blob: 08415b2a6d36f321bf67b5a697c4db676cb15f78 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
63 * See 'Documentation/mandatory.txt' for details.
64 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -0400119#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000128#include <linux/pid_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130#include <asm/uaccess.h>
131
132#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
133#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
134#define IS_LEASE(fl) (fl->fl_flags & FL_LEASE)
135
136int leases_enable = 1;
137int lease_break_time = 45;
138
139#define for_each_lock(inode, lockp) \
140 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
141
Christoph Hellwig26bcbf92006-03-20 13:44:40 -0500142static LIST_HEAD(file_lock_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static LIST_HEAD(blocked_list);
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200144static DEFINE_SPINLOCK(file_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200146/*
147 * Protects the two list heads above, plus the inode->i_flock list
148 * FIXME: should use a spinlock, once lockd and ceph are ready.
149 */
150void lock_flocks(void)
151{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200152 spin_lock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200153}
154EXPORT_SYMBOL_GPL(lock_flocks);
155
156void unlock_flocks(void)
157{
Arnd Bergmann72f98e72010-10-27 21:39:58 +0200158 spin_unlock(&file_lock_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200159}
160EXPORT_SYMBOL_GPL(unlock_flocks);
161
Christoph Lametere18b8902006-12-06 20:33:20 -0800162static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200165struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
Christoph Lametere94b1762006-12-06 20:33:17 -0800167 return kmem_cache_alloc(filelock_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200169EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500171void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500172{
173 if (fl->fl_ops) {
174 if (fl->fl_ops->fl_release_private)
175 fl->fl_ops->fl_release_private(fl);
176 fl->fl_ops = NULL;
177 }
178 if (fl->fl_lmops) {
179 if (fl->fl_lmops->fl_release_private)
180 fl->fl_lmops->fl_release_private(fl);
181 fl->fl_lmops = NULL;
182 }
183
184}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500185EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400188void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800190 BUG_ON(waitqueue_active(&fl->fl_wait));
191 BUG_ON(!list_empty(&fl->fl_block));
192 BUG_ON(!list_empty(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
Trond Myklebust47831f32006-03-20 13:44:05 -0500194 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 kmem_cache_free(filelock_cache, fl);
196}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400197EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198
199void locks_init_lock(struct file_lock *fl)
200{
201 INIT_LIST_HEAD(&fl->fl_link);
202 INIT_LIST_HEAD(&fl->fl_block);
203 init_waitqueue_head(&fl->fl_wait);
204 fl->fl_next = NULL;
205 fl->fl_fasync = NULL;
206 fl->fl_owner = NULL;
207 fl->fl_pid = 0;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000208 fl->fl_nspid = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 fl->fl_file = NULL;
210 fl->fl_flags = 0;
211 fl->fl_type = 0;
212 fl->fl_start = fl->fl_end = 0;
213 fl->fl_ops = NULL;
214 fl->fl_lmops = NULL;
215}
216
217EXPORT_SYMBOL(locks_init_lock);
218
219/*
220 * Initialises the fields of the file lock which are invariant for
221 * free file_locks.
222 */
Alexey Dobriyan51cc5062008-07-25 19:45:34 -0700223static void init_once(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224{
225 struct file_lock *lock = (struct file_lock *) foo;
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 locks_init_lock(lock);
228}
229
Trond Myklebust47831f32006-03-20 13:44:05 -0500230static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
231{
232 if (fl->fl_ops) {
233 if (fl->fl_ops->fl_copy_lock)
234 fl->fl_ops->fl_copy_lock(new, fl);
235 new->fl_ops = fl->fl_ops;
236 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400237 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500238 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500239}
240
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241/*
242 * Initialize a new lock from an existing file_lock structure.
243 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400244void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245{
246 new->fl_owner = fl->fl_owner;
247 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500248 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 new->fl_flags = fl->fl_flags;
250 new->fl_type = fl->fl_type;
251 new->fl_start = fl->fl_start;
252 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500253 new->fl_ops = NULL;
254 new->fl_lmops = NULL;
255}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700256EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500257
258void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
259{
260 locks_release_private(new);
261
262 __locks_copy_lock(new, fl);
263 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 new->fl_ops = fl->fl_ops;
265 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500266
267 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268}
269
270EXPORT_SYMBOL(locks_copy_lock);
271
272static inline int flock_translate_cmd(int cmd) {
273 if (cmd & LOCK_MAND)
274 return cmd & (LOCK_MAND | LOCK_RW);
275 switch (cmd) {
276 case LOCK_SH:
277 return F_RDLCK;
278 case LOCK_EX:
279 return F_WRLCK;
280 case LOCK_UN:
281 return F_UNLCK;
282 }
283 return -EINVAL;
284}
285
286/* Fill in a file_lock structure with an appropriate FLOCK lock. */
287static int flock_make_lock(struct file *filp, struct file_lock **lock,
288 unsigned int cmd)
289{
290 struct file_lock *fl;
291 int type = flock_translate_cmd(cmd);
292 if (type < 0)
293 return type;
294
295 fl = locks_alloc_lock();
296 if (fl == NULL)
297 return -ENOMEM;
298
299 fl->fl_file = filp;
300 fl->fl_pid = current->tgid;
301 fl->fl_flags = FL_FLOCK;
302 fl->fl_type = type;
303 fl->fl_end = OFFSET_MAX;
304
305 *lock = fl;
306 return 0;
307}
308
309static int assign_type(struct file_lock *fl, int type)
310{
311 switch (type) {
312 case F_RDLCK:
313 case F_WRLCK:
314 case F_UNLCK:
315 fl->fl_type = type;
316 break;
317 default:
318 return -EINVAL;
319 }
320 return 0;
321}
322
323/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
324 * style lock.
