blob: 0e62dd35d088486ee2c04c62f312ffb2b0ca5149 [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>
125#include <linux/smp_lock.h>
126#include <linux/syscalls.h>
127#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700128#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000129#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131#include <asm/uaccess.h>
132
133#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
134#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
135#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
136
137int leases_enable = 1;
138int lease_break_time = 45;
139
140#define for_each_lock(inode, lockp) \
141 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
142
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500143static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144static LIST_HEAD(blocked_list);
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200145static DEFINE_SPINLOCK(file_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200147/*
148 * Protects the two list heads above, plus the inode->i_flock list
149 * FIXME: should use a spinlock, once lockd and ceph are ready.
150 */
151void lock_flocks(void)
152{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200153 spin_lock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200154}
155EXPORT_SYMBOL_GPL(lock_flocks);
156
157void unlock_flocks(void)
158{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200159 spin_unlock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200160}
161EXPORT_SYMBOL_GPL(unlock_flocks);
162
Christoph Lametere18b8902006-12-06 20:33:20 -0800163static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200166struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
Christoph Lametere94b1762006-12-06 20:33:17 -0800168 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200170EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500172void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500173{
174 if (fl->fl_ops) {
175 if (fl->fl_ops->fl_release_private)
176 fl->fl_ops->fl_release_private(fl);
177 fl->fl_ops = NULL;
178 }
179 if (fl->fl_lmops) {
180 if (fl->fl_lmops->fl_release_private)
181 fl->fl_lmops->fl_release_private(fl);
182 fl->fl_lmops = NULL;
183 }
184
185}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500186EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400189void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800191 BUG_ON(waitqueue_active(&fl->fl_wait));
192 BUG_ON(!list_empty(&fl->fl_block));
193 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Trond Myklebust47831f32006-03-20 13:44:05 -0500195 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 kmem_cache_free(filelock_cache, fl);
197}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400198EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200void locks_init_lock(struct file_lock *fl)
201{
202 INIT_LIST_HEAD(&fl->fl_link);
203 INIT_LIST_HEAD(&fl->fl_block);
204 init_waitqueue_head(&fl->fl_wait);
205 fl->fl_next = NULL;
206 fl->fl_fasync = NULL;
207 fl->fl_owner = NULL;
208 fl->fl_pid = 0;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000209 fl->fl_nspid = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 fl->fl_file = NULL;
211 fl->fl_flags = 0;
212 fl->fl_type = 0;
213 fl->fl_start = fl->fl_end = 0;
214 fl->fl_ops = NULL;
215 fl->fl_lmops = NULL;
216}
217
218EXPORT_SYMBOL(locks_init_lock);
219
220/*
221 * Initialises the fields of the file lock which are invariant for
222 * free file_locks.
223 */
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700224static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225{
226 struct file_lock *lock = (struct file_lock *) foo;
227
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 locks_init_lock(lock);
229}
230
Trond Myklebust47831f32006-03-20 13:44:05 -0500231static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
232{
233 if (fl->fl_ops) {
234 if (fl->fl_ops->fl_copy_lock)
235 fl->fl_ops->fl_copy_lock(new, fl);
236 new->fl_ops = fl->fl_ops;
237 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400238 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500239 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500240}
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242/*
243 * Initialize a new lock from an existing file_lock structure.
244 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400245void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
247 new->fl_owner = fl->fl_owner;
248 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500249 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 new->fl_flags = fl->fl_flags;
251 new->fl_type = fl->fl_type;
252 new->fl_start = fl->fl_start;
253 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500254 new->fl_ops = NULL;
255 new->fl_lmops = NULL;
256}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700257EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500258
259void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
260{
261 locks_release_private(new);
262
263 __locks_copy_lock(new, fl);
264 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 new->fl_ops = fl->fl_ops;
266 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500267
268 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269}
270
271EXPORT_SYMBOL(locks_copy_lock);
272
273static inline int flock_translate_cmd(int cmd) {
274 if (cmd & LOCK_MAND)
275 return cmd & (LOCK_MAND | LOCK_RW);
276 switch (cmd) {
277 case LOCK_SH:
278 return F_RDLCK;
279 case LOCK_EX:
280 return F_WRLCK;
281 case LOCK_UN:
282 return F_UNLCK;
283 }
284 return -EINVAL;
285}
286
287/* Fill in a file_lock structure with an appropriate FLOCK lock. */
288static int flock_make_lock(struct file *filp, struct file_lock **lock,
289 unsigned int cmd)
290{
291 struct file_lock *fl;
292 int type = flock_translate_cmd(cmd);
293 if (type < 0)
294 return type;
295
296 fl = locks_alloc_lock();
297 if (fl == NULL)
298 return -ENOMEM;
299
300 fl->fl_file = filp;
301 fl->fl_pid = current->tgid;
302 fl->fl_flags = FL_FLOCK;
303 fl->fl_type = type;
304 fl->fl_end = OFFSET_MAX;
305
306 *lock = fl;
307 return 0;
308}
309
310static int assign_type(struct file_lock *fl, int type)
311{
312 switch (type) {
313 case F_RDLCK:
314 case F_WRLCK:
315 case F_UNLCK:
316 fl->fl_type = type;
317 break;
318 default:
319 return -EINVAL;
320 }
321 return 0;
322}
323
324/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
325 * style lock.
326 */
327static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
328 struct flock *l)
329{
330 off_t start, end;
331
332 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700333 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 start = 0;
335 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700336 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 start = filp->f_pos;
338 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700339 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800340 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 break;
342 default:
343 return -EINVAL;
344 }
345
346 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
347 POSIX-2001 defines it. */
348 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 if (start < 0)
350 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700351 fl->fl_end = OFFSET_MAX;
352 if (l->l_len > 0) {
353 end = start + l->l_len - 1;
354 fl->fl_end = end;
355 } else if (l->l_len < 0) {
356 end = start - 1;
357 fl->fl_end = end;
358 start += l->l_len;
359 if (start < 0)
360 return -EINVAL;
361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700363 if (fl->fl_end < fl->fl_start)
364 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
366 fl->fl_owner = current->files;
367 fl->fl_pid = current->tgid;
368 fl->fl_file = filp;
369 fl->fl_flags = FL_POSIX;
370 fl->fl_ops = NULL;
371 fl->fl_lmops = NULL;
372
373 return assign_type(fl, l->l_type);
374}
375
376#if BITS_PER_LONG == 32
377static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
378 struct flock64 *l)
379{
380 loff_t start;
381
382 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700383 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 start = 0;
385 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700386 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 start = filp->f_pos;
388 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700389 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800390 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 break;
392 default:
393 return -EINVAL;
394 }
395
Trond Myklebust4c780a42005-10-18 14:20:21 -0700396 start += l->l_start;
397 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700399 fl->fl_end = OFFSET_MAX;
400 if (l->l_len > 0) {
401 fl->fl_end = start + l->l_len - 1;
402 } else if (l->l_len < 0) {
403 fl->fl_end = start - 1;
404 start += l->l_len;
405 if (start < 0)
406 return -EINVAL;
407 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700409 if (fl->fl_end < fl->fl_start)
410 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 fl->fl_owner = current->files;
413 fl->fl_pid = current->tgid;
414 fl->fl_file = filp;
415 fl->fl_flags = FL_POSIX;
416 fl->fl_ops = NULL;
417 fl->fl_lmops = NULL;
418
419 switch (l->l_type) {
420 case F_RDLCK:
421 case F_WRLCK:
422 case F_UNLCK:
423 fl->fl_type = l->l_type;
424 break;
425 default:
426 return -EINVAL;
427 }
428
429 return (0);
430}
431#endif
432
433/* default lease lock manager operations */
434static void lease_break_callback(struct file_lock *fl)
435{
436 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
437}
438
439static void lease_release_private_callback(struct file_lock *fl)
440{
441 if (!fl->fl_file)
442 return;
443
444 f_delown(fl->fl_file);
445 fl->fl_file->f_owner.signum = 0;
446}
447
Adrian Bunk75c96f82005-05-05 16:16:09 -0700448static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449{
450 return fl->fl_file == try->fl_file;
451}
452
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700453static const struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 .fl_break = lease_break_callback,
455 .fl_release_private = lease_release_private_callback,
456 .fl_mylease = lease_mylease_callback,
457 .fl_change = lease_modify,
458};
459
460/*
461 * Initialize a lease, use the default lock manager operations
462 */
463static int lease_init(struct file *filp, int type, struct file_lock *fl)
464 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400465 if (assign_type(fl, type) != 0)
466 return -EINVAL;
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 fl->fl_owner = current->files;
469 fl->fl_pid = current->tgid;
470
471 fl->fl_file = filp;
472 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 fl->fl_start = 0;
474 fl->fl_end = OFFSET_MAX;
475 fl->fl_ops = NULL;
476 fl->fl_lmops = &lease_manager_ops;
477 return 0;
478}
479
480/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500481static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
483 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400484 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485
486 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500487 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
489 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400490 if (error) {
491 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500492 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400493 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500494 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495}
496
497/* Check if two locks overlap each other.
