blob: 0f3998291f78e4fa08b5d3c102f6d8dcc120247a [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
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700447static const struct lock_manager_operations lease_manager_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 .fl_break = lease_break_callback,
449 .fl_release_private = lease_release_private_callback,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 .fl_change = lease_modify,
451};
452
453/*
454 * Initialize a lease, use the default lock manager operations
455 */
456static int lease_init(struct file *filp, int type, struct file_lock *fl)
457 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400458 if (assign_type(fl, type) != 0)
459 return -EINVAL;
460
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 fl->fl_owner = current->files;
462 fl->fl_pid = current->tgid;
463
464 fl->fl_file = filp;
465 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 fl->fl_start = 0;
467 fl->fl_end = OFFSET_MAX;
468 fl->fl_ops = NULL;
469 fl->fl_lmops = &lease_manager_ops;
470 return 0;
471}
472
473/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500474static struct file_lock *lease_alloc(struct file *filp, int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
476 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400477 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
479 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500480 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400483 if (error) {
484 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500485 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400486 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500487 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488}
489
490/* Check if two locks overlap each other.
491 */
492static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
493{
494 return ((fl1->fl_end >= fl2->fl_start) &&
495 (fl2->fl_end >= fl1->fl_start));
496}
497
498/*
499 * Check whether two locks have the same owner.
500 */
Matt Mackall33443c42006-01-08 01:05:22 -0800501static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502{
503 if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
504 return fl2->fl_lmops == fl1->fl_lmops &&
505 fl1->fl_lmops->fl_compare_owner(fl1, fl2);
506 return fl1->fl_owner == fl2->fl_owner;
507}
508
509/* Remove waiter from blocker's block list.
510 * When blocker ends up pointing to itself then the list is empty.
511 */
Matt Mackall33443c42006-01-08 01:05:22 -0800512static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513{
514 list_del_init(&waiter->fl_block);
515 list_del_init(&waiter->fl_link);
516 waiter->fl_next = NULL;
517}
518
519/*
520 */
521static void locks_delete_block(struct file_lock *waiter)
522{
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200523 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 __locks_delete_block(waiter);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200525 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526}
527
528/* Insert waiter into blocker's block list.
529 * We use a circular list so that processes can be easily woken up in
530 * the order they blocked. The documentation doesn't require this but
531 * it seems like the reasonable thing to do.
532 */
533static void locks_insert_block(struct file_lock *blocker,
534 struct file_lock *waiter)
535{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800536 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 list_add_tail(&waiter->fl_block, &blocker->fl_block);
538 waiter->fl_next = blocker;
539 if (IS_POSIX(blocker))
540 list_add(&waiter->fl_link, &blocked_list);
541}
542
543/* Wake up processes blocked waiting for blocker.
544 * If told to wait then schedule the processes until the block list
545 * is empty, otherwise empty the block list ourselves.
546 */
547static void locks_wake_up_blocks(struct file_lock *blocker)
548{
549 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400550 struct file_lock *waiter;
551
552 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 struct file_lock, fl_block);
554 __locks_delete_block(waiter);
555 if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
556 waiter->fl_lmops->fl_notify(waiter);
557 else
558 wake_up(&waiter->fl_wait);
559 }
560}
561
562/* Insert file lock fl into an inode's lock list at the position indicated
563 * by pos. At the same time add the lock to the global file lock list.
564 */
565static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
566{
567 list_add(&fl->fl_link, &file_lock_list);
568
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000569 fl->fl_nspid = get_pid(task_tgid(current));
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 /* insert into file's list */
572 fl->fl_next = *pos;
573 *pos = fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574}
575
576/*
577 * Delete a lock and then free it.
578 * Wake up processes that are blocked waiting for this lock,
579 * notify the FS that the lock has been cleared and
580 * finally free the lock.
581 */
582static void locks_delete_lock(struct file_lock **thisfl_p)
583{
584 struct file_lock *fl = *thisfl_p;
585
586 *thisfl_p = fl->fl_next;
587 fl->fl_next = NULL;
588 list_del_init(&fl->fl_link);
589
590 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
591 if (fl->fl_fasync != NULL) {
592 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
593 fl->fl_fasync = NULL;
594 }
595
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000596 if (fl->fl_nspid) {
597 put_pid(fl->fl_nspid);
598 fl->fl_nspid = NULL;
599 }
600
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 locks_wake_up_blocks(fl);
602 locks_free_lock(fl);
603}
604
605/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
606 * checks for shared/exclusive status of overlapping locks.
607 */
608static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
609{
610 if (sys_fl->fl_type == F_WRLCK)
611 return 1;
612 if (caller_fl->fl_type == F_WRLCK)
613 return 1;
614 return 0;
615}
616
617/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
618 * checking before calling the locks_conflict().
619 */
620static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
621{
622 /* POSIX locks owned by the same process do not conflict with
623 * each other.
624 */
625 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
626 return (0);
627
628 /* Check whether they overlap */
629 if (!locks_overlap(caller_fl, sys_fl))
630 return 0;
631
632 return (locks_conflict(caller_fl, sys_fl));
633}
634
635/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
636 * checking before calling the locks_conflict().
637 */
638static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
639{
640 /* FLOCK locks referring to the same filp do not conflict with
641 * each other.
642 */
643 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
644 return (0);
645 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
646 return 0;
647
648 return (locks_conflict(caller_fl, sys_fl));
649}
650
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400651void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500652posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
654 struct file_lock *cfl;
655
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200656 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800657 for (cfl = filp->f_path.dentry->d_inode->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 if (!IS_POSIX(cfl))
659 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400660 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 break;
662 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000663 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500664 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000665 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800666 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000667 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400668 fl->fl_type = F_UNLCK;
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200669 unlock_flocks();
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400670 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672EXPORT_SYMBOL(posix_test_lock);
673
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400674/*
675 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400677 * We attempt to detect deadlocks that are due purely to posix file
678 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400680 * We assume that a task can be waiting for at most one lock at a time.
681 * So for any acquired lock, the process holding that lock may be
682 * waiting on at most one other lock. That lock in turns may be held by
683 * someone waiting for at most one other lock. Given a requested lock
684 * caller_fl which is about to wait for a conflicting lock block_fl, we
685 * follow this chain of waiters to ensure we are not about to create a
686 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400687 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400688 * Since we do this before we ever put a process to sleep on a lock, we
689 * are ensured that there is never a cycle; that is what guarantees that
690 * the while() loop in posix_locks_deadlock() eventually completes.
691 *
692 * Note: the above assumption may not be true when handling lock
693 * requests from a broken NFS client. It may also fail in the presence
694 * of tasks (such as posix threads) sharing the same open file table.
