blob: e663aeac579e5d8aaa2596a177114b17f262bd6c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
Paul Bolle395cf962011-08-15 02:02:26 +020063 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -0400119#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000128#include <linux/pid_namespace.h>
Jeff Layton48f74182013-06-21 08:58:18 -0400129#include <linux/hashtable.h>
Jeff Layton7012b022013-06-21 08:58:22 -0400130#include <linux/percpu.h>
131#include <linux/lglock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133#include <asm/uaccess.h>
134
135#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
136#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
J. Bruce Fields617588d2011-07-01 15:18:34 -0400137#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400138#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400140static bool lease_breaking(struct file_lock *fl)
141{
J. Bruce Fields778fc542011-07-26 18:25:49 -0400142 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
143}
144
145static int target_leasetype(struct file_lock *fl)
146{
147 if (fl->fl_flags & FL_UNLOCK_PENDING)
148 return F_UNLCK;
149 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
150 return F_RDLCK;
151 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400152}
153
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154int leases_enable = 1;
155int lease_break_time = 45;
156
157#define for_each_lock(inode, lockp) \
158 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
159
Jeff Layton1c8c6012013-06-21 08:58:15 -0400160/*
Jeff Layton7012b022013-06-21 08:58:22 -0400161 * The global file_lock_list is only used for displaying /proc/locks, so we
162 * keep a list on each CPU, with each list protected by its own spinlock via
163 * the file_lock_lglock. Note that alterations to the list also require that
164 * the relevant i_lock is held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400165 */
Jeff Layton7012b022013-06-21 08:58:22 -0400166DEFINE_STATIC_LGLOCK(file_lock_lglock);
167static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
Jeff Layton88974692013-06-21 08:58:14 -0400168
Jeff Layton1c8c6012013-06-21 08:58:15 -0400169/*
Jeff Layton48f74182013-06-21 08:58:18 -0400170 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400171 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400172 *
173 * We hash locks by lockowner in order to optimize searching for the lock a
174 * particular lockowner is waiting on.
175 *
176 * FIXME: make this value scale via some heuristic? We generally will want more
177 * buckets when we have more lockowners holding locks, but that's a little
178 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400179 */
Jeff Layton48f74182013-06-21 08:58:18 -0400180#define BLOCKED_HASH_BITS 7
181static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400182
Jeff Layton1c8c6012013-06-21 08:58:15 -0400183/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400184 * This lock protects the blocked_hash. Generally, if you're accessing it, you
185 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400186 *
187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
188 * pointer for file_lock structures that are acting as lock requests (in
189 * contrast to those that are acting as records of acquired locks).
190 *
191 * Note that when we acquire this lock in order to change the above fields,
192 * we often hold the i_lock as well. In certain cases, when reading the fields
193 * protected by this lock, we can skip acquiring it iff we already hold the
194 * i_lock.
195 *
196 * In particular, adding an entry to the fl_block list requires that you hold
197 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
198 * an entry from the list however only requires the file_lock_lock.
199 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400200static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Christoph Lametere18b8902006-12-06 20:33:20 -0800202static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Miklos Szerediee19cc42011-07-07 13:06:09 +0200204static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200205{
Jeff Layton139ca042013-06-21 08:58:17 -0400206 INIT_HLIST_NODE(&fl->fl_link);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200207 INIT_LIST_HEAD(&fl->fl_block);
208 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200209}
210
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200212struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200214 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200215
216 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200217 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200218
219 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200221EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500223void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500224{
225 if (fl->fl_ops) {
226 if (fl->fl_ops->fl_release_private)
227 fl->fl_ops->fl_release_private(fl);
228 fl->fl_ops = NULL;
229 }
J. Bruce Fields068535f2012-08-01 07:56:16 -0400230 fl->fl_lmops = NULL;
Trond Myklebust47831f32006-03-20 13:44:05 -0500231
232}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500233EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500234
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400236void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800238 BUG_ON(waitqueue_active(&fl->fl_wait));
239 BUG_ON(!list_empty(&fl->fl_block));
Jeff Layton139ca042013-06-21 08:58:17 -0400240 BUG_ON(!hlist_unhashed(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
Trond Myklebust47831f32006-03-20 13:44:05 -0500242 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 kmem_cache_free(filelock_cache, fl);
244}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400245EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247void locks_init_lock(struct file_lock *fl)
248{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200249 memset(fl, 0, sizeof(struct file_lock));
250 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
253EXPORT_SYMBOL(locks_init_lock);
254
Trond Myklebust47831f32006-03-20 13:44:05 -0500255static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
256{
257 if (fl->fl_ops) {
258 if (fl->fl_ops->fl_copy_lock)
259 fl->fl_ops->fl_copy_lock(new, fl);
260 new->fl_ops = fl->fl_ops;
261 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400262 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500263 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500264}
265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266/*
267 * Initialize a new lock from an existing file_lock structure.
268 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400269void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270{
271 new->fl_owner = fl->fl_owner;
272 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500273 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 new->fl_flags = fl->fl_flags;
275 new->fl_type = fl->fl_type;
276 new->fl_start = fl->fl_start;
277 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500278 new->fl_ops = NULL;
279 new->fl_lmops = NULL;
280}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700281EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500282
283void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
284{
285 locks_release_private(new);
286
287 __locks_copy_lock(new, fl);
288 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 new->fl_ops = fl->fl_ops;
290 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500291
292 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293}
294
295EXPORT_SYMBOL(locks_copy_lock);
296
297static inline int flock_translate_cmd(int cmd) {
298 if (cmd & LOCK_MAND)
299 return cmd & (LOCK_MAND | LOCK_RW);
300 switch (cmd) {
301 case LOCK_SH:
302 return F_RDLCK;
303 case LOCK_EX:
304 return F_WRLCK;
305 case LOCK_UN:
306 return F_UNLCK;
307 }
308 return -EINVAL;
309}
310
311/* Fill in a file_lock structure with an appropriate FLOCK lock. */
312static int flock_make_lock(struct file *filp, struct file_lock **lock,
313 unsigned int cmd)
314{
315 struct file_lock *fl;
316 int type = flock_translate_cmd(cmd);
317 if (type < 0)
318 return type;
319
320 fl = locks_alloc_lock();
321 if (fl == NULL)
322 return -ENOMEM;
323
324 fl->fl_file = filp;
325 fl->fl_pid = current->tgid;
326 fl->fl_flags = FL_FLOCK;
327 fl->fl_type = type;
328 fl->fl_end = OFFSET_MAX;
329
330 *lock = fl;
331 return 0;
332}
333
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400334static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335{
336 switch (type) {
337 case F_RDLCK:
338 case F_WRLCK:
339 case F_UNLCK:
340 fl->fl_type = type;
341 break;
342 default:
343 return -EINVAL;
344 }
345 return 0;
346}
347
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500348static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
349 struct flock64 *l)
350{
351 switch (l->l_whence) {
352 case SEEK_SET:
353 fl->fl_start = 0;
354 break;
355 case SEEK_CUR:
356 fl->fl_start = filp->f_pos;
357 break;
358 case SEEK_END:
359 fl->fl_start = i_size_read(file_inode(filp));
360 break;
361 default:
362 return -EINVAL;
363 }
364 if (l->l_start > OFFSET_MAX - fl->fl_start)
365 return -EOVERFLOW;
366 fl->fl_start += l->l_start;
367 if (fl->fl_start < 0)
368 return -EINVAL;
369
370 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
371 POSIX-2001 defines it. */
372 if (l->l_len > 0) {
373 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
374 return -EOVERFLOW;
375 fl->fl_end = fl->fl_start + l->l_len - 1;
376
377 } else if (l->l_len < 0) {
378 if (fl->fl_start + l->l_len < 0)
379 return -EINVAL;
380 fl->fl_end = fl->fl_start - 1;
381 fl->fl_start += l->l_len;
382 } else
383 fl->fl_end = OFFSET_MAX;
384
385 fl->fl_owner = current->files;
386 fl->fl_pid = current->tgid;
387 fl->fl_file = filp;
388 fl->fl_flags = FL_POSIX;
389 fl->fl_ops = NULL;
390 fl->fl_lmops = NULL;
391
Jeff Laytonbce75602014-02-03 12:13:08 -0500392 /* Ensure that fl->fl_filp has compatible f_mode */
393 switch (l->l_type) {
394 case F_RDLCK:
395 if (!(filp->f_mode & FMODE_READ))
396 return -EBADF;
397 break;
398 case F_WRLCK:
399 if (!(filp->f_mode & FMODE_WRITE))
400 return -EBADF;
401 break;
402 }
403
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500404 return assign_type(fl, l->l_type);
405}
406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
408 * style lock.
