blob: 7ddab3f9df07e7955a396547d9f81e538dbfe3c8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
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
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200144static DEFINE_SPINLOCK(file_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200146/*
147 * Protects the two list heads above, plus the inode->i_flock list
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200148 */
149void lock_flocks(void)
150{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200151 spin_lock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200152}
153EXPORT_SYMBOL_GPL(lock_flocks);
154
155void unlock_flocks(void)
156{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200157 spin_unlock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200158}
159EXPORT_SYMBOL_GPL(unlock_flocks);
160
Christoph Lametere18b8902006-12-06 20:33:20 -0800161static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Miklos Szerediee19cc42011-07-07 13:06:09 +0200163static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200164{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200165 INIT_LIST_HEAD(&fl->fl_link);
166 INIT_LIST_HEAD(&fl->fl_block);
167 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200168}
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200171struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200173 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200174
175 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200176 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200177
178 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200180EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500182void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500183{
184 if (fl->fl_ops) {
185 if (fl->fl_ops->fl_release_private)
186 fl->fl_ops->fl_release_private(fl);
187 fl->fl_ops = NULL;
188 }
189 if (fl->fl_lmops) {
190 if (fl->fl_lmops->fl_release_private)
191 fl->fl_lmops->fl_release_private(fl);
192 fl->fl_lmops = NULL;
193 }
194
195}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500196EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500197
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400199void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800201 BUG_ON(waitqueue_active(&fl->fl_wait));
202 BUG_ON(!list_empty(&fl->fl_block));
203 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Trond Myklebust47831f32006-03-20 13:44:05 -0500205 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 kmem_cache_free(filelock_cache, fl);
207}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400208EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210void locks_init_lock(struct file_lock *fl)
211{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200212 memset(fl, 0, sizeof(struct file_lock));
213 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
216EXPORT_SYMBOL(locks_init_lock);
217
Trond Myklebust47831f32006-03-20 13:44:05 -0500218static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
219{
220 if (fl->fl_ops) {
221 if (fl->fl_ops->fl_copy_lock)
222 fl->fl_ops->fl_copy_lock(new, fl);
223 new->fl_ops = fl->fl_ops;
224 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400225 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500226 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500227}
228
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229/*
230 * Initialize a new lock from an existing file_lock structure.
231 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400232void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233{
234 new->fl_owner = fl->fl_owner;
235 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500236 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 new->fl_flags = fl->fl_flags;
238 new->fl_type = fl->fl_type;
239 new->fl_start = fl->fl_start;
240 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500241 new->fl_ops = NULL;
242 new->fl_lmops = NULL;
243}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700244EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500245
246void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
247{
248 locks_release_private(new);
249
250 __locks_copy_lock(new, fl);
251 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 new->fl_ops = fl->fl_ops;
253 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500254
255 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
258EXPORT_SYMBOL(locks_copy_lock);
259
260static inline int flock_translate_cmd(int cmd) {
261 if (cmd & LOCK_MAND)
262 return cmd & (LOCK_MAND | LOCK_RW);
263 switch (cmd) {
264 case LOCK_SH:
265 return F_RDLCK;
266 case LOCK_EX:
267 return F_WRLCK;
268 case LOCK_UN:
269 return F_UNLCK;
270 }
271 return -EINVAL;
272}
273
274/* Fill in a file_lock structure with an appropriate FLOCK lock. */
275static int flock_make_lock(struct file *filp, struct file_lock **lock,
276 unsigned int cmd)
277{
278 struct file_lock *fl;
279 int type = flock_translate_cmd(cmd);
280 if (type < 0)
281 return type;
282
283 fl = locks_alloc_lock();
284 if (fl == NULL)
285 return -ENOMEM;
286
287 fl->fl_file = filp;
288 fl->fl_pid = current->tgid;
289 fl->fl_flags = FL_FLOCK;
290 fl->fl_type = type;
291 fl->fl_end = OFFSET_MAX;
292
293 *lock = fl;
294 return 0;
295}
296
297static int assign_type(struct file_lock *fl, int type)
298{
299 switch (type) {
300 case F_RDLCK:
301 case F_WRLCK:
302 case F_UNLCK:
303 fl->fl_type = type;
304 break;
305 default:
306 return -EINVAL;
307 }
308 return 0;
309}
310
311/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
312 * style lock.
313 */
314static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
315 struct flock *l)
316{
317 off_t start, end;
318
319 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700320 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 start = 0;
322 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700323 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 start = filp->f_pos;
325 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700326 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800327 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 break;
329 default:
330 return -EINVAL;
331 }
332
333 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
334 POSIX-2001 defines it. */
335 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 if (start < 0)
337 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700338 fl->fl_end = OFFSET_MAX;
339 if (l->l_len > 0) {
340 end = start + l->l_len - 1;
341 fl->fl_end = end;
342 } else if (l->l_len < 0) {
343 end = start - 1;
344 fl->fl_end = end;
345 start += l->l_len;
346 if (start < 0)
347 return -EINVAL;
348 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700350 if (fl->fl_end < fl->fl_start)
351 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352
353 fl->fl_owner = current->files;
354 fl->fl_pid = current->tgid;
355 fl->fl_file = filp;
356 fl->fl_flags = FL_POSIX;
357 fl->fl_ops = NULL;
358 fl->fl_lmops = NULL;
359
360 return assign_type(fl, l->l_type);
361}
362
363#if BITS_PER_LONG == 32
364static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
365 struct flock64 *l)
366{
367 loff_t start;
368
369 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700370 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 start = 0;
372 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700373 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 start = filp->f_pos;
375 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700376 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800377 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 break;
379 default:
380 return -EINVAL;
381 }
382
Trond Myklebust4c780a42005-10-18 14:20:21 -0700383 start += l->l_start;
384 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700386 fl->fl_end = OFFSET_MAX;
387 if (l->l_len > 0) {
388 fl->fl_end = start + l->l_len - 1;
389 } else if (l->l_len < 0) {
390 fl->fl_end = start - 1;
391 start += l->l_len;
392 if (start < 0)
393 return -EINVAL;
394 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700396 if (fl->fl_end < fl->fl_start)
397 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 fl->fl_owner = current->files;
400 fl->fl_pid = current->tgid;
401 fl->fl_file = filp;
402 fl->fl_flags = FL_POSIX;
403 fl->fl_ops = NULL;
404 fl->fl_lmops = NULL;
405
Namhyung Kimf32cb532011-01-17 15:45:59 +0900406 return assign_type(fl, l->l_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407}
408#endif
409
410/* default lease lock manager operations */
411static void lease_break_callback(struct file_lock *fl)
412{
413 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
414}
415
416static void lease_release_private_callback(struct file_lock *fl)
417{
418 if (!fl->fl_file)
419 return;
420
421 f_delown(fl->fl_file);
422 fl->fl_file->f_owner.signum = 0;
423}
424
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700425static const struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 .fl_break = lease_break_callback,
427 .fl_release_private = lease_release_private_callback,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 .fl_change = lease_modify,
429};
430
431/*
432 * Initialize a lease, use the default lock manager operations
433 */
434static int lease_init(struct file *filp, int type, struct file_lock *fl)
435 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400436 if (assign_type(fl, type) != 0)
437 return -EINVAL;
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 fl->fl_owner = current->files;
440 fl->fl_pid = current->tgid;
441
442 fl->fl_file = filp;
443 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 fl->fl_start = 0;
445 fl->fl_end = OFFSET_MAX;
446 fl->fl_ops = NULL;
447 fl->fl_lmops = &lease_manager_ops;
448 return 0;
449}
450
451/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500452static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453{
454 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400455 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
457 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500458 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400461 if (error) {
462 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500463 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400464 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500465 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466}
467
468/* Check if two locks overlap each other.