325 */
326static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
327 struct flock *l)
328{
329 off_t start, end;
330
331 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700332 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 start = 0;
334 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700335 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 start = filp->f_pos;
337 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700338 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800339 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 break;
341 default:
342 return -EINVAL;
343 }
344
345 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
346 POSIX-2001 defines it. */
347 start += l->l_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 if (start < 0)
349 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700350 fl->fl_end = OFFSET_MAX;
351 if (l->l_len > 0) {
352 end = start + l->l_len - 1;
353 fl->fl_end = end;
354 } else if (l->l_len < 0) {
355 end = start - 1;
356 fl->fl_end = end;
357 start += l->l_len;
358 if (start < 0)
359 return -EINVAL;
360 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700362 if (fl->fl_end < fl->fl_start)
363 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364
365 fl->fl_owner = current->files;
366 fl->fl_pid = current->tgid;
367 fl->fl_file = filp;
368 fl->fl_flags = FL_POSIX;
369 fl->fl_ops = NULL;
370 fl->fl_lmops = NULL;
371
372 return assign_type(fl, l->l_type);
373}
374
375#if BITS_PER_LONG == 32
376static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
377 struct flock64 *l)
378{
379 loff_t start;
380
381 switch (l->l_whence) {
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700382 case SEEK_SET:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 start = 0;
384 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700385 case SEEK_CUR:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 start = filp->f_pos;
387 break;
Josef 'Jeff' Sipekf5579f82006-09-30 23:27:35 -0700388 case SEEK_END:
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800389 start = i_size_read(filp->f_path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 break;
391 default:
392 return -EINVAL;
393 }
394
Trond Myklebust4c780a42005-10-18 14:20:21 -0700395 start += l->l_start;
396 if (start < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 return -EINVAL;
Trond Myklebust4c780a42005-10-18 14:20:21 -0700398 fl->fl_end = OFFSET_MAX;
399 if (l->l_len > 0) {
400 fl->fl_end = start + l->l_len - 1;
401 } else if (l->l_len < 0) {
402 fl->fl_end = start - 1;
403 start += l->l_len;
404 if (start < 0)
405 return -EINVAL;
406 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 fl->fl_start = start; /* we record the absolute position */
Trond Myklebust4c780a42005-10-18 14:20:21 -0700408 if (fl->fl_end < fl->fl_start)
409 return -EOVERFLOW;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411 fl->fl_owner = current->files;
412 fl->fl_pid = current->tgid;
413 fl->fl_file = filp;
414 fl->fl_flags = FL_POSIX;
415 fl->fl_ops = NULL;
416 fl->fl_lmops = NULL;
417
418 switch (l->l_type) {
419 case F_RDLCK:
420 case F_WRLCK:
421 case F_UNLCK:
422 fl->fl_type = l->l_type;
423 break;
424 default:
425 return -EINVAL;
426 }
427
428 return (0);
429}
430#endif
431
432/* default lease lock manager operations */
433static void lease_break_callback(struct file_lock *fl)
434{
435 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
436}
437
438static void lease_release_private_callback(struct file_lock *fl)
439{
440 if (!fl->fl_file)
441 return;
442
443 f_delown(fl->fl_file);
444 fl->fl_file->f_owner.signum = 0;
445}
446
Adrian Bunk75c96f82005-05-05 16:16:09 -0700447static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448{
449 return fl->fl_file == try->fl_file;
450}
451
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700452static const struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 .fl_break = lease_break_callback,
454 .fl_release_private = lease_release_private_callback,
455 .fl_mylease = lease_mylease_callback,
456 .fl_change = lease_modify,
457};
458
459/*
460 * Initialize a lease, use the default lock manager operations
461 */
462static int lease_init(struct file *filp, int type, struct file_lock *fl)
463 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400464 if (assign_type(fl, type) != 0)
465 return -EINVAL;
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 fl->fl_owner = current->files;
468 fl->fl_pid = current->tgid;
469
470 fl->fl_file = filp;
471 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 fl->fl_start = 0;
473 fl->fl_end = OFFSET_MAX;
474 fl->fl_ops = NULL;
475 fl->fl_lmops = &lease_manager_ops;
476 return 0;
477}
478
479/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500480static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
482 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400483 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
485 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500486 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400489 if (error) {
490 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500491 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400492 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500493 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494}
495
496/* Check if two locks overlap each other.
497 */
498static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
499{
500 return ((fl1->fl_end >= fl2->fl_start) &&
501 (fl2->fl_end >= fl1->fl_start));
502}
503
504/*
505 * Check whether two locks have the same owner.
506 */
Matt Mackall33443c42006-01-08 01:05:22 -0800507static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
510 return fl2->fl_lmops == fl1->fl_lmops &&
511 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
512 return fl1->fl_owner == fl2->fl_owner;
513}
514
515/* Remove waiter from blocker's block list.
516 * When blocker ends up pointing to itself then the list is empty.
517 */
Matt Mackall33443c42006-01-08 01:05:22 -0800518static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
520 list_del_init(&waiter->fl_block);
521 list_del_init(&waiter->fl_link);
522 waiter->fl_next = NULL;
523}
524
525/*
526 */
527static void locks_delete_block(struct file_lock *waiter)
528{
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200529 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 __locks_delete_block(waiter);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200531 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532}
533
534/* Insert waiter into blocker's block list.
535 * We use a circular list so that processes can be easily woken up in
536 * the order they blocked. The documentation doesn't require this but
537 * it seems like the reasonable thing to do.
538 */
539static void locks_insert_block(struct file_lock *blocker,
540 struct file_lock *waiter)
541{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800542 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 list_add_tail(&waiter->fl_block, &blocker->fl_block);
544 waiter->fl_next = blocker;
545 if (IS_POSIX(blocker))
546 list_add(&waiter->fl_link, &blocked_list);
547}
548
549/* Wake up processes blocked waiting for blocker.
550 * If told to wait then schedule the processes until the block list
551 * is empty, otherwise empty the block list ourselves.
552 */
553static void locks_wake_up_blocks(struct file_lock *blocker)
554{
555 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400556 struct file_lock *waiter;
557
558 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 struct file_lock, fl_block);
560 __locks_delete_block(waiter);
561 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
562 waiter->fl_lmops->fl_notify(waiter);
563 else
564 wake_up(&waiter->fl_wait);
565 }
566}
567
568/* Insert file lock fl into an inode's lock list at the position indicated
569 * by pos. At the same time add the lock to the global file lock list.
570 */
571static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
572{
573 list_add(&fl->fl_link, &file_lock_list);
574
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000575 fl->fl_nspid = get_pid(task_tgid(current));
576
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 /* insert into file's list */
578 fl->fl_next = *pos;
579 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580}
581
582/*
583 * Delete a lock and then free it.
584 * Wake up processes that are blocked waiting for this lock,
585 * notify the FS that the lock has been cleared and
586 * finally free the lock.
587 */
588static void locks_delete_lock(struct file_lock **thisfl_p)
589{
590 struct file_lock *fl = *thisfl_p;
591
592 *thisfl_p = fl->fl_next;
593 fl->fl_next = NULL;
594 list_del_init(&fl->fl_link);
595
596 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
597 if (fl->fl_fasync != NULL) {
598 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
599 fl->fl_fasync = NULL;
600 }
601
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000602 if (fl->fl_nspid) {
603 put_pid(fl->fl_nspid);
604 fl->fl_nspid = NULL;
605 }
606
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 locks_wake_up_blocks(fl);
608 locks_free_lock(fl);
609}
610
611/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
612 * checks for shared/exclusive status of overlapping locks.
613 */
614static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
615{
616 if (sys_fl->fl_type == F_WRLCK)
617 return 1;
618 if (caller_fl->fl_type == F_WRLCK)
619 return 1;
620 return 0;
621}
622
623/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
624 * checking before calling the locks_conflict().
625 */
626static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
627{
628 /* POSIX locks owned by the same process do not conflict with
629 * each other.
630 */
631 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
632 return (0);
633
634 /* Check whether they overlap */
635 if (!locks_overlap(caller_fl, sys_fl))
636 return 0;
637
638 return (locks_conflict(caller_fl, sys_fl));
639}
640
641/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
642 * checking before calling the locks_conflict().
643 */
644static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
645{
646 /* FLOCK locks referring to the same filp do not conflict with
647 * each other.
648 */
649 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
650 return (0);
651 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
652 return 0;
653
654 return (locks_conflict(caller_fl, sys_fl));
655}
656
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400657void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500658posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659{
660 struct file_lock *cfl;
661
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200662 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800663 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 if (!IS_POSIX(cfl))
665 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400666 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 break;
668 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000669 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500670 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000671 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800672 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000673 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400674 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200675 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400676 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678EXPORT_SYMBOL(posix_test_lock);
679
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400680/*
681 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400683 * We attempt to detect deadlocks that are due purely to posix file
684 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400686 * We assume that a task can be waiting for at most one lock at a time.