409 */
410static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
411 struct flock *l)
412{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500413 struct flock64 ll = {
414 .l_type = l->l_type,
415 .l_whence = l->l_whence,
416 .l_start = l->l_start,
417 .l_len = l->l_len,
418 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500420 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421}
422
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423/* default lease lock manager operations */
424static void lease_break_callback(struct file_lock *fl)
425{
426 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
427}
428
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700429static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400430 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400431 .lm_change = lease_modify,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432};
433
434/*
435 * Initialize a lease, use the default lock manager operations
436 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400437static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400439 if (assign_type(fl, type) != 0)
440 return -EINVAL;
441
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 fl->fl_owner = current->files;
443 fl->fl_pid = current->tgid;
444
445 fl->fl_file = filp;
446 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 fl->fl_start = 0;
448 fl->fl_end = OFFSET_MAX;
449 fl->fl_ops = NULL;
450 fl->fl_lmops = &lease_manager_ops;
451 return 0;
452}
453
454/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400455static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
457 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400458 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
460 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500461 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400464 if (error) {
465 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500466 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400467 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500468 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}
470
471/* Check if two locks overlap each other.
472 */
473static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
474{
475 return ((fl1->fl_end >= fl2->fl_start) &&
476 (fl2->fl_end >= fl1->fl_start));
477}
478
479/*
480 * Check whether two locks have the same owner.
481 */
Matt Mackall33443c42006-01-08 01:05:22 -0800482static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400484 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400486 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return fl1->fl_owner == fl2->fl_owner;
488}
489
Jeff Layton7012b022013-06-21 08:58:22 -0400490/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500491static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400492{
Jeff Layton7012b022013-06-21 08:58:22 -0400493 lg_local_lock(&file_lock_lglock);
494 fl->fl_link_cpu = smp_processor_id();
495 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
496 lg_local_unlock(&file_lock_lglock);
Jeff Layton88974692013-06-21 08:58:14 -0400497}
498
Jeff Layton7012b022013-06-21 08:58:22 -0400499/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500500static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400501{
Jeff Layton7012b022013-06-21 08:58:22 -0400502 /*
503 * Avoid taking lock if already unhashed. This is safe since this check
504 * is done while holding the i_lock, and new insertions into the list
505 * also require that it be held.
506 */
507 if (hlist_unhashed(&fl->fl_link))
508 return;
509 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton139ca042013-06-21 08:58:17 -0400510 hlist_del_init(&fl->fl_link);
Jeff Layton7012b022013-06-21 08:58:22 -0400511 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton88974692013-06-21 08:58:14 -0400512}
513
Jeff Layton3999e492013-06-21 08:58:19 -0400514static unsigned long
515posix_owner_key(struct file_lock *fl)
516{
517 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
518 return fl->fl_lmops->lm_owner_key(fl);
519 return (unsigned long)fl->fl_owner;
520}
521
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500522static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400523{
Jeff Layton3999e492013-06-21 08:58:19 -0400524 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400525}
526
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500527static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400528{
Jeff Layton48f74182013-06-21 08:58:18 -0400529 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400530}
531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532/* Remove waiter from blocker's block list.
533 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400534 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400535 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 */
Matt Mackall33443c42006-01-08 01:05:22 -0800537static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538{
Jeff Layton88974692013-06-21 08:58:14 -0400539 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 waiter->fl_next = NULL;
542}
543
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400544static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400546 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400548 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549}
550
551/* Insert waiter into blocker's block list.
552 * We use a circular list so that processes can be easily woken up in
553 * the order they blocked. The documentation doesn't require this but
554 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400555 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400556 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
Jeff Layton46dad762014-02-03 12:13:06 -0500557 * list itself is protected by the blocked_lock_lock, but by ensuring that the
Jeff Layton7b2296a2013-06-21 08:58:20 -0400558 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
Jeff Layton4e8c7652013-06-21 08:58:16 -0400559 * in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400561static void __locks_insert_block(struct file_lock *blocker,
562 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800564 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400566 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400567 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400568 locks_insert_global_blocked(waiter);
569}
570
571/* Must be called with i_lock held. */
572static void locks_insert_block(struct file_lock *blocker,
573 struct file_lock *waiter)
574{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400575 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400576 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400577 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Jeff Layton1cb36012013-06-21 08:58:12 -0400580/*
581 * Wake up processes blocked waiting for blocker.
582 *
Jeff Layton1c8c6012013-06-21 08:58:15 -0400583 * Must be called with the inode->i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 */
585static void locks_wake_up_blocks(struct file_lock *blocker)
586{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400587 /*
588 * Avoid taking global lock if list is empty. This is safe since new
589 * blocked requests are only added to the list under the i_lock, and
590 * the i_lock is always held here. Note that removal from the fl_block
591 * list does not require the i_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400592 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400593 */
594 if (list_empty(&blocker->fl_block))
595 return;
596
Jeff Layton7b2296a2013-06-21 08:58:20 -0400597 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400599 struct file_lock *waiter;
600
601 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 struct file_lock, fl_block);
603 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400604 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
605 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 else
607 wake_up(&waiter->fl_wait);
608 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400609 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610}
611
612/* Insert file lock fl into an inode's lock list at the position indicated
613 * by pos. At the same time add the lock to the global file lock list.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400614 *
615 * Must be called with the i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 */
617static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
618{
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000619 fl->fl_nspid = get_pid(task_tgid(current));
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 /* insert into file's list */
622 fl->fl_next = *pos;
623 *pos = fl;
Jeff Layton88974692013-06-21 08:58:14 -0400624
625 locks_insert_global_locks(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626}
627
Jeff Layton24cbe782014-02-03 12:13:06 -0500628/**
629 * locks_delete_lock - Delete a lock and then free it.
630 * @thisfl_p: pointer that points to the fl_next field of the previous
631 * inode->i_flock list entry
632 *
633 * Unlink a lock from all lists and free the namespace reference, but don't
634 * free it yet. Wake up processes that are blocked waiting for this lock and
635 * notify the FS that the lock has been cleared.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400636 *
637 * Must be called with the i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 */
Jeff Layton24cbe782014-02-03 12:13:06 -0500639static void locks_unlink_lock(struct file_lock **thisfl_p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640{
641 struct file_lock *fl = *thisfl_p;
642
Jeff Layton88974692013-06-21 08:58:14 -0400643 locks_delete_global_locks(fl);
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 *thisfl_p = fl->fl_next;
646 fl->fl_next = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000648 if (fl->fl_nspid) {
649 put_pid(fl->fl_nspid);
650 fl->fl_nspid = NULL;
651 }
652
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500654}
655
656/*
657 * Unlink a lock from all lists and free it.
658 *
659 * Must be called with i_lock held!
660 */
661static void locks_delete_lock(struct file_lock **thisfl_p)
662{
663 struct file_lock *fl = *thisfl_p;
664
665 locks_unlink_lock(thisfl_p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 locks_free_lock(fl);
667}
668
669/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
670 * checks for shared/exclusive status of overlapping locks.
671 */
672static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
673{
674 if (sys_fl->fl_type == F_WRLCK)
675 return 1;
676 if (caller_fl->fl_type == F_WRLCK)
677 return 1;
678 return 0;
679}
680
681/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
682 * checking before calling the locks_conflict().
683 */
684static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
685{
686 /* POSIX locks owned by the same process do not conflict with
687 * each other.
688 */
689 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
690 return (0);
691
692 /* Check whether they overlap */
693 if (!locks_overlap(caller_fl, sys_fl))
694 return 0;
695
696 return (locks_conflict(caller_fl, sys_fl));
697}
698
699/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
700 * checking before calling the locks_conflict().
701 */
702static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
703{
704 /* FLOCK locks referring to the same filp do not conflict with
705 * each other.
706 */
707 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
708 return (0);
709 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
710 return 0;
711
712 return (locks_conflict(caller_fl, sys_fl));
713}
714
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400715void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500716posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717{
718 struct file_lock *cfl;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400719 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720
Jeff Layton1c8c6012013-06-21 08:58:15 -0400721 spin_lock(&inode->i_lock);
Al Viro496ad9a2013-01-23 17:07:38 -0500722 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 if (!IS_POSIX(cfl))
724 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400725 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 break;
727 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000728 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500729 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000730 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800731 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000732 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400733 fl->fl_type = F_UNLCK;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400734 spin_unlock(&inode->i_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400735 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737EXPORT_SYMBOL(posix_test_lock);
738
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400739/*
740 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400742 * We attempt to detect deadlocks that are due purely to posix file
743 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400745 * We assume that a task can be waiting for at most one lock at a time.
