blob: ec9becd02d3d79495ba7fdbe3031bf37cb23544c [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
Jeff Layton62af4f12014-05-09 14:13:05 -0400133#define CREATE_TRACE_POINTS
134#include <trace/events/filelock.h>
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136#include <asm/uaccess.h>
137
138#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
139#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
J. Bruce Fields617588d2011-07-01 15:18:34 -0400140#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400141#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400143static bool lease_breaking(struct file_lock *fl)
144{
J. Bruce Fields778fc542011-07-26 18:25:49 -0400145 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
146}
147
148static int target_leasetype(struct file_lock *fl)
149{
150 if (fl->fl_flags & FL_UNLOCK_PENDING)
151 return F_UNLCK;
152 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
153 return F_RDLCK;
154 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400155}
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157int leases_enable = 1;
158int lease_break_time = 45;
159
160#define for_each_lock(inode, lockp) \
161 for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
162
Jeff Layton1c8c6012013-06-21 08:58:15 -0400163/*
Jeff Layton7012b022013-06-21 08:58:22 -0400164 * The global file_lock_list is only used for displaying /proc/locks, so we
165 * keep a list on each CPU, with each list protected by its own spinlock via
166 * the file_lock_lglock. Note that alterations to the list also require that
167 * the relevant i_lock is held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400168 */
Jeff Layton7012b022013-06-21 08:58:22 -0400169DEFINE_STATIC_LGLOCK(file_lock_lglock);
170static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
Jeff Layton88974692013-06-21 08:58:14 -0400171
Jeff Layton1c8c6012013-06-21 08:58:15 -0400172/*
Jeff Layton48f74182013-06-21 08:58:18 -0400173 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400174 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400175 *
176 * We hash locks by lockowner in order to optimize searching for the lock a
177 * particular lockowner is waiting on.
178 *
179 * FIXME: make this value scale via some heuristic? We generally will want more
180 * buckets when we have more lockowners holding locks, but that's a little
181 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400182 */
Jeff Layton48f74182013-06-21 08:58:18 -0400183#define BLOCKED_HASH_BITS 7
184static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400185
Jeff Layton1c8c6012013-06-21 08:58:15 -0400186/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400187 * This lock protects the blocked_hash. Generally, if you're accessing it, you
188 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400189 *
190 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
191 * pointer for file_lock structures that are acting as lock requests (in
192 * contrast to those that are acting as records of acquired locks).
193 *
194 * Note that when we acquire this lock in order to change the above fields,
195 * we often hold the i_lock as well. In certain cases, when reading the fields
196 * protected by this lock, we can skip acquiring it iff we already hold the
197 * i_lock.
198 *
199 * In particular, adding an entry to the fl_block list requires that you hold
200 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
201 * an entry from the list however only requires the file_lock_lock.
202 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400203static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Christoph Lametere18b8902006-12-06 20:33:20 -0800205static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
Miklos Szerediee19cc42011-07-07 13:06:09 +0200207static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200208{
Jeff Layton139ca042013-06-21 08:58:17 -0400209 INIT_HLIST_NODE(&fl->fl_link);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200210 INIT_LIST_HEAD(&fl->fl_block);
211 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200212}
213
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200215struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200217 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200218
219 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200220 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200221
222 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200224EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500226void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500227{
228 if (fl->fl_ops) {
229 if (fl->fl_ops->fl_release_private)
230 fl->fl_ops->fl_release_private(fl);
231 fl->fl_ops = NULL;
232 }
J. Bruce Fields068535f2012-08-01 07:56:16 -0400233 fl->fl_lmops = NULL;
Trond Myklebust47831f32006-03-20 13:44:05 -0500234
235}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500236EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500237
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400239void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800241 BUG_ON(waitqueue_active(&fl->fl_wait));
242 BUG_ON(!list_empty(&fl->fl_block));
Jeff Layton139ca042013-06-21 08:58:17 -0400243 BUG_ON(!hlist_unhashed(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
Trond Myklebust47831f32006-03-20 13:44:05 -0500245 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 kmem_cache_free(filelock_cache, fl);
247}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400248EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Jeff Laytoned9814d2014-08-11 14:20:31 -0400250static void
251locks_dispose_list(struct list_head *dispose)
252{
253 struct file_lock *fl;
254
255 while (!list_empty(dispose)) {
256 fl = list_first_entry(dispose, struct file_lock, fl_block);
257 list_del_init(&fl->fl_block);
258 locks_free_lock(fl);
259 }
260}
261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262void locks_init_lock(struct file_lock *fl)
263{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200264 memset(fl, 0, sizeof(struct file_lock));
265 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
267
268EXPORT_SYMBOL(locks_init_lock);
269
Trond Myklebust47831f32006-03-20 13:44:05 -0500270static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
271{
272 if (fl->fl_ops) {
273 if (fl->fl_ops->fl_copy_lock)
274 fl->fl_ops->fl_copy_lock(new, fl);
275 new->fl_ops = fl->fl_ops;
276 }
Christoph Hellwigbb8430a2010-10-31 08:35:31 -0400277 if (fl->fl_lmops)
Trond Myklebust47831f32006-03-20 13:44:05 -0500278 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500279}
280
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281/*
282 * Initialize a new lock from an existing file_lock structure.