469 */
470static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
471{
472 return ((fl1->fl_end >= fl2->fl_start) &&
473 (fl2->fl_end >= fl1->fl_start));
474}
475
476/*
477 * Check whether two locks have the same owner.
478 */
Matt Mackall33443c42006-01-08 01:05:22 -0800479static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
481 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
482 return fl2->fl_lmops == fl1->fl_lmops &&
483 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
484 return fl1->fl_owner == fl2->fl_owner;
485}
486
487/* Remove waiter from blocker's block list.
488 * When blocker ends up pointing to itself then the list is empty.
489 */
Matt Mackall33443c42006-01-08 01:05:22 -0800490static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491{
492 list_del_init(&waiter->fl_block);
493 list_del_init(&waiter->fl_link);
494 waiter->fl_next = NULL;
495}
496
497/*
498 */
499static void locks_delete_block(struct file_lock *waiter)
500{
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200501 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 __locks_delete_block(waiter);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200503 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504}
505
506/* Insert waiter into blocker's block list.
507 * We use a circular list so that processes can be easily woken up in
508 * the order they blocked. The documentation doesn't require this but
509 * it seems like the reasonable thing to do.
510 */
511static void locks_insert_block(struct file_lock *blocker,
512 struct file_lock *waiter)
513{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800514 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 list_add_tail(&waiter->fl_block, &blocker->fl_block);
516 waiter->fl_next = blocker;
517 if (IS_POSIX(blocker))
518 list_add(&waiter->fl_link, &blocked_list);
519}
520
521/* Wake up processes blocked waiting for blocker.
522 * If told to wait then schedule the processes until the block list
523 * is empty, otherwise empty the block list ourselves.
524 */
525static void locks_wake_up_blocks(struct file_lock *blocker)
526{
527 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400528 struct file_lock *waiter;
529
530 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 struct file_lock, fl_block);
532 __locks_delete_block(waiter);
533 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
534 waiter->fl_lmops->fl_notify(waiter);
535 else
536 wake_up(&waiter->fl_wait);
537 }
538}
539
540/* Insert file lock fl into an inode's lock list at the position indicated
541 * by pos. At the same time add the lock to the global file lock list.
542 */
543static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
544{
545 list_add(&fl->fl_link, &file_lock_list);
546
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000547 fl->fl_nspid = get_pid(task_tgid(current));
548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 /* insert into file's list */
550 fl->fl_next = *pos;
551 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
554/*
555 * Delete a lock and then free it.
556 * Wake up processes that are blocked waiting for this lock,
557 * notify the FS that the lock has been cleared and
558 * finally free the lock.
559 */
560static void locks_delete_lock(struct file_lock **thisfl_p)
561{
562 struct file_lock *fl = *thisfl_p;
563
564 *thisfl_p = fl->fl_next;
565 fl->fl_next = NULL;
566 list_del_init(&fl->fl_link);
567
568 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
569 if (fl->fl_fasync != NULL) {
570 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
571 fl->fl_fasync = NULL;
572 }
573
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000574 if (fl->fl_nspid) {
575 put_pid(fl->fl_nspid);
576 fl->fl_nspid = NULL;
577 }
578
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 locks_wake_up_blocks(fl);
580 locks_free_lock(fl);
581}
582
583/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
584 * checks for shared/exclusive status of overlapping locks.
585 */
586static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
587{
588 if (sys_fl->fl_type == F_WRLCK)
589 return 1;
590 if (caller_fl->fl_type == F_WRLCK)
591 return 1;
592 return 0;
593}
594
595/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
596 * checking before calling the locks_conflict().
597 */
598static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
599{
600 /* POSIX locks owned by the same process do not conflict with
601 * each other.
602 */
603 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
604 return (0);
605
606 /* Check whether they overlap */
607 if (!locks_overlap(caller_fl, sys_fl))
608 return 0;
609
610 return (locks_conflict(caller_fl, sys_fl));
611}
612
613/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
614 * checking before calling the locks_conflict().
615 */
616static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
617{
618 /* FLOCK locks referring to the same filp do not conflict with
619 * each other.
620 */
621 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
622 return (0);
623 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
624 return 0;
625
626 return (locks_conflict(caller_fl, sys_fl));
627}
628
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400629void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500630posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
632 struct file_lock *cfl;
633
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200634 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800635 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (!IS_POSIX(cfl))
637 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400638 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 break;
640 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000641 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500642 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000643 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800644 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000645 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400646 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200647 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400648 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650EXPORT_SYMBOL(posix_test_lock);
651
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400652/*
653 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400655 * We attempt to detect deadlocks that are due purely to posix file
656 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400658 * We assume that a task can be waiting for at most one lock at a time.
659 * So for any acquired lock, the process holding that lock may be
660 * waiting on at most one other lock. That lock in turns may be held by
661 * someone waiting for at most one other lock. Given a requested lock
662 * caller_fl which is about to wait for a conflicting lock block_fl, we
663 * follow this chain of waiters to ensure we are not about to create a
664 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400665 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400666 * Since we do this before we ever put a process to sleep on a lock, we
667 * are ensured that there is never a cycle; that is what guarantees that
668 * the while() loop in posix_locks_deadlock() eventually completes.
669 *
670 * Note: the above assumption may not be true when handling lock
671 * requests from a broken NFS client. It may also fail in the presence
672 * of tasks (such as posix threads) sharing the same open file table.
673 *
674 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400676
677#define MAX_DEADLK_ITERATIONS 10
678
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400679/* Find a lock that the owner of the given block_fl is blocking on. */
680static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
681{
682 struct file_lock *fl;
683
684 list_for_each_entry(fl, &blocked_list, fl_link) {
685 if (posix_same_owner(fl, block_fl))
686 return fl->fl_next;
687 }
688 return NULL;
689}
690
Adrian Bunkb0904e12006-06-23 02:05:13 -0700691static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 struct file_lock *block_fl)
693{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400694 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400696 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
697 if (i++ > MAX_DEADLK_ITERATIONS)
698 return 0;
699 if (posix_same_owner(caller_fl, block_fl))
700 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 }
702 return 0;
703}
704
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400706 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400707 *
708 * Note that if called with an FL_EXISTS argument, the caller may determine
709 * whether or not a lock was successfully freed by testing the return
710 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800712static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800714 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800716 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 int error = 0;
718 int found = 0;
719
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200720 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
721 new_fl = locks_alloc_lock();
722 if (!new_fl)
723 return -ENOMEM;
724 }
725
726 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400727 if (request->fl_flags & FL_ACCESS)
728 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 for_each_lock(inode, before) {
731 struct file_lock *fl = *before;
732 if (IS_POSIX(fl))
733 break;
734 if (IS_LEASE(fl))
735 continue;
736 if (filp != fl->fl_file)
737 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800738 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 goto out;
740 found = 1;
741 locks_delete_lock(before);
742 break;
743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Trond Myklebustf475ae92006-06-29 16:38:32 -0400745 if (request->fl_type == F_UNLCK) {
746 if ((request->fl_flags & FL_EXISTS) && !found)
747 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800748 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751 /*
752 * If a higher-priority process was blocked on the old file lock,
753 * give it the opportunity to lock the file.