687 * So for any acquired lock, the process holding that lock may be
688 * waiting on at most one other lock. That lock in turns may be held by
689 * someone waiting for at most one other lock. Given a requested lock
690 * caller_fl which is about to wait for a conflicting lock block_fl, we
691 * follow this chain of waiters to ensure we are not about to create a
692 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400693 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400694 * Since we do this before we ever put a process to sleep on a lock, we
695 * are ensured that there is never a cycle; that is what guarantees that
696 * the while() loop in posix_locks_deadlock() eventually completes.
697 *
698 * Note: the above assumption may not be true when handling lock
699 * requests from a broken NFS client. It may also fail in the presence
700 * of tasks (such as posix threads) sharing the same open file table.
701 *
702 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400704
705#define MAX_DEADLK_ITERATIONS 10
706
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400707/* Find a lock that the owner of the given block_fl is blocking on. */
708static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
709{
710 struct file_lock *fl;
711
712 list_for_each_entry(fl, &blocked_list, fl_link) {
713 if (posix_same_owner(fl, block_fl))
714 return fl->fl_next;
715 }
716 return NULL;
717}
718
Adrian Bunkb0904e12006-06-23 02:05:13 -0700719static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 struct file_lock *block_fl)
721{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400722 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400724 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
725 if (i++ > MAX_DEADLK_ITERATIONS)
726 return 0;
727 if (posix_same_owner(caller_fl, block_fl))
728 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 }
730 return 0;
731}
732
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400734 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400735 *
736 * Note that if called with an FL_EXISTS argument, the caller may determine
737 * whether or not a lock was successfully freed by testing the return
738 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800740static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800742 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800744 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 int error = 0;
746 int found = 0;
747
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200748 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
749 new_fl = locks_alloc_lock();
750 if (!new_fl)
751 return -ENOMEM;
752 }
753
754 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400755 if (request->fl_flags & FL_ACCESS)
756 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400757
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 for_each_lock(inode, before) {
759 struct file_lock *fl = *before;
760 if (IS_POSIX(fl))
761 break;
762 if (IS_LEASE(fl))
763 continue;
764 if (filp != fl->fl_file)
765 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800766 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 goto out;
768 found = 1;
769 locks_delete_lock(before);
770 break;
771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Trond Myklebustf475ae92006-06-29 16:38:32 -0400773 if (request->fl_type == F_UNLCK) {
774 if ((request->fl_flags & FL_EXISTS) && !found)
775 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800776 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778
779 /*
780 * If a higher-priority process was blocked on the old file lock,
781 * give it the opportunity to lock the file.
782 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200783 if (found) {
784 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200785 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200786 lock_flocks();
787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400789find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 for_each_lock(inode, before) {
791 struct file_lock *fl = *before;
792 if (IS_POSIX(fl))
793 break;
794 if (IS_LEASE(fl))
795 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800796 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 continue;
798 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700799 if (!(request->fl_flags & FL_SLEEP))
800 goto out;
801 error = FILE_LOCK_DEFERRED;
802 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 goto out;
804 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400805 if (request->fl_flags & FL_ACCESS)
806 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800807 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700808 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800809 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400810 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200813 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800814 if (new_fl)
815 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 return error;
817}
818
Marc Eshel150b3932007-01-18 16:15:35 -0500819static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
821 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700822 struct file_lock *new_fl = NULL;
823 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 struct file_lock *left = NULL;
825 struct file_lock *right = NULL;
826 struct file_lock **before;
827 int error, added = 0;
828
829 /*
830 * We may need two file_lock structures for this operation,
831 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700832 *
833 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700835 if (!(request->fl_flags & FL_ACCESS) &&
836 (request->fl_type != F_UNLCK ||
837 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
838 new_fl = locks_alloc_lock();
839 new_fl2 = locks_alloc_lock();
840 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200842 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 if (request->fl_type != F_UNLCK) {
844 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500845 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 if (!IS_POSIX(fl))
847 continue;
848 if (!posix_locks_conflict(request, fl))
849 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800850 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400851 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 error = -EAGAIN;
853 if (!(request->fl_flags & FL_SLEEP))
854 goto out;
855 error = -EDEADLK;
856 if (posix_locks_deadlock(request, fl))
857 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700858 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 locks_insert_block(fl, request);
860 goto out;
861 }
862 }
863
864 /* If we're just looking for a conflict, we're done. */
865 error = 0;
866 if (request->fl_flags & FL_ACCESS)
867 goto out;
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 * Find the first old lock with the same owner as the new lock.
871 */
872
873 before = &inode->i_flock;
874
875 /* First skip locks owned by other processes. */
876 while ((fl = *before) && (!IS_POSIX(fl) ||
877 !posix_same_owner(request, fl))) {
878 before = &fl->fl_next;
879 }
880
881 /* Process locks with this owner. */
882 while ((fl = *before) && posix_same_owner(request, fl)) {
883 /* Detect adjacent or overlapping regions (if same lock type)
884 */
885 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700886 /* In all comparisons of start vs end, use
887 * "start - 1" rather than "end + 1". If end
888 * is OFFSET_MAX, end + 1 will become negative.
889 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (fl->fl_end < request->fl_start - 1)
891 goto next_lock;
892 /* If the next lock in the list has entirely bigger
893 * addresses than the new one, insert the lock here.
894 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700895 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 break;
897
898 /* If we come here, the new and old lock are of the
899 * same type and adjacent or overlapping. Make one
900 * lock yielding from the lower start address of both
901 * locks to the higher end address.
902 */
903 if (fl->fl_start > request->fl_start)
904 fl->fl_start = request->fl_start;
905 else
906 request->fl_start = fl->fl_start;
907 if (fl->fl_end < request->fl_end)
908 fl->fl_end = request->fl_end;
909 else
910 request->fl_end = fl->fl_end;
911 if (added) {
912 locks_delete_lock(before);
913 continue;
914 }
915 request = fl;
916 added = 1;
917 }
918 else {
919 /* Processing for different lock types is a bit
920 * more complex.
921 */
922 if (fl->fl_end < request->fl_start)
923 goto next_lock;
924 if (fl->fl_start > request->fl_end)
925 break;
926 if (request->fl_type == F_UNLCK)
927 added = 1;
928 if (fl->fl_start < request->fl_start)
929 left = fl;
930 /* If the next lock in the list has a higher end
931 * address than the new one, insert the new one here.
932 */
933 if (fl->fl_end > request->fl_end) {
934 right = fl;
935 break;
936 }
937 if (fl->fl_start >= request->fl_start) {
938 /* The new lock completely replaces an old
939 * one (This may happen several times).
940 */
941 if (added) {
942 locks_delete_lock(before);
943 continue;
944 }
945 /* Replace the old lock with the new one.
946 * Wake up anybody waiting for the old one,
947 * as the change in lock type might satisfy
948 * their needs.
949 */
950 locks_wake_up_blocks(fl);
951 fl->fl_start = request->fl_start;
952 fl->fl_end = request->fl_end;
953 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500954 locks_release_private(fl);
955 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 request = fl;
957 added = 1;
958 }
959 }
960 /* Go on to next lock.
961 */
962 next_lock:
963 before = &fl->fl_next;
964 }
965
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700966 /*
967 * The above code only modifies existing locks in case of
968 * merging or replacing. If new lock(s) need to be inserted
969 * all modifications are done bellow this, so it's safe yet to
970 * bail out.
