blob: 4d4e790150e02d985477857939e0bb6db3bdae69 [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))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400139static bool lease_breaking(struct file_lock *fl)
140{
J. Bruce Fields778fc542011-07-26 18:25:49 -0400141 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
142}
143
144static int target_leasetype(struct file_lock *fl)
145{
146 if (fl->fl_flags & FL_UNLOCK_PENDING)
147 return F_UNLCK;
148 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
149 return F_RDLCK;
150 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400151}
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153int leases_enable = 1;
154int lease_break_time = 45;
155
156#define for_each_lock(inode, lockp) \
157 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
158
Jeff Layton1c8c6012013-06-21 08:58:15 -0400159/*
Jeff Layton7012b022013-06-21 08:58:22 -0400160 * The global file_lock_list is only used for displaying /proc/locks, so we
161 * keep a list on each CPU, with each list protected by its own spinlock via
162 * the file_lock_lglock. Note that alterations to the list also require that
163 * the relevant i_lock is held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400164 */
Jeff Layton7012b022013-06-21 08:58:22 -0400165DEFINE_STATIC_LGLOCK(file_lock_lglock);
166static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
Jeff Layton88974692013-06-21 08:58:14 -0400167
Jeff Layton1c8c6012013-06-21 08:58:15 -0400168/*
Jeff Layton48f74182013-06-21 08:58:18 -0400169 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400170 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400171 *
172 * We hash locks by lockowner in order to optimize searching for the lock a
173 * particular lockowner is waiting on.
174 *
175 * FIXME: make this value scale via some heuristic? We generally will want more
176 * buckets when we have more lockowners holding locks, but that's a little
177 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400178 */
Jeff Layton48f74182013-06-21 08:58:18 -0400179#define BLOCKED_HASH_BITS 7
180static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400181
Jeff Layton1c8c6012013-06-21 08:58:15 -0400182/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400183 * This lock protects the blocked_hash. Generally, if you're accessing it, you
184 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400185 *
186 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
187 * pointer for file_lock structures that are acting as lock requests (in
188 * contrast to those that are acting as records of acquired locks).
189 *
190 * Note that when we acquire this lock in order to change the above fields,
191 * we often hold the i_lock as well. In certain cases, when reading the fields
192 * protected by this lock, we can skip acquiring it iff we already hold the
193 * i_lock.
194 *
195 * In particular, adding an entry to the fl_block list requires that you hold
196 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
197 * an entry from the list however only requires the file_lock_lock.
198 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400199static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Christoph Lametere18b8902006-12-06 20:33:20 -0800201static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
Miklos Szerediee19cc42011-07-07 13:06:09 +0200203static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200204{
Jeff Layton139ca042013-06-21 08:58:17 -0400205 INIT_HLIST_NODE(&fl->fl_link);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200206 INIT_LIST_HEAD(&fl->fl_block);
207 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200208}
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200211struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200213 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200214
215 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200216 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200217
218 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200220EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500222void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500223{
224 if (fl->fl_ops) {
225 if (fl->fl_ops->fl_release_private)
226 fl->fl_ops->fl_release_private(fl);
227 fl->fl_ops = NULL;
228 }
J. Bruce Fields068535f2012-08-01 07:56:16 -0400229 fl->fl_lmops = NULL;
Trond Myklebust47831f32006-03-20 13:44:05 -0500230
231}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500232EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400235void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800237 BUG_ON(waitqueue_active(&fl->fl_wait));
238 BUG_ON(!list_empty(&fl->fl_block));
Jeff Layton139ca042013-06-21 08:58:17 -0400239 BUG_ON(!hlist_unhashed(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
Trond Myklebust47831f32006-03-20 13:44:05 -0500241 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 kmem_cache_free(filelock_cache, fl);
243}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400244EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246void locks_init_lock(struct file_lock *fl)
247{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200248 memset(fl, 0, sizeof(struct file_lock));
249 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250}
251
252EXPORT_SYMBOL(locks_init_lock);
253
Trond Myklebust47831f32006-03-20 13:44:05 -0500254static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
255{
256 if (fl->fl_ops) {
257 if (fl->fl_ops->fl_copy_lock)
258 fl->fl_ops->fl_copy_lock(new, fl);
259 new->fl_ops = fl->fl_ops;
260 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400261 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500262 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500263}
264
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265/*
266 * Initialize a new lock from an existing file_lock structure.
267 */
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400268void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269{
270 new->fl_owner = fl->fl_owner;
271 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500272 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 new->fl_flags = fl->fl_flags;
274 new->fl_type = fl->fl_type;
275 new->fl_start = fl->fl_start;
276 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500277 new->fl_ops = NULL;
278 new->fl_lmops = NULL;
279}
Roland Dreier3dd7b712008-04-25 15:32:51 -0700280EXPORT_SYMBOL(__locks_copy_lock);
Trond Myklebust09969052006-03-20 13:44:38 -0500281
282void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
283{
284 locks_release_private(new);
285
286 __locks_copy_lock(new, fl);
287 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 new->fl_ops = fl->fl_ops;
289 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500290
291 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292}
293
294EXPORT_SYMBOL(locks_copy_lock);
295
296static inline int flock_translate_cmd(int cmd) {
297 if (cmd & LOCK_MAND)
298 return cmd & (LOCK_MAND | LOCK_RW);
299 switch (cmd) {
300 case LOCK_SH:
301 return F_RDLCK;
302 case LOCK_EX:
303 return F_WRLCK;
304 case LOCK_UN:
305 return F_UNLCK;
306 }
307 return -EINVAL;
308}
309
310/* Fill in a file_lock structure with an appropriate FLOCK lock. */
311static int flock_make_lock(struct file *filp, struct file_lock **lock,
312 unsigned int cmd)
313{
314 struct file_lock *fl;
315 int type = flock_translate_cmd(cmd);
316 if (type < 0)
317 return type;
318
319 fl = locks_alloc_lock();
320 if (fl == NULL)
321 return -ENOMEM;
322
323 fl->fl_file = filp;
324 fl->fl_pid = current->tgid;
325 fl->fl_flags = FL_FLOCK;
326 fl->fl_type = type;
327 fl->fl_end = OFFSET_MAX;
328
329 *lock = fl;
330 return 0;
331}
332
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400333static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334{
335 switch (type) {
336 case F_RDLCK:
337 case F_WRLCK:
338 case F_UNLCK:
339 fl->fl_type = type;
340 break;
341 default:
342 return -EINVAL;
343 }
344 return 0;
345}
346
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500347static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
348 struct flock64 *l)
349{
350 switch (l->l_whence) {
351 case SEEK_SET:
352 fl->fl_start = 0;
353 break;
354 case SEEK_CUR:
355 fl->fl_start = filp->f_pos;
356 break;
357 case SEEK_END:
358 fl->fl_start = i_size_read(file_inode(filp));
359 break;
360 default:
361 return -EINVAL;
362 }
363 if (l->l_start > OFFSET_MAX - fl->fl_start)
364 return -EOVERFLOW;
365 fl->fl_start += l->l_start;
366 if (fl->fl_start < 0)
367 return -EINVAL;
368
369 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
370 POSIX-2001 defines it. */
371 if (l->l_len > 0) {
372 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
373 return -EOVERFLOW;
374 fl->fl_end = fl->fl_start + l->l_len - 1;
375
376 } else if (l->l_len < 0) {
377 if (fl->fl_start + l->l_len < 0)
378 return -EINVAL;
379 fl->fl_end = fl->fl_start - 1;
380 fl->fl_start += l->l_len;
381 } else
382 fl->fl_end = OFFSET_MAX;
383
384 fl->fl_owner = current->files;
385 fl->fl_pid = current->tgid;
386 fl->fl_file = filp;
387 fl->fl_flags = FL_POSIX;
388 fl->fl_ops = NULL;
389 fl->fl_lmops = NULL;
390
Jeff Laytonbce75602014-02-03 12:13:08 -0500391 /* Ensure that fl->fl_filp has compatible f_mode */
392 switch (l->l_type) {
393 case F_RDLCK:
394 if (!(filp->f_mode & FMODE_READ))
395 return -EBADF;
396 break;
397 case F_WRLCK:
398 if (!(filp->f_mode & FMODE_WRITE))
399 return -EBADF;
400 break;
401 }
402
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500403 return assign_type(fl, l->l_type);
404}
405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
407 * style lock.