754 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200755 if (found) {
756 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200757 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200758 lock_flocks();
759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400761find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 for_each_lock(inode, before) {
763 struct file_lock *fl = *before;
764 if (IS_POSIX(fl))
765 break;
766 if (IS_LEASE(fl))
767 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800768 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 continue;
770 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700771 if (!(request->fl_flags & FL_SLEEP))
772 goto out;
773 error = FILE_LOCK_DEFERRED;
774 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 goto out;
776 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400777 if (request->fl_flags & FL_ACCESS)
778 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800779 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700780 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800781 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400782 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
784out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200785 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800786 if (new_fl)
787 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 return error;
789}
790
Marc Eshel150b3932007-01-18 16:15:35 -0500791static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792{
793 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700794 struct file_lock *new_fl = NULL;
795 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 struct file_lock *left = NULL;
797 struct file_lock *right = NULL;
798 struct file_lock **before;
799 int error, added = 0;
800
801 /*
802 * We may need two file_lock structures for this operation,
803 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700804 *
805 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700807 if (!(request->fl_flags & FL_ACCESS) &&
808 (request->fl_type != F_UNLCK ||
809 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
810 new_fl = locks_alloc_lock();
811 new_fl2 = locks_alloc_lock();
812 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200814 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815 if (request->fl_type != F_UNLCK) {
816 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500817 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (!IS_POSIX(fl))
819 continue;
820 if (!posix_locks_conflict(request, fl))
821 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800822 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400823 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 error = -EAGAIN;
825 if (!(request->fl_flags & FL_SLEEP))
826 goto out;
827 error = -EDEADLK;
828 if (posix_locks_deadlock(request, fl))
829 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700830 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 locks_insert_block(fl, request);
832 goto out;
833 }
834 }
835
836 /* If we're just looking for a conflict, we're done. */
837 error = 0;
838 if (request->fl_flags & FL_ACCESS)
839 goto out;
840
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 * Find the first old lock with the same owner as the new lock.
843 */
844
845 before = &inode->i_flock;
846
847 /* First skip locks owned by other processes. */
848 while ((fl = *before) && (!IS_POSIX(fl) ||
849 !posix_same_owner(request, fl))) {
850 before = &fl->fl_next;
851 }
852
853 /* Process locks with this owner. */
854 while ((fl = *before) && posix_same_owner(request, fl)) {
855 /* Detect adjacent or overlapping regions (if same lock type)
856 */
857 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700858 /* In all comparisons of start vs end, use
859 * "start - 1" rather than "end + 1". If end
860 * is OFFSET_MAX, end + 1 will become negative.
861 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 if (fl->fl_end < request->fl_start - 1)
863 goto next_lock;
864 /* If the next lock in the list has entirely bigger
865 * addresses than the new one, insert the lock here.
866 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700867 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 break;
869
870 /* If we come here, the new and old lock are of the
871 * same type and adjacent or overlapping. Make one
872 * lock yielding from the lower start address of both
873 * locks to the higher end address.
874 */
875 if (fl->fl_start > request->fl_start)
876 fl->fl_start = request->fl_start;
877 else
878 request->fl_start = fl->fl_start;
879 if (fl->fl_end < request->fl_end)
880 fl->fl_end = request->fl_end;
881 else
882 request->fl_end = fl->fl_end;
883 if (added) {
884 locks_delete_lock(before);
885 continue;
886 }
887 request = fl;
888 added = 1;
889 }
890 else {
891 /* Processing for different lock types is a bit
892 * more complex.
893 */
894 if (fl->fl_end < request->fl_start)
895 goto next_lock;
896 if (fl->fl_start > request->fl_end)
897 break;
898 if (request->fl_type == F_UNLCK)
899 added = 1;
900 if (fl->fl_start < request->fl_start)
901 left = fl;
902 /* If the next lock in the list has a higher end
903 * address than the new one, insert the new one here.
904 */
905 if (fl->fl_end > request->fl_end) {
906 right = fl;
907 break;
908 }
909 if (fl->fl_start >= request->fl_start) {
910 /* The new lock completely replaces an old
911 * one (This may happen several times).
912 */
913 if (added) {
914 locks_delete_lock(before);
915 continue;
916 }
917 /* Replace the old lock with the new one.
918 * Wake up anybody waiting for the old one,
919 * as the change in lock type might satisfy
920 * their needs.
921 */
922 locks_wake_up_blocks(fl);
923 fl->fl_start = request->fl_start;
924 fl->fl_end = request->fl_end;
925 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500926 locks_release_private(fl);
927 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 request = fl;
929 added = 1;
930 }
931 }
932 /* Go on to next lock.
933 */
934 next_lock:
935 before = &fl->fl_next;
936 }
937
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700938 /*
939 * The above code only modifies existing locks in case of
940 * merging or replacing. If new lock(s) need to be inserted
941 * all modifications are done bellow this, so it's safe yet to
942 * bail out.
943 */
944 error = -ENOLCK; /* "no luck" */
945 if (right && left == right && !new_fl2)
946 goto out;
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 error = 0;
949 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400950 if (request->fl_type == F_UNLCK) {
951 if (request->fl_flags & FL_EXISTS)
952 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400954 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700955
956 if (!new_fl) {
957 error = -ENOLCK;
958 goto out;
959 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 locks_copy_lock(new_fl, request);
961 locks_insert_lock(before, new_fl);
962 new_fl = NULL;
963 }
964 if (right) {
965 if (left == right) {
966 /* The new lock breaks the old one in two pieces,
967 * so we have to use the second new lock.
968 */
969 left = new_fl2;
970 new_fl2 = NULL;
971 locks_copy_lock(left, right);
972 locks_insert_lock(before, left);
973 }
974 right->fl_start = request->fl_end + 1;
975 locks_wake_up_blocks(right);
976 }
977 if (left) {
978 left->fl_end = request->fl_start - 1;
979 locks_wake_up_blocks(left);
980 }
981 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200982 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 /*
984 * Free any unused locks.
985 */
986 if (new_fl)
987 locks_free_lock(new_fl);
988 if (new_fl2)
989 locks_free_lock(new_fl2);
990 return error;
991}
992
993/**
994 * posix_lock_file - Apply a POSIX-style lock to a file
995 * @filp: The file to apply the lock to
996 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -0500997 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 *
999 * Add a POSIX style lock to a file.