971 */
972 error = -ENOLCK; /* "no luck" */
973 if (right && left == right && !new_fl2)
974 goto out;
975
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 error = 0;
977 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400978 if (request->fl_type == F_UNLCK) {
979 if (request->fl_flags & FL_EXISTS)
980 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400982 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700983
984 if (!new_fl) {
985 error = -ENOLCK;
986 goto out;
987 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 locks_copy_lock(new_fl, request);
989 locks_insert_lock(before, new_fl);
990 new_fl = NULL;
991 }
992 if (right) {
993 if (left == right) {
994 /* The new lock breaks the old one in two pieces,
995 * so we have to use the second new lock.
996 */
997 left = new_fl2;
998 new_fl2 = NULL;
999 locks_copy_lock(left, right);
1000 locks_insert_lock(before, left);
1001 }
1002 right->fl_start = request->fl_end + 1;
1003 locks_wake_up_blocks(right);
1004 }
1005 if (left) {
1006 left->fl_end = request->fl_start - 1;
1007 locks_wake_up_blocks(left);
1008 }
1009 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001010 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 /*
1012 * Free any unused locks.
1013 */
1014 if (new_fl)
1015 locks_free_lock(new_fl);
1016 if (new_fl2)
1017 locks_free_lock(new_fl2);
1018 return error;
1019}
1020
1021/**
1022 * posix_lock_file - Apply a POSIX-style lock to a file
1023 * @filp: The file to apply the lock to
1024 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001025 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 *
1027 * Add a POSIX style lock to a file.
1028 * We merge adjacent & overlapping locks whenever possible.
1029 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001030 *
1031 * Note that if called with an FL_EXISTS argument, the caller may determine
1032 * whether or not a lock was successfully freed by testing the return
1033 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 */
Marc Eshel150b3932007-01-18 16:15:35 -05001035int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001036 struct file_lock *conflock)
1037{
Marc Eshel150b3932007-01-18 16:15:35 -05001038 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001039}
Marc Eshel150b3932007-01-18 16:15:35 -05001040EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042/**
1043 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1044 * @filp: The file to apply the lock to
1045 * @fl: The lock to be applied
1046 *
1047 * Add a POSIX style lock to a file.
1048 * We merge adjacent & overlapping locks whenever possible.
1049 * POSIX locks are sorted by owner task, then by starting address
1050 */
1051int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1052{
1053 int error;
1054 might_sleep ();
1055 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001056 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001057 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 break;
1059 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1060 if (!error)
1061 continue;
1062
1063 locks_delete_block(fl);
1064 break;
1065 }
1066 return error;
1067}
1068EXPORT_SYMBOL(posix_lock_file_wait);
1069
1070/**
1071 * locks_mandatory_locked - Check for an active lock
1072 * @inode: the file to check
1073 *
1074 * Searches the inode's list of locks to find any POSIX locks which conflict.
1075 * This function is called from locks_verify_locked() only.
1076 */
1077int locks_mandatory_locked(struct inode *inode)
1078{
1079 fl_owner_t owner = current->files;
1080 struct file_lock *fl;
1081
1082 /*
1083 * Search the lock list for this inode for any POSIX locks.
1084 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001085 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1087 if (!IS_POSIX(fl))
1088 continue;
1089 if (fl->fl_owner != owner)
1090 break;
1091 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001092 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 return fl ? -EAGAIN : 0;
1094}
1095
1096/**
1097 * locks_mandatory_area - Check for a conflicting lock
1098 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1099 * for shared
1100 * @inode: the file to check
1101 * @filp: how the file was opened (if it was)
1102 * @offset: start of area to check
1103 * @count: length of area to check
1104 *
1105 * Searches the inode's list of locks to find any POSIX locks which conflict.
1106 * This function is called from rw_verify_area() and
1107 * locks_verify_truncate().
1108 */
1109int locks_mandatory_area(int read_write, struct inode *inode,
1110 struct file *filp, loff_t offset,
1111 size_t count)
1112{
1113 struct file_lock fl;
1114 int error;
1115
1116 locks_init_lock(&fl);
1117 fl.fl_owner = current->files;
1118 fl.fl_pid = current->tgid;
1119 fl.fl_file = filp;
1120 fl.fl_flags = FL_POSIX | FL_ACCESS;
1121 if (filp && !(filp->f_flags & O_NONBLOCK))
1122 fl.fl_flags |= FL_SLEEP;
1123 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1124 fl.fl_start = offset;
1125 fl.fl_end = offset + count - 1;
1126
1127 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001128 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001129 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 break;
1131 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1132 if (!error) {
1133 /*
1134 * If we've been sleeping someone might have
1135 * changed the permissions behind our back.
1136 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001137 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 continue;
1139 }
1140
1141 locks_delete_block(&fl);
1142 break;
1143 }
1144
1145 return error;
1146}
1147
1148EXPORT_SYMBOL(locks_mandatory_area);
1149
1150/* We already had a lease on this file; just change its type */
1151int lease_modify(struct file_lock **before, int arg)
1152{
1153 struct file_lock *fl = *before;
1154 int error = assign_type(fl, arg);
1155
1156 if (error)
1157 return error;
1158 locks_wake_up_blocks(fl);
1159 if (arg == F_UNLCK)
1160 locks_delete_lock(before);
1161 return 0;
1162}
1163
1164EXPORT_SYMBOL(lease_modify);
1165
1166static void time_out_leases(struct inode *inode)
1167{
1168 struct file_lock **before;
1169 struct file_lock *fl;
1170
1171 before = &inode->i_flock;
1172 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1173 if ((fl->fl_break_time == 0)
1174 || time_before(jiffies, fl->fl_break_time)) {
1175 before = &fl->fl_next;
1176 continue;
1177 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1179 if (fl == *before) /* lease_modify may have freed fl */
1180 before = &fl->fl_next;
1181 }
1182}
1183
1184/**
1185 * __break_lease - revoke all outstanding leases on file
1186 * @inode: the inode of the file to return
1187 * @mode: the open mode (read or write)
1188 *
david m. richter87250dd2007-05-09 16:10:27 -04001189 * break_lease (inlined for speed) has checked there already is at least
1190 * some kind of lock (maybe a lease) on this file. Leases are broken on
1191 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 * specified %O_NONBLOCK to your open().