498 */
499static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
500{
501 return ((fl1->fl_end >= fl2->fl_start) &&
502 (fl2->fl_end >= fl1->fl_start));
503}
504
505/*
506 * Check whether two locks have the same owner.
507 */
Matt Mackall33443c42006-01-08 01:05:22 -0800508static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509{
510 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
511 return fl2->fl_lmops == fl1->fl_lmops &&
512 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
513 return fl1->fl_owner == fl2->fl_owner;
514}
515
516/* Remove waiter from blocker's block list.
517 * When blocker ends up pointing to itself then the list is empty.
518 */
Matt Mackall33443c42006-01-08 01:05:22 -0800519static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
521 list_del_init(&waiter->fl_block);
522 list_del_init(&waiter->fl_link);
523 waiter->fl_next = NULL;
524}
525
526/*
527 */
528static void locks_delete_block(struct file_lock *waiter)
529{
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200530 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 __locks_delete_block(waiter);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200532 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533}
534
535/* Insert waiter into blocker's block list.
536 * We use a circular list so that processes can be easily woken up in
537 * the order they blocked. The documentation doesn't require this but
538 * it seems like the reasonable thing to do.
539 */
540static void locks_insert_block(struct file_lock *blocker,
541 struct file_lock *waiter)
542{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800543 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 list_add_tail(&waiter->fl_block, &blocker->fl_block);
545 waiter->fl_next = blocker;
546 if (IS_POSIX(blocker))
547 list_add(&waiter->fl_link, &blocked_list);
548}
549
550/* Wake up processes blocked waiting for blocker.
551 * If told to wait then schedule the processes until the block list
552 * is empty, otherwise empty the block list ourselves.
553 */
554static void locks_wake_up_blocks(struct file_lock *blocker)
555{
556 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400557 struct file_lock *waiter;
558
559 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 struct file_lock, fl_block);
561 __locks_delete_block(waiter);
562 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
563 waiter->fl_lmops->fl_notify(waiter);
564 else
565 wake_up(&waiter->fl_wait);
566 }
567}
568
569/* Insert file lock fl into an inode's lock list at the position indicated
570 * by pos. At the same time add the lock to the global file lock list.
571 */
572static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
573{
574 list_add(&fl->fl_link, &file_lock_list);
575
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000576 fl->fl_nspid = get_pid(task_tgid(current));
577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 /* insert into file's list */
579 fl->fl_next = *pos;
580 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581}
582
583/*
584 * Delete a lock and then free it.
585 * Wake up processes that are blocked waiting for this lock,
586 * notify the FS that the lock has been cleared and
587 * finally free the lock.
588 */
589static void locks_delete_lock(struct file_lock **thisfl_p)
590{
591 struct file_lock *fl = *thisfl_p;
592
593 *thisfl_p = fl->fl_next;
594 fl->fl_next = NULL;
595 list_del_init(&fl->fl_link);
596
597 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
598 if (fl->fl_fasync != NULL) {
599 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
600 fl->fl_fasync = NULL;
601 }
602
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000603 if (fl->fl_nspid) {
604 put_pid(fl->fl_nspid);
605 fl->fl_nspid = NULL;
606 }
607
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 locks_wake_up_blocks(fl);
609 locks_free_lock(fl);
610}
611
612/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
613 * checks for shared/exclusive status of overlapping locks.
614 */
615static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
616{
617 if (sys_fl->fl_type == F_WRLCK)
618 return 1;
619 if (caller_fl->fl_type == F_WRLCK)
620 return 1;
621 return 0;
622}
623
624/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
625 * checking before calling the locks_conflict().
626 */
627static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
628{
629 /* POSIX locks owned by the same process do not conflict with
630 * each other.
631 */
632 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
633 return (0);
634
635 /* Check whether they overlap */
636 if (!locks_overlap(caller_fl, sys_fl))
637 return 0;
638
639 return (locks_conflict(caller_fl, sys_fl));
640}
641
642/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
643 * checking before calling the locks_conflict().
644 */
645static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
646{
647 /* FLOCK locks referring to the same filp do not conflict with
648 * each other.
649 */
650 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
651 return (0);
652 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
653 return 0;
654
655 return (locks_conflict(caller_fl, sys_fl));
656}
657
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400658void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500659posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 struct file_lock *cfl;
662
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200663 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800664 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 if (!IS_POSIX(cfl))
666 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400667 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 break;
669 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000670 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500671 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000672 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800673 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000674 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400675 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200676 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400677 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679EXPORT_SYMBOL(posix_test_lock);
680
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400681/*
682 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400684 * We attempt to detect deadlocks that are due purely to posix file
685 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400687 * We assume that a task can be waiting for at most one lock at a time.
688 * So for any acquired lock, the process holding that lock may be
689 * waiting on at most one other lock. That lock in turns may be held by
690 * someone waiting for at most one other lock. Given a requested lock
691 * caller_fl which is about to wait for a conflicting lock block_fl, we
692 * follow this chain of waiters to ensure we are not about to create a
693 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400694 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400695 * Since we do this before we ever put a process to sleep on a lock, we
696 * are ensured that there is never a cycle; that is what guarantees that
697 * the while() loop in posix_locks_deadlock() eventually completes.
698 *
699 * Note: the above assumption may not be true when handling lock
700 * requests from a broken NFS client. It may also fail in the presence
701 * of tasks (such as posix threads) sharing the same open file table.