1000 * We merge adjacent & overlapping locks whenever possible.
1001 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001002 *
1003 * Note that if called with an FL_EXISTS argument, the caller may determine
1004 * whether or not a lock was successfully freed by testing the return
1005 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 */
Marc Eshel150b3932007-01-18 16:15:35 -05001007int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001008 struct file_lock *conflock)
1009{
Marc Eshel150b3932007-01-18 16:15:35 -05001010 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001011}
Marc Eshel150b3932007-01-18 16:15:35 -05001012EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013
1014/**
1015 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1016 * @filp: The file to apply the lock to
1017 * @fl: The lock to be applied
1018 *
1019 * Add a POSIX style lock to a file.
1020 * We merge adjacent & overlapping locks whenever possible.
1021 * POSIX locks are sorted by owner task, then by starting address
1022 */
1023int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1024{
1025 int error;
1026 might_sleep ();
1027 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001028 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001029 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 break;
1031 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1032 if (!error)
1033 continue;
1034
1035 locks_delete_block(fl);
1036 break;
1037 }
1038 return error;
1039}
1040EXPORT_SYMBOL(posix_lock_file_wait);
1041
1042/**
1043 * locks_mandatory_locked - Check for an active lock
1044 * @inode: the file to check
1045 *
1046 * Searches the inode's list of locks to find any POSIX locks which conflict.
1047 * This function is called from locks_verify_locked() only.
1048 */
1049int locks_mandatory_locked(struct inode *inode)
1050{
1051 fl_owner_t owner = current->files;
1052 struct file_lock *fl;
1053
1054 /*
1055 * Search the lock list for this inode for any POSIX locks.
1056 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001057 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1059 if (!IS_POSIX(fl))
1060 continue;
1061 if (fl->fl_owner != owner)
1062 break;
1063 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001064 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 return fl ? -EAGAIN : 0;
1066}
1067
1068/**
1069 * locks_mandatory_area - Check for a conflicting lock
1070 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1071 * for shared
1072 * @inode: the file to check
1073 * @filp: how the file was opened (if it was)
1074 * @offset: start of area to check
1075 * @count: length of area to check
1076 *
1077 * Searches the inode's list of locks to find any POSIX locks which conflict.
1078 * This function is called from rw_verify_area() and
1079 * locks_verify_truncate().
1080 */
1081int locks_mandatory_area(int read_write, struct inode *inode,
1082 struct file *filp, loff_t offset,
1083 size_t count)
1084{
1085 struct file_lock fl;
1086 int error;
1087
1088 locks_init_lock(&fl);
1089 fl.fl_owner = current->files;
1090 fl.fl_pid = current->tgid;
1091 fl.fl_file = filp;
1092 fl.fl_flags = FL_POSIX | FL_ACCESS;
1093 if (filp && !(filp->f_flags & O_NONBLOCK))
1094 fl.fl_flags |= FL_SLEEP;
1095 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1096 fl.fl_start = offset;
1097 fl.fl_end = offset + count - 1;
1098
1099 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001100 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001101 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 break;
1103 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1104 if (!error) {
1105 /*
1106 * If we've been sleeping someone might have
1107 * changed the permissions behind our back.
1108 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001109 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 continue;
1111 }
1112
1113 locks_delete_block(&fl);
1114 break;
1115 }
1116
1117 return error;
1118}
1119
1120EXPORT_SYMBOL(locks_mandatory_area);
1121
1122/* We already had a lease on this file; just change its type */
1123int lease_modify(struct file_lock **before, int arg)
1124{
1125 struct file_lock *fl = *before;
1126 int error = assign_type(fl, arg);
1127
1128 if (error)
1129 return error;
1130 locks_wake_up_blocks(fl);
1131 if (arg == F_UNLCK)
1132 locks_delete_lock(before);
1133 return 0;
1134}
1135
1136EXPORT_SYMBOL(lease_modify);
1137
1138static void time_out_leases(struct inode *inode)
1139{
1140 struct file_lock **before;
1141 struct file_lock *fl;
1142
1143 before = &inode->i_flock;
1144 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1145 if ((fl->fl_break_time == 0)
1146 || time_before(jiffies, fl->fl_break_time)) {
1147 before = &fl->fl_next;
1148 continue;
1149 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1151 if (fl == *before) /* lease_modify may have freed fl */
1152 before = &fl->fl_next;
1153 }
1154}
1155
1156/**
1157 * __break_lease - revoke all outstanding leases on file
1158 * @inode: the inode of the file to return
1159 * @mode: the open mode (read or write)
1160 *
david m. richter87250dd2007-05-09 16:10:27 -04001161 * break_lease (inlined for speed) has checked there already is at least
1162 * some kind of lock (maybe a lease) on this file. Leases are broken on
1163 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 * specified %O_NONBLOCK to your open().
1165 */
1166int __break_lease(struct inode *inode, unsigned int mode)
1167{
1168 int error = 0, future;
1169 struct file_lock *new_fl, *flock;
1170 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 unsigned long break_time;
1172 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001173 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
Al Viro8737c932009-12-24 06:47:55 -05001175 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001177 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178
1179 time_out_leases(inode);
1180
1181 flock = inode->i_flock;
1182 if ((flock == NULL) || !IS_LEASE(flock))
1183 goto out;
1184
1185 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1186 if (fl->fl_owner == current->files)
1187 i_have_this_lease = 1;
1188
Al Viro8737c932009-12-24 06:47:55 -05001189 if (want_write) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 /* If we want write access, we have to revoke any lease. */
1191 future = F_UNLCK | F_INPROGRESS;
1192 } else if (flock->fl_type & F_INPROGRESS) {
1193 /* If the lease is already being broken, we just leave it */
1194 future = flock->fl_type;
1195 } else if (flock->fl_type & F_WRLCK) {
1196 /* Downgrade the exclusive lease to a read-only lease. */
1197 future = F_RDLCK | F_INPROGRESS;
1198 } else {
1199 /* the existing lease was read-only, so we can read too. */
1200 goto out;
1201 }
1202
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001203 if (IS_ERR(new_fl) && !i_have_this_lease
1204 && ((mode & O_NONBLOCK) == 0)) {
1205 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 goto out;
1207 }
1208
1209 break_time = 0;
1210 if (lease_break_time > 0) {
1211 break_time = jiffies + lease_break_time * HZ;
1212 if (break_time == 0)
1213 break_time++; /* so that 0 means no break time */
1214 }
1215
1216 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1217 if (fl->fl_type != future) {
1218 fl->fl_type = future;
1219 fl->fl_break_time = break_time;
1220 /* lease must have lmops break callback */
1221 fl->fl_lmops->fl_break(fl);
1222 }
1223 }
1224
1225 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1226 error = -EWOULDBLOCK;
1227 goto out;
1228 }
1229
1230restart:
1231 break_time = flock->fl_break_time;
1232 if (break_time != 0) {
1233 break_time -= jiffies;
1234 if (break_time == 0)
1235 break_time++;
1236 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001237 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001238 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001239 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1240 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001241 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001242 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243 if (error >= 0) {
1244 if (error == 0)
1245 time_out_leases(inode);
1246 /* Wait for the next lease that has not been broken yet */
1247 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1248 flock = flock->fl_next) {
1249 if (flock->fl_type & F_INPROGRESS)
1250 goto restart;
1251 }
1252 error = 0;
1253 }
1254
1255out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001256 unlock_flocks();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001257 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258 locks_free_lock(new_fl);
1259 return error;
1260}
1261
1262EXPORT_SYMBOL(__break_lease);
1263
1264/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001265 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 * @inode: the inode
1267 * @time: pointer to a timespec which will contain the last modified time
1268 *
1269 * This is to force NFS clients to flush their caches for files with
1270 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001271 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 */
1273void lease_get_mtime(struct inode *inode, struct timespec *time)
1274{
1275 struct file_lock *flock = inode->i_flock;
1276 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1277 *time = current_fs_time(inode->i_sb);
1278 else
1279 *time = inode->i_mtime;
1280}
1281
1282EXPORT_SYMBOL(lease_get_mtime);
1283
1284/**
1285 * fcntl_getlease - Enquire what lease is currently active
1286 * @filp: the file
1287 *
1288 * The value returned by this function will be one of
1289 * (if no lease break is pending):
1290 *
1291 * %F_RDLCK to indicate a shared lease is held.