1193 */
1194int __break_lease(struct inode *inode, unsigned int mode)
1195{
1196 int error = 0, future;
1197 struct file_lock *new_fl, *flock;
1198 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 unsigned long break_time;
1200 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001201 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
Al Viro8737c932009-12-24 06:47:55 -05001203 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001205 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206
1207 time_out_leases(inode);
1208
1209 flock = inode->i_flock;
1210 if ((flock == NULL) || !IS_LEASE(flock))
1211 goto out;
1212
1213 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1214 if (fl->fl_owner == current->files)
1215 i_have_this_lease = 1;
1216
Al Viro8737c932009-12-24 06:47:55 -05001217 if (want_write) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 /* If we want write access, we have to revoke any lease. */
1219 future = F_UNLCK | F_INPROGRESS;
1220 } else if (flock->fl_type & F_INPROGRESS) {
1221 /* If the lease is already being broken, we just leave it */
1222 future = flock->fl_type;
1223 } else if (flock->fl_type & F_WRLCK) {
1224 /* Downgrade the exclusive lease to a read-only lease. */
1225 future = F_RDLCK | F_INPROGRESS;
1226 } else {
1227 /* the existing lease was read-only, so we can read too. */
1228 goto out;
1229 }
1230
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001231 if (IS_ERR(new_fl) && !i_have_this_lease
1232 && ((mode & O_NONBLOCK) == 0)) {
1233 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 goto out;
1235 }
1236
1237 break_time = 0;
1238 if (lease_break_time > 0) {
1239 break_time = jiffies + lease_break_time * HZ;
1240 if (break_time == 0)
1241 break_time++; /* so that 0 means no break time */
1242 }
1243
1244 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1245 if (fl->fl_type != future) {
1246 fl->fl_type = future;
1247 fl->fl_break_time = break_time;
1248 /* lease must have lmops break callback */
1249 fl->fl_lmops->fl_break(fl);
1250 }
1251 }
1252
1253 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1254 error = -EWOULDBLOCK;
1255 goto out;
1256 }
1257
1258restart:
1259 break_time = flock->fl_break_time;
1260 if (break_time != 0) {
1261 break_time -= jiffies;
1262 if (break_time == 0)
1263 break_time++;
1264 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001265 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001266 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001267 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1268 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001269 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001270 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (error >= 0) {
1272 if (error == 0)
1273 time_out_leases(inode);
1274 /* Wait for the next lease that has not been broken yet */
1275 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1276 flock = flock->fl_next) {
1277 if (flock->fl_type & F_INPROGRESS)
1278 goto restart;
1279 }
1280 error = 0;
1281 }
1282
1283out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001284 unlock_flocks();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001285 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 locks_free_lock(new_fl);
1287 return error;
1288}
1289
1290EXPORT_SYMBOL(__break_lease);
1291
1292/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001293 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 * @inode: the inode
1295 * @time: pointer to a timespec which will contain the last modified time
1296 *
1297 * This is to force NFS clients to flush their caches for files with
1298 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001299 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 */
1301void lease_get_mtime(struct inode *inode, struct timespec *time)
1302{
1303 struct file_lock *flock = inode->i_flock;
1304 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1305 *time = current_fs_time(inode->i_sb);
1306 else
1307 *time = inode->i_mtime;
1308}
1309
1310EXPORT_SYMBOL(lease_get_mtime);
1311
1312/**
1313 * fcntl_getlease - Enquire what lease is currently active
1314 * @filp: the file
1315 *
1316 * The value returned by this function will be one of
1317 * (if no lease break is pending):
1318 *
1319 * %F_RDLCK to indicate a shared lease is held.
1320 *
1321 * %F_WRLCK to indicate an exclusive lease is held.
1322 *
1323 * %F_UNLCK to indicate no lease is held.
1324 *
1325 * (if a lease break is pending):
1326 *
1327 * %F_RDLCK to indicate an exclusive lease needs to be
1328 * changed to a shared lease (or removed).
1329 *
1330 * %F_UNLCK to indicate the lease needs to be removed.
1331 *
1332 * XXX: sfr & willy disagree over whether F_INPROGRESS
1333 * should be returned to userspace.
1334 */
1335int fcntl_getlease(struct file *filp)
1336{
1337 struct file_lock *fl;
1338 int type = F_UNLCK;
1339
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001340 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001341 time_out_leases(filp->f_path.dentry->d_inode);
1342 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343 fl = fl->fl_next) {
1344 if (fl->fl_file == filp) {
1345 type = fl->fl_type & ~F_INPROGRESS;
1346 break;
1347 }
1348 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001349 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 return type;
1351}
1352
1353/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001354 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 * @filp: file pointer
1356 * @arg: type of lease to obtain
1357 * @flp: input - file_lock to use, output - file_lock inserted
1358 *
1359 * The (input) flp->fl_lmops->fl_break function is required
1360 * by break_lease().
1361 *
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001362 * Called with file_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001364int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001366 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001367 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 struct inode *inode = dentry->d_inode;
1369 int error, rdlease_count = 0, wrlease_count = 0;
1370
J. Bruce Fields096657b2010-10-30 17:31:14 -04001371 lease = *flp;
1372
1373 error = -EACCES;
David Howellsda9592e2008-11-14 10:39:05 +11001374 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001375 goto out;
1376 error = -EINVAL;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001377 if (!S_ISREG(inode->i_mode))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001378 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001379 error = security_file_lock(filp, arg);
1380 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001381 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001382
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 time_out_leases(inode);
1384
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001385 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
David M. Richter288b2fd2008-04-23 16:29:00 -04001387 if (arg != F_UNLCK) {
1388 error = -EAGAIN;
1389 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1390 goto out;
1391 if ((arg == F_WRLCK)
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001392 && ((dentry->d_count > 1)
David M. Richter288b2fd2008-04-23 16:29:00 -04001393 || (atomic_read(&inode->i_count) > 1)))
1394 goto out;
David M. Richter288b2fd2008-04-23 16:29:00 -04001395 }
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001396
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 /*
1398 * At this point, we know that if there is an exclusive
1399 * lease on this file, then we hold it on this filp
1400 * (otherwise our open of this file would have blocked).
1401 * And if we are trying to acquire an exclusive lease,
1402 * then the file is not open by anyone (including us)
1403 * except for this filp.
1404 */
1405 for (before = &inode->i_flock;
1406 ((fl = *before) != NULL) && IS_LEASE(fl);
1407 before = &fl->fl_next) {
1408 if (lease->fl_lmops->fl_mylease(fl, lease))
1409 my_before = before;
1410 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1411 /*
1412 * Someone is in the process of opening this
1413 * file for writing so we may not take an
1414 * exclusive lease on it.
1415 */
1416 wrlease_count++;
1417 else
1418 rdlease_count++;
1419 }
1420
David M. Richter5fcc60c2008-04-23 16:28:59 -04001421 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1423 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1424 goto out;
1425
1426 if (my_before != NULL) {
1427 error = lease->fl_lmops->fl_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001428 if (!error)
1429 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 goto out;
1431 }
1432
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 if (arg == F_UNLCK)
1434 goto out;
1435
1436 error = -EINVAL;
1437 if (!leases_enable)
1438 goto out;
1439
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001440 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001441 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443out:
1444 return error;
1445}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001446EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001448static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1449{
1450 if (filp->f_op && filp->f_op->setlease)
1451 return filp->f_op->setlease(filp, arg, lease);
1452 else
1453 return generic_setlease(filp, arg, lease);
1454}
1455
1456/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001457 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 * @filp: file pointer
1459 * @arg: type of lease to obtain
1460 * @lease: file_lock to use
1461 *
1462 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001463 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1464 * break_lease will oops!
1465 *
1466 * This will call the filesystem's setlease file method, if
1467 * defined. Note that there is no getlease method; instead, the
1468 * filesystem setlease method should call back to setlease() to
1469 * add a lease to the inode's lease list, where fcntl_getlease() can
1470 * find it. Since fcntl_getlease() only reports whether the current
1471 * task holds a lease, a cluster filesystem need only do this for
1472 * leases held by processes on this node.
1473 *
1474 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001475 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001476 * filesystem's open, create, and (on truncate) setattr methods.