408 */
409static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
410 struct flock *l)
411{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500412 struct flock64 ll = {
413 .l_type = l->l_type,
414 .l_whence = l->l_whence,
415 .l_start = l->l_start,
416 .l_len = l->l_len,
417 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500419 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422/* default lease lock manager operations */
423static void lease_break_callback(struct file_lock *fl)
424{
425 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
426}
427
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700428static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400429 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400430 .lm_change = lease_modify,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431};
432
433/*
434 * Initialize a lease, use the default lock manager operations
435 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400436static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400438 if (assign_type(fl, type) != 0)
439 return -EINVAL;
440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 fl->fl_owner = current->files;
442 fl->fl_pid = current->tgid;
443
444 fl->fl_file = filp;
445 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 fl->fl_start = 0;
447 fl->fl_end = OFFSET_MAX;
448 fl->fl_ops = NULL;
449 fl->fl_lmops = &lease_manager_ops;
450 return 0;
451}
452
453/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400454static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455{
456 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400457 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458
459 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500460 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
462 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400463 if (error) {
464 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500465 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400466 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500467 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468}
469
470/* Check if two locks overlap each other.
471 */
472static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
473{
474 return ((fl1->fl_end >= fl2->fl_start) &&
475 (fl2->fl_end >= fl1->fl_start));
476}
477
478/*
479 * Check whether two locks have the same owner.
480 */
Matt Mackall33443c42006-01-08 01:05:22 -0800481static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400483 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400485 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 return fl1->fl_owner == fl2->fl_owner;
487}
488
Jeff Layton7012b022013-06-21 08:58:22 -0400489/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500490static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400491{
Jeff Layton7012b022013-06-21 08:58:22 -0400492 lg_local_lock(&file_lock_lglock);
493 fl->fl_link_cpu = smp_processor_id();
494 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
495 lg_local_unlock(&file_lock_lglock);
Jeff Layton88974692013-06-21 08:58:14 -0400496}
497
Jeff Layton7012b022013-06-21 08:58:22 -0400498/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500499static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400500{
Jeff Layton7012b022013-06-21 08:58:22 -0400501 /*
502 * Avoid taking lock if already unhashed. This is safe since this check
503 * is done while holding the i_lock, and new insertions into the list
504 * also require that it be held.
505 */
506 if (hlist_unhashed(&fl->fl_link))
507 return;
508 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton139ca042013-06-21 08:58:17 -0400509 hlist_del_init(&fl->fl_link);
Jeff Layton7012b022013-06-21 08:58:22 -0400510 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton88974692013-06-21 08:58:14 -0400511}
512
Jeff Layton3999e492013-06-21 08:58:19 -0400513static unsigned long
514posix_owner_key(struct file_lock *fl)
515{
516 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
517 return fl->fl_lmops->lm_owner_key(fl);
518 return (unsigned long)fl->fl_owner;
519}
520
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500521static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400522{
Jeff Layton3999e492013-06-21 08:58:19 -0400523 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400524}
525
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500526static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400527{
Jeff Layton48f74182013-06-21 08:58:18 -0400528 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400529}
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531/* Remove waiter from blocker's block list.
532 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400533 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400534 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 */
Matt Mackall33443c42006-01-08 01:05:22 -0800536static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537{
Jeff Layton88974692013-06-21 08:58:14 -0400538 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 waiter->fl_next = NULL;
541}
542
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400543static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400545 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400547 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548}
549
550/* Insert waiter into blocker's block list.
551 * We use a circular list so that processes can be easily woken up in
552 * the order they blocked. The documentation doesn't require this but
553 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400554 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400555 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
Jeff Layton46dad762014-02-03 12:13:06 -0500556 * list itself is protected by the blocked_lock_lock, but by ensuring that the
Jeff Layton7b2296a2013-06-21 08:58:20 -0400557 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
Jeff Layton4e8c7652013-06-21 08:58:16 -0400558 * in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400560static void __locks_insert_block(struct file_lock *blocker,
561 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800563 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400565 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 if (IS_POSIX(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400567 locks_insert_global_blocked(waiter);
568}
569
570/* Must be called with i_lock held. */
571static void locks_insert_block(struct file_lock *blocker,
572 struct file_lock *waiter)
573{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400574 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400575 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400576 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577}
578
Jeff Layton1cb36012013-06-21 08:58:12 -0400579/*
580 * Wake up processes blocked waiting for blocker.
581 *
Jeff Layton1c8c6012013-06-21 08:58:15 -0400582 * Must be called with the inode->i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 */
584static void locks_wake_up_blocks(struct file_lock *blocker)
585{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400586 /*
587 * Avoid taking global lock if list is empty. This is safe since new
588 * blocked requests are only added to the list under the i_lock, and
589 * the i_lock is always held here. Note that removal from the fl_block
590 * list does not require the i_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400591 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400592 */
593 if (list_empty(&blocker->fl_block))
594 return;
595
Jeff Layton7b2296a2013-06-21 08:58:20 -0400596 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400598 struct file_lock *waiter;
599
600 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 struct file_lock, fl_block);
602 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400603 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
604 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 else
606 wake_up(&waiter->fl_wait);
607 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400608 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
611/* Insert file lock fl into an inode's lock list at the position indicated
612 * by pos. At the same time add the lock to the global file lock list.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400613 *
614 * Must be called with the i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 */
616static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
617{
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000618 fl->fl_nspid = get_pid(task_tgid(current));
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 /* insert into file's list */
621 fl->fl_next = *pos;
622 *pos = fl;
Jeff Layton88974692013-06-21 08:58:14 -0400623
624 locks_insert_global_locks(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625}
626
Jeff Layton24cbe782014-02-03 12:13:06 -0500627/**
628 * locks_delete_lock - Delete a lock and then free it.
629 * @thisfl_p: pointer that points to the fl_next field of the previous
630 * inode->i_flock list entry
631 *
632 * Unlink a lock from all lists and free the namespace reference, but don't
633 * free it yet. Wake up processes that are blocked waiting for this lock and
634 * notify the FS that the lock has been cleared.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400635 *
636 * Must be called with the i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 */
Jeff Layton24cbe782014-02-03 12:13:06 -0500638static void locks_unlink_lock(struct file_lock **thisfl_p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639{
640 struct file_lock *fl = *thisfl_p;
641
Jeff Layton88974692013-06-21 08:58:14 -0400642 locks_delete_global_locks(fl);
643
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 *thisfl_p = fl->fl_next;
645 fl->fl_next = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000647 if (fl->fl_nspid) {
648 put_pid(fl->fl_nspid);
649 fl->fl_nspid = NULL;
650 }
651
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500653}
654
655/*
656 * Unlink a lock from all lists and free it.
657 *
658 * Must be called with i_lock held!
659 */
660static void locks_delete_lock(struct file_lock **thisfl_p)
661{
662 struct file_lock *fl = *thisfl_p;
663
664 locks_unlink_lock(thisfl_p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 locks_free_lock(fl);
666}
667
668/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
669 * checks for shared/exclusive status of overlapping locks.
670 */
671static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
672{
673 if (sys_fl->fl_type == F_WRLCK)
674 return 1;
675 if (caller_fl->fl_type == F_WRLCK)
676 return 1;
677 return 0;
678}
679
680/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
681 * checking before calling the locks_conflict().
682 */
683static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
684{
685 /* POSIX locks owned by the same process do not conflict with
686 * each other.
687 */
688 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
689 return (0);
690
691 /* Check whether they overlap */
692 if (!locks_overlap(caller_fl, sys_fl))
693 return 0;
694
695 return (locks_conflict(caller_fl, sys_fl));
696}
697
698/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
699 * checking before calling the locks_conflict().
700 */
701static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
702{
703 /* FLOCK locks referring to the same filp do not conflict with
704 * each other.
705 */
706 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
707 return (0);
708 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
709 return 0;
710
711 return (locks_conflict(caller_fl, sys_fl));
712}
713
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400714void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500715posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
717 struct file_lock *cfl;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400718 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
Jeff Layton1c8c6012013-06-21 08:58:15 -0400720 spin_lock(&inode->i_lock);
Al Viro496ad9a2013-01-23 17:07:38 -0500721 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (!IS_POSIX(cfl))
723 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400724 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 break;
726 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000727 if (cfl) {
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500728 __locks_copy_lock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000729 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800730 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000731 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400732 fl->fl_type = F_UNLCK;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400733 spin_unlock(&inode->i_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400734 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736EXPORT_SYMBOL(posix_test_lock);
737
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400738/*
739 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400741 * We attempt to detect deadlocks that are due purely to posix file
742 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400744 * We assume that a task can be waiting for at most one lock at a time.