1292 *
1293 * %F_WRLCK to indicate an exclusive lease is held.
1294 *
1295 * %F_UNLCK to indicate no lease is held.
1296 *
1297 * (if a lease break is pending):
1298 *
1299 * %F_RDLCK to indicate an exclusive lease needs to be
1300 * changed to a shared lease (or removed).
1301 *
1302 * %F_UNLCK to indicate the lease needs to be removed.
1303 *
1304 * XXX: sfr & willy disagree over whether F_INPROGRESS
1305 * should be returned to userspace.
1306 */
1307int fcntl_getlease(struct file *filp)
1308{
1309 struct file_lock *fl;
1310 int type = F_UNLCK;
1311
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001312 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001313 time_out_leases(filp->f_path.dentry->d_inode);
1314 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 fl = fl->fl_next) {
1316 if (fl->fl_file == filp) {
1317 type = fl->fl_type & ~F_INPROGRESS;
1318 break;
1319 }
1320 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001321 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 return type;
1323}
1324
1325/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001326 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 * @filp: file pointer
1328 * @arg: type of lease to obtain
1329 * @flp: input - file_lock to use, output - file_lock inserted
1330 *
1331 * The (input) flp->fl_lmops->fl_break function is required
1332 * by break_lease().
1333 *
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001334 * Called with file_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001336int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001338 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001339 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 struct inode *inode = dentry->d_inode;
1341 int error, rdlease_count = 0, wrlease_count = 0;
1342
J. Bruce Fields096657b2010-10-30 17:31:14 -04001343 lease = *flp;
1344
1345 error = -EACCES;
David Howellsda9592e2008-11-14 10:39:05 +11001346 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001347 goto out;
1348 error = -EINVAL;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001349 if (!S_ISREG(inode->i_mode))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001350 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001351 error = security_file_lock(filp, arg);
1352 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001353 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001354
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 time_out_leases(inode);
1356
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001357 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358
David M. Richter288b2fd2008-04-23 16:29:00 -04001359 if (arg != F_UNLCK) {
1360 error = -EAGAIN;
1361 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1362 goto out;
1363 if ((arg == F_WRLCK)
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001364 && ((dentry->d_count > 1)
David M. Richter288b2fd2008-04-23 16:29:00 -04001365 || (atomic_read(&inode->i_count) > 1)))
1366 goto out;
David M. Richter288b2fd2008-04-23 16:29:00 -04001367 }
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001368
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 /*
1370 * At this point, we know that if there is an exclusive
1371 * lease on this file, then we hold it on this filp
1372 * (otherwise our open of this file would have blocked).
1373 * And if we are trying to acquire an exclusive lease,
1374 * then the file is not open by anyone (including us)
1375 * except for this filp.
1376 */
1377 for (before = &inode->i_flock;
1378 ((fl = *before) != NULL) && IS_LEASE(fl);
1379 before = &fl->fl_next) {
J. Bruce Fields255c7cf2010-10-31 12:35:48 -04001380 if (fl->fl_file == filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 my_before = before;
1382 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1383 /*
1384 * Someone is in the process of opening this
1385 * file for writing so we may not take an
1386 * exclusive lease on it.
1387 */
1388 wrlease_count++;
1389 else
1390 rdlease_count++;
1391 }
1392
David M. Richter5fcc60c2008-04-23 16:28:59 -04001393 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1395 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1396 goto out;
1397
1398 if (my_before != NULL) {
1399 error = lease->fl_lmops->fl_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001400 if (!error)
1401 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 goto out;
1403 }
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 if (arg == F_UNLCK)
1406 goto out;
1407
1408 error = -EINVAL;
1409 if (!leases_enable)
1410 goto out;
1411
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001412 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001413 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415out:
1416 return error;
1417}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001418EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001420static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1421{
1422 if (filp->f_op && filp->f_op->setlease)
1423 return filp->f_op->setlease(filp, arg, lease);
1424 else
1425 return generic_setlease(filp, arg, lease);
1426}
1427
1428/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001429 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 * @filp: file pointer
1431 * @arg: type of lease to obtain
1432 * @lease: file_lock to use
1433 *
1434 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001435 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1436 * break_lease will oops!
1437 *
1438 * This will call the filesystem's setlease file method, if
1439 * defined. Note that there is no getlease method; instead, the
1440 * filesystem setlease method should call back to setlease() to
1441 * add a lease to the inode's lease list, where fcntl_getlease() can
1442 * find it. Since fcntl_getlease() only reports whether the current
1443 * task holds a lease, a cluster filesystem need only do this for
1444 * leases held by processes on this node.
1445 *
1446 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001447 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001448 * filesystem's open, create, and (on truncate) setattr methods.
1449 *
1450 * Warning: the only current setlease methods exist only to disable
1451 * leases in certain cases. More vfs changes may be required to
1452 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 */
1454
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001455int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 int error;
1458
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001459 lock_flocks();
1460 error = __vfs_setlease(filp, arg, lease);
1461 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
1463 return error;
1464}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001465EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001467static int do_fcntl_delete_lease(struct file *filp)
1468{
1469 struct file_lock fl, *flp = &fl;
1470
1471 lease_init(filp, F_UNLCK, flp);
1472
1473 return vfs_setlease(filp, F_UNLCK, &flp);
1474}
1475
1476static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001478 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001479 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 int error;
1481
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001482 fl = lease_alloc(filp, arg);
1483 if (IS_ERR(fl))
1484 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001486 new = fasync_alloc();
1487 if (!new) {
1488 locks_free_lock(fl);
1489 return -ENOMEM;
1490 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001491 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001492 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001493 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001494 if (error) {
1495 unlock_flocks();
1496 locks_free_lock(fl);
1497 goto out_free_fasync;
1498 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001499 if (ret != fl)
1500 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001502 /*
1503 * fasync_insert_entry() returns the old entry if any.