695 *
696 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400698
699#define MAX_DEADLK_ITERATIONS 10
700
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400701/* Find a lock that the owner of the given block_fl is blocking on. */
702static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
703{
704 struct file_lock *fl;
705
706 list_for_each_entry(fl, &blocked_list, fl_link) {
707 if (posix_same_owner(fl, block_fl))
708 return fl->fl_next;
709 }
710 return NULL;
711}
712
Adrian Bunkb0904e12006-06-23 02:05:13 -0700713static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 struct file_lock *block_fl)
715{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400716 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400718 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
719 if (i++ > MAX_DEADLK_ITERATIONS)
720 return 0;
721 if (posix_same_owner(caller_fl, block_fl))
722 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 }
724 return 0;
725}
726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400728 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400729 *
730 * Note that if called with an FL_EXISTS argument, the caller may determine
731 * whether or not a lock was successfully freed by testing the return
732 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800734static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800736 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 struct file_lock **before;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800738 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 int error = 0;
740 int found = 0;
741
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200742 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
743 new_fl = locks_alloc_lock();
744 if (!new_fl)
745 return -ENOMEM;
746 }
747
748 lock_flocks();
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400749 if (request->fl_flags & FL_ACCESS)
750 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 for_each_lock(inode, before) {
753 struct file_lock *fl = *before;
754 if (IS_POSIX(fl))
755 break;
756 if (IS_LEASE(fl))
757 continue;
758 if (filp != fl->fl_file)
759 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800760 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 goto out;
762 found = 1;
763 locks_delete_lock(before);
764 break;
765 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766
Trond Myklebustf475ae92006-06-29 16:38:32 -0400767 if (request->fl_type == F_UNLCK) {
768 if ((request->fl_flags & FL_EXISTS) && !found)
769 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800770 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400771 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
773 /*
774 * If a higher-priority process was blocked on the old file lock,
775 * give it the opportunity to lock the file.
776 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200777 if (found) {
778 unlock_flocks();
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200779 cond_resched();
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200780 lock_flocks();
781 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400783find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 for_each_lock(inode, before) {
785 struct file_lock *fl = *before;
786 if (IS_POSIX(fl))
787 break;
788 if (IS_LEASE(fl))
789 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800790 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 continue;
792 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700793 if (!(request->fl_flags & FL_SLEEP))
794 goto out;
795 error = FILE_LOCK_DEFERRED;
796 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 goto out;
798 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400799 if (request->fl_flags & FL_ACCESS)
800 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800801 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700802 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800803 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400804 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805
806out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200807 unlock_flocks();
Trond Myklebust993dfa82006-03-31 02:30:55 -0800808 if (new_fl)
809 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 return error;
811}
812
Marc Eshel150b3932007-01-18 16:15:35 -0500813static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814{
815 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700816 struct file_lock *new_fl = NULL;
817 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 struct file_lock *left = NULL;
819 struct file_lock *right = NULL;
820 struct file_lock **before;
821 int error, added = 0;
822
823 /*
824 * We may need two file_lock structures for this operation,
825 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700826 *
827 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700829 if (!(request->fl_flags & FL_ACCESS) &&
830 (request->fl_type != F_UNLCK ||
831 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
832 new_fl = locks_alloc_lock();
833 new_fl2 = locks_alloc_lock();
834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200836 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 if (request->fl_type != F_UNLCK) {
838 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500839 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 if (!IS_POSIX(fl))
841 continue;
842 if (!posix_locks_conflict(request, fl))
843 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800844 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400845 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 error = -EAGAIN;
847 if (!(request->fl_flags & FL_SLEEP))
848 goto out;
849 error = -EDEADLK;
850 if (posix_locks_deadlock(request, fl))
851 goto out;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700852 error = FILE_LOCK_DEFERRED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 locks_insert_block(fl, request);
854 goto out;
855 }
856 }
857
858 /* If we're just looking for a conflict, we're done. */
859 error = 0;
860 if (request->fl_flags & FL_ACCESS)
861 goto out;
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 * Find the first old lock with the same owner as the new lock.
865 */
866
867 before = &inode->i_flock;
868
869 /* First skip locks owned by other processes. */
870 while ((fl = *before) && (!IS_POSIX(fl) ||
871 !posix_same_owner(request, fl))) {
872 before = &fl->fl_next;
873 }
874
875 /* Process locks with this owner. */
876 while ((fl = *before) && posix_same_owner(request, fl)) {
877 /* Detect adjacent or overlapping regions (if same lock type)
878 */
879 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700880 /* In all comparisons of start vs end, use
881 * "start - 1" rather than "end + 1". If end
882 * is OFFSET_MAX, end + 1 will become negative.
883 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 if (fl->fl_end < request->fl_start - 1)
885 goto next_lock;
886 /* If the next lock in the list has entirely bigger
887 * addresses than the new one, insert the lock here.
888 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700889 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 break;
891
892 /* If we come here, the new and old lock are of the
893 * same type and adjacent or overlapping. Make one
894 * lock yielding from the lower start address of both
895 * locks to the higher end address.
896 */
897 if (fl->fl_start > request->fl_start)
898 fl->fl_start = request->fl_start;
899 else
900 request->fl_start = fl->fl_start;
901 if (fl->fl_end < request->fl_end)
902 fl->fl_end = request->fl_end;
903 else
904 request->fl_end = fl->fl_end;
905 if (added) {
906 locks_delete_lock(before);
907 continue;
908 }
909 request = fl;
910 added = 1;
911 }
912 else {
913 /* Processing for different lock types is a bit
914 * more complex.
915 */
916 if (fl->fl_end < request->fl_start)
917 goto next_lock;
918 if (fl->fl_start > request->fl_end)
919 break;
920 if (request->fl_type == F_UNLCK)
921 added = 1;
922 if (fl->fl_start < request->fl_start)
923 left = fl;
924 /* If the next lock in the list has a higher end
925 * address than the new one, insert the new one here.
926 */
927 if (fl->fl_end > request->fl_end) {
928 right = fl;
929 break;
930 }
931 if (fl->fl_start >= request->fl_start) {
932 /* The new lock completely replaces an old
933 * one (This may happen several times).
934 */
935 if (added) {
936 locks_delete_lock(before);
937 continue;
938 }
939 /* Replace the old lock with the new one.
940 * Wake up anybody waiting for the old one,
941 * as the change in lock type might satisfy
942 * their needs.
943 */
944 locks_wake_up_blocks(fl);
945 fl->fl_start = request->fl_start;
946 fl->fl_end = request->fl_end;
947 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -0500948 locks_release_private(fl);
949 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 request = fl;
951 added = 1;
952 }
953 }
954 /* Go on to next lock.