745 * So for any acquired lock, the process holding that lock may be
746 * waiting on at most one other lock. That lock in turns may be held by
747 * someone waiting for at most one other lock. Given a requested lock
748 * caller_fl which is about to wait for a conflicting lock block_fl, we
749 * follow this chain of waiters to ensure we are not about to create a
750 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400751 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400752 * Since we do this before we ever put a process to sleep on a lock, we
753 * are ensured that there is never a cycle; that is what guarantees that
754 * the while() loop in posix_locks_deadlock() eventually completes.
755 *
756 * Note: the above assumption may not be true when handling lock
757 * requests from a broken NFS client. It may also fail in the presence
758 * of tasks (such as posix threads) sharing the same open file table.
759 *
760 * To handle those cases, we just bail out after a few iterations.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400762
763#define MAX_DEADLK_ITERATIONS 10
764
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400765/* Find a lock that the owner of the given block_fl is blocking on. */
766static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
767{
768 struct file_lock *fl;
769
Jeff Layton3999e492013-06-21 08:58:19 -0400770 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400771 if (posix_same_owner(fl, block_fl))
772 return fl->fl_next;
773 }
774 return NULL;
775}
776
Jeff Layton7b2296a2013-06-21 08:58:20 -0400777/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700778static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 struct file_lock *block_fl)
780{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400781 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400783 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
784 if (i++ > MAX_DEADLK_ITERATIONS)
785 return 0;
786 if (posix_same_owner(caller_fl, block_fl))
787 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 }
789 return 0;
790}
791
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400793 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400794 *
795 * Note that if called with an FL_EXISTS argument, the caller may determine
796 * whether or not a lock was successfully freed by testing the return
797 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800799static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800801 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 struct file_lock **before;
Al Viro496ad9a2013-01-23 17:07:38 -0500803 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 int error = 0;
805 int found = 0;
806
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200807 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
808 new_fl = locks_alloc_lock();
809 if (!new_fl)
810 return -ENOMEM;
811 }
812
Jeff Layton1c8c6012013-06-21 08:58:15 -0400813 spin_lock(&inode->i_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400814 if (request->fl_flags & FL_ACCESS)
815 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 for_each_lock(inode, before) {
818 struct file_lock *fl = *before;
819 if (IS_POSIX(fl))
820 break;
821 if (IS_LEASE(fl))
822 continue;
823 if (filp != fl->fl_file)
824 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800825 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 goto out;
827 found = 1;
828 locks_delete_lock(before);
829 break;
830 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831
Trond Myklebustf475ae92006-06-29 16:38:32 -0400832 if (request->fl_type == F_UNLCK) {
833 if ((request->fl_flags & FL_EXISTS) && !found)
834 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800835 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400836 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837
838 /*
839 * If a higher-priority process was blocked on the old file lock,
840 * give it the opportunity to lock the file.
841 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200842 if (found) {
Jeff Layton1c8c6012013-06-21 08:58:15 -0400843 spin_unlock(&inode->i_lock);
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200844 cond_resched();
Jeff Layton1c8c6012013-06-21 08:58:15 -0400845 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400848find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 for_each_lock(inode, before) {
850 struct file_lock *fl = *before;
851 if (IS_POSIX(fl))
852 break;
853 if (IS_LEASE(fl))
854 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800855 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 continue;
857 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700858 if (!(request->fl_flags & FL_SLEEP))
859 goto out;
860 error = FILE_LOCK_DEFERRED;
861 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 goto out;
863 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400864 if (request->fl_flags & FL_ACCESS)
865 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800866 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700867 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800868 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400869 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871out:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400872 spin_unlock(&inode->i_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800873 if (new_fl)
874 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 return error;
876}
877
Marc Eshel150b3932007-01-18 16:15:35 -0500878static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879{
880 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700881 struct file_lock *new_fl = NULL;
882 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 struct file_lock *left = NULL;
884 struct file_lock *right = NULL;
885 struct file_lock **before;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400886 int error;
887 bool added = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
889 /*
890 * We may need two file_lock structures for this operation,
891 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700892 *
893 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700895 if (!(request->fl_flags & FL_ACCESS) &&
896 (request->fl_type != F_UNLCK ||
897 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
898 new_fl = locks_alloc_lock();
899 new_fl2 = locks_alloc_lock();
900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Jeff Layton1c8c6012013-06-21 08:58:15 -0400902 spin_lock(&inode->i_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400903 /*
904 * New lock request. Walk all POSIX locks and look for conflicts. If
905 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400906 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400907 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 if (request->fl_type != F_UNLCK) {
909 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500910 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (!IS_POSIX(fl))
912 continue;
913 if (!posix_locks_conflict(request, fl))
914 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800915 if (conflock)
J. Bruce Fields1a747ee2008-04-24 10:08:22 -0400916 __locks_copy_lock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 error = -EAGAIN;
918 if (!(request->fl_flags & FL_SLEEP))
919 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400920 /*
921 * Deadlock detection and insertion into the blocked
922 * locks list must be done while holding the same lock!
923 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400925 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400926 if (likely(!posix_locks_deadlock(request, fl))) {
927 error = FILE_LOCK_DEFERRED;
928 __locks_insert_block(fl, request);
929 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400930 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 goto out;
932 }
933 }
934
935 /* If we're just looking for a conflict, we're done. */
936 error = 0;
937 if (request->fl_flags & FL_ACCESS)
938 goto out;
939
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 * Find the first old lock with the same owner as the new lock.
942 */
943
944 before = &inode->i_flock;
945
946 /* First skip locks owned by other processes. */
947 while ((fl = *before) && (!IS_POSIX(fl) ||
948 !posix_same_owner(request, fl))) {
949 before = &fl->fl_next;
950 }
951
Jeff Layton1cb36012013-06-21 08:58:12 -0400952 /* Process locks with this owner. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 while ((fl = *before) && posix_same_owner(request, fl)) {
954 /* Detect adjacent or overlapping regions (if same lock type)
955 */
956 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700957 /* In all comparisons of start vs end, use
958 * "start - 1" rather than "end + 1". If end
959 * is OFFSET_MAX, end + 1 will become negative.
960 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 if (fl->fl_end < request->fl_start - 1)
962 goto next_lock;
963 /* If the next lock in the list has entirely bigger
964 * addresses than the new one, insert the lock here.
965 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700966 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 break;
968
969 /* If we come here, the new and old lock are of the
970 * same type and adjacent or overlapping. Make one
971 * lock yielding from the lower start address of both
972 * locks to the higher end address.
973 */
974 if (fl->fl_start > request->fl_start)
975 fl->fl_start = request->fl_start;
976 else
977 request->fl_start = fl->fl_start;
978 if (fl->fl_end < request->fl_end)
979 fl->fl_end = request->fl_end;
980 else
981 request->fl_end = fl->fl_end;
982 if (added) {
983 locks_delete_lock(before);
984 continue;
985 }
986 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400987 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 }
989 else {
990 /* Processing for different lock types is a bit
991 * more complex.
992 */
993 if (fl->fl_end < request->fl_start)
994 goto next_lock;
995 if (fl->fl_start > request->fl_end)
996 break;
997 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400998 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 if (fl->fl_start < request->fl_start)
1000 left = fl;
1001 /* If the next lock in the list has a higher end
1002 * address than the new one, insert the new one here.
1003 */
1004 if (fl->fl_end > request->fl_end) {
1005 right = fl;
1006 break;
1007 }
1008 if (fl->fl_start >= request->fl_start) {
1009 /* The new lock completely replaces an old
1010 * one (This may happen several times).
1011 */
1012 if (added) {
1013 locks_delete_lock(before);
1014 continue;
1015 }
1016 /* Replace the old lock with the new one.
1017 * Wake up anybody waiting for the old one,
1018 * as the change in lock type might satisfy
1019 * their needs.
1020 */
1021 locks_wake_up_blocks(fl);
1022 fl->fl_start = request->fl_start;
1023 fl->fl_end = request->fl_end;
1024 fl->fl_type = request->fl_type;
Trond Myklebust47831f32006-03-20 13:44:05 -05001025 locks_release_private(fl);
1026 locks_copy_private(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001028 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 }
1030 }
1031 /* Go on to next lock.