1504 * If there was no old entry, then it used 'new' and
1505 * inserted it into the fasync list. Clear new so that
1506 * we don't release it here.
1507 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001508 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001509 new = NULL;
1510
J. Bruce Fields8896b932010-11-03 18:09:18 -04001511 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001512 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001513
1514out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001515 if (new)
1516 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517 return error;
1518}
1519
1520/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001521 * fcntl_setlease - sets a lease on an open file
1522 * @fd: open file descriptor
1523 * @filp: file pointer
1524 * @arg: type of lease to obtain
1525 *
1526 * Call this fcntl to establish a lease on the file.
1527 * Note that you also need to call %F_SETSIG to
1528 * receive a signal when the lease is broken.
1529 */
1530int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1531{
1532 if (arg == F_UNLCK)
1533 return do_fcntl_delete_lease(filp);
1534 return do_fcntl_add_lease(fd, filp, arg);
1535}
1536
1537/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1539 * @filp: The file to apply the lock to
1540 * @fl: The lock to be applied
1541 *
1542 * Add a FLOCK style lock to a file.
1543 */
1544int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1545{
1546 int error;
1547 might_sleep();
1548 for (;;) {
1549 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001550 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 break;
1552 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1553 if (!error)
1554 continue;
1555
1556 locks_delete_block(fl);
1557 break;
1558 }
1559 return error;
1560}
1561
1562EXPORT_SYMBOL(flock_lock_file_wait);
1563
1564/**
1565 * sys_flock: - flock() system call.
1566 * @fd: the file descriptor to lock.
1567 * @cmd: the type of lock to apply.
1568 *
1569 * Apply a %FL_FLOCK style lock to an open file descriptor.
1570 * The @cmd can be one of
1571 *
1572 * %LOCK_SH -- a shared lock.
1573 *
1574 * %LOCK_EX -- an exclusive lock.
1575 *
1576 * %LOCK_UN -- remove an existing lock.
1577 *
1578 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1579 *
1580 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1581 * processes read and write access respectively.
1582 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001583SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584{
1585 struct file *filp;
1586 struct file_lock *lock;
1587 int can_sleep, unlock;
1588 int error;
1589
1590 error = -EBADF;
1591 filp = fget(fd);
1592 if (!filp)
1593 goto out;
1594
1595 can_sleep = !(cmd & LOCK_NB);
1596 cmd &= ~LOCK_NB;
1597 unlock = (cmd == LOCK_UN);
1598
Al Viroaeb5d722008-09-02 15:28:45 -04001599 if (!unlock && !(cmd & LOCK_MAND) &&
1600 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 goto out_putf;
1602
1603 error = flock_make_lock(filp, &lock, cmd);
1604 if (error)
1605 goto out_putf;
1606 if (can_sleep)
1607 lock->fl_flags |= FL_SLEEP;
1608
Sten Spans713c0ec2009-07-16 09:41:39 +02001609 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 if (error)
1611 goto out_free;
1612
1613 if (filp->f_op && filp->f_op->flock)
1614 error = filp->f_op->flock(filp,
1615 (can_sleep) ? F_SETLKW : F_SETLK,
1616 lock);
1617 else
1618 error = flock_lock_file_wait(filp, lock);
1619
1620 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001621 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
1623 out_putf:
1624 fput(filp);
1625 out:
1626 return error;
1627}
1628
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001629/**
1630 * vfs_test_lock - test file byte range lock
1631 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001632 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001633 *
1634 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1635 * setting conf->fl_type to something other than F_UNLCK.
1636 */
1637int vfs_test_lock(struct file *filp, struct file_lock *fl)
1638{
1639 if (filp->f_op && filp->f_op->lock)
1640 return filp->f_op->lock(filp, F_GETLK, fl);
1641 posix_test_lock(filp, fl);
1642 return 0;
1643}
1644EXPORT_SYMBOL_GPL(vfs_test_lock);
1645
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001646static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1647{
1648 flock->l_pid = fl->fl_pid;
1649#if BITS_PER_LONG == 32
1650 /*
1651 * Make sure we can represent the posix lock via
1652 * legacy 32bit flock.
1653 */
1654 if (fl->fl_start > OFFT_OFFSET_MAX)
1655 return -EOVERFLOW;
1656 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1657 return -EOVERFLOW;
1658#endif
1659 flock->l_start = fl->fl_start;
1660 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1661 fl->fl_end - fl->fl_start + 1;
1662 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001663 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001664 return 0;
1665}
1666
1667#if BITS_PER_LONG == 32
1668static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1669{
1670 flock->l_pid = fl->fl_pid;
1671 flock->l_start = fl->fl_start;
1672 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1673 fl->fl_end - fl->fl_start + 1;
1674 flock->l_whence = 0;
1675 flock->l_type = fl->fl_type;
1676}
1677#endif
1678
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679/* Report the first existing lock that would conflict with l.
1680 * This implements the F_GETLK command of fcntl().
1681 */
1682int fcntl_getlk(struct file *filp, struct flock __user *l)
1683{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001684 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685 struct flock flock;
1686 int error;
1687
1688 error = -EFAULT;
1689 if (copy_from_user(&flock, l, sizeof(flock)))
1690 goto out;
1691 error = -EINVAL;
1692 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1693 goto out;
1694
1695 error = flock_to_posix_lock(filp, &file_lock, &flock);
1696 if (error)
1697 goto out;
1698
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001699 error = vfs_test_lock(filp, &file_lock);
1700 if (error)
1701 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001703 flock.l_type = file_lock.fl_type;
1704 if (file_lock.fl_type != F_UNLCK) {
1705 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001706 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 }
1709 error = -EFAULT;
1710 if (!copy_to_user(l, &flock, sizeof(flock)))
1711 error = 0;
1712out:
1713 return error;
1714}
1715
Marc Eshel7723ec92007-01-18 15:08:55 -05001716/**
1717 * vfs_lock_file - file byte range lock
1718 * @filp: The file to apply the lock to
1719 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1720 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001721 * @conf: Place to return a copy of the conflicting lock, if found.
1722 *
1723 * A caller that doesn't care about the conflicting lock may pass NULL
1724 * as the final argument.
1725 *
1726 * If the filesystem defines a private ->lock() method, then @conf will
1727 * be left unchanged; so a caller that cares should initialize it to
1728 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001729 *
1730 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1731 * locks, the ->lock() interface may return asynchronously, before the lock has
1732 * been granted or denied by the underlying filesystem, if (and only if)
1733 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1734 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1735 * the request is for a blocking lock. When ->lock() does return asynchronously,
Miklos Szeredibde74e42008-07-25 01:48:57 -07001736 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001737 * request completes.
1738 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001739 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1740 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001741 * nonzero return code and the file system should release the lock. The file
1742 * system is also responsible to keep a corresponding posix lock when it
1743 * grants a lock so the VFS can find out which locks are locally held and do
1744 * the correct lock cleanup when required.