746 * So for any acquired lock, the process holding that lock may be
747 * waiting on at most one other lock. That lock in turns may be held by
748 * someone waiting for at most one other lock. Given a requested lock
749 * caller_fl which is about to wait for a conflicting lock block_fl, we
750 * follow this chain of waiters to ensure we are not about to create a
751 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400752 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400753 * Since we do this before we ever put a process to sleep on a lock, we
754 * are ensured that there is never a cycle; that is what guarantees that
755 * the while() loop in posix_locks_deadlock() eventually completes.
756 *
757 * Note: the above assumption may not be true when handling lock
758 * requests from a broken NFS client. It may also fail in the presence
759 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400760 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500761 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400762 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500763 * Because the owner is not even nominally tied to a thread of
764 * execution, the deadlock detection below can't reasonably work well. Just
765 * skip it for those.
766 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400767 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500768 * locks that just checks for the case where two tasks are attempting to
769 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400771
772#define MAX_DEADLK_ITERATIONS 10
773
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400774/* Find a lock that the owner of the given block_fl is blocking on. */
775static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
776{
777 struct file_lock *fl;
778
Jeff Layton3999e492013-06-21 08:58:19 -0400779 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400780 if (posix_same_owner(fl, block_fl))
781 return fl->fl_next;
782 }
783 return NULL;
784}
785
Jeff Layton7b2296a2013-06-21 08:58:20 -0400786/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700787static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 struct file_lock *block_fl)
789{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400790 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
Jeff Layton57b65322014-02-03 12:13:09 -0500792 /*
793 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400794 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500795 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400796 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500797 return 0;
798
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400799 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
800 if (i++ > MAX_DEADLK_ITERATIONS)
801 return 0;
802 if (posix_same_owner(caller_fl, block_fl))
803 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 }
805 return 0;
806}
807
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400809 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400810 *
811 * Note that if called with an FL_EXISTS argument, the caller may determine
812 * whether or not a lock was successfully freed by testing the return
813 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800815static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800817 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 struct file_lock **before;
Al Viro496ad9a2013-01-23 17:07:38 -0500819 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 int error = 0;
821 int found = 0;
822
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200823 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
824 new_fl = locks_alloc_lock();
825 if (!new_fl)
826 return -ENOMEM;
827 }
828
Jeff Layton1c8c6012013-06-21 08:58:15 -0400829 spin_lock(&inode->i_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400830 if (request->fl_flags & FL_ACCESS)
831 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 for_each_lock(inode, before) {
834 struct file_lock *fl = *before;
835 if (IS_POSIX(fl))
836 break;
837 if (IS_LEASE(fl))
838 continue;
839 if (filp != fl->fl_file)
840 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800841 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 goto out;
843 found = 1;
844 locks_delete_lock(before);
845 break;
846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Trond Myklebustf475ae92006-06-29 16:38:32 -0400848 if (request->fl_type == F_UNLCK) {
849 if ((request->fl_flags & FL_EXISTS) && !found)
850 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800851 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
854 /*
855 * If a higher-priority process was blocked on the old file lock,
856 * give it the opportunity to lock the file.
857 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200858 if (found) {
Jeff Layton1c8c6012013-06-21 08:58:15 -0400859 spin_unlock(&inode->i_lock);
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200860 cond_resched();
Jeff Layton1c8c6012013-06-21 08:58:15 -0400861 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200862 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400864find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 for_each_lock(inode, before) {
866 struct file_lock *fl = *before;
867 if (IS_POSIX(fl))
868 break;
869 if (IS_LEASE(fl))
870 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800871 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 continue;
873 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700874 if (!(request->fl_flags & FL_SLEEP))
875 goto out;
876 error = FILE_LOCK_DEFERRED;
877 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 goto out;
879 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400880 if (request->fl_flags & FL_ACCESS)
881 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800882 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700883 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800884 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400885 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886
887out:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400888 spin_unlock(&inode->i_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800889 if (new_fl)
890 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 return error;
892}
893
Marc Eshel150b3932007-01-18 16:15:35 -0500894static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895{
896 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700897 struct file_lock *new_fl = NULL;
898 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 struct file_lock *left = NULL;
900 struct file_lock *right = NULL;
901 struct file_lock **before;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400902 int error;
903 bool added = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
905 /*
906 * We may need two file_lock structures for this operation,
907 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700908 *
909 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700911 if (!(request->fl_flags & FL_ACCESS) &&
912 (request->fl_type != F_UNLCK ||
913 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
914 new_fl = locks_alloc_lock();
915 new_fl2 = locks_alloc_lock();
916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Jeff Layton1c8c6012013-06-21 08:58:15 -0400918 spin_lock(&inode->i_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400919 /*
920 * New lock request. Walk all POSIX locks and look for conflicts. If
921 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400922 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400923 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 if (request->fl_type != F_UNLCK) {
925 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500926 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 if (!IS_POSIX(fl))
928 continue;
929 if (!posix_locks_conflict(request, fl))
930 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800931 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400932 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 error = -EAGAIN;
934 if (!(request->fl_flags & FL_SLEEP))
935 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400936 /*
937 * Deadlock detection and insertion into the blocked
938 * locks list must be done while holding the same lock!
939 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400941 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400942 if (likely(!posix_locks_deadlock(request, fl))) {
943 error = FILE_LOCK_DEFERRED;
944 __locks_insert_block(fl, request);
945 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400946 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 goto out;
948 }
949 }
950
951 /* If we're just looking for a conflict, we're done. */
952 error = 0;
953 if (request->fl_flags & FL_ACCESS)
954 goto out;
955
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * Find the first old lock with the same owner as the new lock.
958 */
959
960 before = &inode->i_flock;
961
962 /* First skip locks owned by other processes. */
963 while ((fl = *before) && (!IS_POSIX(fl) ||
964 !posix_same_owner(request, fl))) {
965 before = &fl->fl_next;
966 }
967
Jeff Layton1cb36012013-06-21 08:58:12 -0400968 /* Process locks with this owner. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 while ((fl = *before) && posix_same_owner(request, fl)) {
970 /* Detect adjacent or overlapping regions (if same lock type)
971 */
972 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700973 /* In all comparisons of start vs end, use
974 * "start - 1" rather than "end + 1". If end
975 * is OFFSET_MAX, end + 1 will become negative.
976 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (fl->fl_end < request->fl_start - 1)
978 goto next_lock;
979 /* If the next lock in the list has entirely bigger
980 * addresses than the new one, insert the lock here.
981 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700982 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 break;
984
985 /* If we come here, the new and old lock are of the
986 * same type and adjacent or overlapping. Make one
987 * lock yielding from the lower start address of both
988 * locks to the higher end address.
989 */
990 if (fl->fl_start > request->fl_start)
991 fl->fl_start = request->fl_start;
992 else
993 request->fl_start = fl->fl_start;
994 if (fl->fl_end < request->fl_end)
995 fl->fl_end = request->fl_end;
996 else
997 request->fl_end = fl->fl_end;
998 if (added) {
999 locks_delete_lock(before);
1000 continue;
1001 }
1002 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001003 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 }
1005 else {
1006 /* Processing for different lock types is a bit
1007 * more complex.
1008 */
1009 if (fl->fl_end < request->fl_start)
1010 goto next_lock;
1011 if (fl->fl_start > request->fl_end)
1012 break;
1013 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001014 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 if (fl->fl_start < request->fl_start)
1016 left = fl;
1017 /* If the next lock in the list has a higher end
1018 * address than the new one, insert the new one here.
1019 */
1020 if (fl->fl_end > request->fl_end) {
1021 right = fl;
1022 break;
1023 }
1024 if (fl->fl_start >= request->fl_start) {
1025 /* The new lock completely replaces an old
1026 * one (This may happen several times).
1027 */
1028 if (added) {
1029 locks_delete_lock(before);
1030 continue;
1031 }
1032 /* Replace the old lock with the new one.
1033 * Wake up anybody waiting for the old one,
1034 * as the change in lock type might satisfy
1035 * their needs.
1036 */
1037 locks_wake_up_blocks(fl);
1038 fl->fl_start = request->fl_start;
1039 fl->fl_end = request->fl_end;
1040 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -05001041 locks_release_private(fl);
1042 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001044 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 }
1046 }
1047 /* Go on to next lock.