702 *
703 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400705
706#define MAX_DEADLK_ITERATIONS 10
707
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400708/* Find a lock that the owner of the given block_fl is blocking on. */
709static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
710{
711 struct file_lock *fl;
712
713 list_for_each_entry(fl, &blocked_list, fl_link) {
714 if (posix_same_owner(fl, block_fl))
715 return fl->fl_next;
716 }
717 return NULL;
718}
719
Adrian Bunkb0904e12006-06-23 02:05:13 -0700720static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 struct file_lock *block_fl)
722{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400723 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400725 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
726 if (i++ > MAX_DEADLK_ITERATIONS)
727 return 0;
728 if (posix_same_owner(caller_fl, block_fl))
729 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 }
731 return 0;
732}
733
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400735 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400736 *
737 * Note that if called with an FL_EXISTS argument, the caller may determine
738 * whether or not a lock was successfully freed by testing the return
739 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800741static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800743 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800745 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 int error = 0;
747 int found = 0;
748
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200749 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
750 new_fl = locks_alloc_lock();
751 if (!new_fl)
752 return -ENOMEM;
753 }
754
755 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400756 if (request->fl_flags & FL_ACCESS)
757 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400758
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 for_each_lock(inode, before) {
760 struct file_lock *fl = *before;
761 if (IS_POSIX(fl))
762 break;
763 if (IS_LEASE(fl))
764 continue;
765 if (filp != fl->fl_file)
766 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800767 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 goto out;
769 found = 1;
770 locks_delete_lock(before);
771 break;
772 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773
Trond Myklebustf475ae92006-06-29 16:38:32 -0400774 if (request->fl_type == F_UNLCK) {
775 if ((request->fl_flags & FL_EXISTS) && !found)
776 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800777 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 /*
781 * If a higher-priority process was blocked on the old file lock,
782 * give it the opportunity to lock the file.
783 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200784 if (found) {
785 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200786 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200787 lock_flocks();
788 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400790find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 for_each_lock(inode, before) {
792 struct file_lock *fl = *before;
793 if (IS_POSIX(fl))
794 break;
795 if (IS_LEASE(fl))
796 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800797 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 continue;
799 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700800 if (!(request->fl_flags & FL_SLEEP))
801 goto out;
802 error = FILE_LOCK_DEFERRED;
803 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 goto out;
805 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400806 if (request->fl_flags & FL_ACCESS)
807 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800808 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700809 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800810 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400811 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812
813out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200814 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800815 if (new_fl)
816 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 return error;
818}
819
Marc Eshel150b3932007-01-18 16:15:35 -0500820static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821{
822 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700823 struct file_lock *new_fl = NULL;
824 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 struct file_lock *left = NULL;
826 struct file_lock *right = NULL;
827 struct file_lock **before;
828 int error, added = 0;
829
830 /*
831 * We may need two file_lock structures for this operation,
832 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700833 *
834 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700836 if (!(request->fl_flags & FL_ACCESS) &&
837 (request->fl_type != F_UNLCK ||
838 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
839 new_fl = locks_alloc_lock();
840 new_fl2 = locks_alloc_lock();
841 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200843 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (request->fl_type != F_UNLCK) {
845 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500846 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 if (!IS_POSIX(fl))
848 continue;
849 if (!posix_locks_conflict(request, fl))
850 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800851 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400852 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 error = -EAGAIN;
854 if (!(request->fl_flags & FL_SLEEP))
855 goto out;
856 error = -EDEADLK;
857 if (posix_locks_deadlock(request, fl))
858 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700859 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 locks_insert_block(fl, request);
861 goto out;
862 }
863 }
864
865 /* If we're just looking for a conflict, we're done. */
866 error = 0;
867 if (request->fl_flags & FL_ACCESS)
868 goto out;
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 * Find the first old lock with the same owner as the new lock.
872 */
873
874 before = &inode->i_flock;
875
876 /* First skip locks owned by other processes. */
877 while ((fl = *before) && (!IS_POSIX(fl) ||
878 !posix_same_owner(request, fl))) {
879 before = &fl->fl_next;
880 }
881
882 /* Process locks with this owner. */
883 while ((fl = *before) && posix_same_owner(request, fl)) {
884 /* Detect adjacent or overlapping regions (if same lock type)
885 */
886 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700887 /* In all comparisons of start vs end, use
888 * "start - 1" rather than "end + 1". If end
889 * is OFFSET_MAX, end + 1 will become negative.
890 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 if (fl->fl_end < request->fl_start - 1)
892 goto next_lock;
893 /* If the next lock in the list has entirely bigger
894 * addresses than the new one, insert the lock here.
895 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700896 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 break;
898
899 /* If we come here, the new and old lock are of the
900 * same type and adjacent or overlapping. Make one
901 * lock yielding from the lower start address of both
902 * locks to the higher end address.
903 */
904 if (fl->fl_start > request->fl_start)
905 fl->fl_start = request->fl_start;
906 else
907 request->fl_start = fl->fl_start;
908 if (fl->fl_end < request->fl_end)
909 fl->fl_end = request->fl_end;
910 else
911 request->fl_end = fl->fl_end;
912 if (added) {
913 locks_delete_lock(before);
914 continue;
915 }
916 request = fl;
917 added = 1;
918 }
919 else {
920 /* Processing for different lock types is a bit
921 * more complex.
922 */
923 if (fl->fl_end < request->fl_start)
924 goto next_lock;
925 if (fl->fl_start > request->fl_end)
926 break;
927 if (request->fl_type == F_UNLCK)
928 added = 1;
929 if (fl->fl_start < request->fl_start)
930 left = fl;
931 /* If the next lock in the list has a higher end
932 * address than the new one, insert the new one here.
933 */
934 if (fl->fl_end > request->fl_end) {
935 right = fl;
936 break;
937 }
938 if (fl->fl_start >= request->fl_start) {
939 /* The new lock completely replaces an old
940 * one (This may happen several times).
941 */
942 if (added) {
943 locks_delete_lock(before);
944 continue;
945 }
946 /* Replace the old lock with the new one.
947 * Wake up anybody waiting for the old one,
948 * as the change in lock type might satisfy
949 * their needs.
950 */
951 locks_wake_up_blocks(fl);
952 fl->fl_start = request->fl_start;
953 fl->fl_end = request->fl_end;
954 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500955 locks_release_private(fl);
956 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 request = fl;
958 added = 1;
959 }
960 }
961 /* Go on to next lock.
962 */
963 next_lock:
964 before = &fl->fl_next;
965 }
966
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700967 /*
968 * The above code only modifies existing locks in case of
969 * merging or replacing. If new lock(s) need to be inserted
970 * all modifications are done bellow this, so it's safe yet to
971 * bail out.
972 */
973 error = -ENOLCK; /* "no luck" */
974 if (right && left == right && !new_fl2)
975 goto out;
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 error = 0;
978 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400979 if (request->fl_type == F_UNLCK) {
980 if (request->fl_flags & FL_EXISTS)
981 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400983 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700984
985 if (!new_fl) {
986 error = -ENOLCK;
987 goto out;
988 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 locks_copy_lock(new_fl, request);
990 locks_insert_lock(before, new_fl);
991 new_fl = NULL;
992 }
993 if (right) {
994 if (left == right) {
995 /* The new lock breaks the old one in two pieces,
996 * so we have to use the second new lock.