1032 */
1033 next_lock:
1034 before = &fl->fl_next;
1035 }
1036
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001037 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001038 * The above code only modifies existing locks in case of merging or
1039 * replacing. If new lock(s) need to be inserted all modifications are
1040 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001041 */
1042 error = -ENOLCK; /* "no luck" */
1043 if (right && left == right && !new_fl2)
1044 goto out;
1045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 error = 0;
1047 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001048 if (request->fl_type == F_UNLCK) {
1049 if (request->fl_flags & FL_EXISTS)
1050 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001052 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001053
1054 if (!new_fl) {
1055 error = -ENOLCK;
1056 goto out;
1057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 locks_copy_lock(new_fl, request);
1059 locks_insert_lock(before, new_fl);
1060 new_fl = NULL;
1061 }
1062 if (right) {
1063 if (left == right) {
1064 /* The new lock breaks the old one in two pieces,
1065 * so we have to use the second new lock.
1066 */
1067 left = new_fl2;
1068 new_fl2 = NULL;
1069 locks_copy_lock(left, right);
1070 locks_insert_lock(before, left);
1071 }
1072 right->fl_start = request->fl_end + 1;
1073 locks_wake_up_blocks(right);
1074 }
1075 if (left) {
1076 left->fl_end = request->fl_start - 1;
1077 locks_wake_up_blocks(left);
1078 }
1079 out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001080 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /*
1082 * Free any unused locks.
1083 */
1084 if (new_fl)
1085 locks_free_lock(new_fl);
1086 if (new_fl2)
1087 locks_free_lock(new_fl2);
1088 return error;
1089}
1090
1091/**
1092 * posix_lock_file - Apply a POSIX-style lock to a file
1093 * @filp: The file to apply the lock to
1094 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001095 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 *
1097 * Add a POSIX style lock to a file.
1098 * We merge adjacent & overlapping locks whenever possible.
1099 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001100 *
1101 * Note that if called with an FL_EXISTS argument, the caller may determine
1102 * whether or not a lock was successfully freed by testing the return
1103 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 */
Marc Eshel150b3932007-01-18 16:15:35 -05001105int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001106 struct file_lock *conflock)
1107{
Al Viro496ad9a2013-01-23 17:07:38 -05001108 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001109}
Marc Eshel150b3932007-01-18 16:15:35 -05001110EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112/**
1113 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1114 * @filp: The file to apply the lock to
1115 * @fl: The lock to be applied
1116 *
1117 * Add a POSIX style lock to a file.
1118 * We merge adjacent & overlapping locks whenever possible.
1119 * POSIX locks are sorted by owner task, then by starting address
1120 */
1121int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1122{
1123 int error;
1124 might_sleep ();
1125 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001126 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001127 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 break;
1129 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1130 if (!error)
1131 continue;
1132
1133 locks_delete_block(fl);
1134 break;
1135 }
1136 return error;
1137}
1138EXPORT_SYMBOL(posix_lock_file_wait);
1139
1140/**
1141 * locks_mandatory_locked - Check for an active lock
1142 * @inode: the file to check
1143 *
1144 * Searches the inode's list of locks to find any POSIX locks which conflict.
1145 * This function is called from locks_verify_locked() only.
1146 */
1147int locks_mandatory_locked(struct inode *inode)
1148{
1149 fl_owner_t owner = current->files;
1150 struct file_lock *fl;
1151
1152 /*
1153 * Search the lock list for this inode for any POSIX locks.
1154 */
Jeff Layton1c8c6012013-06-21 08:58:15 -04001155 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1157 if (!IS_POSIX(fl))
1158 continue;
1159 if (fl->fl_owner != owner)
1160 break;
1161 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001162 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 return fl ? -EAGAIN : 0;
1164}
1165
1166/**
1167 * locks_mandatory_area - Check for a conflicting lock
1168 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1169 * for shared
1170 * @inode: the file to check
1171 * @filp: how the file was opened (if it was)
1172 * @offset: start of area to check
1173 * @count: length of area to check
1174 *
1175 * Searches the inode's list of locks to find any POSIX locks which conflict.
1176 * This function is called from rw_verify_area() and
1177 * locks_verify_truncate().
1178 */
1179int locks_mandatory_area(int read_write, struct inode *inode,
1180 struct file *filp, loff_t offset,
1181 size_t count)
1182{
1183 struct file_lock fl;
1184 int error;
1185
1186 locks_init_lock(&fl);
1187 fl.fl_owner = current->files;
1188 fl.fl_pid = current->tgid;
1189 fl.fl_file = filp;
1190 fl.fl_flags = FL_POSIX | FL_ACCESS;
1191 if (filp && !(filp->f_flags & O_NONBLOCK))
1192 fl.fl_flags |= FL_SLEEP;
1193 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1194 fl.fl_start = offset;
1195 fl.fl_end = offset + count - 1;
1196
1197 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001198 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001199 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 break;
1201 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1202 if (!error) {
1203 /*
1204 * If we've been sleeping someone might have
1205 * changed the permissions behind our back.
1206 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001207 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 continue;
1209 }
1210
1211 locks_delete_block(&fl);
1212 break;
1213 }
1214
1215 return error;
1216}
1217
1218EXPORT_SYMBOL(locks_mandatory_area);
1219
J. Bruce Fields778fc542011-07-26 18:25:49 -04001220static void lease_clear_pending(struct file_lock *fl, int arg)
1221{
1222 switch (arg) {
1223 case F_UNLCK:
1224 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1225 /* fall through: */
1226 case F_RDLCK:
1227 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1228 }
1229}
1230
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231/* We already had a lease on this file; just change its type */
1232int lease_modify(struct file_lock **before, int arg)
1233{
1234 struct file_lock *fl = *before;
1235 int error = assign_type(fl, arg);
1236
1237 if (error)
1238 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001239 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001241 if (arg == F_UNLCK) {
1242 struct file *filp = fl->fl_file;
1243
1244 f_delown(filp);
1245 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001246 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1247 if (fl->fl_fasync != NULL) {
1248 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1249 fl->fl_fasync = NULL;
1250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 locks_delete_lock(before);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 return 0;
1254}
1255
1256EXPORT_SYMBOL(lease_modify);
1257
J. Bruce Fields778fc542011-07-26 18:25:49 -04001258static bool past_time(unsigned long then)
1259{
1260 if (!then)
1261 /* 0 is a special value meaning "this never expires": */
1262 return false;
1263 return time_after(jiffies, then);
1264}
1265
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266static void time_out_leases(struct inode *inode)
1267{
1268 struct file_lock **before;
1269 struct file_lock *fl;
1270
1271 before = &inode->i_flock;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04001272 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001273 if (past_time(fl->fl_downgrade_time))
1274 lease_modify(before, F_RDLCK);
1275 if (past_time(fl->fl_break_time))
1276 lease_modify(before, F_UNLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 if (fl == *before) /* lease_modify may have freed fl */
1278 before = &fl->fl_next;
1279 }
1280}
1281
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001282static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1283{
1284 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1285 return false;
1286 return locks_conflict(breaker, lease);
1287}
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289/**
1290 * __break_lease - revoke all outstanding leases on file
1291 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001292 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1293 * break all leases
1294 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1295 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001296 *
david m. richter87250dd2007-05-09 16:10:27 -04001297 * break_lease (inlined for speed) has checked there already is at least
1298 * some kind of lock (maybe a lease) on this file. Leases are broken on
1299 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 * specified %O_NONBLOCK to your open().