1048 */
1049 next_lock:
1050 before = &fl->fl_next;
1051 }
1052
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001053 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001054 * The above code only modifies existing locks in case of merging or
1055 * replacing. If new lock(s) need to be inserted all modifications are
1056 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001057 */
1058 error = -ENOLCK; /* "no luck" */
1059 if (right && left == right && !new_fl2)
1060 goto out;
1061
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 error = 0;
1063 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001064 if (request->fl_type == F_UNLCK) {
1065 if (request->fl_flags & FL_EXISTS)
1066 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001068 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001069
1070 if (!new_fl) {
1071 error = -ENOLCK;
1072 goto out;
1073 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 locks_copy_lock(new_fl, request);
1075 locks_insert_lock(before, new_fl);
1076 new_fl = NULL;
1077 }
1078 if (right) {
1079 if (left == right) {
1080 /* The new lock breaks the old one in two pieces,
1081 * so we have to use the second new lock.
1082 */
1083 left = new_fl2;
1084 new_fl2 = NULL;
1085 locks_copy_lock(left, right);
1086 locks_insert_lock(before, left);
1087 }
1088 right->fl_start = request->fl_end + 1;
1089 locks_wake_up_blocks(right);
1090 }
1091 if (left) {
1092 left->fl_end = request->fl_start - 1;
1093 locks_wake_up_blocks(left);
1094 }
1095 out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001096 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 /*
1098 * Free any unused locks.
1099 */
1100 if (new_fl)
1101 locks_free_lock(new_fl);
1102 if (new_fl2)
1103 locks_free_lock(new_fl2);
1104 return error;
1105}
1106
1107/**
1108 * posix_lock_file - Apply a POSIX-style lock to a file
1109 * @filp: The file to apply the lock to
1110 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001111 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 *
1113 * Add a POSIX style lock to a file.
1114 * We merge adjacent & overlapping locks whenever possible.
1115 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001116 *
1117 * Note that if called with an FL_EXISTS argument, the caller may determine
1118 * whether or not a lock was successfully freed by testing the return
1119 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 */
Marc Eshel150b3932007-01-18 16:15:35 -05001121int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001122 struct file_lock *conflock)
1123{
Al Viro496ad9a2013-01-23 17:07:38 -05001124 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001125}
Marc Eshel150b3932007-01-18 16:15:35 -05001126EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128/**
1129 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1130 * @filp: The file to apply the lock to
1131 * @fl: The lock to be applied
1132 *
1133 * Add a POSIX style lock to a file.
1134 * We merge adjacent & overlapping locks whenever possible.
1135 * POSIX locks are sorted by owner task, then by starting address
1136 */
1137int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1138{
1139 int error;
1140 might_sleep ();
1141 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001142 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001143 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 break;
1145 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1146 if (!error)
1147 continue;
1148
1149 locks_delete_block(fl);
1150 break;
1151 }
1152 return error;
1153}
1154EXPORT_SYMBOL(posix_lock_file_wait);
1155
1156/**
1157 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001158 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 *
1160 * Searches the inode's list of locks to find any POSIX locks which conflict.
1161 * This function is called from locks_verify_locked() only.
1162 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001163int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164{
Jeff Laytond7a06982014-03-10 09:54:15 -04001165 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 fl_owner_t owner = current->files;
1167 struct file_lock *fl;
1168
1169 /*
1170 * Search the lock list for this inode for any POSIX locks.
1171 */
Jeff Layton1c8c6012013-06-21 08:58:15 -04001172 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1174 if (!IS_POSIX(fl))
1175 continue;
Jeff Laytond7a06982014-03-10 09:54:15 -04001176 if (fl->fl_owner != owner && fl->fl_owner != (fl_owner_t)file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 break;
1178 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001179 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 return fl ? -EAGAIN : 0;
1181}
1182
1183/**
1184 * locks_mandatory_area - Check for a conflicting lock
1185 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1186 * for shared
1187 * @inode: the file to check
1188 * @filp: how the file was opened (if it was)
1189 * @offset: start of area to check
1190 * @count: length of area to check
1191 *
1192 * Searches the inode's list of locks to find any POSIX locks which conflict.
1193 * This function is called from rw_verify_area() and
1194 * locks_verify_truncate().
1195 */
1196int locks_mandatory_area(int read_write, struct inode *inode,
1197 struct file *filp, loff_t offset,
1198 size_t count)
1199{
1200 struct file_lock fl;
1201 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001202 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203
1204 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 fl.fl_pid = current->tgid;
1206 fl.fl_file = filp;
1207 fl.fl_flags = FL_POSIX | FL_ACCESS;
1208 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001209 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1211 fl.fl_start = offset;
1212 fl.fl_end = offset + count - 1;
1213
1214 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001215 if (filp) {
1216 fl.fl_owner = (fl_owner_t)filp;
1217 fl.fl_flags &= ~FL_SLEEP;
1218 error = __posix_lock_file(inode, &fl, NULL);
1219 if (!error)
1220 break;
1221 }
1222
1223 if (sleep)
1224 fl.fl_flags |= FL_SLEEP;
1225 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001226 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001227 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 break;
1229 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1230 if (!error) {
1231 /*
1232 * If we've been sleeping someone might have
1233 * changed the permissions behind our back.
1234 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001235 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 continue;
1237 }
1238
1239 locks_delete_block(&fl);
1240 break;
1241 }
1242
1243 return error;
1244}
1245
1246EXPORT_SYMBOL(locks_mandatory_area);
1247
J. Bruce Fields778fc542011-07-26 18:25:49 -04001248static void lease_clear_pending(struct file_lock *fl, int arg)
1249{
1250 switch (arg) {
1251 case F_UNLCK:
1252 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1253 /* fall through: */
1254 case F_RDLCK:
1255 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1256 }
1257}
1258
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259/* We already had a lease on this file; just change its type */
1260int lease_modify(struct file_lock **before, int arg)
1261{
1262 struct file_lock *fl = *before;
1263 int error = assign_type(fl, arg);
1264
1265 if (error)
1266 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001267 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001269 if (arg == F_UNLCK) {
1270 struct file *filp = fl->fl_file;
1271
1272 f_delown(filp);
1273 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001274 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1275 if (fl->fl_fasync != NULL) {
1276 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1277 fl->fl_fasync = NULL;
1278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 locks_delete_lock(before);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001280 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 return 0;
1282}
1283
1284EXPORT_SYMBOL(lease_modify);
1285
J. Bruce Fields778fc542011-07-26 18:25:49 -04001286static bool past_time(unsigned long then)
1287{
1288 if (!then)
1289 /* 0 is a special value meaning "this never expires": */
1290 return false;
1291 return time_after(jiffies, then);
1292}
1293
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294static void time_out_leases(struct inode *inode)
1295{
1296 struct file_lock **before;
1297 struct file_lock *fl;
1298
1299 before = &inode->i_flock;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04001300 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001301 if (past_time(fl->fl_downgrade_time))
1302 lease_modify(before, F_RDLCK);
1303 if (past_time(fl->fl_break_time))
1304 lease_modify(before, F_UNLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 if (fl == *before) /* lease_modify may have freed fl */
1306 before = &fl->fl_next;
1307 }
1308}
1309
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001310static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1311{
1312 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1313 return false;
1314 return locks_conflict(breaker, lease);
1315}
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317/**
1318 * __break_lease - revoke all outstanding leases on file
1319 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001320 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1321 * break all leases
1322 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1323 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 *
david m. richter87250dd2007-05-09 16:10:27 -04001325 * break_lease (inlined for speed) has checked there already is at least
1326 * some kind of lock (maybe a lease) on this file. Leases are broken on
1327 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 * specified %O_NONBLOCK to your open().