955 */
956 next_lock:
957 before = &fl->fl_next;
958 }
959
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700960 /*
961 * The above code only modifies existing locks in case of
962 * merging or replacing. If new lock(s) need to be inserted
963 * all modifications are done bellow this, so it's safe yet to
964 * bail out.
965 */
966 error = -ENOLCK; /* "no luck" */
967 if (right && left == right && !new_fl2)
968 goto out;
969
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 error = 0;
971 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -0400972 if (request->fl_type == F_UNLCK) {
973 if (request->fl_flags & FL_EXISTS)
974 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400976 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -0700977
978 if (!new_fl) {
979 error = -ENOLCK;
980 goto out;
981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 locks_copy_lock(new_fl, request);
983 locks_insert_lock(before, new_fl);
984 new_fl = NULL;
985 }
986 if (right) {
987 if (left == right) {
988 /* The new lock breaks the old one in two pieces,
989 * so we have to use the second new lock.
990 */
991 left = new_fl2;
992 new_fl2 = NULL;
993 locks_copy_lock(left, right);
994 locks_insert_lock(before, left);
995 }
996 right->fl_start = request->fl_end + 1;
997 locks_wake_up_blocks(right);
998 }
999 if (left) {
1000 left->fl_end = request->fl_start - 1;
1001 locks_wake_up_blocks(left);
1002 }
1003 out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001004 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 /*
1006 * Free any unused locks.
1007 */
1008 if (new_fl)
1009 locks_free_lock(new_fl);
1010 if (new_fl2)
1011 locks_free_lock(new_fl2);
1012 return error;
1013}
1014
1015/**
1016 * posix_lock_file - Apply a POSIX-style lock to a file
1017 * @filp: The file to apply the lock to
1018 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001019 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 *
1021 * Add a POSIX style lock to a file.
1022 * We merge adjacent & overlapping locks whenever possible.
1023 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001024 *
1025 * Note that if called with an FL_EXISTS argument, the caller may determine
1026 * whether or not a lock was successfully freed by testing the return
1027 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 */
Marc Eshel150b3932007-01-18 16:15:35 -05001029int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001030 struct file_lock *conflock)
1031{
Marc Eshel150b3932007-01-18 16:15:35 -05001032 return __posix_lock_file(filp->f_path.dentry->d_inode, fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001033}
Marc Eshel150b3932007-01-18 16:15:35 -05001034EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
1036/**
1037 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1038 * @filp: The file to apply the lock to
1039 * @fl: The lock to be applied
1040 *
1041 * Add a POSIX style lock to a file.
1042 * We merge adjacent & overlapping locks whenever possible.
1043 * POSIX locks are sorted by owner task, then by starting address
1044 */
1045int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1046{
1047 int error;
1048 might_sleep ();
1049 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001050 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001051 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 break;
1053 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1054 if (!error)
1055 continue;
1056
1057 locks_delete_block(fl);
1058 break;
1059 }
1060 return error;
1061}
1062EXPORT_SYMBOL(posix_lock_file_wait);
1063
1064/**
1065 * locks_mandatory_locked - Check for an active lock
1066 * @inode: the file to check
1067 *
1068 * Searches the inode's list of locks to find any POSIX locks which conflict.
1069 * This function is called from locks_verify_locked() only.
1070 */
1071int locks_mandatory_locked(struct inode *inode)
1072{
1073 fl_owner_t owner = current->files;
1074 struct file_lock *fl;
1075
1076 /*
1077 * Search the lock list for this inode for any POSIX locks.
1078 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001079 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1081 if (!IS_POSIX(fl))
1082 continue;
1083 if (fl->fl_owner != owner)
1084 break;
1085 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001086 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 return fl ? -EAGAIN : 0;
1088}
1089
1090/**
1091 * locks_mandatory_area - Check for a conflicting lock
1092 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1093 * for shared
1094 * @inode: the file to check
1095 * @filp: how the file was opened (if it was)
1096 * @offset: start of area to check
1097 * @count: length of area to check
1098 *
1099 * Searches the inode's list of locks to find any POSIX locks which conflict.
1100 * This function is called from rw_verify_area() and
1101 * locks_verify_truncate().
1102 */
1103int locks_mandatory_area(int read_write, struct inode *inode,
1104 struct file *filp, loff_t offset,
1105 size_t count)
1106{
1107 struct file_lock fl;
1108 int error;
1109
1110 locks_init_lock(&fl);
1111 fl.fl_owner = current->files;
1112 fl.fl_pid = current->tgid;
1113 fl.fl_file = filp;
1114 fl.fl_flags = FL_POSIX | FL_ACCESS;
1115 if (filp && !(filp->f_flags & O_NONBLOCK))
1116 fl.fl_flags |= FL_SLEEP;
1117 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1118 fl.fl_start = offset;
1119 fl.fl_end = offset + count - 1;
1120
1121 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001122 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001123 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 break;
1125 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1126 if (!error) {
1127 /*
1128 * If we've been sleeping someone might have
1129 * changed the permissions behind our back.
1130 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001131 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 continue;
1133 }
1134
1135 locks_delete_block(&fl);
1136 break;
1137 }
1138
1139 return error;
1140}
1141
1142EXPORT_SYMBOL(locks_mandatory_area);
1143
1144/* We already had a lease on this file; just change its type */
1145int lease_modify(struct file_lock **before, int arg)
1146{
1147 struct file_lock *fl = *before;
1148 int error = assign_type(fl, arg);
1149
1150 if (error)
1151 return error;
1152 locks_wake_up_blocks(fl);
1153 if (arg == F_UNLCK)
1154 locks_delete_lock(before);
1155 return 0;
1156}
1157
1158EXPORT_SYMBOL(lease_modify);
1159
1160static void time_out_leases(struct inode *inode)
1161{
1162 struct file_lock **before;
1163 struct file_lock *fl;
1164
1165 before = &inode->i_flock;
1166 while ((fl = *before) && IS_LEASE(fl) && (fl->fl_type & F_INPROGRESS)) {
1167 if ((fl->fl_break_time == 0)
1168 || time_before(jiffies, fl->fl_break_time)) {
1169 before = &fl->fl_next;
1170 continue;
1171 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 lease_modify(before, fl->fl_type & ~F_INPROGRESS);
1173 if (fl == *before) /* lease_modify may have freed fl */
1174 before = &fl->fl_next;
1175 }
1176}
1177
1178/**
1179 * __break_lease - revoke all outstanding leases on file
1180 * @inode: the inode of the file to return
1181 * @mode: the open mode (read or write)
1182 *
david m. richter87250dd2007-05-09 16:10:27 -04001183 * break_lease (inlined for speed) has checked there already is at least
1184 * some kind of lock (maybe a lease) on this file. Leases are broken on
1185 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 * specified %O_NONBLOCK to your open().