997 */
998 left = new_fl2;
999 new_fl2 = NULL;
1000 locks_copy_lock(left, right);
1001 locks_insert_lock(before, left);
1002 }
1003 right->fl_start = request->fl_end + 1;
1004 locks_wake_up_blocks(right);
1005 }
1006 if (left) {
1007 left->fl_end = request->fl_start - 1;
1008 locks_wake_up_blocks(left);
1009 }
1010 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001011 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /*
1013 * Free any unused locks.
1014 */
1015 if (new_fl)
1016 locks_free_lock(new_fl);
1017 if (new_fl2)
1018 locks_free_lock(new_fl2);
1019 return error;
1020}
1021
1022/**
1023 * posix_lock_file - Apply a POSIX-style lock to a file
1024 * @filp: The file to apply the lock to
1025 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001026 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 *
1028 * Add a POSIX style lock to a file.
1029 * We merge adjacent & overlapping locks whenever possible.
1030 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001031 *
1032 * Note that if called with an FL_EXISTS argument, the caller may determine
1033 * whether or not a lock was successfully freed by testing the return
1034 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 */
Marc Eshel150b3932007-01-18 16:15:35 -05001036int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001037 struct file_lock *conflock)
1038{
Marc Eshel150b3932007-01-18 16:15:35 -05001039 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001040}
Marc Eshel150b3932007-01-18 16:15:35 -05001041EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043/**
1044 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1045 * @filp: The file to apply the lock to
1046 * @fl: The lock to be applied
1047 *
1048 * Add a POSIX style lock to a file.
1049 * We merge adjacent & overlapping locks whenever possible.
1050 * POSIX locks are sorted by owner task, then by starting address
1051 */
1052int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1053{
1054 int error;
1055 might_sleep ();
1056 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001057 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001058 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 break;
1060 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1061 if (!error)
1062 continue;
1063
1064 locks_delete_block(fl);
1065 break;
1066 }
1067 return error;
1068}
1069EXPORT_SYMBOL(posix_lock_file_wait);
1070
1071/**
1072 * locks_mandatory_locked - Check for an active lock
1073 * @inode: the file to check
1074 *
1075 * Searches the inode's list of locks to find any POSIX locks which conflict.
1076 * This function is called from locks_verify_locked() only.
1077 */
1078int locks_mandatory_locked(struct inode *inode)
1079{
1080 fl_owner_t owner = current->files;
1081 struct file_lock *fl;
1082
1083 /*
1084 * Search the lock list for this inode for any POSIX locks.
1085 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001086 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1088 if (!IS_POSIX(fl))
1089 continue;
1090 if (fl->fl_owner != owner)
1091 break;
1092 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001093 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 return fl ? -EAGAIN : 0;
1095}
1096
1097/**
1098 * locks_mandatory_area - Check for a conflicting lock
1099 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1100 * for shared
1101 * @inode: the file to check
1102 * @filp: how the file was opened (if it was)
1103 * @offset: start of area to check
1104 * @count: length of area to check
1105 *
1106 * Searches the inode's list of locks to find any POSIX locks which conflict.
1107 * This function is called from rw_verify_area() and
1108 * locks_verify_truncate().
1109 */
1110int locks_mandatory_area(int read_write, struct inode *inode,
1111 struct file *filp, loff_t offset,
1112 size_t count)
1113{
1114 struct file_lock fl;
1115 int error;
1116
1117 locks_init_lock(&fl);
1118 fl.fl_owner = current->files;
1119 fl.fl_pid = current->tgid;
1120 fl.fl_file = filp;
1121 fl.fl_flags = FL_POSIX | FL_ACCESS;
1122 if (filp && !(filp->f_flags & O_NONBLOCK))
1123 fl.fl_flags |= FL_SLEEP;
1124 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1125 fl.fl_start = offset;
1126 fl.fl_end = offset + count - 1;
1127
1128 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001129 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001130 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 break;
1132 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1133 if (!error) {
1134 /*
1135 * If we've been sleeping someone might have
1136 * changed the permissions behind our back.
1137 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001138 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 continue;
1140 }
1141
1142 locks_delete_block(&fl);
1143 break;
1144 }
1145
1146 return error;
1147}
1148
1149EXPORT_SYMBOL(locks_mandatory_area);
1150
1151/* We already had a lease on this file; just change its type */
1152int lease_modify(struct file_lock **before, int arg)
1153{
1154 struct file_lock *fl = *before;
1155 int error = assign_type(fl, arg);
1156
1157 if (error)
1158 return error;
1159 locks_wake_up_blocks(fl);
1160 if (arg == F_UNLCK)
1161 locks_delete_lock(before);
1162 return 0;
1163}
1164
1165EXPORT_SYMBOL(lease_modify);
1166
1167static void time_out_leases(struct inode *inode)
1168{
1169 struct file_lock **before;
1170 struct file_lock *fl;
1171
1172 before = &inode->i_flock;
1173 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1174 if ((fl->fl_break_time == 0)
1175 || time_before(jiffies, fl->fl_break_time)) {
1176 before = &fl->fl_next;
1177 continue;
1178 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1180 if (fl == *before) /* lease_modify may have freed fl */
1181 before = &fl->fl_next;
1182 }
1183}
1184
1185/**
1186 * __break_lease - revoke all outstanding leases on file
1187 * @inode: the inode of the file to return
1188 * @mode: the open mode (read or write)
1189 *
david m. richter87250dd2007-05-09 16:10:27 -04001190 * break_lease (inlined for speed) has checked there already is at least
1191 * some kind of lock (maybe a lease) on this file. Leases are broken on
1192 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 * specified %O_NONBLOCK to your open().