1301 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001302int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001304 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 struct file_lock *new_fl, *flock;
1306 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 unsigned long break_time;
1308 int i_have_this_lease = 0;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001309 bool lease_conflict = false;
Al Viro8737c932009-12-24 06:47:55 -05001310 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
Al Viro8737c932009-12-24 06:47:55 -05001312 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001313 if (IS_ERR(new_fl))
1314 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001315 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Jeff Layton1c8c6012013-06-21 08:58:15 -04001317 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
1319 time_out_leases(inode);
1320
1321 flock = inode->i_flock;
1322 if ((flock == NULL) || !IS_LEASE(flock))
1323 goto out;
1324
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001325 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1326 if (leases_conflict(fl, new_fl)) {
1327 lease_conflict = true;
1328 if (fl->fl_owner == current->files)
1329 i_have_this_lease = 1;
1330 }
1331 }
1332 if (!lease_conflict)
J. Bruce Fields778fc542011-07-26 18:25:49 -04001333 goto out;
1334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 break_time = 0;
1336 if (lease_break_time > 0) {
1337 break_time = jiffies + lease_break_time * HZ;
1338 if (break_time == 0)
1339 break_time++; /* so that 0 means no break time */
1340 }
1341
1342 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001343 if (!leases_conflict(fl, new_fl))
1344 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001345 if (want_write) {
1346 if (fl->fl_flags & FL_UNLOCK_PENDING)
1347 continue;
1348 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001350 } else {
1351 if (lease_breaking(flock))
1352 continue;
1353 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1354 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 }
J. Bruce Fields778fc542011-07-26 18:25:49 -04001356 fl->fl_lmops->lm_break(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358
1359 if (i_have_this_lease || (mode & O_NONBLOCK)) {
1360 error = -EWOULDBLOCK;
1361 goto out;
1362 }
1363
1364restart:
1365 break_time = flock->fl_break_time;
1366 if (break_time != 0) {
1367 break_time -= jiffies;
1368 if (break_time == 0)
1369 break_time++;
1370 }
Matthew Wilcox4321e012008-01-14 21:28:30 -07001371 locks_insert_block(flock, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001372 spin_unlock(&inode->i_lock);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001373 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1374 !new_fl->fl_next, break_time);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001375 spin_lock(&inode->i_lock);
1376 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 if (error >= 0) {
1378 if (error == 0)
1379 time_out_leases(inode);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001380 /*
1381 * Wait for the next conflicting lease that has not been
1382 * broken yet
1383 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1385 flock = flock->fl_next) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001386 if (leases_conflict(new_fl, flock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 goto restart;
1388 }
1389 error = 0;
1390 }
1391
1392out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001393 spin_unlock(&inode->i_lock);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001394 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 return error;
1396}
1397
1398EXPORT_SYMBOL(__break_lease);
1399
1400/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001401 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 * @inode: the inode
1403 * @time: pointer to a timespec which will contain the last modified time
1404 *
1405 * This is to force NFS clients to flush their caches for files with
1406 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001407 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 */
1409void lease_get_mtime(struct inode *inode, struct timespec *time)
1410{
1411 struct file_lock *flock = inode->i_flock;
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04001412 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 *time = current_fs_time(inode->i_sb);
1414 else
1415 *time = inode->i_mtime;
1416}
1417
1418EXPORT_SYMBOL(lease_get_mtime);
1419
1420/**
1421 * fcntl_getlease - Enquire what lease is currently active
1422 * @filp: the file
1423 *
1424 * The value returned by this function will be one of
1425 * (if no lease break is pending):
1426 *
1427 * %F_RDLCK to indicate a shared lease is held.
1428 *
1429 * %F_WRLCK to indicate an exclusive lease is held.
1430 *
1431 * %F_UNLCK to indicate no lease is held.
1432 *
1433 * (if a lease break is pending):
1434 *
1435 * %F_RDLCK to indicate an exclusive lease needs to be
1436 * changed to a shared lease (or removed).
1437 *
1438 * %F_UNLCK to indicate the lease needs to be removed.
1439 *
1440 * XXX: sfr & willy disagree over whether F_INPROGRESS
1441 * should be returned to userspace.
1442 */
1443int fcntl_getlease(struct file *filp)
1444{
1445 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001446 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 int type = F_UNLCK;
1448
Jeff Layton1c8c6012013-06-21 08:58:15 -04001449 spin_lock(&inode->i_lock);
Al Viro496ad9a2013-01-23 17:07:38 -05001450 time_out_leases(file_inode(filp));
1451 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 fl = fl->fl_next) {
1453 if (fl->fl_file == filp) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001454 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 break;
1456 }
1457 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001458 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 return type;
1460}
1461
Jeff Layton24cbe782014-02-03 12:13:06 -05001462/**
1463 * check_conflicting_open - see if the given dentry points to a file that has
1464 * an existing open that would conflict with the
1465 * desired lease.
1466 * @dentry: dentry to check
1467 * @arg: type of lease that we're trying to acquire
1468 *
1469 * Check to see if there's an existing open fd on this file that would
1470 * conflict with the lease we're trying to set.
1471 */
1472static int
1473check_conflicting_open(const struct dentry *dentry, const long arg)
1474{
1475 int ret = 0;
1476 struct inode *inode = dentry->d_inode;
1477
1478 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1479 return -EAGAIN;
1480
1481 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1482 (atomic_read(&inode->i_count) > 1)))
1483 ret = -EAGAIN;
1484
1485 return ret;
1486}
1487
Jeff Laytond4f22d12013-06-21 08:58:11 -04001488static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001490 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001491 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 struct inode *inode = dentry->d_inode;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001493 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001494 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
J. Bruce Fields096657b2010-10-30 17:31:14 -04001496 lease = *flp;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001497 /*
1498 * In the delegation case we need mutual exclusion with
1499 * a number of operations that take the i_mutex. We trylock
1500 * because delegations are an optional optimization, and if
1501 * there's some chance of a conflict--we'd rather not
1502 * bother, maybe that's a sign this just isn't a good file to
1503 * hand out a delegation on.
1504 */
1505 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1506 return -EAGAIN;
1507
1508 if (is_deleg && arg == F_WRLCK) {
1509 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001510 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001511 WARN_ON_ONCE(1);
1512 return -EINVAL;
1513 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001514
Jeff Layton24cbe782014-02-03 12:13:06 -05001515 error = check_conflicting_open(dentry, arg);
1516 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001517 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 /*
1520 * At this point, we know that if there is an exclusive
1521 * lease on this file, then we hold it on this filp
1522 * (otherwise our open of this file would have blocked).
1523 * And if we are trying to acquire an exclusive lease,
1524 * then the file is not open by anyone (including us)
1525 * except for this filp.
1526 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001527 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 for (before = &inode->i_flock;
1529 ((fl = *before) != NULL) && IS_LEASE(fl);
1530 before = &fl->fl_next) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001531 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 my_before = before;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001533 continue;
1534 }
1535 /*
1536 * No exclusive leases if someone else has a lease on
1537 * this file:
1538 */
1539 if (arg == F_WRLCK)
1540 goto out;
1541 /*
1542 * Modifying our existing lease is OK, but no getting a
1543 * new lease if someone else is opening for write:
1544 */
1545 if (fl->fl_flags & FL_UNLOCK_PENDING)
1546 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 }
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 if (my_before != NULL) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001550 error = lease->fl_lmops->lm_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001551 if (!error)
1552 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 goto out;
1554 }
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 error = -EINVAL;
1557 if (!leases_enable)
1558 goto out;
1559
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001560 locks_insert_lock(before, lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001561 /*
1562 * The check in break_lease() is lockless. It's possible for another
1563 * open to race in after we did the earlier check for a conflicting
1564 * open but before the lease was inserted. Check again for a
1565 * conflicting open and cancel the lease if there is one.
1566 *
1567 * We also add a barrier here to ensure that the insertion of the lock
1568 * precedes these checks.
1569 */
1570 smp_mb();
1571 error = check_conflicting_open(dentry, arg);
1572 if (error)
1573 locks_unlink_lock(flp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574out:
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001575 if (is_deleg)
1576 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 return error;
1578}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001579
Jeff Laytond4f22d12013-06-21 08:58:11 -04001580static int generic_delete_lease(struct file *filp, struct file_lock **flp)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001581{
1582 struct file_lock *fl, **before;
1583 struct dentry *dentry = filp->f_path.dentry;
1584 struct inode *inode = dentry->d_inode;
1585
1586 for (before = &inode->i_flock;
1587 ((fl = *before) != NULL) && IS_LEASE(fl);
1588 before = &fl->fl_next) {
1589 if (fl->fl_file != filp)
1590 continue;
1591 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1592 }
1593 return -EAGAIN;
1594}
1595
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596/**
1597 * generic_setlease - sets a lease on an open file
1598 * @filp: file pointer
1599 * @arg: type of lease to obtain
1600 * @flp: input - file_lock to use, output - file_lock inserted
1601 *
1602 * The (input) flp->fl_lmops->lm_break function is required
1603 * by break_lease().