1187 */
1188int __break_lease(struct inode *inode, unsigned int mode)
1189{
1190 int error = 0, future;
1191 struct file_lock *new_fl, *flock;
1192 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 unsigned long break_time;
1194 int i_have_this_lease = 0;
Al Viro8737c932009-12-24 06:47:55 -05001195 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Al Viro8737c932009-12-24 06:47:55 -05001197 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001199 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201 time_out_leases(inode);
1202
1203 flock = inode->i_flock;
1204 if ((flock == NULL) || !IS_LEASE(flock))
1205 goto out;
1206
1207 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next)
1208 if (fl->fl_owner == current->files)
1209 i_have_this_lease = 1;
1210
Al Viro8737c932009-12-24 06:47:55 -05001211 if (want_write) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 /* If we want write access, we have to revoke any lease. */
1213 future = F_UNLCK | F_INPROGRESS;
1214 } else if (flock->fl_type & F_INPROGRESS) {
1215 /* If the lease is already being broken, we just leave it */
1216 future = flock->fl_type;
1217 } else if (flock->fl_type & F_WRLCK) {
1218 /* Downgrade the exclusive lease to a read-only lease. */
1219 future = F_RDLCK | F_INPROGRESS;
1220 } else {
1221 /* the existing lease was read-only, so we can read too. */
1222 goto out;
1223 }
1224
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001225 if (IS_ERR(new_fl) && !i_have_this_lease
1226 && ((mode & O_NONBLOCK) == 0)) {
1227 error = PTR_ERR(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 goto out;
1229 }
1230
1231 break_time = 0;
1232 if (lease_break_time > 0) {
1233 break_time = jiffies + lease_break_time * HZ;
1234 if (break_time == 0)
1235 break_time++; /* so that 0 means no break time */
1236 }
1237
1238 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1239 if (fl->fl_type != future) {
1240 fl->fl_type = future;
1241 fl->fl_break_time = break_time;
1242 /* lease must have lmops break callback */
1243 fl->fl_lmops->fl_break(fl);
1244 }
1245 }
1246
1247 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1248 error = -EWOULDBLOCK;
1249 goto out;
1250 }
1251
1252restart:
1253 break_time = flock->fl_break_time;
1254 if (break_time != 0) {
1255 break_time -= jiffies;
1256 if (break_time == 0)
1257 break_time++;
1258 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001259 locks_insert_block(flock, new_fl);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001260 unlock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001261 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1262 !new_fl->fl_next, break_time);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001263 lock_flocks();
Matthew Wilcox4321e012008-01-14 21:28:30 -07001264 __locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 if (error >= 0) {
1266 if (error == 0)
1267 time_out_leases(inode);
1268 /* Wait for the next lease that has not been broken yet */
1269 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1270 flock = flock->fl_next) {
1271 if (flock->fl_type & F_INPROGRESS)
1272 goto restart;
1273 }
1274 error = 0;
1275 }
1276
1277out:
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001278 unlock_flocks();
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -05001279 if (!IS_ERR(new_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 locks_free_lock(new_fl);
1281 return error;
1282}
1283
1284EXPORT_SYMBOL(__break_lease);
1285
1286/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001287 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 * @inode: the inode
1289 * @time: pointer to a timespec which will contain the last modified time
1290 *
1291 * This is to force NFS clients to flush their caches for files with
1292 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001293 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 */
1295void lease_get_mtime(struct inode *inode, struct timespec *time)
1296{
1297 struct file_lock *flock = inode->i_flock;
1298 if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
1299 *time = current_fs_time(inode->i_sb);
1300 else
1301 *time = inode->i_mtime;
1302}
1303
1304EXPORT_SYMBOL(lease_get_mtime);
1305
1306/**
1307 * fcntl_getlease - Enquire what lease is currently active
1308 * @filp: the file
1309 *
1310 * The value returned by this function will be one of
1311 * (if no lease break is pending):
1312 *
1313 * %F_RDLCK to indicate a shared lease is held.
1314 *
1315 * %F_WRLCK to indicate an exclusive lease is held.
1316 *
1317 * %F_UNLCK to indicate no lease is held.
1318 *
1319 * (if a lease break is pending):
1320 *
1321 * %F_RDLCK to indicate an exclusive lease needs to be
1322 * changed to a shared lease (or removed).
1323 *
1324 * %F_UNLCK to indicate the lease needs to be removed.
1325 *
1326 * XXX: sfr & willy disagree over whether F_INPROGRESS
1327 * should be returned to userspace.
1328 */
1329int fcntl_getlease(struct file *filp)
1330{
1331 struct file_lock *fl;
1332 int type = F_UNLCK;
1333
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001334 lock_flocks();
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001335 time_out_leases(filp->f_path.dentry->d_inode);
1336 for (fl = filp->f_path.dentry->d_inode->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 fl = fl->fl_next) {
1338 if (fl->fl_file == filp) {
1339 type = fl->fl_type & ~F_INPROGRESS;
1340 break;
1341 }
1342 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001343 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 return type;
1345}
1346
1347/**
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001348 * generic_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 * @filp: file pointer
1350 * @arg: type of lease to obtain
1351 * @flp: input - file_lock to use, output - file_lock inserted
1352 *
1353 * The (input) flp->fl_lmops->fl_break function is required
1354 * by break_lease().
1355 *
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001356 * Called with file_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 */
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001358int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001360 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001361 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 struct inode *inode = dentry->d_inode;
1363 int error, rdlease_count = 0, wrlease_count = 0;
1364
J. Bruce Fields096657b2010-10-30 17:31:14 -04001365 lease = *flp;
1366
1367 error = -EACCES;
David Howellsda9592e2008-11-14 10:39:05 +11001368 if ((current_fsuid() != inode->i_uid) && !capable(CAP_LEASE))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001369 goto out;
1370 error = -EINVAL;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001371 if (!S_ISREG(inode->i_mode))
J. Bruce Fields096657b2010-10-30 17:31:14 -04001372 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001373 error = security_file_lock(filp, arg);
1374 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001375 goto out;
J. Bruce Fields6d5e8b02007-05-31 17:03:46 -04001376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 time_out_leases(inode);
1378
J. Bruce Fieldsd2ab0b0c2007-06-30 12:40:32 -04001379 BUG_ON(!(*flp)->fl_lmops->fl_break);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
David M. Richter288b2fd2008-04-23 16:29:00 -04001381 if (arg != F_UNLCK) {
1382 error = -EAGAIN;
1383 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1384 goto out;
1385 if ((arg == F_WRLCK)
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001386 && ((dentry->d_count > 1)
David M. Richter288b2fd2008-04-23 16:29:00 -04001387 || (atomic_read(&inode->i_count) > 1)))
1388 goto out;
David M. Richter288b2fd2008-04-23 16:29:00 -04001389 }
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 /*
1392 * At this point, we know that if there is an exclusive
1393 * lease on this file, then we hold it on this filp
1394 * (otherwise our open of this file would have blocked).