1477 *
1478 * Warning: the only current setlease methods exist only to disable
1479 * leases in certain cases. More vfs changes may be required to
1480 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 */
1482
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001483int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 int error;
1486
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001487 lock_flocks();
1488 error = __vfs_setlease(filp, arg, lease);
1489 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
1491 return error;
1492}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001493EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001495static int do_fcntl_delete_lease(struct file *filp)
1496{
1497 struct file_lock fl, *flp = &fl;
1498
1499 lease_init(filp, F_UNLCK, flp);
1500
1501 return vfs_setlease(filp, F_UNLCK, &flp);
1502}
1503
1504static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001506 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001507 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 int error;
1509
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001510 fl = lease_alloc(filp, arg);
1511 if (IS_ERR(fl))
1512 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001514 new = fasync_alloc();
1515 if (!new) {
1516 locks_free_lock(fl);
1517 return -ENOMEM;
1518 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001519 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001520 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001521 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001522 if (error) {
1523 unlock_flocks();
1524 locks_free_lock(fl);
1525 goto out_free_fasync;
1526 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001527 if (ret != fl)
1528 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001530 /*
1531 * fasync_insert_entry() returns the old entry if any.
1532 * If there was no old entry, then it used 'new' and
1533 * inserted it into the fasync list. Clear new so that
1534 * we don't release it here.
1535 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001536 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001537 new = NULL;
1538
J. Bruce Fields8896b932010-11-03 18:09:18 -04001539 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001540 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001541
1542out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001543 if (new)
1544 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return error;
1546}
1547
1548/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001549 * fcntl_setlease - sets a lease on an open file
1550 * @fd: open file descriptor
1551 * @filp: file pointer
1552 * @arg: type of lease to obtain
1553 *
1554 * Call this fcntl to establish a lease on the file.
1555 * Note that you also need to call %F_SETSIG to
1556 * receive a signal when the lease is broken.
1557 */
1558int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1559{
1560 if (arg == F_UNLCK)
1561 return do_fcntl_delete_lease(filp);
1562 return do_fcntl_add_lease(fd, filp, arg);
1563}
1564
1565/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1567 * @filp: The file to apply the lock to
1568 * @fl: The lock to be applied
1569 *
1570 * Add a FLOCK style lock to a file.
1571 */
1572int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1573{
1574 int error;
1575 might_sleep();
1576 for (;;) {
1577 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001578 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 break;
1580 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1581 if (!error)
1582 continue;
1583
1584 locks_delete_block(fl);
1585 break;
1586 }
1587 return error;
1588}
1589
1590EXPORT_SYMBOL(flock_lock_file_wait);
1591
1592/**
1593 * sys_flock: - flock() system call.
1594 * @fd: the file descriptor to lock.
1595 * @cmd: the type of lock to apply.
1596 *
1597 * Apply a %FL_FLOCK style lock to an open file descriptor.
1598 * The @cmd can be one of
1599 *
1600 * %LOCK_SH -- a shared lock.
1601 *
1602 * %LOCK_EX -- an exclusive lock.
1603 *
1604 * %LOCK_UN -- remove an existing lock.
1605 *
1606 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1607 *
1608 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1609 * processes read and write access respectively.
1610 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001611SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612{
1613 struct file *filp;
1614 struct file_lock *lock;
1615 int can_sleep, unlock;
1616 int error;
1617
1618 error = -EBADF;
1619 filp = fget(fd);
1620 if (!filp)
1621 goto out;
1622
1623 can_sleep = !(cmd & LOCK_NB);
1624 cmd &= ~LOCK_NB;
1625 unlock = (cmd == LOCK_UN);
1626
Al Viroaeb5d722008-09-02 15:28:45 -04001627 if (!unlock && !(cmd & LOCK_MAND) &&
1628 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 goto out_putf;
1630
1631 error = flock_make_lock(filp, &lock, cmd);
1632 if (error)
1633 goto out_putf;
1634 if (can_sleep)
1635 lock->fl_flags |= FL_SLEEP;
1636
Sten Spans713c0ec2009-07-16 09:41:39 +02001637 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 if (error)
1639 goto out_free;
1640
1641 if (filp->f_op && filp->f_op->flock)
1642 error = filp->f_op->flock(filp,
1643 (can_sleep) ? F_SETLKW : F_SETLK,
1644 lock);
1645 else
1646 error = flock_lock_file_wait(filp, lock);
1647
1648 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001649 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 out_putf:
1652 fput(filp);
1653 out:
1654 return error;
1655}
1656
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001657/**
1658 * vfs_test_lock - test file byte range lock
1659 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001660 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001661 *
1662 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1663 * setting conf->fl_type to something other than F_UNLCK.
1664 */
1665int vfs_test_lock(struct file *filp, struct file_lock *fl)
1666{
1667 if (filp->f_op && filp->f_op->lock)
1668 return filp->f_op->lock(filp, F_GETLK, fl);
1669 posix_test_lock(filp, fl);
1670 return 0;
1671}
1672EXPORT_SYMBOL_GPL(vfs_test_lock);
1673
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001674static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1675{
1676 flock->l_pid = fl->fl_pid;
1677#if BITS_PER_LONG == 32
1678 /*
1679 * Make sure we can represent the posix lock via
1680 * legacy 32bit flock.
1681 */
1682 if (fl->fl_start > OFFT_OFFSET_MAX)
1683 return -EOVERFLOW;
1684 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1685 return -EOVERFLOW;
1686#endif
1687 flock->l_start = fl->fl_start;
1688 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1689 fl->fl_end - fl->fl_start + 1;
1690 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001691 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001692 return 0;
1693}
1694
1695#if BITS_PER_LONG == 32
1696static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1697{
1698 flock->l_pid = fl->fl_pid;
1699 flock->l_start = fl->fl_start;
1700 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1701 fl->fl_end - fl->fl_start + 1;
1702 flock->l_whence = 0;
1703 flock->l_type = fl->fl_type;
1704}
1705#endif
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707/* Report the first existing lock that would conflict with l.
1708 * This implements the F_GETLK command of fcntl().
1709 */
1710int fcntl_getlk(struct file *filp, struct flock __user *l)
1711{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001712 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 struct flock flock;
1714 int error;
1715
1716 error = -EFAULT;
1717 if (copy_from_user(&flock, l, sizeof(flock)))
1718 goto out;
1719 error = -EINVAL;
1720 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1721 goto out;
1722
1723 error = flock_to_posix_lock(filp, &file_lock, &flock);
1724 if (error)
1725 goto out;
1726
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001727 error = vfs_test_lock(filp, &file_lock);
1728 if (error)
1729 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001731 flock.l_type = file_lock.fl_type;
1732 if (file_lock.fl_type != F_UNLCK) {
1733 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001734 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 }
1737 error = -EFAULT;
1738 if (!copy_to_user(l, &flock, sizeof(flock)))
1739 error = 0;
1740out:
1741 return error;
1742}
1743
Marc Eshel7723ec92007-01-18 15:08:55 -05001744/**
1745 * vfs_lock_file - file byte range lock
1746 * @filp: The file to apply the lock to
1747 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1748 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001749 * @conf: Place to return a copy of the conflicting lock, if found.
1750 *
1751 * A caller that doesn't care about the conflicting lock may pass NULL
1752 * as the final argument.
1753 *
1754 * If the filesystem defines a private ->lock() method, then @conf will
1755 * be left unchanged; so a caller that cares should initialize it to
1756 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001757 *
1758 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1759 * locks, the ->lock() interface may return asynchronously, before the lock has
1760 * been granted or denied by the underlying filesystem, if (and only if)
1761 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1762 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1763 * the request is for a blocking lock. When ->lock() does return asynchronously,
Miklos Szeredibde74e42008-07-25 01:48:57 -07001764 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001765 * request completes.