1745 * The underlying filesystem must not drop the kernel lock or call
Miklos Szeredibde74e42008-07-25 01:48:57 -07001746 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001747 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001748 */
Marc Eshel150b3932007-01-18 16:15:35 -05001749int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001750{
1751 if (filp->f_op && filp->f_op->lock)
1752 return filp->f_op->lock(filp, cmd, fl);
1753 else
Marc Eshel150b3932007-01-18 16:15:35 -05001754 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001755}
1756EXPORT_SYMBOL_GPL(vfs_lock_file);
1757
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001758static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1759 struct file_lock *fl)
1760{
1761 int error;
1762
1763 error = security_file_lock(filp, fl->fl_type);
1764 if (error)
1765 return error;
1766
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001767 for (;;) {
1768 error = vfs_lock_file(filp, cmd, fl, NULL);
1769 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001770 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001771 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1772 if (!error)
1773 continue;
1774
1775 locks_delete_block(fl);
1776 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001777 }
1778
1779 return error;
1780}
1781
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782/* Apply the lock described by l to an open file descriptor.
1783 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1784 */
Peter Staubachc2936212005-07-27 11:45:09 -07001785int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1786 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
1788 struct file_lock *file_lock = locks_alloc_lock();
1789 struct flock flock;
1790 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001791 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 int error;
1793
1794 if (file_lock == NULL)
1795 return -ENOLCK;
1796
1797 /*
1798 * This might block, so we do it before checking the inode.
1799 */
1800 error = -EFAULT;
1801 if (copy_from_user(&flock, l, sizeof(flock)))
1802 goto out;
1803
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001804 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
1806 /* Don't allow mandatory locks on files that may be memory mapped
1807 * and shared.
1808 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001809 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 error = -EAGAIN;
1811 goto out;
1812 }
1813
Peter Staubachc2936212005-07-27 11:45:09 -07001814again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 error = flock_to_posix_lock(filp, file_lock, &flock);
1816 if (error)
1817 goto out;
1818 if (cmd == F_SETLKW) {
1819 file_lock->fl_flags |= FL_SLEEP;
1820 }
1821
1822 error = -EBADF;
1823 switch (flock.l_type) {
1824 case F_RDLCK:
1825 if (!(filp->f_mode & FMODE_READ))
1826 goto out;
1827 break;
1828 case F_WRLCK:
1829 if (!(filp->f_mode & FMODE_WRITE))
1830 goto out;
1831 break;
1832 case F_UNLCK:
1833 break;
1834 default:
1835 error = -EINVAL;
1836 goto out;
1837 }
1838
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001839 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Peter Staubachc2936212005-07-27 11:45:09 -07001841 /*
1842 * Attempt to detect a close/fcntl race and recover by
1843 * releasing the lock that was just acquired.
1844 */
Al Viro0b2bac22008-05-06 13:58:34 -04001845 /*
1846 * we need that spin_lock here - it prevents reordering between
1847 * update of inode->i_flock and check for it done in close().
1848 * rcu_read_lock() wouldn't do.
1849 */
1850 spin_lock(&current->files->file_lock);
1851 f = fcheck(fd);
1852 spin_unlock(&current->files->file_lock);
1853 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001854 flock.l_type = F_UNLCK;
1855 goto again;
1856 }
1857
1858out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 locks_free_lock(file_lock);
1860 return error;
1861}
1862
1863#if BITS_PER_LONG == 32
1864/* Report the first existing lock that would conflict with l.
1865 * This implements the F_GETLK command of fcntl().
1866 */
1867int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1868{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001869 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 struct flock64 flock;
1871 int error;
1872
1873 error = -EFAULT;
1874 if (copy_from_user(&flock, l, sizeof(flock)))
1875 goto out;
1876 error = -EINVAL;
1877 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1878 goto out;
1879
1880 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1881 if (error)
1882 goto out;
1883
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001884 error = vfs_test_lock(filp, &file_lock);
1885 if (error)
1886 goto out;
1887
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001888 flock.l_type = file_lock.fl_type;
1889 if (file_lock.fl_type != F_UNLCK)
1890 posix_lock_to_flock64(&flock, &file_lock);
1891
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 error = -EFAULT;
1893 if (!copy_to_user(l, &flock, sizeof(flock)))
1894 error = 0;
1895
1896out:
1897 return error;
1898}
1899
1900/* Apply the lock described by l to an open file descriptor.
1901 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1902 */
Peter Staubachc2936212005-07-27 11:45:09 -07001903int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1904 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905{
1906 struct file_lock *file_lock = locks_alloc_lock();
1907 struct flock64 flock;
1908 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001909 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 int error;
1911
1912 if (file_lock == NULL)
1913 return -ENOLCK;
1914
1915 /*
1916 * This might block, so we do it before checking the inode.
1917 */
1918 error = -EFAULT;
1919 if (copy_from_user(&flock, l, sizeof(flock)))
1920 goto out;
1921
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001922 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923
1924 /* Don't allow mandatory locks on files that may be memory mapped
1925 * and shared.
1926 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001927 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928 error = -EAGAIN;
1929 goto out;
1930 }
1931
Peter Staubachc2936212005-07-27 11:45:09 -07001932again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 error = flock64_to_posix_lock(filp, file_lock, &flock);
1934 if (error)
1935 goto out;
1936 if (cmd == F_SETLKW64) {
1937 file_lock->fl_flags |= FL_SLEEP;
1938 }
1939
1940 error = -EBADF;
1941 switch (flock.l_type) {
1942 case F_RDLCK:
1943 if (!(filp->f_mode & FMODE_READ))
1944 goto out;
1945 break;
1946 case F_WRLCK:
1947 if (!(filp->f_mode & FMODE_WRITE))
1948 goto out;
1949 break;
1950 case F_UNLCK:
1951 break;
1952 default:
1953 error = -EINVAL;
1954 goto out;
1955 }
1956
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001957 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958
Peter Staubachc2936212005-07-27 11:45:09 -07001959 /*
1960 * Attempt to detect a close/fcntl race and recover by
1961 * releasing the lock that was just acquired.
1962 */
Al Viro0b2bac22008-05-06 13:58:34 -04001963 spin_lock(&current->files->file_lock);
1964 f = fcheck(fd);
1965 spin_unlock(&current->files->file_lock);
1966 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001967 flock.l_type = F_UNLCK;
1968 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 }
1970
1971out:
1972 locks_free_lock(file_lock);
1973 return error;
1974}
1975#endif /* BITS_PER_LONG == 32 */
1976
1977/*
1978 * This function is called when the file is being removed
1979 * from the task's fd array. POSIX locks belonging to this task
1980 * are deleted at this time.
1981 */
1982void locks_remove_posix(struct file *filp, fl_owner_t owner)
1983{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07001984 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985
1986 /*
1987 * If there are no locks held on this file, we don't need to call
1988 * posix_lock_file(). Another process could be setting a lock on this
1989 * file at the same time, but we wouldn't remove that lock anyway.