1395 * And if we are trying to acquire an exclusive lease,
1396 * then the file is not open by anyone (including us)
1397 * except for this filp.
1398 */
1399 for (before = &inode->i_flock;
1400 ((fl = *before) != NULL) && IS_LEASE(fl);
1401 before = &fl->fl_next) {
J. Bruce Fields255c7cf2010-10-31 12:35:48 -04001402 if (fl->fl_file == filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 my_before = before;
1404 else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
1405 /*
1406 * Someone is in the process of opening this
1407 * file for writing so we may not take an
1408 * exclusive lease on it.
1409 */
1410 wrlease_count++;
1411 else
1412 rdlease_count++;
1413 }
1414
David M. Richter5fcc60c2008-04-23 16:28:59 -04001415 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 if ((arg == F_RDLCK && (wrlease_count > 0)) ||
1417 (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
1418 goto out;
1419
1420 if (my_before != NULL) {
1421 error = lease->fl_lmops->fl_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001422 if (!error)
1423 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 goto out;
1425 }
1426
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 if (arg == F_UNLCK)
1428 goto out;
1429
1430 error = -EINVAL;
1431 if (!leases_enable)
1432 goto out;
1433
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001434 locks_insert_lock(before, lease);
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001435 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437out:
1438 return error;
1439}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001440EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001442static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1443{
1444 if (filp->f_op && filp->f_op->setlease)
1445 return filp->f_op->setlease(filp, arg, lease);
1446 else
1447 return generic_setlease(filp, arg, lease);
1448}
1449
1450/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001451 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 * @filp: file pointer
1453 * @arg: type of lease to obtain
1454 * @lease: file_lock to use
1455 *
1456 * Call this to establish a lease on the file.
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001457 * The (*lease)->fl_lmops->fl_break operation must be set; if not,
1458 * break_lease will oops!
1459 *
1460 * This will call the filesystem's setlease file method, if
1461 * defined. Note that there is no getlease method; instead, the
1462 * filesystem setlease method should call back to setlease() to
1463 * add a lease to the inode's lease list, where fcntl_getlease() can
1464 * find it. Since fcntl_getlease() only reports whether the current
1465 * task holds a lease, a cluster filesystem need only do this for
1466 * leases held by processes on this node.
1467 *
1468 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001469 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001470 * filesystem's open, create, and (on truncate) setattr methods.
1471 *
1472 * Warning: the only current setlease methods exist only to disable
1473 * leases in certain cases. More vfs changes may be required to
1474 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 */
1476
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001477int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 int error;
1480
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001481 lock_flocks();
1482 error = __vfs_setlease(filp, arg, lease);
1483 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
1485 return error;
1486}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001487EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001489static int do_fcntl_delete_lease(struct file *filp)
1490{
1491 struct file_lock fl, *flp = &fl;
1492
1493 lease_init(filp, F_UNLCK, flp);
1494
1495 return vfs_setlease(filp, F_UNLCK, &flp);
1496}
1497
1498static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001500 struct file_lock *fl, *ret;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001501 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 int error;
1503
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001504 fl = lease_alloc(filp, arg);
1505 if (IS_ERR(fl))
1506 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001508 new = fasync_alloc();
1509 if (!new) {
1510 locks_free_lock(fl);
1511 return -ENOMEM;
1512 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001513 ret = fl;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001514 lock_flocks();
J. Bruce Fields8896b932010-11-03 18:09:18 -04001515 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001516 if (error) {
1517 unlock_flocks();
1518 locks_free_lock(fl);
1519 goto out_free_fasync;
1520 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001521 if (ret != fl)
1522 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001524 /*
1525 * fasync_insert_entry() returns the old entry if any.
1526 * If there was no old entry, then it used 'new' and
1527 * inserted it into the fasync list. Clear new so that
1528 * we don't release it here.
1529 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001530 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001531 new = NULL;
1532
J. Bruce Fields8896b932010-11-03 18:09:18 -04001533 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001534 unlock_flocks();
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001535
1536out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001537 if (new)
1538 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 return error;
1540}
1541
1542/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001543 * fcntl_setlease - sets a lease on an open file
1544 * @fd: open file descriptor
1545 * @filp: file pointer
1546 * @arg: type of lease to obtain
1547 *
1548 * Call this fcntl to establish a lease on the file.
1549 * Note that you also need to call %F_SETSIG to
1550 * receive a signal when the lease is broken.
1551 */
1552int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1553{
1554 if (arg == F_UNLCK)
1555 return do_fcntl_delete_lease(filp);
1556 return do_fcntl_add_lease(fd, filp, arg);
1557}
1558
1559/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1561 * @filp: The file to apply the lock to
1562 * @fl: The lock to be applied
1563 *
1564 * Add a FLOCK style lock to a file.
1565 */
1566int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1567{
1568 int error;
1569 might_sleep();
1570 for (;;) {
1571 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001572 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 break;
1574 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1575 if (!error)
1576 continue;
1577
1578 locks_delete_block(fl);
1579 break;
1580 }
1581 return error;
1582}
1583
1584EXPORT_SYMBOL(flock_lock_file_wait);
1585
1586/**
1587 * sys_flock: - flock() system call.
1588 * @fd: the file descriptor to lock.
1589 * @cmd: the type of lock to apply.
1590 *
1591 * Apply a %FL_FLOCK style lock to an open file descriptor.
1592 * The @cmd can be one of
1593 *
1594 * %LOCK_SH -- a shared lock.
1595 *
1596 * %LOCK_EX -- an exclusive lock.
1597 *
1598 * %LOCK_UN -- remove an existing lock.
1599 *
1600 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1601 *
1602 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1603 * processes read and write access respectively.
1604 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001605SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606{
1607 struct file *filp;
1608 struct file_lock *lock;
1609 int can_sleep, unlock;
1610 int error;
1611
1612 error = -EBADF;
1613 filp = fget(fd);
1614 if (!filp)
1615 goto out;
1616
1617 can_sleep = !(cmd & LOCK_NB);
1618 cmd &= ~LOCK_NB;
1619 unlock = (cmd == LOCK_UN);
1620
Al Viroaeb5d722008-09-02 15:28:45 -04001621 if (!unlock && !(cmd & LOCK_MAND) &&
1622 !(filp->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 goto out_putf;
1624
1625 error = flock_make_lock(filp, &lock, cmd);
1626 if (error)
1627 goto out_putf;
1628 if (can_sleep)
1629 lock->fl_flags |= FL_SLEEP;
1630
Sten Spans713c0ec2009-07-16 09:41:39 +02001631 error = security_file_lock(filp, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (error)
1633 goto out_free;
1634
1635 if (filp->f_op && filp->f_op->flock)
1636 error = filp->f_op->flock(filp,
1637 (can_sleep) ? F_SETLKW : F_SETLK,
1638 lock);
1639 else
1640 error = flock_lock_file_wait(filp, lock);
1641
1642 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001643 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644
1645 out_putf:
1646 fput(filp);
1647 out:
1648 return error;
1649}
1650
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001651/**
1652 * vfs_test_lock - test file byte range lock
1653 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001654 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001655 *
1656 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1657 * setting conf->fl_type to something other than F_UNLCK.