1329 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001330int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001332 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 struct file_lock *new_fl, *flock;
1334 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 unsigned long break_time;
1336 int i_have_this_lease = 0;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001337 bool lease_conflict = false;
Al Viro8737c932009-12-24 06:47:55 -05001338 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Al Viro8737c932009-12-24 06:47:55 -05001340 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001341 if (IS_ERR(new_fl))
1342 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001343 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344
Jeff Layton1c8c6012013-06-21 08:58:15 -04001345 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346
1347 time_out_leases(inode);
1348
1349 flock = inode->i_flock;
1350 if ((flock == NULL) || !IS_LEASE(flock))
1351 goto out;
1352
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001353 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1354 if (leases_conflict(fl, new_fl)) {
1355 lease_conflict = true;
1356 if (fl->fl_owner == current->files)
1357 i_have_this_lease = 1;
1358 }
1359 }
1360 if (!lease_conflict)
J. Bruce Fields778fc542011-07-26 18:25:49 -04001361 goto out;
1362
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 break_time = 0;
1364 if (lease_break_time > 0) {
1365 break_time = jiffies + lease_break_time * HZ;
1366 if (break_time == 0)
1367 break_time++; /* so that 0 means no break time */
1368 }
1369
1370 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001371 if (!leases_conflict(fl, new_fl))
1372 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001373 if (want_write) {
1374 if (fl->fl_flags & FL_UNLOCK_PENDING)
1375 continue;
1376 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001378 } else {
1379 if (lease_breaking(flock))
1380 continue;
1381 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1382 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 }
J. Bruce Fields778fc542011-07-26 18:25:49 -04001384 fl->fl_lmops->lm_break(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 }
1386
1387 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1388 error = -EWOULDBLOCK;
1389 goto out;
1390 }
1391
1392restart:
1393 break_time = flock->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001394 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001396 if (break_time == 0)
1397 break_time++;
Matthew Wilcox4321e012008-01-14 21:28:30 -07001398 locks_insert_block(flock, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001399 spin_unlock(&inode->i_lock);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001400 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1401 !new_fl->fl_next, break_time);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001402 spin_lock(&inode->i_lock);
1403 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 if (error >= 0) {
1405 if (error == 0)
1406 time_out_leases(inode);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001407 /*
1408 * Wait for the next conflicting lease that has not been
1409 * broken yet
1410 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1412 flock = flock->fl_next) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001413 if (leases_conflict(new_fl, flock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 goto restart;
1415 }
1416 error = 0;
1417 }
1418
1419out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001420 spin_unlock(&inode->i_lock);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001421 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 return error;
1423}
1424
1425EXPORT_SYMBOL(__break_lease);
1426
1427/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001428 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 * @inode: the inode
1430 * @time: pointer to a timespec which will contain the last modified time
1431 *
1432 * This is to force NFS clients to flush their caches for files with
1433 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001434 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 */
1436void lease_get_mtime(struct inode *inode, struct timespec *time)
1437{
1438 struct file_lock *flock = inode->i_flock;
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04001439 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 *time = current_fs_time(inode->i_sb);
1441 else
1442 *time = inode->i_mtime;
1443}
1444
1445EXPORT_SYMBOL(lease_get_mtime);
1446
1447/**
1448 * fcntl_getlease - Enquire what lease is currently active
1449 * @filp: the file
1450 *
1451 * The value returned by this function will be one of
1452 * (if no lease break is pending):
1453 *
1454 * %F_RDLCK to indicate a shared lease is held.
1455 *
1456 * %F_WRLCK to indicate an exclusive lease is held.
1457 *
1458 * %F_UNLCK to indicate no lease is held.
1459 *
1460 * (if a lease break is pending):
1461 *
1462 * %F_RDLCK to indicate an exclusive lease needs to be
1463 * changed to a shared lease (or removed).
1464 *
1465 * %F_UNLCK to indicate the lease needs to be removed.
1466 *
1467 * XXX: sfr & willy disagree over whether F_INPROGRESS
1468 * should be returned to userspace.
1469 */
1470int fcntl_getlease(struct file *filp)
1471{
1472 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001473 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 int type = F_UNLCK;
1475
Jeff Layton1c8c6012013-06-21 08:58:15 -04001476 spin_lock(&inode->i_lock);
Al Viro496ad9a2013-01-23 17:07:38 -05001477 time_out_leases(file_inode(filp));
1478 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 fl = fl->fl_next) {
1480 if (fl->fl_file == filp) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001481 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 break;
1483 }
1484 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001485 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 return type;
1487}
1488
Jeff Layton24cbe782014-02-03 12:13:06 -05001489/**
1490 * check_conflicting_open - see if the given dentry points to a file that has
1491 * an existing open that would conflict with the
1492 * desired lease.
1493 * @dentry: dentry to check
1494 * @arg: type of lease that we're trying to acquire
1495 *
1496 * Check to see if there's an existing open fd on this file that would
1497 * conflict with the lease we're trying to set.
1498 */
1499static int
1500check_conflicting_open(const struct dentry *dentry, const long arg)
1501{
1502 int ret = 0;
1503 struct inode *inode = dentry->d_inode;
1504
1505 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1506 return -EAGAIN;
1507
1508 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1509 (atomic_read(&inode->i_count) > 1)))
1510 ret = -EAGAIN;
1511
1512 return ret;
1513}
1514
Jeff Laytond4f22d12013-06-21 08:58:11 -04001515static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001517 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001518 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 struct inode *inode = dentry->d_inode;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001520 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001521 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
J. Bruce Fields096657b2010-10-30 17:31:14 -04001523 lease = *flp;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001524 /*
1525 * In the delegation case we need mutual exclusion with
1526 * a number of operations that take the i_mutex. We trylock
1527 * because delegations are an optional optimization, and if
1528 * there's some chance of a conflict--we'd rather not
1529 * bother, maybe that's a sign this just isn't a good file to
1530 * hand out a delegation on.
1531 */
1532 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1533 return -EAGAIN;
1534
1535 if (is_deleg && arg == F_WRLCK) {
1536 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001537 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001538 WARN_ON_ONCE(1);
1539 return -EINVAL;
1540 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001541
Jeff Layton24cbe782014-02-03 12:13:06 -05001542 error = check_conflicting_open(dentry, arg);
1543 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001544 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 /*
1547 * At this point, we know that if there is an exclusive
1548 * lease on this file, then we hold it on this filp
1549 * (otherwise our open of this file would have blocked).
1550 * And if we are trying to acquire an exclusive lease,
1551 * then the file is not open by anyone (including us)
1552 * except for this filp.
1553 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001554 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 for (before = &inode->i_flock;
1556 ((fl = *before) != NULL) && IS_LEASE(fl);
1557 before = &fl->fl_next) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001558 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 my_before = before;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001560 continue;
1561 }
1562 /*
1563 * No exclusive leases if someone else has a lease on
1564 * this file:
1565 */
1566 if (arg == F_WRLCK)
1567 goto out;
1568 /*
1569 * Modifying our existing lease is OK, but no getting a
1570 * new lease if someone else is opening for write:
1571 */
1572 if (fl->fl_flags & FL_UNLOCK_PENDING)
1573 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 }
1575
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 if (my_before != NULL) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001577 error = lease->fl_lmops->lm_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001578 if (!error)
1579 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 goto out;
1581 }
1582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 error = -EINVAL;
1584 if (!leases_enable)
1585 goto out;
1586
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001587 locks_insert_lock(before, lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001588 /*
1589 * The check in break_lease() is lockless. It's possible for another
1590 * open to race in after we did the earlier check for a conflicting
1591 * open but before the lease was inserted. Check again for a
1592 * conflicting open and cancel the lease if there is one.
1593 *
1594 * We also add a barrier here to ensure that the insertion of the lock
1595 * precedes these checks.
1596 */
1597 smp_mb();
1598 error = check_conflicting_open(dentry, arg);
1599 if (error)
1600 locks_unlink_lock(flp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601out:
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001602 if (is_deleg)
1603 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 return error;
1605}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001606
Jeff Laytond4f22d12013-06-21 08:58:11 -04001607static int generic_delete_lease(struct file *filp, struct file_lock **flp)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001608{
1609 struct file_lock *fl, **before;
1610 struct dentry *dentry = filp->f_path.dentry;
1611 struct inode *inode = dentry->d_inode;
1612
1613 for (before = &inode->i_flock;
1614 ((fl = *before) != NULL) && IS_LEASE(fl);
1615 before = &fl->fl_next) {
1616 if (fl->fl_file != filp)
1617 continue;
1618 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1619 }
1620 return -EAGAIN;
1621}
1622
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623/**
1624 * generic_setlease - sets a lease on an open file
1625 * @filp: file pointer
1626 * @arg: type of lease to obtain
1627 * @flp: input - file_lock to use, output - file_lock inserted
1628 *
1629 * The (input) flp->fl_lmops->lm_break function is required
1630 * by break_lease().
1631 *
Jeff Layton1c8c6012013-06-21 08:58:15 -04001632 * Called with inode->i_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 */
1634int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1635{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 struct dentry *dentry = filp->f_path.dentry;
1637 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001638 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001640 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001641 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001643 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 error = security_file_lock(filp, arg);
1645 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001646 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647
1648 time_out_leases(inode);
1649
1650 BUG_ON(!(*flp)->fl_lmops->lm_break);
1651
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001652 switch (arg) {
1653 case F_UNLCK:
1654 return generic_delete_lease(filp, flp);
1655 case F_RDLCK:
1656 case F_WRLCK:
1657 return generic_add_lease(filp, arg, flp);
1658 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001659 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001662EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001664static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1665{
Al Viro72c2d532013-09-22 16:27:52 -04001666 if (filp->f_op->setlease)
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001667 return filp->f_op->setlease(filp, arg, lease);
1668 else
1669 return generic_setlease(filp, arg, lease);
1670}
1671
1672/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001673 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 * @filp: file pointer
1675 * @arg: type of lease to obtain
1676 * @lease: file_lock to use
1677 *
1678 * Call this to establish a lease on the file.