1766 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001767 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1768 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001769 * nonzero return code and the file system should release the lock. The file
1770 * system is also responsible to keep a corresponding posix lock when it
1771 * grants a lock so the VFS can find out which locks are locally held and do
1772 * the correct lock cleanup when required.
1773 * The underlying filesystem must not drop the kernel lock or call
Miklos Szeredibde74e42008-07-25 01:48:57 -07001774 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001775 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001776 */
Marc Eshel150b3932007-01-18 16:15:35 -05001777int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001778{
1779 if (filp->f_op && filp->f_op->lock)
1780 return filp->f_op->lock(filp, cmd, fl);
1781 else
Marc Eshel150b3932007-01-18 16:15:35 -05001782 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001783}
1784EXPORT_SYMBOL_GPL(vfs_lock_file);
1785
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001786static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1787 struct file_lock *fl)
1788{
1789 int error;
1790
1791 error = security_file_lock(filp, fl->fl_type);
1792 if (error)
1793 return error;
1794
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001795 for (;;) {
1796 error = vfs_lock_file(filp, cmd, fl, NULL);
1797 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001798 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001799 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1800 if (!error)
1801 continue;
1802
1803 locks_delete_block(fl);
1804 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001805 }
1806
1807 return error;
1808}
1809
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810/* Apply the lock described by l to an open file descriptor.
1811 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1812 */
Peter Staubachc2936212005-07-27 11:45:09 -07001813int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1814 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815{
1816 struct file_lock *file_lock = locks_alloc_lock();
1817 struct flock flock;
1818 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001819 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 int error;
1821
1822 if (file_lock == NULL)
1823 return -ENOLCK;
1824
1825 /*
1826 * This might block, so we do it before checking the inode.
1827 */
1828 error = -EFAULT;
1829 if (copy_from_user(&flock, l, sizeof(flock)))
1830 goto out;
1831
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001832 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
1834 /* Don't allow mandatory locks on files that may be memory mapped
1835 * and shared.
1836 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001837 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 error = -EAGAIN;
1839 goto out;
1840 }
1841
Peter Staubachc2936212005-07-27 11:45:09 -07001842again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 error = flock_to_posix_lock(filp, file_lock, &flock);
1844 if (error)
1845 goto out;
1846 if (cmd == F_SETLKW) {
1847 file_lock->fl_flags |= FL_SLEEP;
1848 }
1849
1850 error = -EBADF;
1851 switch (flock.l_type) {
1852 case F_RDLCK:
1853 if (!(filp->f_mode & FMODE_READ))
1854 goto out;
1855 break;
1856 case F_WRLCK:
1857 if (!(filp->f_mode & FMODE_WRITE))
1858 goto out;
1859 break;
1860 case F_UNLCK:
1861 break;
1862 default:
1863 error = -EINVAL;
1864 goto out;
1865 }
1866
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001867 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868
Peter Staubachc2936212005-07-27 11:45:09 -07001869 /*
1870 * Attempt to detect a close/fcntl race and recover by
1871 * releasing the lock that was just acquired.
1872 */
Al Viro0b2bac22008-05-06 13:58:34 -04001873 /*
1874 * we need that spin_lock here - it prevents reordering between
1875 * update of inode->i_flock and check for it done in close().
1876 * rcu_read_lock() wouldn't do.
1877 */
1878 spin_lock(&current->files->file_lock);
1879 f = fcheck(fd);
1880 spin_unlock(&current->files->file_lock);
1881 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001882 flock.l_type = F_UNLCK;
1883 goto again;
1884 }
1885
1886out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 locks_free_lock(file_lock);
1888 return error;
1889}
1890
1891#if BITS_PER_LONG == 32
1892/* Report the first existing lock that would conflict with l.
1893 * This implements the F_GETLK command of fcntl().
1894 */
1895int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1896{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001897 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 struct flock64 flock;
1899 int error;
1900
1901 error = -EFAULT;
1902 if (copy_from_user(&flock, l, sizeof(flock)))
1903 goto out;
1904 error = -EINVAL;
1905 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1906 goto out;
1907
1908 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1909 if (error)
1910 goto out;
1911
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001912 error = vfs_test_lock(filp, &file_lock);
1913 if (error)
1914 goto out;
1915
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001916 flock.l_type = file_lock.fl_type;
1917 if (file_lock.fl_type != F_UNLCK)
1918 posix_lock_to_flock64(&flock, &file_lock);
1919
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 error = -EFAULT;
1921 if (!copy_to_user(l, &flock, sizeof(flock)))
1922 error = 0;
1923
1924out:
1925 return error;
1926}
1927
1928/* Apply the lock described by l to an open file descriptor.
1929 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1930 */
Peter Staubachc2936212005-07-27 11:45:09 -07001931int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1932 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933{
1934 struct file_lock *file_lock = locks_alloc_lock();
1935 struct flock64 flock;
1936 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001937 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 int error;
1939
1940 if (file_lock == NULL)
1941 return -ENOLCK;
1942
1943 /*
1944 * This might block, so we do it before checking the inode.
1945 */
1946 error = -EFAULT;
1947 if (copy_from_user(&flock, l, sizeof(flock)))
1948 goto out;
1949
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001950 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
1952 /* Don't allow mandatory locks on files that may be memory mapped
1953 * and shared.
1954 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001955 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 error = -EAGAIN;
1957 goto out;
1958 }
1959
Peter Staubachc2936212005-07-27 11:45:09 -07001960again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 error = flock64_to_posix_lock(filp, file_lock, &flock);
1962 if (error)
1963 goto out;
1964 if (cmd == F_SETLKW64) {
1965 file_lock->fl_flags |= FL_SLEEP;
1966 }
1967
1968 error = -EBADF;
1969 switch (flock.l_type) {
1970 case F_RDLCK:
1971 if (!(filp->f_mode & FMODE_READ))
1972 goto out;
1973 break;
1974 case F_WRLCK:
1975 if (!(filp->f_mode & FMODE_WRITE))
1976 goto out;
1977 break;
1978 case F_UNLCK:
1979 break;
1980 default:
1981 error = -EINVAL;
1982 goto out;
1983 }
1984
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001985 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
Peter Staubachc2936212005-07-27 11:45:09 -07001987 /*
1988 * Attempt to detect a close/fcntl race and recover by
1989 * releasing the lock that was just acquired.
1990 */
Al Viro0b2bac22008-05-06 13:58:34 -04001991 spin_lock(&current->files->file_lock);
1992 f = fcheck(fd);
1993 spin_unlock(&current->files->file_lock);
1994 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001995 flock.l_type = F_UNLCK;
1996 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 }
1998
1999out:
2000 locks_free_lock(file_lock);
2001 return error;
2002}
2003#endif /* BITS_PER_LONG == 32 */
2004
2005/*
2006 * This function is called when the file is being removed
2007 * from the task's fd array. POSIX locks belonging to this task
2008 * are deleted at this time.
2009 */
2010void locks_remove_posix(struct file *filp, fl_owner_t owner)
2011{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002012 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014 /*
2015 * If there are no locks held on this file, we don't need to call
2016 * posix_lock_file(). Another process could be setting a lock on this
2017 * file at the same time, but we wouldn't remove that lock anyway.