1990 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001991 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return;
1993
1994 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07001995 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 lock.fl_start = 0;
1997 lock.fl_end = OFFSET_MAX;
1998 lock.fl_owner = owner;
1999 lock.fl_pid = current->tgid;
2000 lock.fl_file = filp;
2001 lock.fl_ops = NULL;
2002 lock.fl_lmops = NULL;
2003
Marc Eshel150b3932007-01-18 16:15:35 -05002004 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2007 lock.fl_ops->fl_release_private(&lock);
2008}
2009
2010EXPORT_SYMBOL(locks_remove_posix);
2011
2012/*
2013 * This function is called on the last close of an open file.
2014 */
2015void locks_remove_flock(struct file *filp)
2016{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002017 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 struct file_lock *fl;
2019 struct file_lock **before;
2020
2021 if (!inode->i_flock)
2022 return;
2023
2024 if (filp->f_op && filp->f_op->flock) {
2025 struct file_lock fl = {
2026 .fl_pid = current->tgid,
2027 .fl_file = filp,
2028 .fl_flags = FL_FLOCK,
2029 .fl_type = F_UNLCK,
2030 .fl_end = OFFSET_MAX,
2031 };
2032 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002033 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2034 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 }
2036
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002037 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 before = &inode->i_flock;
2039
2040 while ((fl = *before) != NULL) {
2041 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002042 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 locks_delete_lock(before);
2044 continue;
2045 }
2046 if (IS_LEASE(fl)) {
2047 lease_modify(before, F_UNLCK);
2048 continue;
2049 }
2050 /* What? */
2051 BUG();
2052 }
2053 before = &fl->fl_next;
2054 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002055 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056}
2057
2058/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 * posix_unblock_lock - stop waiting for a file lock
2060 * @filp: how the file was opened
2061 * @waiter: the lock which was waiting
2062 *
2063 * lockd needs to block waiting for locks.
2064 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002065int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2067{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002068 int status = 0;
2069
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002070 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002071 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002073 else
2074 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002075 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002076 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077}
2078
2079EXPORT_SYMBOL(posix_unblock_lock);
2080
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002081/**
2082 * vfs_cancel_lock - file byte range unblock lock
2083 * @filp: The file to apply the unblock to
2084 * @fl: The lock to be unblocked
2085 *
2086 * Used by lock managers to cancel blocked requests
2087 */
2088int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2089{
2090 if (filp->f_op && filp->f_op->lock)
2091 return filp->f_op->lock(filp, F_CANCELLK, fl);
2092 return 0;
2093}
2094
2095EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2096
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002097#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002098#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002099#include <linux/seq_file.h>
2100
2101static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002102 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103{
2104 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002105 unsigned int fl_pid;
2106
2107 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002108 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002109 else
2110 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111
2112 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002113 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114
Jerome Marchand99dc8292010-10-26 14:22:33 -07002115 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002117 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2119 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002120 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 } else if (IS_FLOCK(fl)) {
2122 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002123 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002125 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 }
2127 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002128 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002130 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002132 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002134 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002135 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002136 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 }
2138 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002139 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 (fl->fl_type & LOCK_READ)
2141 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2142 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2143 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002144 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 (fl->fl_type & F_INPROGRESS)
2146 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2147 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2148 }
2149 if (inode) {
2150#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002151 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 inode->i_sb->s_id, inode->i_ino);
2153#else
2154 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002155 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156 MAJOR(inode->i_sb->s_dev),
2157 MINOR(inode->i_sb->s_dev), inode->i_ino);
2158#endif
2159 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002160 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 }
2162 if (IS_POSIX(fl)) {
2163 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002164 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002166 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002168 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 }
2170}
2171
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002172static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002174 struct file_lock *fl, *bfl;
2175
2176 fl = list_entry(v, struct file_lock, fl_link);
2177
Jerome Marchand99dc8292010-10-26 14:22:33 -07002178 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002179
2180 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002181 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002182
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002183 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184}
2185
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002186static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002188 loff_t *p = f->private;
2189
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002190 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002191 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002192 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193}
2194
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002195static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2196{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002197 loff_t *p = f->private;
2198 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002199 return seq_list_next(v, &file_lock_list, pos);
2200}
2201
2202static void locks_stop(struct seq_file *f, void *v)
2203{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002204 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002205}
2206
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002207static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002208 .start = locks_start,
2209 .next = locks_next,
2210 .stop = locks_stop,
2211 .show = locks_show,
2212};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002213
2214static int locks_open(struct inode *inode, struct file *filp)
2215{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002216 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002217}
2218
2219static const struct file_operations proc_locks_operations = {
2220 .open = locks_open,
2221 .read = seq_read,
2222 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002223 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002224};
2225
2226static int __init proc_locks_init(void)
2227{
2228 proc_create("locks", 0, NULL, &proc_locks_operations);
2229 return 0;
2230}
2231module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002232#endif
2233
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234/**
2235 * lock_may_read - checks that the region is free of locks
2236 * @inode: the inode that is being read
2237 * @start: the first byte to read
2238 * @len: the number of bytes to read
2239 *
2240 * Emulates Windows locking requirements. Whole-file
2241 * mandatory locks (share modes) can prohibit a read and
2242 * byte-range POSIX locks can prohibit a read if they overlap.
2243 *
2244 * N.B. this function is only ever called
2245 * from knfsd and ownership of locks is never checked.
2246 */
2247int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2248{
2249 struct file_lock *fl;
2250 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002251 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2253 if (IS_POSIX(fl)) {
2254 if (fl->fl_type == F_RDLCK)
2255 continue;
2256 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2257 continue;
2258 } else if (IS_FLOCK(fl)) {
2259 if (!(fl->fl_type & LOCK_MAND))
2260 continue;
2261 if (fl->fl_type & LOCK_READ)
2262 continue;
2263 } else
2264 continue;
2265 result = 0;
2266 break;
2267 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002268 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 return result;
2270}
2271
2272EXPORT_SYMBOL(lock_may_read);
2273
2274/**
2275 * lock_may_write - checks that the region is free of locks
2276 * @inode: the inode that is being written
2277 * @start: the first byte to write
2278 * @len: the number of bytes to write
2279 *
2280 * Emulates Windows locking requirements. Whole-file
2281 * mandatory locks (share modes) can prohibit a write and
2282 * byte-range POSIX locks can prohibit a write if they overlap.
2283 *
2284 * N.B. this function is only ever called
2285 * from knfsd and ownership of locks is never checked.
2286 */
2287int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2288{
2289 struct file_lock *fl;
2290 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002291 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2293 if (IS_POSIX(fl)) {
2294 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2295 continue;
2296 } else if (IS_FLOCK(fl)) {
2297 if (!(fl->fl_type & LOCK_MAND))
2298 continue;
2299 if (fl->fl_type & LOCK_WRITE)
2300 continue;
2301 } else
2302 continue;
2303 result = 0;
2304 break;
2305 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002306 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 return result;
2308}
2309
2310EXPORT_SYMBOL(lock_may_write);
2311
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312static int __init filelock_init(void)
2313{
2314 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002315 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2316
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 return 0;
2318}
2319
2320core_initcall(filelock_init);