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001679 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001680 * break_lease will oops!
1681 *
1682 * This will call the filesystem's setlease file method, if
1683 * defined. Note that there is no getlease method; instead, the
1684 * filesystem setlease method should call back to setlease() to
1685 * add a lease to the inode's lease list, where fcntl_getlease() can
1686 * find it. Since fcntl_getlease() only reports whether the current
1687 * task holds a lease, a cluster filesystem need only do this for
1688 * leases held by processes on this node.
1689 *
1690 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001691 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001692 * filesystem's open, create, and (on truncate) setattr methods.
1693 *
1694 * Warning: the only current setlease methods exist only to disable
1695 * leases in certain cases. More vfs changes may be required to
1696 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 */
1698
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001699int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700{
Jeff Layton1c8c6012013-06-21 08:58:15 -04001701 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702 int error;
1703
Jeff Layton1c8c6012013-06-21 08:58:15 -04001704 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001705 error = __vfs_setlease(filp, arg, lease);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001706 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707
1708 return error;
1709}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001710EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001712static int do_fcntl_delete_lease(struct file *filp)
1713{
1714 struct file_lock fl, *flp = &fl;
1715
1716 lease_init(filp, F_UNLCK, flp);
1717
1718 return vfs_setlease(filp, F_UNLCK, &flp);
1719}
1720
1721static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001723 struct file_lock *fl, *ret;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001724 struct inode *inode = file_inode(filp);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001725 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 int error;
1727
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001728 fl = lease_alloc(filp, arg);
1729 if (IS_ERR(fl))
1730 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001732 new = fasync_alloc();
1733 if (!new) {
1734 locks_free_lock(fl);
1735 return -ENOMEM;
1736 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001737 ret = fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001738 spin_lock(&inode->i_lock);
J. Bruce Fields8896b932010-11-03 18:09:18 -04001739 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001740 if (error) {
Jeff Layton1c8c6012013-06-21 08:58:15 -04001741 spin_unlock(&inode->i_lock);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001742 locks_free_lock(fl);
1743 goto out_free_fasync;
1744 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001745 if (ret != fl)
1746 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001748 /*
1749 * fasync_insert_entry() returns the old entry if any.
1750 * If there was no old entry, then it used 'new' and
1751 * inserted it into the fasync list. Clear new so that
1752 * we don't release it here.
1753 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001754 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001755 new = NULL;
1756
J. Bruce Fields8896b932010-11-03 18:09:18 -04001757 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001758 spin_unlock(&inode->i_lock);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001759
1760out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001761 if (new)
1762 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 return error;
1764}
1765
1766/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001767 * fcntl_setlease - sets a lease on an open file
1768 * @fd: open file descriptor
1769 * @filp: file pointer
1770 * @arg: type of lease to obtain
1771 *
1772 * Call this fcntl to establish a lease on the file.
1773 * Note that you also need to call %F_SETSIG to
1774 * receive a signal when the lease is broken.
1775 */
1776int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1777{
1778 if (arg == F_UNLCK)
1779 return do_fcntl_delete_lease(filp);
1780 return do_fcntl_add_lease(fd, filp, arg);
1781}
1782
1783/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1785 * @filp: The file to apply the lock to
1786 * @fl: The lock to be applied
1787 *
1788 * Add a FLOCK style lock to a file.
1789 */
1790int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1791{
1792 int error;
1793 might_sleep();
1794 for (;;) {
1795 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001796 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 break;
1798 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1799 if (!error)
1800 continue;
1801
1802 locks_delete_block(fl);
1803 break;
1804 }
1805 return error;
1806}
1807
1808EXPORT_SYMBOL(flock_lock_file_wait);
1809
1810/**
1811 * sys_flock: - flock() system call.
1812 * @fd: the file descriptor to lock.
1813 * @cmd: the type of lock to apply.
1814 *
1815 * Apply a %FL_FLOCK style lock to an open file descriptor.
1816 * The @cmd can be one of
1817 *
1818 * %LOCK_SH -- a shared lock.
1819 *
1820 * %LOCK_EX -- an exclusive lock.
1821 *
1822 * %LOCK_UN -- remove an existing lock.
1823 *
1824 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1825 *
1826 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1827 * processes read and write access respectively.
1828 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001829SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830{
Al Viro2903ff02012-08-28 12:52:22 -04001831 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 struct file_lock *lock;
1833 int can_sleep, unlock;
1834 int error;
1835
1836 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001837 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 goto out;
1839
1840 can_sleep = !(cmd & LOCK_NB);
1841 cmd &= ~LOCK_NB;
1842 unlock = (cmd == LOCK_UN);
1843
Al Viroaeb5d722008-09-02 15:28:45 -04001844 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001845 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 goto out_putf;
1847
Al Viro2903ff02012-08-28 12:52:22 -04001848 error = flock_make_lock(f.file, &lock, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 if (error)
1850 goto out_putf;
1851 if (can_sleep)
1852 lock->fl_flags |= FL_SLEEP;
1853
Al Viro2903ff02012-08-28 12:52:22 -04001854 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 if (error)
1856 goto out_free;
1857
Al Viro72c2d532013-09-22 16:27:52 -04001858 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001859 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 (can_sleep) ? F_SETLKW : F_SETLK,
1861 lock);
1862 else
Al Viro2903ff02012-08-28 12:52:22 -04001863 error = flock_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864
1865 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001866 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867
1868 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001869 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 out:
1871 return error;
1872}
1873
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001874/**
1875 * vfs_test_lock - test file byte range lock
1876 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001877 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001878 *
1879 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1880 * setting conf->fl_type to something other than F_UNLCK.
1881 */
1882int vfs_test_lock(struct file *filp, struct file_lock *fl)
1883{
Al Viro72c2d532013-09-22 16:27:52 -04001884 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001885 return filp->f_op->lock(filp, F_GETLK, fl);
1886 posix_test_lock(filp, fl);
1887 return 0;
1888}
1889EXPORT_SYMBOL_GPL(vfs_test_lock);
1890
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001891static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1892{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001893 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001894#if BITS_PER_LONG == 32
1895 /*
1896 * Make sure we can represent the posix lock via
1897 * legacy 32bit flock.
1898 */
1899 if (fl->fl_start > OFFT_OFFSET_MAX)
1900 return -EOVERFLOW;
1901 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1902 return -EOVERFLOW;
1903#endif
1904 flock->l_start = fl->fl_start;
1905 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1906 fl->fl_end - fl->fl_start + 1;
1907 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001908 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001909 return 0;
1910}
1911
1912#if BITS_PER_LONG == 32
1913static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1914{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001915 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001916 flock->l_start = fl->fl_start;
1917 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1918 fl->fl_end - fl->fl_start + 1;
1919 flock->l_whence = 0;
1920 flock->l_type = fl->fl_type;
1921}
1922#endif
1923
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924/* Report the first existing lock that would conflict with l.
1925 * This implements the F_GETLK command of fcntl().
1926 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05001927int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001929 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 struct flock flock;
1931 int error;
1932
1933 error = -EFAULT;
1934 if (copy_from_user(&flock, l, sizeof(flock)))
1935 goto out;
1936 error = -EINVAL;
1937 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1938 goto out;
1939
1940 error = flock_to_posix_lock(filp, &file_lock, &flock);
1941 if (error)
1942 goto out;
1943
Jeff Layton0d3f7a22014-04-22 08:23:58 -04001944 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05001945 error = -EINVAL;
1946 if (flock.l_pid != 0)
1947 goto out;
1948
Jeff Layton5d50ffd2014-02-03 12:13:10 -05001949 cmd = F_GETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001950 file_lock.fl_flags |= FL_OFDLCK;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05001951 file_lock.fl_owner = (fl_owner_t)filp;
1952 }
1953
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001954 error = vfs_test_lock(filp, &file_lock);
1955 if (error)
1956 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001958 flock.l_type = file_lock.fl_type;
1959 if (file_lock.fl_type != F_UNLCK) {
1960 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001961 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 }
1964 error = -EFAULT;
1965 if (!copy_to_user(l, &flock, sizeof(flock)))
1966 error = 0;
1967out:
1968 return error;
1969}
1970
Marc Eshel7723ec92007-01-18 15:08:55 -05001971/**
1972 * vfs_lock_file - file byte range lock
1973 * @filp: The file to apply the lock to
1974 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1975 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001976 * @conf: Place to return a copy of the conflicting lock, if found.
1977 *
1978 * A caller that doesn't care about the conflicting lock may pass NULL
1979 * as the final argument.