2018 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002019 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 return;
2021
2022 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002023 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 lock.fl_start = 0;
2025 lock.fl_end = OFFSET_MAX;
2026 lock.fl_owner = owner;
2027 lock.fl_pid = current->tgid;
2028 lock.fl_file = filp;
2029 lock.fl_ops = NULL;
2030 lock.fl_lmops = NULL;
2031
Marc Eshel150b3932007-01-18 16:15:35 -05002032 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2035 lock.fl_ops->fl_release_private(&lock);
2036}
2037
2038EXPORT_SYMBOL(locks_remove_posix);
2039
2040/*
2041 * This function is called on the last close of an open file.
2042 */
2043void locks_remove_flock(struct file *filp)
2044{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002045 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 struct file_lock *fl;
2047 struct file_lock **before;
2048
2049 if (!inode->i_flock)
2050 return;
2051
2052 if (filp->f_op && filp->f_op->flock) {
2053 struct file_lock fl = {
2054 .fl_pid = current->tgid,
2055 .fl_file = filp,
2056 .fl_flags = FL_FLOCK,
2057 .fl_type = F_UNLCK,
2058 .fl_end = OFFSET_MAX,
2059 };
2060 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002061 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2062 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 }
2064
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002065 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 before = &inode->i_flock;
2067
2068 while ((fl = *before) != NULL) {
2069 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002070 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 locks_delete_lock(before);
2072 continue;
2073 }
2074 if (IS_LEASE(fl)) {
2075 lease_modify(before, F_UNLCK);
2076 continue;
2077 }
2078 /* What? */
2079 BUG();
2080 }
2081 before = &fl->fl_next;
2082 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002083 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084}
2085
2086/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 * posix_unblock_lock - stop waiting for a file lock
2088 * @filp: how the file was opened
2089 * @waiter: the lock which was waiting
2090 *
2091 * lockd needs to block waiting for locks.
2092 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002093int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2095{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002096 int status = 0;
2097
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002098 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002099 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002101 else
2102 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002103 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002104 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105}
2106
2107EXPORT_SYMBOL(posix_unblock_lock);
2108
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002109/**
2110 * vfs_cancel_lock - file byte range unblock lock
2111 * @filp: The file to apply the unblock to
2112 * @fl: The lock to be unblocked
2113 *
2114 * Used by lock managers to cancel blocked requests
2115 */
2116int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2117{
2118 if (filp->f_op && filp->f_op->lock)
2119 return filp->f_op->lock(filp, F_CANCELLK, fl);
2120 return 0;
2121}
2122
2123EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2124
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002125#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002126#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002127#include <linux/seq_file.h>
2128
2129static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002130 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131{
2132 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002133 unsigned int fl_pid;
2134
2135 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002136 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002137 else
2138 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139
2140 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002141 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
Jerome Marchand99dc8292010-10-26 14:22:33 -07002143 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002145 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2147 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002148 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149 } else if (IS_FLOCK(fl)) {
2150 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002151 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002153 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 }
2155 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002156 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002158 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002160 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002162 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002164 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 }
2166 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002167 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 (fl->fl_type & LOCK_READ)
2169 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2170 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2171 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002172 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 (fl->fl_type & F_INPROGRESS)
2174 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2175 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2176 }
2177 if (inode) {
2178#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002179 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 inode->i_sb->s_id, inode->i_ino);
2181#else
2182 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002183 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 MAJOR(inode->i_sb->s_dev),
2185 MINOR(inode->i_sb->s_dev), inode->i_ino);
2186#endif
2187 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002188 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 }
2190 if (IS_POSIX(fl)) {
2191 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002192 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002194 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002196 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 }
2198}
2199
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002200static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002202 struct file_lock *fl, *bfl;
2203
2204 fl = list_entry(v, struct file_lock, fl_link);
2205
Jerome Marchand99dc8292010-10-26 14:22:33 -07002206 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002207
2208 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002209 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002210
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002211 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212}
2213
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002214static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002216 loff_t *p = f->private;
2217
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002218 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002219 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002220 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221}
2222
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002223static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2224{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002225 loff_t *p = f->private;
2226 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002227 return seq_list_next(v, &file_lock_list, pos);
2228}
2229
2230static void locks_stop(struct seq_file *f, void *v)
2231{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002232 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002233}
2234
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002235static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002236 .start = locks_start,
2237 .next = locks_next,
2238 .stop = locks_stop,
2239 .show = locks_show,
2240};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002241
2242static int locks_open(struct inode *inode, struct file *filp)
2243{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002244 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002245}
2246
2247static const struct file_operations proc_locks_operations = {
2248 .open = locks_open,
2249 .read = seq_read,
2250 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002251 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002252};
2253
2254static int __init proc_locks_init(void)
2255{
2256 proc_create("locks", 0, NULL, &proc_locks_operations);
2257 return 0;
2258}
2259module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002260#endif
2261
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262/**
2263 * lock_may_read - checks that the region is free of locks
2264 * @inode: the inode that is being read
2265 * @start: the first byte to read
2266 * @len: the number of bytes to read
2267 *
2268 * Emulates Windows locking requirements. Whole-file
2269 * mandatory locks (share modes) can prohibit a read and
2270 * byte-range POSIX locks can prohibit a read if they overlap.
2271 *
2272 * N.B. this function is only ever called
2273 * from knfsd and ownership of locks is never checked.
2274 */
2275int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2276{
2277 struct file_lock *fl;
2278 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002279 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2281 if (IS_POSIX(fl)) {
2282 if (fl->fl_type == F_RDLCK)
2283 continue;
2284 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2285 continue;
2286 } else if (IS_FLOCK(fl)) {
2287 if (!(fl->fl_type & LOCK_MAND))
2288 continue;
2289 if (fl->fl_type & LOCK_READ)
2290 continue;
2291 } else
2292 continue;
2293 result = 0;
2294 break;
2295 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002296 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 return result;
2298}
2299
2300EXPORT_SYMBOL(lock_may_read);
2301
2302/**
2303 * lock_may_write - checks that the region is free of locks
2304 * @inode: the inode that is being written
2305 * @start: the first byte to write
2306 * @len: the number of bytes to write
2307 *
2308 * Emulates Windows locking requirements. Whole-file
2309 * mandatory locks (share modes) can prohibit a write and
2310 * byte-range POSIX locks can prohibit a write if they overlap.
2311 *
2312 * N.B. this function is only ever called
2313 * from knfsd and ownership of locks is never checked.
2314 */
2315int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2316{
2317 struct file_lock *fl;
2318 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002319 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2321 if (IS_POSIX(fl)) {
2322 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2323 continue;
2324 } else if (IS_FLOCK(fl)) {
2325 if (!(fl->fl_type & LOCK_MAND))
2326 continue;
2327 if (fl->fl_type & LOCK_WRITE)
2328 continue;
2329 } else
2330 continue;
2331 result = 0;
2332 break;
2333 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002334 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 return result;
2336}
2337
2338EXPORT_SYMBOL(lock_may_write);
2339
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340static int __init filelock_init(void)
2341{
2342 filelock_cache = kmem_cache_create("file_lock_cache",
2343 sizeof(struct file_lock), 0, SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002344 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 return 0;
2346}
2347
2348core_initcall(filelock_init);