1658 */
1659int vfs_test_lock(struct file *filp, struct file_lock *fl)
1660{
1661 if (filp->f_op && filp->f_op->lock)
1662 return filp->f_op->lock(filp, F_GETLK, fl);
1663 posix_test_lock(filp, fl);
1664 return 0;
1665}
1666EXPORT_SYMBOL_GPL(vfs_test_lock);
1667
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001668static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1669{
1670 flock->l_pid = fl->fl_pid;
1671#if BITS_PER_LONG == 32
1672 /*
1673 * Make sure we can represent the posix lock via
1674 * legacy 32bit flock.
1675 */
1676 if (fl->fl_start > OFFT_OFFSET_MAX)
1677 return -EOVERFLOW;
1678 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1679 return -EOVERFLOW;
1680#endif
1681 flock->l_start = fl->fl_start;
1682 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1683 fl->fl_end - fl->fl_start + 1;
1684 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001685 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001686 return 0;
1687}
1688
1689#if BITS_PER_LONG == 32
1690static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1691{
1692 flock->l_pid = fl->fl_pid;
1693 flock->l_start = fl->fl_start;
1694 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1695 fl->fl_end - fl->fl_start + 1;
1696 flock->l_whence = 0;
1697 flock->l_type = fl->fl_type;
1698}
1699#endif
1700
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701/* Report the first existing lock that would conflict with l.
1702 * This implements the F_GETLK command of fcntl().
1703 */
1704int fcntl_getlk(struct file *filp, struct flock __user *l)
1705{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001706 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 struct flock flock;
1708 int error;
1709
1710 error = -EFAULT;
1711 if (copy_from_user(&flock, l, sizeof(flock)))
1712 goto out;
1713 error = -EINVAL;
1714 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1715 goto out;
1716
1717 error = flock_to_posix_lock(filp, &file_lock, &flock);
1718 if (error)
1719 goto out;
1720
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001721 error = vfs_test_lock(filp, &file_lock);
1722 if (error)
1723 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001725 flock.l_type = file_lock.fl_type;
1726 if (file_lock.fl_type != F_UNLCK) {
1727 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001728 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 }
1731 error = -EFAULT;
1732 if (!copy_to_user(l, &flock, sizeof(flock)))
1733 error = 0;
1734out:
1735 return error;
1736}
1737
Marc Eshel7723ec92007-01-18 15:08:55 -05001738/**
1739 * vfs_lock_file - file byte range lock
1740 * @filp: The file to apply the lock to
1741 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1742 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001743 * @conf: Place to return a copy of the conflicting lock, if found.
1744 *
1745 * A caller that doesn't care about the conflicting lock may pass NULL
1746 * as the final argument.
1747 *
1748 * If the filesystem defines a private ->lock() method, then @conf will
1749 * be left unchanged; so a caller that cares should initialize it to
1750 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001751 *
1752 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1753 * locks, the ->lock() interface may return asynchronously, before the lock has
1754 * been granted or denied by the underlying filesystem, if (and only if)
1755 * fl_grant is set. Callers expecting ->lock() to return asynchronously
1756 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1757 * the request is for a blocking lock. When ->lock() does return asynchronously,
Miklos Szeredibde74e42008-07-25 01:48:57 -07001758 * it must return FILE_LOCK_DEFERRED, and call ->fl_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001759 * request completes.
1760 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001761 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1762 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001763 * nonzero return code and the file system should release the lock. The file
1764 * system is also responsible to keep a corresponding posix lock when it
1765 * grants a lock so the VFS can find out which locks are locally held and do
1766 * the correct lock cleanup when required.
1767 * The underlying filesystem must not drop the kernel lock or call
Miklos Szeredibde74e42008-07-25 01:48:57 -07001768 * ->fl_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001769 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001770 */
Marc Eshel150b3932007-01-18 16:15:35 -05001771int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001772{
1773 if (filp->f_op && filp->f_op->lock)
1774 return filp->f_op->lock(filp, cmd, fl);
1775 else
Marc Eshel150b3932007-01-18 16:15:35 -05001776 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001777}
1778EXPORT_SYMBOL_GPL(vfs_lock_file);
1779
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001780static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1781 struct file_lock *fl)
1782{
1783 int error;
1784
1785 error = security_file_lock(filp, fl->fl_type);
1786 if (error)
1787 return error;
1788
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001789 for (;;) {
1790 error = vfs_lock_file(filp, cmd, fl, NULL);
1791 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001792 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001793 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1794 if (!error)
1795 continue;
1796
1797 locks_delete_block(fl);
1798 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001799 }
1800
1801 return error;
1802}
1803
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804/* Apply the lock described by l to an open file descriptor.
1805 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1806 */
Peter Staubachc2936212005-07-27 11:45:09 -07001807int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
1808 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809{
1810 struct file_lock *file_lock = locks_alloc_lock();
1811 struct flock flock;
1812 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001813 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 int error;
1815
1816 if (file_lock == NULL)
1817 return -ENOLCK;
1818
1819 /*
1820 * This might block, so we do it before checking the inode.
1821 */
1822 error = -EFAULT;
1823 if (copy_from_user(&flock, l, sizeof(flock)))
1824 goto out;
1825
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001826 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827
1828 /* Don't allow mandatory locks on files that may be memory mapped
1829 * and shared.
1830 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001831 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 error = -EAGAIN;
1833 goto out;
1834 }
1835
Peter Staubachc2936212005-07-27 11:45:09 -07001836again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 error = flock_to_posix_lock(filp, file_lock, &flock);
1838 if (error)
1839 goto out;
1840 if (cmd == F_SETLKW) {
1841 file_lock->fl_flags |= FL_SLEEP;
1842 }
1843
1844 error = -EBADF;
1845 switch (flock.l_type) {
1846 case F_RDLCK:
1847 if (!(filp->f_mode & FMODE_READ))
1848 goto out;
1849 break;
1850 case F_WRLCK:
1851 if (!(filp->f_mode & FMODE_WRITE))
1852 goto out;
1853 break;
1854 case F_UNLCK:
1855 break;
1856 default:
1857 error = -EINVAL;
1858 goto out;
1859 }
1860
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001861 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862
Peter Staubachc2936212005-07-27 11:45:09 -07001863 /*
1864 * Attempt to detect a close/fcntl race and recover by
1865 * releasing the lock that was just acquired.