1980 *
1981 * If the filesystem defines a private ->lock() method, then @conf will
1982 * be left unchanged; so a caller that cares should initialize it to
1983 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001984 *
1985 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1986 * locks, the ->lock() interface may return asynchronously, before the lock has
1987 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001988 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05001989 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1990 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001991 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001992 * request completes.
1993 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001994 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1995 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001996 * nonzero return code and the file system should release the lock. The file
1997 * system is also responsible to keep a corresponding posix lock when it
1998 * grants a lock so the VFS can find out which locks are locally held and do
1999 * the correct lock cleanup when required.
2000 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002001 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002002 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002003 */
Marc Eshel150b3932007-01-18 16:15:35 -05002004int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002005{
Al Viro72c2d532013-09-22 16:27:52 -04002006 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002007 return filp->f_op->lock(filp, cmd, fl);
2008 else
Marc Eshel150b3932007-01-18 16:15:35 -05002009 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002010}
2011EXPORT_SYMBOL_GPL(vfs_lock_file);
2012
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002013static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2014 struct file_lock *fl)
2015{
2016 int error;
2017
2018 error = security_file_lock(filp, fl->fl_type);
2019 if (error)
2020 return error;
2021
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002022 for (;;) {
2023 error = vfs_lock_file(filp, cmd, fl, NULL);
2024 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002025 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002026 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2027 if (!error)
2028 continue;
2029
2030 locks_delete_block(fl);
2031 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002032 }
2033
2034 return error;
2035}
2036
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037/* Apply the lock described by l to an open file descriptor.
2038 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2039 */
Peter Staubachc2936212005-07-27 11:45:09 -07002040int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2041 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042{
2043 struct file_lock *file_lock = locks_alloc_lock();
2044 struct flock flock;
2045 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002046 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 int error;
2048
2049 if (file_lock == NULL)
2050 return -ENOLCK;
2051
2052 /*
2053 * This might block, so we do it before checking the inode.
2054 */
2055 error = -EFAULT;
2056 if (copy_from_user(&flock, l, sizeof(flock)))
2057 goto out;
2058
Al Viro496ad9a2013-01-23 17:07:38 -05002059 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060
2061 /* Don't allow mandatory locks on files that may be memory mapped
2062 * and shared.
2063 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002064 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 error = -EAGAIN;
2066 goto out;
2067 }
2068
Peter Staubachc2936212005-07-27 11:45:09 -07002069again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 error = flock_to_posix_lock(filp, file_lock, &flock);
2071 if (error)
2072 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002073
2074 /*
2075 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002076 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002077 */
2078 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002079 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002080 error = -EINVAL;
2081 if (flock.l_pid != 0)
2082 goto out;
2083
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002084 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002085 file_lock->fl_flags |= FL_OFDLCK;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002086 file_lock->fl_owner = (fl_owner_t)filp;
2087 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002088 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002089 error = -EINVAL;
2090 if (flock.l_pid != 0)
2091 goto out;
2092
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002093 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002094 file_lock->fl_flags |= FL_OFDLCK;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002095 file_lock->fl_owner = (fl_owner_t)filp;
2096 /* Fallthrough */
2097 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 file_lock->fl_flags |= FL_SLEEP;
2099 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002100
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002101 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102
Peter Staubachc2936212005-07-27 11:45:09 -07002103 /*
2104 * Attempt to detect a close/fcntl race and recover by
2105 * releasing the lock that was just acquired.
2106 */
Al Viro0b2bac22008-05-06 13:58:34 -04002107 /*
2108 * we need that spin_lock here - it prevents reordering between
2109 * update of inode->i_flock and check for it done in close().
2110 * rcu_read_lock() wouldn't do.
2111 */
2112 spin_lock(&current->files->file_lock);
2113 f = fcheck(fd);
2114 spin_unlock(&current->files->file_lock);
2115 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002116 flock.l_type = F_UNLCK;
2117 goto again;
2118 }
2119
2120out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 locks_free_lock(file_lock);
2122 return error;
2123}
2124
2125#if BITS_PER_LONG == 32
2126/* Report the first existing lock that would conflict with l.
2127 * This implements the F_GETLK command of fcntl().
2128 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002129int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002131 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132 struct flock64 flock;
2133 int error;
2134
2135 error = -EFAULT;
2136 if (copy_from_user(&flock, l, sizeof(flock)))
2137 goto out;
2138 error = -EINVAL;
2139 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2140 goto out;
2141
2142 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2143 if (error)
2144 goto out;
2145
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002146 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002147 error = -EINVAL;
2148 if (flock.l_pid != 0)
2149 goto out;
2150
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002151 cmd = F_GETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002152 file_lock.fl_flags |= FL_OFDLCK;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002153 file_lock.fl_owner = (fl_owner_t)filp;
2154 }
2155
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002156 error = vfs_test_lock(filp, &file_lock);
2157 if (error)
2158 goto out;
2159
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002160 flock.l_type = file_lock.fl_type;
2161 if (file_lock.fl_type != F_UNLCK)
2162 posix_lock_to_flock64(&flock, &file_lock);
2163
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 error = -EFAULT;
2165 if (!copy_to_user(l, &flock, sizeof(flock)))
2166 error = 0;
2167
2168out:
2169 return error;
2170}
2171
2172/* Apply the lock described by l to an open file descriptor.
2173 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2174 */
Peter Staubachc2936212005-07-27 11:45:09 -07002175int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2176 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177{
2178 struct file_lock *file_lock = locks_alloc_lock();
2179 struct flock64 flock;
2180 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002181 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 int error;
2183
2184 if (file_lock == NULL)
2185 return -ENOLCK;
2186
2187 /*
2188 * This might block, so we do it before checking the inode.
2189 */
2190 error = -EFAULT;
2191 if (copy_from_user(&flock, l, sizeof(flock)))
2192 goto out;
2193
Al Viro496ad9a2013-01-23 17:07:38 -05002194 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
2196 /* Don't allow mandatory locks on files that may be memory mapped
2197 * and shared.
2198 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002199 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 error = -EAGAIN;
2201 goto out;
2202 }
2203
Peter Staubachc2936212005-07-27 11:45:09 -07002204again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 error = flock64_to_posix_lock(filp, file_lock, &flock);
2206 if (error)
2207 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002208
2209 /*
2210 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002211 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002212 */
2213 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002214 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002215 error = -EINVAL;
2216 if (flock.l_pid != 0)
2217 goto out;
2218
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002219 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002220 file_lock->fl_flags |= FL_OFDLCK;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002221 file_lock->fl_owner = (fl_owner_t)filp;
2222 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002223 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002224 error = -EINVAL;
2225 if (flock.l_pid != 0)
2226 goto out;
2227
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002228 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002229 file_lock->fl_flags |= FL_OFDLCK;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002230 file_lock->fl_owner = (fl_owner_t)filp;
2231 /* Fallthrough */
2232 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233 file_lock->fl_flags |= FL_SLEEP;
2234 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002235
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002236 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
Peter Staubachc2936212005-07-27 11:45:09 -07002238 /*
2239 * Attempt to detect a close/fcntl race and recover by
2240 * releasing the lock that was just acquired.
2241 */
Al Viro0b2bac22008-05-06 13:58:34 -04002242 spin_lock(&current->files->file_lock);
2243 f = fcheck(fd);
2244 spin_unlock(&current->files->file_lock);
2245 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002246 flock.l_type = F_UNLCK;
2247 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 }
2249
2250out:
2251 locks_free_lock(file_lock);
2252 return error;
2253}
2254#endif /* BITS_PER_LONG == 32 */
2255
2256/*
2257 * This function is called when the file is being removed
2258 * from the task's fd array. POSIX locks belonging to this task
2259 * are deleted at this time.
2260 */
2261void locks_remove_posix(struct file *filp, fl_owner_t owner)
2262{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002263 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264
2265 /*
2266 * If there are no locks held on this file, we don't need to call
2267 * posix_lock_file(). Another process could be setting a lock on this
2268 * file at the same time, but we wouldn't remove that lock anyway.
2269 */
Al Viro496ad9a2013-01-23 17:07:38 -05002270 if (!file_inode(filp)->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271 return;
2272
2273 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002274 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 lock.fl_start = 0;
2276 lock.fl_end = OFFSET_MAX;
2277 lock.fl_owner = owner;
2278 lock.fl_pid = current->tgid;
2279 lock.fl_file = filp;
2280 lock.fl_ops = NULL;
2281 lock.fl_lmops = NULL;
2282
Marc Eshel150b3932007-01-18 16:15:35 -05002283 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2286 lock.fl_ops->fl_release_private(&lock);
2287}
2288
2289EXPORT_SYMBOL(locks_remove_posix);
2290
2291/*
2292 * This function is called on the last close of an open file.