1604 *
Jeff Layton1c8c6012013-06-21 08:58:15 -04001605 * Called with inode->i_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 */
1607int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1608{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 struct dentry *dentry = filp->f_path.dentry;
1610 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001611 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001613 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001614 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001616 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 error = security_file_lock(filp, arg);
1618 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001619 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
1621 time_out_leases(inode);
1622
1623 BUG_ON(!(*flp)->fl_lmops->lm_break);
1624
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001625 switch (arg) {
1626 case F_UNLCK:
1627 return generic_delete_lease(filp, flp);
1628 case F_RDLCK:
1629 case F_WRLCK:
1630 return generic_add_lease(filp, arg, flp);
1631 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001632 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001635EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001637static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1638{
Al Viro72c2d532013-09-22 16:27:52 -04001639 if (filp->f_op->setlease)
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001640 return filp->f_op->setlease(filp, arg, lease);
1641 else
1642 return generic_setlease(filp, arg, lease);
1643}
1644
1645/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001646 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 * @filp: file pointer
1648 * @arg: type of lease to obtain
1649 * @lease: file_lock to use
1650 *
1651 * Call this to establish a lease on the file.
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001652 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001653 * break_lease will oops!
1654 *
1655 * This will call the filesystem's setlease file method, if
1656 * defined. Note that there is no getlease method; instead, the
1657 * filesystem setlease method should call back to setlease() to
1658 * add a lease to the inode's lease list, where fcntl_getlease() can
1659 * find it. Since fcntl_getlease() only reports whether the current
1660 * task holds a lease, a cluster filesystem need only do this for
1661 * leases held by processes on this node.
1662 *
1663 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001664 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001665 * filesystem's open, create, and (on truncate) setattr methods.
1666 *
1667 * Warning: the only current setlease methods exist only to disable
1668 * leases in certain cases. More vfs changes may be required to
1669 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 */
1671
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001672int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673{
Jeff Layton1c8c6012013-06-21 08:58:15 -04001674 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 int error;
1676
Jeff Layton1c8c6012013-06-21 08:58:15 -04001677 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001678 error = __vfs_setlease(filp, arg, lease);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001679 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680
1681 return error;
1682}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001683EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001685static int do_fcntl_delete_lease(struct file *filp)
1686{
1687 struct file_lock fl, *flp = &fl;
1688
1689 lease_init(filp, F_UNLCK, flp);
1690
1691 return vfs_setlease(filp, F_UNLCK, &flp);
1692}
1693
1694static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001696 struct file_lock *fl, *ret;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001697 struct inode *inode = file_inode(filp);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001698 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 int error;
1700
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001701 fl = lease_alloc(filp, arg);
1702 if (IS_ERR(fl))
1703 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001705 new = fasync_alloc();
1706 if (!new) {
1707 locks_free_lock(fl);
1708 return -ENOMEM;
1709 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001710 ret = fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001711 spin_lock(&inode->i_lock);
J. Bruce Fields8896b932010-11-03 18:09:18 -04001712 error = __vfs_setlease(filp, arg, &ret);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001713 if (error) {
Jeff Layton1c8c6012013-06-21 08:58:15 -04001714 spin_unlock(&inode->i_lock);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001715 locks_free_lock(fl);
1716 goto out_free_fasync;
1717 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001718 if (ret != fl)
1719 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001721 /*
1722 * fasync_insert_entry() returns the old entry if any.
1723 * If there was no old entry, then it used 'new' and
1724 * inserted it into the fasync list. Clear new so that
1725 * we don't release it here.
1726 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001727 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001728 new = NULL;
1729
J. Bruce Fields8896b932010-11-03 18:09:18 -04001730 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001731 spin_unlock(&inode->i_lock);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001732
1733out_free_fasync:
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001734 if (new)
1735 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 return error;
1737}
1738
1739/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001740 * fcntl_setlease - sets a lease on an open file
1741 * @fd: open file descriptor
1742 * @filp: file pointer
1743 * @arg: type of lease to obtain
1744 *
1745 * Call this fcntl to establish a lease on the file.
1746 * Note that you also need to call %F_SETSIG to
1747 * receive a signal when the lease is broken.
1748 */
1749int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1750{
1751 if (arg == F_UNLCK)
1752 return do_fcntl_delete_lease(filp);
1753 return do_fcntl_add_lease(fd, filp, arg);
1754}
1755
1756/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1758 * @filp: The file to apply the lock to
1759 * @fl: The lock to be applied
1760 *
1761 * Add a FLOCK style lock to a file.
1762 */
1763int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1764{
1765 int error;
1766 might_sleep();
1767 for (;;) {
1768 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001769 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 break;
1771 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1772 if (!error)
1773 continue;
1774
1775 locks_delete_block(fl);
1776 break;
1777 }
1778 return error;
1779}
1780
1781EXPORT_SYMBOL(flock_lock_file_wait);
1782
1783/**
1784 * sys_flock: - flock() system call.
1785 * @fd: the file descriptor to lock.
1786 * @cmd: the type of lock to apply.
1787 *
1788 * Apply a %FL_FLOCK style lock to an open file descriptor.
1789 * The @cmd can be one of
1790 *
1791 * %LOCK_SH -- a shared lock.
1792 *
1793 * %LOCK_EX -- an exclusive lock.
1794 *
1795 * %LOCK_UN -- remove an existing lock.
1796 *
1797 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1798 *
1799 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1800 * processes read and write access respectively.
1801 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001802SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803{
Al Viro2903ff02012-08-28 12:52:22 -04001804 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 struct file_lock *lock;
1806 int can_sleep, unlock;
1807 int error;
1808
1809 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001810 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 goto out;
1812
1813 can_sleep = !(cmd & LOCK_NB);
1814 cmd &= ~LOCK_NB;
1815 unlock = (cmd == LOCK_UN);
1816
Al Viroaeb5d722008-09-02 15:28:45 -04001817 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001818 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 goto out_putf;
1820
Al Viro2903ff02012-08-28 12:52:22 -04001821 error = flock_make_lock(f.file, &lock, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 if (error)
1823 goto out_putf;
1824 if (can_sleep)
1825 lock->fl_flags |= FL_SLEEP;
1826
Al Viro2903ff02012-08-28 12:52:22 -04001827 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 if (error)
1829 goto out_free;
1830
Al Viro72c2d532013-09-22 16:27:52 -04001831 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001832 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 (can_sleep) ? F_SETLKW : F_SETLK,
1834 lock);
1835 else
Al Viro2903ff02012-08-28 12:52:22 -04001836 error = flock_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837
1838 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001839 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
1841 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001842 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 out:
1844 return error;
1845}
1846
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001847/**
1848 * vfs_test_lock - test file byte range lock
1849 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001850 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001851 *
1852 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1853 * setting conf->fl_type to something other than F_UNLCK.
1854 */
1855int vfs_test_lock(struct file *filp, struct file_lock *fl)
1856{
Al Viro72c2d532013-09-22 16:27:52 -04001857 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001858 return filp->f_op->lock(filp, F_GETLK, fl);
1859 posix_test_lock(filp, fl);
1860 return 0;
1861}
1862EXPORT_SYMBOL_GPL(vfs_test_lock);
1863
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001864static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1865{
1866 flock->l_pid = fl->fl_pid;
1867#if BITS_PER_LONG == 32
1868 /*
1869 * Make sure we can represent the posix lock via
1870 * legacy 32bit flock.
1871 */
1872 if (fl->fl_start > OFFT_OFFSET_MAX)
1873 return -EOVERFLOW;
1874 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1875 return -EOVERFLOW;
1876#endif
1877 flock->l_start = fl->fl_start;
1878 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1879 fl->fl_end - fl->fl_start + 1;
1880 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001881 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001882 return 0;
1883}
1884
1885#if BITS_PER_LONG == 32
1886static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1887{
1888 flock->l_pid = fl->fl_pid;
1889 flock->l_start = fl->fl_start;
1890 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1891 fl->fl_end - fl->fl_start + 1;
1892 flock->l_whence = 0;
1893 flock->l_type = fl->fl_type;
1894}
1895#endif
1896
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897/* Report the first existing lock that would conflict with l.
1898 * This implements the F_GETLK command of fcntl().
