(filp, &lock, cmd);
1656 if (error)
1657 goto out_putf;
1658 if (can_sleep)
1659 lock->fl_flags |= FL_SLEEP;
1660
Sten Spans713c0ec2009-07-16 09:41:39 +02001661 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 if (error)
1663 goto out_free;
1664
1665 if (filp->f_op && filp->f_op->flock)
1666 error = filp->f_op->flock(filp,
1667 (can_sleep) ? F_SETLKW : F_SETLK,
1668 lock);
1669 else
1670 error = flock_lock_file_wait(filp, lock);
1671
1672 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001673 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674
1675 out_putf:
1676 fput(filp);
1677 out:
1678 return error;
1679}
1680
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001681/**
1682 * vfs_test_lock - test file byte range lock
1683 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001684 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001685 *
1686 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1687 * setting conf->fl_type to something other than F_UNLCK.
1688 */
1689int vfs_test_lock(struct file *filp, struct file_lock *fl)
1690{
1691 if (filp->f_op && filp->f_op->lock)
1692 return filp->f_op->lock(filp, F_GETLK, fl);
1693 posix_test_lock(filp, fl);
1694 return 0;
1695}
1696EXPORT_SYMBOL_GPL(vfs_test_lock);
1697
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001698static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1699{
1700 flock->l_pid = fl->fl_pid;
1701#if BITS_PER_LONG == 32
1702 /*
1703 * Make sure we can represent the posix lock via
1704 * legacy 32bit flock.
1705 */
1706 if (fl->fl_start > OFFT_OFFSET_MAX)
1707 return -EOVERFLOW;
1708 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1709 return -EOVERFLOW;
1710#endif
1711 flock->l_start = fl->fl_start;
1712 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1713 fl->fl_end - fl->fl_start + 1;
1714 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001715 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001716 return 0;
1717}
1718
1719#if BITS_PER_LONG == 32
1720static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1721{
1722 flock->l_pid = fl->fl_pid;
1723 flock->l_start = fl->fl_start;
1724 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1725 fl->fl_end - fl->fl_start + 1;
1726 flock->l_whence = 0;
1727 flock->l_type = fl->fl_type;
1728}
1729#endif
1730
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731/* Report the first existing lock that would conflict with l.
1732 * This implements the F_GETLK command of fcntl().
1733 */
1734int fcntl_getlk(struct file *filp, struct flock __user *l)
1735{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001736 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 struct flock flock;
1738 int error;
1739
1740 error = -EFAULT;
1741 if (copy_from_user(&flock, l, sizeof(flock)))
1742 goto out;
1743 error = -EINVAL;
1744 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1745 goto out;
1746
1747 error = flock_to_posix_lock(filp, &file_lock, &flock);
1748 if (error)
1749 goto out;
1750
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001751 error = vfs_test_lock(filp, &file_lock);
1752 if (error)
1753 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001755 flock.l_type = file_lock.fl_type;
1756 if (file_lock.fl_type != F_UNLCK) {
1757 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001758 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 }
1761 error = -EFAULT;
1762 if (!copy_to_user(l, &flock, sizeof(flock)))
1763 error = 0;
1764out:
1765 return error;
1766}
1767
Marc Eshel7723ec92007-01-18 15:08:55 -05001768/**
1769 * vfs_lock_file - file byte range lock
1770 * @filp: The file to apply the lock to
1771 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1772 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001773 * @conf: Place to return a copy of the conflicting lock, if found.
1774 *
1775 * A caller that doesn't care about the conflicting lock may pass NULL
1776 * as the final argument.
1777 *
1778 * If the filesystem defines a private ->lock() method, then @conf will
1779 * be left unchanged; so a caller that cares should initialize it to
1780 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001781 *
1782 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1783 * locks, the ->lock() interface may return asynchronously, before the lock has
1784 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001785 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05001786 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1787 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001788 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001789 * request completes.
1790 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001791 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1792 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001793 * nonzero return code and the file system should release the lock. The file
1794 * system is also responsible to keep a corresponding posix lock when it
1795 * grants a lock so the VFS can find out which locks are locally held and do
1796 * the correct lock cleanup when required.