1194 */
1195int __break_lease(struct inode *inode, unsigned int mode)
1196{
1197 int error = 0, future;
1198 struct file_lock *new_fl, *flock;
1199 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 unsigned long break_time;
1201 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001202 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
Al Viro8737c932009-12-24 06:47:55 -05001204 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001206 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207
1208 time_out_leases(inode);
1209
1210 flock = inode->i_flock;
1211 if ((flock == NULL) || !IS_LEASE(flock))
1212 goto out;
1213
1214 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1215 if (fl->fl_owner == current->files)
1216 i_have_this_lease = 1;
1217
Al Viro8737c932009-12-24 06:47:55 -05001218 if (want_write) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 /* If we want write access, we have to revoke any lease. */
1220 future = F_UNLCK | F_INPROGRESS;
1221 } else if (flock->fl_type & F_INPROGRESS) {
1222 /* If the lease is already being broken, we just leave it */
1223 future = flock->fl_type;
1224 } else if (flock->fl_type & F_WRLCK) {
1225 /* Downgrade the exclusive lease to a read-only lease. */
1226 future = F_RDLCK | F_INPROGRESS;
1227 } else {
1228 /* the existing lease was read-only, so we can read too. */
1229 goto out;
1230 }
1231
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001232 if (IS_ERR(new_fl) && !i_have_this_lease
1233 && ((mode & O_NONBLOCK) == 0)) {
1234 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 goto out;
1236 }
1237
1238 break_time = 0;
1239 if (lease_break_time > 0) {
1240 break_time = jiffies + lease_break_time * HZ;
1241 if (break_time == 0)
1242 break_time++; /* so that 0 means no break time */
1243 }
1244
1245 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1246 if (fl->fl_type != future) {
1247 fl->fl_type = future;
1248 fl->fl_break_time = break_time;
1249 /* lease must have lmops break callback */
1250 fl->fl_lmops->fl_break(fl);
1251 }
1252 }
1253
1254 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1255 error = -EWOULDBLOCK;
1256 goto out;
1257 }
1258
1259restart:
1260 break_time = flock->fl_break_time;
1261 if (break_time != 0) {
1262 break_time -= jiffies;
1263 if (break_time == 0)
1264 break_time++;
1265 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001266 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001267 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001268 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1269 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001270 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001271 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 if (error >= 0) {
1273 if (error == 0)
1274 time_out_leases(inode);
1275 /* Wait for the next lease that has not been broken yet */
1276 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1277 flock = flock->fl_next) {
1278 if (flock->fl_type & F_INPROGRESS)
1279 goto restart;
1280 }
1281 error = 0;
1282 }
1283
1284out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001285 unlock_flocks();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001286 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 locks_free_lock(new_fl);
1288 return error;
1289}
1290
1291EXPORT_SYMBOL(__break_lease);
1292
1293/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001294 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 * @inode: the inode
1296 * @time: pointer to a timespec which will contain the last modified time
1297 *
1298 * This is to force NFS clients to flush their caches for files with
1299 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001300 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301 */
1302void lease_get_mtime(struct inode *inode, struct timespec *time)
1303{
1304 struct file_lock *flock = inode->i_flock;
1305 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1306 *time = current_fs_time(inode->i_sb);
1307 else
1308 *time = inode->i_mtime;
1309}
1310
1311EXPORT_SYMBOL(lease_get_mtime);
1312
1313/**
1314 * fcntl_getlease - Enquire what lease is currently active
1315 * @filp: the file
1316 *
1317 * The value returned by this function will be one of
1318 * (if no lease break is pending):
1319 *
1320 * %F_RDLCK to indicate a shared lease is held.
1321 *
1322 * %F_WRLCK to indicate an exclusive lease is held.
1323 *
1324 * %F_UNLCK to indicate no lease is held.
1325 *
1326 * (if a lease break is pending):
1327 *
1328 * %F_RDLCK to indicate an exclusive lease needs to be
1329 * changed to a shared lease (or removed).
1330 *
1331 * %F_UNLCK to indicate the lease needs to be removed.
1332 *
1333 * XXX: sfr & willy disagree over whether F_INPROGRESS
1334 * should be returned to userspace.
1335 */
1336int fcntl_getlease(struct file *filp)
1337{
1338 struct file_lock *fl;
1339 int type = F_UNLCK;
1340
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001341 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001342 time_out_leases(filp->f_path.dentry->d_inode);
1343 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 fl = fl->fl_next) {
1345 if (fl->fl_file == filp) {
1346 type = fl->fl_type & ~F_INPROGRESS;
1347 break;
1348 }
1349 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001350 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 return type;
1352}
1353
1354/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001355 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 * @filp: file pointer
1357 * @arg: type of lease to obtain
1358 * @flp: input - file_lock to use, output - file_lock inserted
1359 *
1360 * The (input) flp->fl_lmops->fl_break function is required
1361 * by break_lease().
1362 *
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001363 * Called with file_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001365int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001367 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001368 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 struct inode *inode = dentry->d_inode;
1370 int error, rdlease_count = 0, wrlease_count = 0;
1371
J. Bruce Fields096657b2010-10-30 17:31:14 -04001372 lease = *flp;
1373
1374 error = -EACCES;
David Howellsda9592e2008-11-14 10:39:05 +11001375 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001376 goto out;
1377 error = -EINVAL;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001378 if (!S_ISREG(inode->i_mode))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001379 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001380 error = security_file_lock(filp, arg);
1381 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001382 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001383
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 time_out_leases(inode);
1385
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001386 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387
David M. Richter288b2fd2008-04-23 16:29:00 -04001388 if (arg != F_UNLCK) {
1389 error = -EAGAIN;
1390 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1391 goto out;
1392 if ((arg == F_WRLCK)
1393 && ((atomic_read(&dentry->d_count) > 1)
1394 || (atomic_read(&inode->i_count) > 1)))
1395 goto out;
David M. Richter288b2fd2008-04-23 16:29:00 -04001396 }
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001397
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 /*
1399 * At this point, we know that if there is an exclusive
1400 * lease on this file, then we hold it on this filp
1401 * (otherwise our open of this file would have blocked).
1402 * And if we are trying to acquire an exclusive lease,
1403 * then the file is not open by anyone (including us)
1404 * except for this filp.
1405 */
1406 for (before = &inode->i_flock;
1407 ((fl = *before) != NULL) && IS_LEASE(fl);
1408 before = &fl->fl_next) {
1409 if (lease->fl_lmops->fl_mylease(fl, lease))
1410 my_before = before;
1411 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1412 /*
1413 * Someone is in the process of opening this
1414 * file for writing so we may not take an
1415 * exclusive lease on it.
1416 */
1417 wrlease_count++;
1418 else
1419 rdlease_count++;
1420 }
1421
David M. Richter5fcc60c2008-04-23 16:28:59 -04001422 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1424 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1425 goto out;
1426
1427 if (my_before != NULL) {
1428 error = lease->fl_lmops->fl_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001429 if (!error)
1430 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 goto out;
1432 }
1433
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 if (arg == F_UNLCK)
1435 goto out;
1436
1437 error = -EINVAL;
1438 if (!leases_enable)
1439 goto out;
1440
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001441 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001442 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444out:
1445 return error;
1446}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001447EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001449static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1450{
1451 if (filp->f_op && filp->f_op->setlease)
1452 return filp->f_op->setlease(filp, arg, lease);
1453 else
1454 return generic_setlease(filp, arg, lease);
1455}
1456
1457/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001458 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 * @filp: file pointer
1460 * @arg: type of lease to obtain
1461 * @lease: file_lock to use
1462 *
1463 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001464 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1465 * break_lease will oops!
1466 *
1467 * This will call the filesystem's setlease file method, if
1468 * defined. Note that there is no getlease method; instead, the
1469 * filesystem setlease method should call back to setlease() to
1470 * add a lease to the inode's lease list, where fcntl_getlease() can
1471 * find it. Since fcntl_getlease() only reports whether the current
1472 * task holds a lease, a cluster filesystem need only do this for
1473 * leases held by processes on this node.
1474 *
1475 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001476 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001477 * filesystem's open, create, and (on truncate) setattr methods.