1899 */
1900int fcntl_getlk(struct file *filp, struct flock __user *l)
1901{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001902 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 struct flock flock;
1904 int error;
1905
1906 error = -EFAULT;
1907 if (copy_from_user(&flock, l, sizeof(flock)))
1908 goto out;
1909 error = -EINVAL;
1910 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1911 goto out;
1912
1913 error = flock_to_posix_lock(filp, &file_lock, &flock);
1914 if (error)
1915 goto out;
1916
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001917 error = vfs_test_lock(filp, &file_lock);
1918 if (error)
1919 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001921 flock.l_type = file_lock.fl_type;
1922 if (file_lock.fl_type != F_UNLCK) {
1923 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001924 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 }
1927 error = -EFAULT;
1928 if (!copy_to_user(l, &flock, sizeof(flock)))
1929 error = 0;
1930out:
1931 return error;
1932}
1933
Marc Eshel7723ec92007-01-18 15:08:55 -05001934/**
1935 * vfs_lock_file - file byte range lock
1936 * @filp: The file to apply the lock to
1937 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1938 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001939 * @conf: Place to return a copy of the conflicting lock, if found.
1940 *
1941 * A caller that doesn't care about the conflicting lock may pass NULL
1942 * as the final argument.
1943 *
1944 * If the filesystem defines a private ->lock() method, then @conf will
1945 * be left unchanged; so a caller that cares should initialize it to
1946 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05001947 *
1948 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
1949 * locks, the ->lock() interface may return asynchronously, before the lock has
1950 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001951 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05001952 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
1953 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001954 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05001955 * request completes.
1956 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07001957 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
1958 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05001959 * nonzero return code and the file system should release the lock. The file
1960 * system is also responsible to keep a corresponding posix lock when it
1961 * grants a lock so the VFS can find out which locks are locally held and do
1962 * the correct lock cleanup when required.
1963 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001964 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05001965 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05001966 */
Marc Eshel150b3932007-01-18 16:15:35 -05001967int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05001968{
Al Viro72c2d532013-09-22 16:27:52 -04001969 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05001970 return filp->f_op->lock(filp, cmd, fl);
1971 else
Marc Eshel150b3932007-01-18 16:15:35 -05001972 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05001973}
1974EXPORT_SYMBOL_GPL(vfs_lock_file);
1975
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001976static int do_lock_file_wait(struct file *filp, unsigned int cmd,
1977 struct file_lock *fl)
1978{
1979 int error;
1980
1981 error = security_file_lock(filp, fl->fl_type);
1982 if (error)
1983 return error;
1984
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001985 for (;;) {
1986 error = vfs_lock_file(filp, cmd, fl, NULL);
1987 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001988 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07001989 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1990 if (!error)
1991 continue;
1992
1993 locks_delete_block(fl);
1994 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07001995 }
1996
1997 return error;
1998}
1999
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000/* Apply the lock described by l to an open file descriptor.
2001 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2002 */
Peter Staubachc2936212005-07-27 11:45:09 -07002003int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2004 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005{
2006 struct file_lock *file_lock = locks_alloc_lock();
2007 struct flock flock;
2008 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002009 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 int error;
2011
2012 if (file_lock == NULL)
2013 return -ENOLCK;
2014
2015 /*
2016 * This might block, so we do it before checking the inode.
2017 */
2018 error = -EFAULT;
2019 if (copy_from_user(&flock, l, sizeof(flock)))
2020 goto out;
2021
Al Viro496ad9a2013-01-23 17:07:38 -05002022 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
2024 /* Don't allow mandatory locks on files that may be memory mapped
2025 * and shared.
2026 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002027 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028 error = -EAGAIN;
2029 goto out;
2030 }
2031
Peter Staubachc2936212005-07-27 11:45:09 -07002032again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 error = flock_to_posix_lock(filp, file_lock, &flock);
2034 if (error)
2035 goto out;
2036 if (cmd == F_SETLKW) {
2037 file_lock->fl_flags |= FL_SLEEP;
2038 }
2039
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002040 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Peter Staubachc2936212005-07-27 11:45:09 -07002042 /*
2043 * Attempt to detect a close/fcntl race and recover by
2044 * releasing the lock that was just acquired.
2045 */
Al Viro0b2bac22008-05-06 13:58:34 -04002046 /*
2047 * we need that spin_lock here - it prevents reordering between
2048 * update of inode->i_flock and check for it done in close().
2049 * rcu_read_lock() wouldn't do.
2050 */
2051 spin_lock(&current->files->file_lock);
2052 f = fcheck(fd);
2053 spin_unlock(&current->files->file_lock);
2054 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002055 flock.l_type = F_UNLCK;
2056 goto again;
2057 }
2058
2059out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060 locks_free_lock(file_lock);
2061 return error;
2062}
2063
2064#if BITS_PER_LONG == 32
2065/* Report the first existing lock that would conflict with l.
2066 * This implements the F_GETLK command of fcntl().
2067 */
2068int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
2069{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002070 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 struct flock64 flock;
2072 int error;
2073
2074 error = -EFAULT;
2075 if (copy_from_user(&flock, l, sizeof(flock)))
2076 goto out;
2077 error = -EINVAL;
2078 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2079 goto out;
2080
2081 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2082 if (error)
2083 goto out;
2084
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002085 error = vfs_test_lock(filp, &file_lock);
2086 if (error)
2087 goto out;
2088
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002089 flock.l_type = file_lock.fl_type;
2090 if (file_lock.fl_type != F_UNLCK)
2091 posix_lock_to_flock64(&flock, &file_lock);
2092
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 error = -EFAULT;
2094 if (!copy_to_user(l, &flock, sizeof(flock)))
2095 error = 0;
2096
2097out:
2098 return error;
2099}
2100
2101/* Apply the lock described by l to an open file descriptor.
2102 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2103 */
Peter Staubachc2936212005-07-27 11:45:09 -07002104int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2105 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106{
2107 struct file_lock *file_lock = locks_alloc_lock();
2108 struct flock64 flock;
2109 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002110 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 int error;
2112
2113 if (file_lock == NULL)
2114 return -ENOLCK;
2115
2116 /*
2117 * This might block, so we do it before checking the inode.
2118 */
2119 error = -EFAULT;
2120 if (copy_from_user(&flock, l, sizeof(flock)))
2121 goto out;
2122
Al Viro496ad9a2013-01-23 17:07:38 -05002123 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124
2125 /* Don't allow mandatory locks on files that may be memory mapped
2126 * and shared.
2127 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002128 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 error = -EAGAIN;
2130 goto out;
2131 }
2132
Peter Staubachc2936212005-07-27 11:45:09 -07002133again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134 error = flock64_to_posix_lock(filp, file_lock, &flock);
2135 if (error)
2136 goto out;
2137 if (cmd == F_SETLKW64) {
2138 file_lock->fl_flags |= FL_SLEEP;
2139 }
2140
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002141 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142
Peter Staubachc2936212005-07-27 11:45:09 -07002143 /*
2144 * Attempt to detect a close/fcntl race and recover by
2145 * releasing the lock that was just acquired.
2146 */
Al Viro0b2bac22008-05-06 13:58:34 -04002147 spin_lock(&current->files->file_lock);
2148 f = fcheck(fd);
2149 spin_unlock(&current->files->file_lock);
2150 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002151 flock.l_type = F_UNLCK;
2152 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 }
2154
2155out:
2156 locks_free_lock(file_lock);
2157 return error;
2158}
2159#endif /* BITS_PER_LONG == 32 */
2160
2161/*
2162 * This function is called when the file is being removed
2163 * from the task's fd array. POSIX locks belonging to this task
2164 * are deleted at this time.
2165 */
2166void locks_remove_posix(struct file *filp, fl_owner_t owner)
2167{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002168 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169
2170 /*
2171 * If there are no locks held on this file, we don't need to call
2172 * posix_lock_file(). Another process could be setting a lock on this
2173 * file at the same time, but we wouldn't remove that lock anyway.
2174 */
Al Viro496ad9a2013-01-23 17:07:38 -05002175 if (!file_inode(filp)->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 return;
2177
2178 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002179 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 lock.fl_start = 0;
2181 lock.fl_end = OFFSET_MAX;
2182 lock.fl_owner = owner;
2183 lock.fl_pid = current->tgid;
2184 lock.fl_file = filp;
2185 lock.fl_ops = NULL;
2186 lock.fl_lmops = NULL;
2187
Marc Eshel150b3932007-01-18 16:15:35 -05002188 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2191 lock.fl_ops->fl_release_private(&lock);
2192}
2193
2194EXPORT_SYMBOL(locks_remove_posix);
2195
2196/*
2197 * This function is called on the last close of an open file.