2293 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002294void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295{
Al Viro496ad9a2013-01-23 17:07:38 -05002296 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 struct file_lock *fl;
2298 struct file_lock **before;
2299
2300 if (!inode->i_flock)
2301 return;
2302
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002303 locks_remove_posix(filp, (fl_owner_t)filp);
2304
Al Viro72c2d532013-09-22 16:27:52 -04002305 if (filp->f_op->flock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 struct file_lock fl = {
2307 .fl_pid = current->tgid,
2308 .fl_file = filp,
2309 .fl_flags = FL_FLOCK,
2310 .fl_type = F_UNLCK,
2311 .fl_end = OFFSET_MAX,
2312 };
2313 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002314 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2315 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 }
2317
Jeff Layton1c8c6012013-06-21 08:58:15 -04002318 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 before = &inode->i_flock;
2320
2321 while ((fl = *before) != NULL) {
2322 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 if (IS_LEASE(fl)) {
2324 lease_modify(before, F_UNLCK);
2325 continue;
2326 }
Jeff Layton8c3cac52014-02-03 12:13:07 -05002327
2328 /*
2329 * There's a leftover lock on the list of a type that
2330 * we didn't expect to see. Most likely a classic
2331 * POSIX lock that ended up not getting released
2332 * properly, or that raced onto the list somehow. Log
2333 * some info about it and then just remove it from
2334 * the list.
2335 */
2336 WARN(!IS_FLOCK(fl),
2337 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2338 MAJOR(inode->i_sb->s_dev),
2339 MINOR(inode->i_sb->s_dev), inode->i_ino,
2340 fl->fl_type, fl->fl_flags,
2341 fl->fl_start, fl->fl_end);
2342
2343 locks_delete_lock(before);
2344 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 }
2346 before = &fl->fl_next;
2347 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002348 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349}
2350
2351/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 * @waiter: the lock which was waiting
2354 *
2355 * lockd needs to block waiting for locks.
2356 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002357int
Jeff Laytonf891a292013-06-21 08:58:09 -04002358posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002360 int status = 0;
2361
Jeff Layton7b2296a2013-06-21 08:58:20 -04002362 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002363 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002365 else
2366 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002367 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002368 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370EXPORT_SYMBOL(posix_unblock_lock);
2371
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002372/**
2373 * vfs_cancel_lock - file byte range unblock lock
2374 * @filp: The file to apply the unblock to
2375 * @fl: The lock to be unblocked
2376 *
2377 * Used by lock managers to cancel blocked requests
2378 */
2379int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2380{
Al Viro72c2d532013-09-22 16:27:52 -04002381 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002382 return filp->f_op->lock(filp, F_CANCELLK, fl);
2383 return 0;
2384}
2385
2386EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2387
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002388#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002389#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002390#include <linux/seq_file.h>
2391
Jeff Layton7012b022013-06-21 08:58:22 -04002392struct locks_iterator {
2393 int li_cpu;
2394 loff_t li_pos;
2395};
2396
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002397static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002398 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399{
2400 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002401 unsigned int fl_pid;
2402
2403 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002404 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002405 else
2406 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407
2408 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002409 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Jerome Marchand99dc8292010-10-26 14:22:33 -07002411 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002413 if (fl->fl_flags & FL_ACCESS)
2414 seq_printf(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002415 else if (IS_OFDLCK(fl))
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002416 seq_printf(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002417 else
2418 seq_printf(f, "POSIX ");
2419
2420 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002422 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 } else if (IS_FLOCK(fl)) {
2424 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002425 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002426 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002427 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 }
2429 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002430 seq_printf(f, "LEASE ");
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002431 if (lease_breaking(fl))
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002432 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002434 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002436 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002438 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 }
2440 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002441 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 (fl->fl_type & LOCK_READ)
2443 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2444 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2445 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002446 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002447 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002448 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2449 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 }
2451 if (inode) {
2452#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002453 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 inode->i_sb->s_id, inode->i_ino);
2455#else
2456 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002457 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 MAJOR(inode->i_sb->s_dev),
2459 MINOR(inode->i_sb->s_dev), inode->i_ino);
2460#endif
2461 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002462 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 }
2464 if (IS_POSIX(fl)) {
2465 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002466 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002468 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002470 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 }
2472}
2473
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002474static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475{
Jeff Layton7012b022013-06-21 08:58:22 -04002476 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002477 struct file_lock *fl, *bfl;
2478
Jeff Layton139ca042013-06-21 08:58:17 -04002479 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002480
Jeff Layton7012b022013-06-21 08:58:22 -04002481 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002482
2483 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002484 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002485
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002486 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487}
2488
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002489static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002490 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491{
Jeff Layton7012b022013-06-21 08:58:22 -04002492 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002493
Jeff Layton7012b022013-06-21 08:58:22 -04002494 iter->li_pos = *pos + 1;
2495 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002496 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002497 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498}
2499
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002500static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2501{
Jeff Layton7012b022013-06-21 08:58:22 -04002502 struct locks_iterator *iter = f->private;
2503
2504 ++iter->li_pos;
2505 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002506}
2507
2508static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002509 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002510{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002511 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002512 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002513}
2514
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002515static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002516 .start = locks_start,
2517 .next = locks_next,
2518 .stop = locks_stop,
2519 .show = locks_show,
2520};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002521
2522static int locks_open(struct inode *inode, struct file *filp)
2523{
Jeff Layton7012b022013-06-21 08:58:22 -04002524 return seq_open_private(filp, &locks_seq_operations,
2525 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002526}
2527
2528static const struct file_operations proc_locks_operations = {
2529 .open = locks_open,
2530 .read = seq_read,
2531 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002532 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002533};
2534
2535static int __init proc_locks_init(void)
2536{
2537 proc_create("locks", 0, NULL, &proc_locks_operations);
2538 return 0;
2539}
2540module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002541#endif
2542
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543/**
2544 * lock_may_read - checks that the region is free of locks
2545 * @inode: the inode that is being read
2546 * @start: the first byte to read
2547 * @len: the number of bytes to read
2548 *
2549 * Emulates Windows locking requirements. Whole-file
2550 * mandatory locks (share modes) can prohibit a read and
2551 * byte-range POSIX locks can prohibit a read if they overlap.
2552 *
2553 * N.B. this function is only ever called
2554 * from knfsd and ownership of locks is never checked.
2555 */
2556int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2557{
2558 struct file_lock *fl;
2559 int result = 1;
Jeff Layton1c8c6012013-06-21 08:58:15 -04002560
2561 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2563 if (IS_POSIX(fl)) {
2564 if (fl->fl_type == F_RDLCK)
2565 continue;
2566 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2567 continue;
2568 } else if (IS_FLOCK(fl)) {
2569 if (!(fl->fl_type & LOCK_MAND))
2570 continue;
2571 if (fl->fl_type & LOCK_READ)
2572 continue;
2573 } else
2574 continue;
2575 result = 0;
2576 break;
2577 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002578 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 return result;
2580}
2581
2582EXPORT_SYMBOL(lock_may_read);
2583
2584/**
2585 * lock_may_write - checks that the region is free of locks
2586 * @inode: the inode that is being written
2587 * @start: the first byte to write
2588 * @len: the number of bytes to write
2589 *
2590 * Emulates Windows locking requirements. Whole-file
2591 * mandatory locks (share modes) can prohibit a write and
2592 * byte-range POSIX locks can prohibit a write if they overlap.
2593 *
2594 * N.B. this function is only ever called
2595 * from knfsd and ownership of locks is never checked.
2596 */
2597int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2598{
2599 struct file_lock *fl;
2600 int result = 1;
Jeff Layton1c8c6012013-06-21 08:58:15 -04002601
2602 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2604 if (IS_POSIX(fl)) {
2605 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2606 continue;
2607 } else if (IS_FLOCK(fl)) {
2608 if (!(fl->fl_type & LOCK_MAND))
2609 continue;
2610 if (fl->fl_type & LOCK_WRITE)
2611 continue;
2612 } else
2613 continue;
2614 result = 0;
2615 break;
2616 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002617 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 return result;
2619}
2620
2621EXPORT_SYMBOL(lock_may_write);
2622
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623static int __init filelock_init(void)
2624{
Jeff Layton7012b022013-06-21 08:58:22 -04002625 int i;
2626
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002628 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2629
Jeff Layton7012b022013-06-21 08:58:22 -04002630 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2631
2632 for_each_possible_cpu(i)
2633 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2634
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 return 0;
2636}
2637
2638core_initcall(filelock_init);