1478 *
1479 * Warning: the only current setlease methods exist only to disable
1480 * leases in certain cases. More vfs changes may be required to
1481 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 */
1483
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001484int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 int error;
1487
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001488 lock_flocks();
1489 error = __vfs_setlease(filp, arg, lease);
1490 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
1492 return error;
1493}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001494EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001496static int do_fcntl_delete_lease(struct file *filp)
1497{
1498 struct file_lock fl, *flp = &fl;
1499
1500 lease_init(filp, F_UNLCK, flp);
1501
1502 return vfs_setlease(filp, F_UNLCK, &flp);
1503}
1504
1505static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001507 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001508 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 int error;
1510
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001511 fl = lease_alloc(filp, arg);
1512 if (IS_ERR(fl))
1513 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001515 new = fasync_alloc();
1516 if (!new) {
1517 locks_free_lock(fl);
1518 return -ENOMEM;
1519 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001520 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001521 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001522 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001523 if (error) {
1524 unlock_flocks();
1525 locks_free_lock(fl);
1526 goto out_free_fasync;
1527 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001528 if (ret != fl)
1529 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001530
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001531 /*
1532 * fasync_insert_entry() returns the old entry if any.
1533 * If there was no old entry, then it used 'new' and
1534 * inserted it into the fasync list. Clear new so that
1535 * we don't release it here.
1536 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001537 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001538 new = NULL;
1539
J. Bruce Fields8896b932010-11-03 18:09:18 -04001540 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001541 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001542
1543out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001544 if (new)
1545 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 return error;
1547}
1548
1549/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001550 * fcntl_setlease - sets a lease on an open file
1551 * @fd: open file descriptor
1552 * @filp: file pointer
1553 * @arg: type of lease to obtain
1554 *
1555 * Call this fcntl to establish a lease on the file.
1556 * Note that you also need to call %F_SETSIG to
1557 * receive a signal when the lease is broken.
1558 */
1559int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1560{
1561 if (arg == F_UNLCK)
1562 return do_fcntl_delete_lease(filp);
1563 return do_fcntl_add_lease(fd, filp, arg);
1564}
1565
1566/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1568 * @filp: The file to apply the lock to
1569 * @fl: The lock to be applied
1570 *
1571 * Add a FLOCK style lock to a file.
1572 */
1573int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1574{
1575 int error;
1576 might_sleep();
1577 for (;;) {
1578 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001579 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 break;
1581 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1582 if (!error)
1583 continue;
1584
1585 locks_delete_block(fl);
1586 break;
1587 }
1588 return error;
1589}
1590
1591EXPORT_SYMBOL(flock_lock_file_wait);
1592
1593/**
1594 * sys_flock: - flock() system call.
1595 * @fd: the file descriptor to lock.
1596 * @cmd: the type of lock to apply.
1597 *
1598 * Apply a %FL_FLOCK style lock to an open file descriptor.
1599 * The @cmd can be one of
1600 *
1601 * %LOCK_SH -- a shared lock.
1602 *
1603 * %LOCK_EX -- an exclusive lock.
1604 *
1605 * %LOCK_UN -- remove an existing lock.
1606 *
1607 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1608 *
1609 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1610 * processes read and write access respectively.
1611 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001612SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613{
1614 struct file *filp;
1615 struct file_lock *lock;
1616 int can_sleep, unlock;
1617 int error;
1618
1619 error = -EBADF;
1620 filp = fget(fd);
1621 if (!filp)
1622 goto out;
1623
1624 can_sleep = !(cmd & LOCK_NB);
1625 cmd &= ~LOCK_NB;
1626 unlock = (cmd == LOCK_UN);
1627
Al Viroaeb5d722008-09-02 15:28:45 -04001628 if (!unlock && !(cmd & LOCK_MAND) &&
1629 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 goto out_putf;
1631
1632 error = flock_make_lock(filp, &lock, cmd);
1633 if (error)
1634 goto out_putf;
1635 if (can_sleep)
1636 lock->fl_flags |= FL_SLEEP;
1637
Sten Spans713c0ec2009-07-16 09:41:39 +02001638 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 if (error)
1640 goto out_free;
1641
1642 if (filp->f_op && filp->f_op->flock)
1643 error = filp->f_op->flock(filp,
1644 (can_sleep) ? F_SETLKW : F_SETLK,
1645 lock);
1646 else
1647 error = flock_lock_file_wait(filp, lock);
1648
1649 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001650 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652 out_putf:
1653 fput(filp);
1654 out:
1655 return error;
1656}
1657
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001658/**
1659 * vfs_test_lock - test file byte range lock
1660 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001661 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001662 *
1663 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1664 * setting conf->fl_type to something other than F_UNLCK.
1665 */
1666int vfs_test_lock(struct file *filp, struct file_lock *fl)
1667{
1668 if (filp->f_op && filp->f_op->lock)
1669 return filp->f_op->lock(filp, F_GETLK, fl);
1670 posix_test_lock(filp, fl);
1671 return 0;
1672}
1673EXPORT_SYMBOL_GPL(vfs_test_lock);
1674
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001675static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1676{
1677 flock->l_pid = fl->fl_pid;
1678#if BITS_PER_LONG == 32
1679 /*
1680 * Make sure we can represent the posix lock via
1681 * legacy 32bit flock.
1682 */
1683 if (fl->fl_start > OFFT_OFFSET_MAX)
1684 return -EOVERFLOW;
1685 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1686 return -EOVERFLOW;
1687#endif
1688 flock->l_start = fl->fl_start;
1689 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1690 fl->fl_end - fl->fl_start + 1;
1691 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001692 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001693 return 0;
1694}
1695
1696#if BITS_PER_LONG == 32
1697static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1698{
1699 flock->l_pid = fl->fl_pid;
1700 flock->l_start = fl->fl_start;
1701 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1702 fl->fl_end - fl->fl_start + 1;
1703 flock->l_whence = 0;
1704 flock->l_type = fl->fl_type;
1705}
1706#endif
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708/* Report the first existing lock that would conflict with l.
1709 * This implements the F_GETLK command of fcntl().
1710 */
1711int fcntl_getlk(struct file *filp, struct flock __user *l)
1712{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001713 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 struct flock flock;
1715 int error;
1716
1717 error = -EFAULT;
1718 if (copy_from_user(&flock, l, sizeof(flock)))
1719 goto out;
1720 error = -EINVAL;
1721 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1722 goto out;
1723
1724 error = flock_to_posix_lock(filp, &file_lock, &flock);
1725 if (error)
1726 goto out;
1727
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001728 error = vfs_test_lock(filp, &file_lock);
1729 if (error)
1730 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001732 flock.l_type = file_lock.fl_type;
1733 if (file_lock.fl_type != F_UNLCK) {
1734 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001735 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 }
1738 error = -EFAULT;
1739 if (!copy_to_user(l, &flock, sizeof(flock)))
1740 error = 0;
1741out:
1742 return error;
1743}
1744
Marc Eshel7723ec92007-01-18 15:08:55 -05001745/**
1746 * vfs_lock_file - file byte range lock
1747 * @filp: The file to apply the lock to
1748 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1749 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001750 * @conf: Place to return a copy of the conflicting lock, if found.
1751 *
1752 * A caller that doesn't care about the conflicting lock may pass NULL
1753 * as the final argument.
1754 *
1755 * If the filesystem defines a private ->lock() method, then @conf will
1756 * be left unchanged; so a caller that cares should initialize it to
1757 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001758 *
1759 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1760 * locks, the ->lock() interface may return asynchronously, before the lock has
1761 * been granted or denied by the underlying filesystem, if (and only if)
1762 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1763 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1764 * the request is for a blocking lock. When ->lock() does return asynchronously,
Miklos Szeredibde74e42008-07-25 01:48:57 -07001765 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001766 * request completes.