1797 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001798 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001799 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001800 */
Marc Eshel150b3932007-01-18 16:15:35 -05001801int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001802{
1803 if (filp->f_op && filp->f_op->lock)
1804 return filp->f_op->lock(filp, cmd, fl);
1805 else
Marc Eshel150b3932007-01-18 16:15:35 -05001806 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001807}
1808EXPORT_SYMBOL_GPL(vfs_lock_file);
1809
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001810static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1811 struct file_lock *fl)
1812{
1813 int error;
1814
1815 error = security_file_lock(filp, fl->fl_type);
1816 if (error)
1817 return error;
1818
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001819 for (;;) {
1820 error = vfs_lock_file(filp, cmd, fl, NULL);
1821 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001822 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001823 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1824 if (!error)
1825 continue;
1826
1827 locks_delete_block(fl);
1828 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001829 }
1830
1831 return error;
1832}
1833
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834/* Apply the lock described by l to an open file descriptor.
1835 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1836 */
Peter Staubachc2936212005-07-27 11:45:09 -07001837int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1838 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839{
1840 struct file_lock *file_lock = locks_alloc_lock();
1841 struct flock flock;
1842 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001843 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 int error;
1845
1846 if (file_lock == NULL)
1847 return -ENOLCK;
1848
1849 /*
1850 * This might block, so we do it before checking the inode.
1851 */
1852 error = -EFAULT;
1853 if (copy_from_user(&flock, l, sizeof(flock)))
1854 goto out;
1855
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001856 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857
1858 /* Don't allow mandatory locks on files that may be memory mapped
1859 * and shared.
1860 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001861 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 error = -EAGAIN;
1863 goto out;
1864 }
1865
Peter Staubachc2936212005-07-27 11:45:09 -07001866again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 error = flock_to_posix_lock(filp, file_lock, &flock);
1868 if (error)
1869 goto out;
1870 if (cmd == F_SETLKW) {
1871 file_lock->fl_flags |= FL_SLEEP;
1872 }
1873
1874 error = -EBADF;
1875 switch (flock.l_type) {
1876 case F_RDLCK:
1877 if (!(filp->f_mode & FMODE_READ))
1878 goto out;
1879 break;
1880 case F_WRLCK:
1881 if (!(filp->f_mode & FMODE_WRITE))
1882 goto out;
1883 break;
1884 case F_UNLCK:
1885 break;
1886 default:
1887 error = -EINVAL;
1888 goto out;
1889 }
1890
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001891 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
Peter Staubachc2936212005-07-27 11:45:09 -07001893 /*
1894 * Attempt to detect a close/fcntl race and recover by
1895 * releasing the lock that was just acquired.
1896 */
Al Viro0b2bac22008-05-06 13:58:34 -04001897 /*
1898 * we need that spin_lock here - it prevents reordering between
1899 * update of inode->i_flock and check for it done in close().
1900 * rcu_read_lock() wouldn't do.
1901 */
1902 spin_lock(&current->files->file_lock);
1903 f = fcheck(fd);
1904 spin_unlock(&current->files->file_lock);
1905 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001906 flock.l_type = F_UNLCK;
1907 goto again;
1908 }
1909
1910out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 locks_free_lock(file_lock);
1912 return error;
1913}
1914
1915#if BITS_PER_LONG == 32
1916/* Report the first existing lock that would conflict with l.
1917 * This implements the F_GETLK command of fcntl().
1918 */
1919int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1920{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001921 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 struct flock64 flock;
1923 int error;
1924
1925 error = -EFAULT;
1926 if (copy_from_user(&flock, l, sizeof(flock)))
1927 goto out;
1928 error = -EINVAL;
1929 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1930 goto out;
1931
1932 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1933 if (error)
1934 goto out;
1935
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001936 error = vfs_test_lock(filp, &file_lock);
1937 if (error)
1938 goto out;
1939
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001940 flock.l_type = file_lock.fl_type;
1941 if (file_lock.fl_type != F_UNLCK)
1942 posix_lock_to_flock64(&flock, &file_lock);
1943
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 error = -EFAULT;
1945 if (!copy_to_user(l, &flock, sizeof(flock)))
1946 error = 0;
1947
1948out:
1949 return error;
1950}
1951
1952/* Apply the lock described by l to an open file descriptor.
1953 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1954 */
Peter Staubachc2936212005-07-27 11:45:09 -07001955int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1956 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957{
1958 struct file_lock *file_lock = locks_alloc_lock();
1959 struct flock64 flock;
1960 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001961 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 int error;
1963
1964 if (file_lock == NULL)
1965 return -ENOLCK;
1966
1967 /*
1968 * This might block, so we do it before checking the inode.
1969 */
1970 error = -EFAULT;
1971 if (copy_from_user(&flock, l, sizeof(flock)))
1972 goto out;
1973
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001974 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
1976 /* Don't allow mandatory locks on files that may be memory mapped
1977 * and shared.