1866 */
Al Viro0b2bac22008-05-06 13:58:34 -04001867 /*
1868 * we need that spin_lock here - it prevents reordering between
1869 * update of inode->i_flock and check for it done in close().
1870 * rcu_read_lock() wouldn't do.
1871 */
1872 spin_lock(&current->files->file_lock);
1873 f = fcheck(fd);
1874 spin_unlock(&current->files->file_lock);
1875 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001876 flock.l_type = F_UNLCK;
1877 goto again;
1878 }
1879
1880out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 locks_free_lock(file_lock);
1882 return error;
1883}
1884
1885#if BITS_PER_LONG == 32
1886/* Report the first existing lock that would conflict with l.
1887 * This implements the F_GETLK command of fcntl().
1888 */
1889int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
1890{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001891 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 struct flock64 flock;
1893 int error;
1894
1895 error = -EFAULT;
1896 if (copy_from_user(&flock, l, sizeof(flock)))
1897 goto out;
1898 error = -EINVAL;
1899 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1900 goto out;
1901
1902 error = flock64_to_posix_lock(filp, &file_lock, &flock);
1903 if (error)
1904 goto out;
1905
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001906 error = vfs_test_lock(filp, &file_lock);
1907 if (error)
1908 goto out;
1909
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001910 flock.l_type = file_lock.fl_type;
1911 if (file_lock.fl_type != F_UNLCK)
1912 posix_lock_to_flock64(&flock, &file_lock);
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 error = -EFAULT;
1915 if (!copy_to_user(l, &flock, sizeof(flock)))
1916 error = 0;
1917
1918out:
1919 return error;
1920}
1921
1922/* Apply the lock described by l to an open file descriptor.
1923 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
1924 */
Peter Staubachc2936212005-07-27 11:45:09 -07001925int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
1926 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927{
1928 struct file_lock *file_lock = locks_alloc_lock();
1929 struct flock64 flock;
1930 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04001931 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932 int error;
1933
1934 if (file_lock == NULL)
1935 return -ENOLCK;
1936
1937 /*
1938 * This might block, so we do it before checking the inode.
1939 */
1940 error = -EFAULT;
1941 if (copy_from_user(&flock, l, sizeof(flock)))
1942 goto out;
1943
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001944 inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
1946 /* Don't allow mandatory locks on files that may be memory mapped
1947 * and shared.
1948 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001949 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 error = -EAGAIN;
1951 goto out;
1952 }
1953
Peter Staubachc2936212005-07-27 11:45:09 -07001954again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955 error = flock64_to_posix_lock(filp, file_lock, &flock);
1956 if (error)
1957 goto out;
1958 if (cmd == F_SETLKW64) {
1959 file_lock->fl_flags |= FL_SLEEP;
1960 }
1961
1962 error = -EBADF;
1963 switch (flock.l_type) {
1964 case F_RDLCK:
1965 if (!(filp->f_mode & FMODE_READ))
1966 goto out;
1967 break;
1968 case F_WRLCK:
1969 if (!(filp->f_mode & FMODE_WRITE))
1970 goto out;
1971 break;
1972 case F_UNLCK:
1973 break;
1974 default:
1975 error = -EINVAL;
1976 goto out;
1977 }
1978
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001979 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
Peter Staubachc2936212005-07-27 11:45:09 -07001981 /*
1982 * Attempt to detect a close/fcntl race and recover by
1983 * releasing the lock that was just acquired.
1984 */
Al Viro0b2bac22008-05-06 13:58:34 -04001985 spin_lock(&current->files->file_lock);
1986 f = fcheck(fd);
1987 spin_unlock(&current->files->file_lock);
1988 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07001989 flock.l_type = F_UNLCK;
1990 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 }
1992
1993out:
1994 locks_free_lock(file_lock);
1995 return error;
1996}
1997#endif /* BITS_PER_LONG == 32 */
1998
1999/*
2000 * This function is called when the file is being removed
2001 * from the task's fd array. POSIX locks belonging to this task
2002 * are deleted at this time.
2003 */
2004void locks_remove_posix(struct file *filp, fl_owner_t owner)
2005{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002006 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
2008 /*
2009 * If there are no locks held on this file, we don't need to call
2010 * posix_lock_file(). Another process could be setting a lock on this
2011 * file at the same time, but we wouldn't remove that lock anyway.
2012 */
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002013 if (!filp->f_path.dentry->d_inode->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014 return;
2015
2016 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002017 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 lock.fl_start = 0;
2019 lock.fl_end = OFFSET_MAX;
2020 lock.fl_owner = owner;
2021 lock.fl_pid = current->tgid;
2022 lock.fl_file = filp;
2023 lock.fl_ops = NULL;
2024 lock.fl_lmops = NULL;
2025
Marc Eshel150b3932007-01-18 16:15:35 -05002026 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2029 lock.fl_ops->fl_release_private(&lock);
2030}
2031
2032EXPORT_SYMBOL(locks_remove_posix);
2033
2034/*
2035 * This function is called on the last close of an open file.
2036 */
2037void locks_remove_flock(struct file *filp)
2038{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002039 struct inode * inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 struct file_lock *fl;
2041 struct file_lock **before;
2042
2043 if (!inode->i_flock)
2044 return;
2045
2046 if (filp->f_op && filp->f_op->flock) {
2047 struct file_lock fl = {
2048 .fl_pid = current->tgid,
2049 .fl_file = filp,
2050 .fl_flags = FL_FLOCK,
2051 .fl_type = F_UNLCK,
2052 .fl_end = OFFSET_MAX,
2053 };
2054 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002055 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2056 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002057 }
2058
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002059 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 before = &inode->i_flock;
2061
2062 while ((fl = *before) != NULL) {
2063 if (fl->fl_file == filp) {
Peter Staubachc2936212005-07-27 11:45:09 -07002064 if (IS_FLOCK(fl)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 locks_delete_lock(before);
2066 continue;
2067 }
2068 if (IS_LEASE(fl)) {
2069 lease_modify(before, F_UNLCK);
2070 continue;
2071 }
2072 /* What? */
2073 BUG();
2074 }
2075 before = &fl->fl_next;
2076 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002077 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078}
2079
2080/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 * posix_unblock_lock - stop waiting for a file lock
2082 * @filp: how the file was opened
2083 * @waiter: the lock which was waiting
2084 *
2085 * lockd needs to block waiting for locks.