1767 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001768 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1769 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001770 * nonzero return code and the file system should release the lock. The file
1771 * system is also responsible to keep a corresponding posix lock when it
1772 * grants a lock so the VFS can find out which locks are locally held and do
1773 * the correct lock cleanup when required.
1774 * The underlying filesystem must not drop the kernel lock or call
Miklos Szeredibde74e42008-07-25 01:48:57 -07001775 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001776 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001777 */
Marc Eshel150b3932007-01-18 16:15:35 -05001778int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001779{
1780 if (filp->f_op && filp->f_op->lock)
1781 return filp->f_op->lock(filp, cmd, fl);
1782 else
Marc Eshel150b3932007-01-18 16:15:35 -05001783 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001784}
1785EXPORT_SYMBOL_GPL(vfs_lock_file);
1786
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001787static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1788 struct file_lock *fl)
1789{
1790 int error;
1791
1792 error = security_file_lock(filp, fl->fl_type);
1793 if (error)
1794 return error;
1795
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001796 for (;;) {
1797 error = vfs_lock_file(filp, cmd, fl, NULL);
1798 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001799 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001800 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1801 if (!error)
1802 continue;
1803
1804 locks_delete_block(fl);
1805 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001806 }
1807
1808 return error;
1809}
1810
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811/* Apply the lock described by l to an open file descriptor.
1812 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1813 */
Peter Staubachc2936212005-07-27 11:45:09 -07001814int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1815 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816{
1817 struct file_lock *file_lock = locks_alloc_lock();
1818 struct flock flock;
1819 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001820 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 int error;
1822
1823 if (file_lock == NULL)
1824 return -ENOLCK;
1825
1826 /*
1827 * This might block, so we do it before checking the inode.
1828 */
1829 error = -EFAULT;
1830 if (copy_from_user(&flock, l, sizeof(flock)))
1831 goto out;
1832
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001833 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001834
1835 /* Don't allow mandatory locks on files that may be memory mapped
1836 * and shared.
1837 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001838 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 error = -EAGAIN;
1840 goto out;
1841 }
1842
Peter Staubachc2936212005-07-27 11:45:09 -07001843again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 error = flock_to_posix_lock(filp, file_lock, &flock);
1845 if (error)
1846 goto out;
1847 if (cmd == F_SETLKW) {
1848 file_lock->fl_flags |= FL_SLEEP;
1849 }
1850
1851 error = -EBADF;
1852 switch (flock.l_type) {
1853 case F_RDLCK:
1854 if (!(filp->f_mode & FMODE_READ))
1855 goto out;
1856 break;
1857 case F_WRLCK:
1858 if (!(filp->f_mode & FMODE_WRITE))
1859 goto out;
1860 break;
1861 case F_UNLCK:
1862 break;
1863 default:
1864 error = -EINVAL;
1865 goto out;
1866 }
1867
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001868 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869
Peter Staubachc2936212005-07-27 11:45:09 -07001870 /*
1871 * Attempt to detect a close/fcntl race and recover by
1872 * releasing the lock that was just acquired.
1873 */
Al Viro0b2bac22008-05-06 13:58:34 -04001874 /*
1875 * we need that spin_lock here - it prevents reordering between
1876 * update of inode->i_flock and check for it done in close().
1877 * rcu_read_lock() wouldn't do.
1878 */
1879 spin_lock(&current->files->file_lock);
1880 f = fcheck(fd);
1881 spin_unlock(&current->files->file_lock);
1882 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001883 flock.l_type = F_UNLCK;
1884 goto again;
1885 }
1886
1887out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 locks_free_lock(file_lock);
1889 return error;
1890}
1891
1892#if BITS_PER_LONG == 32
1893/* Report the first existing lock that would conflict with l.
1894 * This implements the F_GETLK command of fcntl().
1895 */
1896int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1897{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001898 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 struct flock64 flock;
1900 int error;
1901
1902 error = -EFAULT;
1903 if (copy_from_user(&flock, l, sizeof(flock)))
1904 goto out;
1905 error = -EINVAL;
1906 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1907 goto out;
1908
1909 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1910 if (error)
1911 goto out;
1912
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001913 error = vfs_test_lock(filp, &file_lock);
1914 if (error)
1915 goto out;
1916
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001917 flock.l_type = file_lock.fl_type;
1918 if (file_lock.fl_type != F_UNLCK)
1919 posix_lock_to_flock64(&flock, &file_lock);
1920
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 error = -EFAULT;
1922 if (!copy_to_user(l, &flock, sizeof(flock)))
1923 error = 0;
1924
1925out:
1926 return error;
1927}
1928
1929/* Apply the lock described by l to an open file descriptor.
1930 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1931 */
Peter Staubachc2936212005-07-27 11:45:09 -07001932int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1933 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
1935 struct file_lock *file_lock = locks_alloc_lock();
1936 struct flock64 flock;
1937 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001938 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 int error;
1940
1941 if (file_lock == NULL)
1942 return -ENOLCK;
1943
1944 /*
1945 * This might block, so we do it before checking the inode.
1946 */
1947 error = -EFAULT;
1948 if (copy_from_user(&flock, l, sizeof(flock)))
1949 goto out;
1950
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001951 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
1953 /* Don't allow mandatory locks on files that may be memory mapped
1954 * and shared.
1955 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001956 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 error = -EAGAIN;
1958 goto out;
1959 }
1960
Peter Staubachc2936212005-07-27 11:45:09 -07001961again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 error = flock64_to_posix_lock(filp, file_lock, &flock);
1963 if (error)
1964 goto out;
1965 if (cmd == F_SETLKW64) {
1966 file_lock->fl_flags |= FL_SLEEP;
1967 }
1968
1969 error = -EBADF;
1970 switch (flock.l_type) {
1971 case F_RDLCK:
1972 if (!(filp->f_mode & FMODE_READ))
1973 goto out;
1974 break;
1975 case F_WRLCK:
1976 if (!(filp->f_mode & FMODE_WRITE))
1977 goto out;
1978 break;
1979 case F_UNLCK:
1980 break;
1981 default:
1982 error = -EINVAL;
1983 goto out;
1984 }
1985
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001986 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Peter Staubachc2936212005-07-27 11:45:09 -07001988 /*
1989 * Attempt to detect a close/fcntl race and recover by
1990 * releasing the lock that was just acquired.
1991 */
Al Viro0b2bac22008-05-06 13:58:34 -04001992 spin_lock(&current->files->file_lock);
1993 f = fcheck(fd);
1994 spin_unlock(&current->files->file_lock);
1995 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001996 flock.l_type = F_UNLCK;
1997 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 }
1999
2000out:
2001 locks_free_lock(file_lock);
2002 return error;
2003}
2004#endif /* BITS_PER_LONG == 32 */
2005
2006/*
2007 * This function is called when the file is being removed
2008 * from the task's fd array. POSIX locks belonging to this task
2009 * are deleted at this time.
2010 */
2011void locks_remove_posix(struct file *filp, fl_owner_t owner)
2012{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002013 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
2015 /*
2016 * If there are no locks held on this file, we don't need to call
2017 * posix_lock_file(). Another process could be setting a lock on this
2018 * file at the same time, but we wouldn't remove that lock anyway.