1978 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001979 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 error = -EAGAIN;
1981 goto out;
1982 }
1983
Peter Staubachc2936212005-07-27 11:45:09 -07001984again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 error = flock64_to_posix_lock(filp, file_lock, &flock);
1986 if (error)
1987 goto out;
1988 if (cmd == F_SETLKW64) {
1989 file_lock->fl_flags |= FL_SLEEP;
1990 }
1991
1992 error = -EBADF;
1993 switch (flock.l_type) {
1994 case F_RDLCK:
1995 if (!(filp->f_mode & FMODE_READ))
1996 goto out;
1997 break;
1998 case F_WRLCK:
1999 if (!(filp->f_mode & FMODE_WRITE))
2000 goto out;
2001 break;
2002 case F_UNLCK:
2003 break;
2004 default:
2005 error = -EINVAL;
2006 goto out;
2007 }
2008
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002009 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010
Peter Staubachc2936212005-07-27 11:45:09 -07002011 /*
2012 * Attempt to detect a close/fcntl race and recover by
2013 * releasing the lock that was just acquired.
2014 */
Al Viro0b2bac22008-05-06 13:58:34 -04002015 spin_lock(&current->files->file_lock);
2016 f = fcheck(fd);
2017 spin_unlock(&current->files->file_lock);
2018 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002019 flock.l_type = F_UNLCK;
2020 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 }
2022
2023out:
2024 locks_free_lock(file_lock);
2025 return error;
2026}
2027#endif /* BITS_PER_LONG == 32 */
2028
2029/*
2030 * This function is called when the file is being removed
2031 * from the task's fd array. POSIX locks belonging to this task
2032 * are deleted at this time.
2033 */
2034void locks_remove_posix(struct file *filp, fl_owner_t owner)
2035{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002036 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038 /*
2039 * If there are no locks held on this file, we don't need to call
2040 * posix_lock_file(). Another process could be setting a lock on this
2041 * file at the same time, but we wouldn't remove that lock anyway.
2042 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002043 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002044 return;
2045
2046 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002047 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 lock.fl_start = 0;
2049 lock.fl_end = OFFSET_MAX;
2050 lock.fl_owner = owner;
2051 lock.fl_pid = current->tgid;
2052 lock.fl_file = filp;
2053 lock.fl_ops = NULL;
2054 lock.fl_lmops = NULL;
2055
Marc Eshel150b3932007-01-18 16:15:35 -05002056 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2059 lock.fl_ops->fl_release_private(&lock);
2060}
2061
2062EXPORT_SYMBOL(locks_remove_posix);
2063
2064/*
2065 * This function is called on the last close of an open file.
2066 */
2067void locks_remove_flock(struct file *filp)
2068{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002069 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 struct file_lock *fl;
2071 struct file_lock **before;
2072
2073 if (!inode->i_flock)
2074 return;
2075
2076 if (filp->f_op && filp->f_op->flock) {
2077 struct file_lock fl = {
2078 .fl_pid = current->tgid,
2079 .fl_file = filp,
2080 .fl_flags = FL_FLOCK,
2081 .fl_type = F_UNLCK,
2082 .fl_end = OFFSET_MAX,
2083 };
2084 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002085 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2086 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 }
2088
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002089 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090 before = &inode->i_flock;
2091
2092 while ((fl = *before) != NULL) {
2093 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002094 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 locks_delete_lock(before);
2096 continue;
2097 }
2098 if (IS_LEASE(fl)) {
2099 lease_modify(before, F_UNLCK);
2100 continue;
2101 }
2102 /* What? */
2103 BUG();
2104 }
2105 before = &fl->fl_next;
2106 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002107 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108}
2109
2110/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 * posix_unblock_lock - stop waiting for a file lock
2112 * @filp: how the file was opened
2113 * @waiter: the lock which was waiting
2114 *
2115 * lockd needs to block waiting for locks.