2086 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002087int
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088posix_unblock_lock(struct file *filp, struct file_lock *waiter)
2089{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002090 int status = 0;
2091
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002092 lock_flocks();
J. Bruce Fields5996a292006-01-03 09:55:44 +01002093 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002095 else
2096 status = -ENOENT;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002097 unlock_flocks();
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002098 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099}
2100
2101EXPORT_SYMBOL(posix_unblock_lock);
2102
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002103/**
2104 * vfs_cancel_lock - file byte range unblock lock
2105 * @filp: The file to apply the unblock to
2106 * @fl: The lock to be unblocked
2107 *
2108 * Used by lock managers to cancel blocked requests
2109 */
2110int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2111{
2112 if (filp->f_op && filp->f_op->lock)
2113 return filp->f_op->lock(filp, F_CANCELLK, fl);
2114 return 0;
2115}
2116
2117EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2118
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002119#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002120#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002121#include <linux/seq_file.h>
2122
2123static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002124 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
2126 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002127 unsigned int fl_pid;
2128
2129 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002130 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002131 else
2132 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
2134 if (fl->fl_file != NULL)
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08002135 inode = fl->fl_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
Jerome Marchand99dc8292010-10-26 14:22:33 -07002137 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002139 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2141 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002142 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 } else if (IS_FLOCK(fl)) {
2144 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002145 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002147 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148 }
2149 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002150 seq_printf(f, "LEASE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 if (fl->fl_type & F_INPROGRESS)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002152 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002154 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002156 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002157 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002158 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 }
2160 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002161 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 (fl->fl_type & LOCK_READ)
2163 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2164 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2165 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002166 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167 (fl->fl_type & F_INPROGRESS)
2168 ? (fl->fl_type & F_UNLCK) ? "UNLCK" : "READ "
2169 : (fl->fl_type & F_WRLCK) ? "WRITE" : "READ ");
2170 }
2171 if (inode) {
2172#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002173 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 inode->i_sb->s_id, inode->i_ino);
2175#else
2176 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002177 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178 MAJOR(inode->i_sb->s_dev),
2179 MINOR(inode->i_sb->s_dev), inode->i_ino);
2180#endif
2181 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002182 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 }
2184 if (IS_POSIX(fl)) {
2185 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002186 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002188 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002190 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 }
2192}
2193
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002194static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195{
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002196 struct file_lock *fl, *bfl;
2197
2198 fl = list_entry(v, struct file_lock, fl_link);
2199
Jerome Marchand99dc8292010-10-26 14:22:33 -07002200 lock_get_status(f, fl, *((loff_t *)f->private), "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002201
2202 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jerome Marchand99dc8292010-10-26 14:22:33 -07002203 lock_get_status(f, bfl, *((loff_t *)f->private), " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002204
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002205 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206}
2207
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002208static void *locks_start(struct seq_file *f, loff_t *pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002210 loff_t *p = f->private;
2211
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002212 lock_flocks();
Jerome Marchand99dc8292010-10-26 14:22:33 -07002213 *p = (*pos + 1);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002214 return seq_list_start(&file_lock_list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215}
2216
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002217static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2218{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002219 loff_t *p = f->private;
2220 ++*p;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002221 return seq_list_next(v, &file_lock_list, pos);
2222}
2223
2224static void locks_stop(struct seq_file *f, void *v)
2225{
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002226 unlock_flocks();
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002227}
2228
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002229static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002230 .start = locks_start,
2231 .next = locks_next,
2232 .stop = locks_stop,
2233 .show = locks_show,
2234};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002235
2236static int locks_open(struct inode *inode, struct file *filp)
2237{
Jerome Marchand99dc8292010-10-26 14:22:33 -07002238 return seq_open_private(filp, &locks_seq_operations, sizeof(loff_t));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002239}
2240
2241static const struct file_operations proc_locks_operations = {
2242 .open = locks_open,
2243 .read = seq_read,
2244 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002245 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002246};
2247
2248static int __init proc_locks_init(void)
2249{
2250 proc_create("locks", 0, NULL, &proc_locks_operations);
2251 return 0;
2252}
2253module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002254#endif
2255
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256/**
2257 * lock_may_read - checks that the region is free of locks
2258 * @inode: the inode that is being read
2259 * @start: the first byte to read
2260 * @len: the number of bytes to read
2261 *
2262 * Emulates Windows locking requirements. Whole-file
2263 * mandatory locks (share modes) can prohibit a read and
2264 * byte-range POSIX locks can prohibit a read if they overlap.
2265 *
2266 * N.B. this function is only ever called
2267 * from knfsd and ownership of locks is never checked.
2268 */
2269int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2270{
2271 struct file_lock *fl;
2272 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002273 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2275 if (IS_POSIX(fl)) {
2276 if (fl->fl_type == F_RDLCK)
2277 continue;
2278 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2279 continue;
2280 } else if (IS_FLOCK(fl)) {
2281 if (!(fl->fl_type & LOCK_MAND))
2282 continue;
2283 if (fl->fl_type & LOCK_READ)
2284 continue;
2285 } else
2286 continue;
2287 result = 0;
2288 break;
2289 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002290 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 return result;
2292}
2293
2294EXPORT_SYMBOL(lock_may_read);
2295
2296/**
2297 * lock_may_write - checks that the region is free of locks
2298 * @inode: the inode that is being written
2299 * @start: the first byte to write
2300 * @len: the number of bytes to write
2301 *
2302 * Emulates Windows locking requirements. Whole-file
2303 * mandatory locks (share modes) can prohibit a write and
2304 * byte-range POSIX locks can prohibit a write if they overlap.
2305 *
2306 * N.B. this function is only ever called
2307 * from knfsd and ownership of locks is never checked.
2308 */
2309int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2310{
2311 struct file_lock *fl;
2312 int result = 1;
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002313 lock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2315 if (IS_POSIX(fl)) {
2316 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2317 continue;
2318 } else if (IS_FLOCK(fl)) {
2319 if (!(fl->fl_type & LOCK_MAND))
2320 continue;
2321 if (fl->fl_type & LOCK_WRITE)
2322 continue;
2323 } else
2324 continue;
2325 result = 0;
2326 break;
2327 }
Arnd Bergmannb89f4322010-09-18 15:09:31 +02002328 unlock_flocks();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 return result;
2330}
2331
2332EXPORT_SYMBOL(lock_may_write);
2333
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334static int __init filelock_init(void)
2335{
2336 filelock_cache = kmem_cache_create("file_lock_cache",
2337 sizeof(struct file_lock), 0, SLAB_PANIC,
Paul Mundt20c2df82007-07-20 10:11:58 +09002338 init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 return 0;
2340}
2341
2342core_initcall(filelock_init);