2019 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002020 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 return;
2022
2023 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002024 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 lock.fl_start = 0;
2026 lock.fl_end = OFFSET_MAX;
2027 lock.fl_owner = owner;
2028 lock.fl_pid = current->tgid;
2029 lock.fl_file = filp;
2030 lock.fl_ops = NULL;
2031 lock.fl_lmops = NULL;
2032
Marc Eshel150b3932007-01-18 16:15:35 -05002033 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2036 lock.fl_ops->fl_release_private(&lock);
2037}
2038
2039EXPORT_SYMBOL(locks_remove_posix);
2040
2041/*
2042 * This function is called on the last close of an open file.
2043 */
2044void locks_remove_flock(struct file *filp)
2045{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002046 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 struct file_lock *fl;
2048 struct file_lock **before;
2049
2050 if (!inode->i_flock)
2051 return;
2052
2053 if (filp->f_op && filp->f_op->flock) {
2054 struct file_lock fl = {
2055 .fl_pid = current->tgid,
2056 .fl_file = filp,
2057 .fl_flags = FL_FLOCK,
2058 .fl_type = F_UNLCK,
2059 .fl_end = OFFSET_MAX,
2060 };
2061 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002062 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2063 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 }
2065
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002066 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002067 before = &inode->i_flock;
2068
2069 while ((fl = *before) != NULL) {
2070 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002071 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 locks_delete_lock(before);
2073 continue;
2074 }
2075 if (IS_LEASE(fl)) {
2076 lease_modify(before, F_UNLCK);
2077 continue;
2078 }
2079 /* What? */
2080 BUG();
2081 }
2082 before = &fl->fl_next;
2083 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002084 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085}
2086
2087/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088 * posix_unblock_lock - stop waiting for a file lock
2089 * @filp: how the file was opened
2090 * @waiter: the lock which was waiting
2091 *
2092 * lockd needs to block waiting for locks.
2093 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002094int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2096{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002097 int status = 0;
2098
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002099 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002100 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002102 else
2103 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002104 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002105 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106}
2107
2108EXPORT_SYMBOL(posix_unblock_lock);
2109
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002110/**
2111 * vfs_cancel_lock - file byte range unblock lock
2112 * @filp: The file to apply the unblock to
2113 * @fl: The lock to be unblocked
2114 *
2115 * Used by lock managers to cancel blocked requests
2116 */
2117int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2118{
2119 if (filp->f_op && filp->f_op->lock)
2120 return filp->f_op->lock(filp, F_CANCELLK, fl);
2121 return 0;
2122}
2123
2124EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2125
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002126#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002127#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002128#include <linux/seq_file.h>
2129
2130static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002131 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132{
2133 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002134 unsigned int fl_pid;
2135
2136 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002137 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002138 else
2139 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002142 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143
Jerome Marchand99dc8292010-10-26 14:22:33 -07002144 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002146 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2148 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002149 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 } else if (IS_FLOCK(fl)) {
2151 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002152 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002154 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 }
2156 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002157 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002159 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002160 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002161 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002163 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002165 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 }
2167 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002168 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 (fl->fl_type & LOCK_READ)
2170 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2171 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2172 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002173 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 (fl->fl_type & F_INPROGRESS)
2175 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2176 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2177 }
2178 if (inode) {
2179#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002180 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 inode->i_sb->s_id, inode->i_ino);
2182#else
2183 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002184 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002185 MAJOR(inode->i_sb->s_dev),
2186 MINOR(inode->i_sb->s_dev), inode->i_ino);
2187#endif
2188 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002189 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 }
2191 if (IS_POSIX(fl)) {
2192 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002193 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002195 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002197 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 }
2199}
2200
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002201static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002203 struct file_lock *fl, *bfl;
2204
2205 fl = list_entry(v, struct file_lock, fl_link);
2206
Jerome Marchand99dc8292010-10-26 14:22:33 -07002207 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002208
2209 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002210 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002211
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002212 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213}
2214
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002215static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002217 loff_t *p = f->private;
2218
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002219 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002220 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002221 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222}
2223
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002224static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2225{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002226 loff_t *p = f->private;
2227 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002228 return seq_list_next(v, &file_lock_list, pos);
2229}
2230
2231static void locks_stop(struct seq_file *f, void *v)
2232{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002233 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002234}
2235
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002236static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002237 .start = locks_start,
2238 .next = locks_next,
2239 .stop = locks_stop,
2240 .show = locks_show,
2241};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002242
2243static int locks_open(struct inode *inode, struct file *filp)
2244{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002245 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002246}
2247
2248static const struct file_operations proc_locks_operations = {
2249 .open = locks_open,
2250 .read = seq_read,
2251 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002252 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002253};
2254
2255static int __init proc_locks_init(void)
2256{
2257 proc_create("locks", 0, NULL, &proc_locks_operations);
2258 return 0;
2259}
2260module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002261#endif
2262
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263/**
2264 * lock_may_read - checks that the region is free of locks
2265 * @inode: the inode that is being read
2266 * @start: the first byte to read
2267 * @len: the number of bytes to read
2268 *
2269 * Emulates Windows locking requirements. Whole-file
2270 * mandatory locks (share modes) can prohibit a read and
2271 * byte-range POSIX locks can prohibit a read if they overlap.
2272 *
2273 * N.B. this function is only ever called
2274 * from knfsd and ownership of locks is never checked.
2275 */
2276int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2277{
2278 struct file_lock *fl;
2279 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002280 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2282 if (IS_POSIX(fl)) {
2283 if (fl->fl_type == F_RDLCK)
2284 continue;
2285 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2286 continue;
2287 } else if (IS_FLOCK(fl)) {
2288 if (!(fl->fl_type & LOCK_MAND))
2289 continue;
2290 if (fl->fl_type & LOCK_READ)
2291 continue;
2292 } else
2293 continue;
2294 result = 0;
2295 break;
2296 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002297 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 return result;
2299}
2300
2301EXPORT_SYMBOL(lock_may_read);
2302
2303/**
2304 * lock_may_write - checks that the region is free of locks
2305 * @inode: the inode that is being written
2306 * @start: the first byte to write
2307 * @len: the number of bytes to write
2308 *
2309 * Emulates Windows locking requirements. Whole-file
2310 * mandatory locks (share modes) can prohibit a write and
2311 * byte-range POSIX locks can prohibit a write if they overlap.
2312 *
2313 * N.B. this function is only ever called
2314 * from knfsd and ownership of locks is never checked.
2315 */
2316int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2317{
2318 struct file_lock *fl;
2319 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002320 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2322 if (IS_POSIX(fl)) {
2323 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2324 continue;
2325 } else if (IS_FLOCK(fl)) {
2326 if (!(fl->fl_type & LOCK_MAND))
2327 continue;
2328 if (fl->fl_type & LOCK_WRITE)
2329 continue;
2330 } else
2331 continue;
2332 result = 0;
2333 break;
2334 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002335 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 return result;
2337}
2338
2339EXPORT_SYMBOL(lock_may_write);
2340
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341static int __init filelock_init(void)
2342{
2343 filelock_cache = kmem_cache_create("file_lock_cache",
2344 sizeof(struct file_lock), 0, SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002345 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 return 0;
2347}
2348
2349core_initcall(filelock_init);