2116 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002117int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2119{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002120 int status = 0;
2121
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002122 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002123 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002125 else
2126 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002127 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002128 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129}
2130
2131EXPORT_SYMBOL(posix_unblock_lock);
2132
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002133/**
2134 * vfs_cancel_lock - file byte range unblock lock
2135 * @filp: The file to apply the unblock to
2136 * @fl: The lock to be unblocked
2137 *
2138 * Used by lock managers to cancel blocked requests
2139 */
2140int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2141{
2142 if (filp->f_op && filp->f_op->lock)
2143 return filp->f_op->lock(filp, F_CANCELLK, fl);
2144 return 0;
2145}
2146
2147EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2148
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002149#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002150#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002151#include <linux/seq_file.h>
2152
2153static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002154 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
2156 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002157 unsigned int fl_pid;
2158
2159 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002160 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002161 else
2162 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163
2164 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002165 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166
Jerome Marchand99dc8292010-10-26 14:22:33 -07002167 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002169 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2171 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002172 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 } else if (IS_FLOCK(fl)) {
2174 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002175 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002177 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 }
2179 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002180 seq_printf(f, "LEASE ");
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002181 if (lease_breaking(fl))
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002182 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002184 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002186 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002188 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 }
2190 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002191 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 (fl->fl_type & LOCK_READ)
2193 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2194 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2195 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002196 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002197 (lease_breaking(fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2199 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2200 }
2201 if (inode) {
2202#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002203 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 inode->i_sb->s_id, inode->i_ino);
2205#else
2206 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002207 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 MAJOR(inode->i_sb->s_dev),
2209 MINOR(inode->i_sb->s_dev), inode->i_ino);
2210#endif
2211 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002212 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 }
2214 if (IS_POSIX(fl)) {
2215 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002216 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002218 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002220 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 }
2222}
2223
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002224static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002226 struct file_lock *fl, *bfl;
2227
2228 fl = list_entry(v, struct file_lock, fl_link);
2229
Jerome Marchand99dc8292010-10-26 14:22:33 -07002230 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002231
2232 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002233 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002234
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002235 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236}
2237
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002238static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002240 loff_t *p = f->private;
2241
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002242 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002243 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002244 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245}
2246
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002247static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2248{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002249 loff_t *p = f->private;
2250 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002251 return seq_list_next(v, &file_lock_list, pos);
2252}
2253
2254static void locks_stop(struct seq_file *f, void *v)
2255{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002256 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002257}
2258
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002259static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002260 .start = locks_start,
2261 .next = locks_next,
2262 .stop = locks_stop,
2263 .show = locks_show,
2264};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002265
2266static int locks_open(struct inode *inode, struct file *filp)
2267{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002268 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002269}
2270
2271static const struct file_operations proc_locks_operations = {
2272 .open = locks_open,
2273 .read = seq_read,
2274 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002275 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002276};
2277
2278static int __init proc_locks_init(void)
2279{
2280 proc_create("locks", 0, NULL, &proc_locks_operations);
2281 return 0;
2282}
2283module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002284#endif
2285
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286/**
2287 * lock_may_read - checks that the region is free of locks
2288 * @inode: the inode that is being read
2289 * @start: the first byte to read
2290 * @len: the number of bytes to read
2291 *
2292 * Emulates Windows locking requirements. Whole-file
2293 * mandatory locks (share modes) can prohibit a read and
2294 * byte-range POSIX locks can prohibit a read if they overlap.
2295 *
2296 * N.B. this function is only ever called
2297 * from knfsd and ownership of locks is never checked.
2298 */
2299int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2300{
2301 struct file_lock *fl;
2302 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002303 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2305 if (IS_POSIX(fl)) {
2306 if (fl->fl_type == F_RDLCK)
2307 continue;
2308 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2309 continue;
2310 } else if (IS_FLOCK(fl)) {
2311 if (!(fl->fl_type & LOCK_MAND))
2312 continue;
2313 if (fl->fl_type & LOCK_READ)
2314 continue;
2315 } else
2316 continue;
2317 result = 0;
2318 break;
2319 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002320 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 return result;
2322}
2323
2324EXPORT_SYMBOL(lock_may_read);
2325
2326/**
2327 * lock_may_write - checks that the region is free of locks
2328 * @inode: the inode that is being written
2329 * @start: the first byte to write
2330 * @len: the number of bytes to write
2331 *
2332 * Emulates Windows locking requirements. Whole-file
2333 * mandatory locks (share modes) can prohibit a write and
2334 * byte-range POSIX locks can prohibit a write if they overlap.
2335 *
2336 * N.B. this function is only ever called
2337 * from knfsd and ownership of locks is never checked.
2338 */
2339int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2340{
2341 struct file_lock *fl;
2342 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002343 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2345 if (IS_POSIX(fl)) {
2346 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2347 continue;
2348 } else if (IS_FLOCK(fl)) {
2349 if (!(fl->fl_type & LOCK_MAND))
2350 continue;
2351 if (fl->fl_type & LOCK_WRITE)
2352 continue;
2353 } else
2354 continue;
2355 result = 0;
2356 break;
2357 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002358 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 return result;
2360}
2361
2362EXPORT_SYMBOL(lock_may_write);
2363
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364static int __init filelock_init(void)
2365{
2366 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002367 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2368
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 return 0;
2370}
2371
2372core_initcall(filelock_init);