2198 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002199void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200{
Al Viro496ad9a2013-01-23 17:07:38 -05002201 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 struct file_lock *fl;
2203 struct file_lock **before;
2204
2205 if (!inode->i_flock)
2206 return;
2207
Al Viro72c2d532013-09-22 16:27:52 -04002208 if (filp->f_op->flock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002209 struct file_lock fl = {
2210 .fl_pid = current->tgid,
2211 .fl_file = filp,
2212 .fl_flags = FL_FLOCK,
2213 .fl_type = F_UNLCK,
2214 .fl_end = OFFSET_MAX,
2215 };
2216 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002217 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2218 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 }
2220
Jeff Layton1c8c6012013-06-21 08:58:15 -04002221 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222 before = &inode->i_flock;
2223
2224 while ((fl = *before) != NULL) {
2225 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 if (IS_LEASE(fl)) {
2227 lease_modify(before, F_UNLCK);
2228 continue;
2229 }
Jeff Layton8c3cac52014-02-03 12:13:07 -05002230
2231 /*
2232 * There's a leftover lock on the list of a type that
2233 * we didn't expect to see. Most likely a classic
2234 * POSIX lock that ended up not getting released
2235 * properly, or that raced onto the list somehow. Log
2236 * some info about it and then just remove it from
2237 * the list.
2238 */
2239 WARN(!IS_FLOCK(fl),
2240 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2241 MAJOR(inode->i_sb->s_dev),
2242 MINOR(inode->i_sb->s_dev), inode->i_ino,
2243 fl->fl_type, fl->fl_flags,
2244 fl->fl_start, fl->fl_end);
2245
2246 locks_delete_lock(before);
2247 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 }
2249 before = &fl->fl_next;
2250 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002251 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252}
2253
2254/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256 * @waiter: the lock which was waiting
2257 *
2258 * lockd needs to block waiting for locks.
2259 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002260int
Jeff Laytonf891a292013-06-21 08:58:09 -04002261posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002263 int status = 0;
2264
Jeff Layton7b2296a2013-06-21 08:58:20 -04002265 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002266 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002268 else
2269 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002270 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002271 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273EXPORT_SYMBOL(posix_unblock_lock);
2274
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002275/**
2276 * vfs_cancel_lock - file byte range unblock lock
2277 * @filp: The file to apply the unblock to
2278 * @fl: The lock to be unblocked
2279 *
2280 * Used by lock managers to cancel blocked requests
2281 */
2282int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2283{
Al Viro72c2d532013-09-22 16:27:52 -04002284 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002285 return filp->f_op->lock(filp, F_CANCELLK, fl);
2286 return 0;
2287}
2288
2289EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2290
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002291#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002292#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002293#include <linux/seq_file.h>
2294
Jeff Layton7012b022013-06-21 08:58:22 -04002295struct locks_iterator {
2296 int li_cpu;
2297 loff_t li_pos;
2298};
2299
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002300static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002301 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302{
2303 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002304 unsigned int fl_pid;
2305
2306 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002307 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002308 else
2309 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
2311 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002312 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
Jerome Marchand99dc8292010-10-26 14:22:33 -07002314 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 if (IS_POSIX(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002316 seq_printf(f, "%6s %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 (fl->fl_flags & FL_ACCESS) ? "ACCESS" : "POSIX ",
2318 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002319 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320 } else if (IS_FLOCK(fl)) {
2321 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002322 seq_printf(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002324 seq_printf(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 }
2326 } else if (IS_LEASE(fl)) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002327 seq_printf(f, "LEASE ");
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002328 if (lease_breaking(fl))
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002329 seq_printf(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330 else if (fl->fl_file)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002331 seq_printf(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002333 seq_printf(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002335 seq_printf(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 }
2337 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002338 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339 (fl->fl_type & LOCK_READ)
2340 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2341 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2342 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002343 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002344 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002345 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2346 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 }
2348 if (inode) {
2349#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002350 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002351 inode->i_sb->s_id, inode->i_ino);
2352#else
2353 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002354 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 MAJOR(inode->i_sb->s_dev),
2356 MINOR(inode->i_sb->s_dev), inode->i_ino);
2357#endif
2358 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002359 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 }
2361 if (IS_POSIX(fl)) {
2362 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002363 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002365 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002367 seq_printf(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368 }
2369}
2370
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002371static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372{
Jeff Layton7012b022013-06-21 08:58:22 -04002373 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002374 struct file_lock *fl, *bfl;
2375
Jeff Layton139ca042013-06-21 08:58:17 -04002376 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002377
Jeff Layton7012b022013-06-21 08:58:22 -04002378 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002379
2380 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002381 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002382
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002383 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384}
2385
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002386static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002387 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388{
Jeff Layton7012b022013-06-21 08:58:22 -04002389 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002390
Jeff Layton7012b022013-06-21 08:58:22 -04002391 iter->li_pos = *pos + 1;
2392 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002393 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002394 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395}
2396
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002397static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2398{
Jeff Layton7012b022013-06-21 08:58:22 -04002399 struct locks_iterator *iter = f->private;
2400
2401 ++iter->li_pos;
2402 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002403}
2404
2405static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002406 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002407{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002408 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002409 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002410}
2411
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002412static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002413 .start = locks_start,
2414 .next = locks_next,
2415 .stop = locks_stop,
2416 .show = locks_show,
2417};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002418
2419static int locks_open(struct inode *inode, struct file *filp)
2420{
Jeff Layton7012b022013-06-21 08:58:22 -04002421 return seq_open_private(filp, &locks_seq_operations,
2422 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002423}
2424
2425static const struct file_operations proc_locks_operations = {
2426 .open = locks_open,
2427 .read = seq_read,
2428 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002429 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002430};
2431
2432static int __init proc_locks_init(void)
2433{
2434 proc_create("locks", 0, NULL, &proc_locks_operations);
2435 return 0;
2436}
2437module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002438#endif
2439
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440/**
2441 * lock_may_read - checks that the region is free of locks
2442 * @inode: the inode that is being read
2443 * @start: the first byte to read
2444 * @len: the number of bytes to read
2445 *
2446 * Emulates Windows locking requirements. Whole-file
2447 * mandatory locks (share modes) can prohibit a read and
2448 * byte-range POSIX locks can prohibit a read if they overlap.
2449 *
2450 * N.B. this function is only ever called
2451 * from knfsd and ownership of locks is never checked.
2452 */
2453int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2454{
2455 struct file_lock *fl;
2456 int result = 1;
Jeff Layton1c8c6012013-06-21 08:58:15 -04002457
2458 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2460 if (IS_POSIX(fl)) {
2461 if (fl->fl_type == F_RDLCK)
2462 continue;
2463 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2464 continue;
2465 } else if (IS_FLOCK(fl)) {
2466 if (!(fl->fl_type & LOCK_MAND))
2467 continue;
2468 if (fl->fl_type & LOCK_READ)
2469 continue;
2470 } else
2471 continue;
2472 result = 0;
2473 break;
2474 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002475 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 return result;
2477}
2478
2479EXPORT_SYMBOL(lock_may_read);
2480
2481/**
2482 * lock_may_write - checks that the region is free of locks
2483 * @inode: the inode that is being written
2484 * @start: the first byte to write
2485 * @len: the number of bytes to write
2486 *
2487 * Emulates Windows locking requirements. Whole-file
2488 * mandatory locks (share modes) can prohibit a write and
2489 * byte-range POSIX locks can prohibit a write if they overlap.
2490 *
2491 * N.B. this function is only ever called
2492 * from knfsd and ownership of locks is never checked.
2493 */
2494int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2495{
2496 struct file_lock *fl;
2497 int result = 1;
Jeff Layton1c8c6012013-06-21 08:58:15 -04002498
2499 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2501 if (IS_POSIX(fl)) {
2502 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2503 continue;
2504 } else if (IS_FLOCK(fl)) {
2505 if (!(fl->fl_type & LOCK_MAND))
2506 continue;
2507 if (fl->fl_type & LOCK_WRITE)
2508 continue;
2509 } else
2510 continue;
2511 result = 0;
2512 break;
2513 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002514 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 return result;
2516}
2517
2518EXPORT_SYMBOL(lock_may_write);
2519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520static int __init filelock_init(void)
2521{
Jeff Layton7012b022013-06-21 08:58:22 -04002522 int i;
2523
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002525 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2526
Jeff Layton7012b022013-06-21 08:58:22 -04002527 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2528
2529 for_each_possible_cpu(i)
2530 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2531
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 return 0;
2533}
2534
2535core_initcall(filelock_init);