283 */
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400284void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285{
286 new->fl_owner = fl->fl_owner;
287 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500288 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 new->fl_flags = fl->fl_flags;
290 new->fl_type = fl->fl_type;
291 new->fl_start = fl->fl_start;
292 new->fl_end = fl->fl_end;
Trond Myklebust09969052006-03-20 13:44:38 -0500293 new->fl_ops = NULL;
294 new->fl_lmops = NULL;
295}
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400296EXPORT_SYMBOL(locks_copy_conflock);
Trond Myklebust09969052006-03-20 13:44:38 -0500297
298void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
299{
Jeff Layton566709b2014-08-11 14:09:35 -0400300 /* "new" must be a freshly-initialized lock */
301 WARN_ON_ONCE(new->fl_ops);
Trond Myklebust09969052006-03-20 13:44:38 -0500302
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400303 locks_copy_conflock(new, fl);
Trond Myklebust09969052006-03-20 13:44:38 -0500304 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 new->fl_ops = fl->fl_ops;
306 new->fl_lmops = fl->fl_lmops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500307
308 locks_copy_private(new, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309}
310
311EXPORT_SYMBOL(locks_copy_lock);
312
313static inline int flock_translate_cmd(int cmd) {
314 if (cmd & LOCK_MAND)
315 return cmd & (LOCK_MAND | LOCK_RW);
316 switch (cmd) {
317 case LOCK_SH:
318 return F_RDLCK;
319 case LOCK_EX:
320 return F_WRLCK;
321 case LOCK_UN:
322 return F_UNLCK;
323 }
324 return -EINVAL;
325}
326
327/* Fill in a file_lock structure with an appropriate FLOCK lock. */
328static int flock_make_lock(struct file *filp, struct file_lock **lock,
329 unsigned int cmd)
330{
331 struct file_lock *fl;
332 int type = flock_translate_cmd(cmd);
333 if (type < 0)
334 return type;
335
336 fl = locks_alloc_lock();
337 if (fl == NULL)
338 return -ENOMEM;
339
340 fl->fl_file = filp;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +0200341 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 fl->fl_pid = current->tgid;
343 fl->fl_flags = FL_FLOCK;
344 fl->fl_type = type;
345 fl->fl_end = OFFSET_MAX;
346
347 *lock = fl;
348 return 0;
349}
350
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400351static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
353 switch (type) {
354 case F_RDLCK:
355 case F_WRLCK:
356 case F_UNLCK:
357 fl->fl_type = type;
358 break;
359 default:
360 return -EINVAL;
361 }
362 return 0;
363}
364
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500365static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
366 struct flock64 *l)
367{
368 switch (l->l_whence) {
369 case SEEK_SET:
370 fl->fl_start = 0;
371 break;
372 case SEEK_CUR:
373 fl->fl_start = filp->f_pos;
374 break;
375 case SEEK_END:
376 fl->fl_start = i_size_read(file_inode(filp));
377 break;
378 default:
379 return -EINVAL;
380 }
381 if (l->l_start > OFFSET_MAX - fl->fl_start)
382 return -EOVERFLOW;
383 fl->fl_start += l->l_start;
384 if (fl->fl_start < 0)
385 return -EINVAL;
386
387 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
388 POSIX-2001 defines it. */
389 if (l->l_len > 0) {
390 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
391 return -EOVERFLOW;
392 fl->fl_end = fl->fl_start + l->l_len - 1;
393
394 } else if (l->l_len < 0) {
395 if (fl->fl_start + l->l_len < 0)
396 return -EINVAL;
397 fl->fl_end = fl->fl_start - 1;
398 fl->fl_start += l->l_len;
399 } else
400 fl->fl_end = OFFSET_MAX;
401
402 fl->fl_owner = current->files;
403 fl->fl_pid = current->tgid;
404 fl->fl_file = filp;
405 fl->fl_flags = FL_POSIX;
406 fl->fl_ops = NULL;
407 fl->fl_lmops = NULL;
408
409 return assign_type(fl, l->l_type);
410}
411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
413 * style lock.
414 */
415static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
416 struct flock *l)
417{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500418 struct flock64 ll = {
419 .l_type = l->l_type,
420 .l_whence = l->l_whence,
421 .l_start = l->l_start,
422 .l_len = l->l_len,
423 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500425 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426}
427
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428/* default lease lock manager operations */
429static void lease_break_callback(struct file_lock *fl)
430{
431 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
432}
433
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700434static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400435 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400436 .lm_change = lease_modify,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437};
438
439/*
440 * Initialize a lease, use the default lock manager operations
441 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400442static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400444 if (assign_type(fl, type) != 0)
445 return -EINVAL;
446
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +0200447 fl->fl_owner = current->files;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 fl->fl_pid = current->tgid;
449
450 fl->fl_file = filp;
451 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 fl->fl_start = 0;
453 fl->fl_end = OFFSET_MAX;
454 fl->fl_ops = NULL;
455 fl->fl_lmops = &lease_manager_ops;
456 return 0;
457}
458
459/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400460static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
462 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400463 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
465 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500466 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
468 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400469 if (error) {
470 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500471 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400472 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500473 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474}
475
476/* Check if two locks overlap each other.
477 */
478static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
479{
480 return ((fl1->fl_end >= fl2->fl_start) &&
481 (fl2->fl_end >= fl1->fl_start));
482}
483
484/*
485 * Check whether two locks have the same owner.
486 */
Matt Mackall33443c42006-01-08 01:05:22 -0800487static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400489 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400491 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 return fl1->fl_owner == fl2->fl_owner;
493}
494
Jeff Layton7012b022013-06-21 08:58:22 -0400495/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500496static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400497{
Jeff Layton7012b022013-06-21 08:58:22 -0400498 lg_local_lock(&file_lock_lglock);
499 fl->fl_link_cpu = smp_processor_id();
500 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
501 lg_local_unlock(&file_lock_lglock);
Jeff Layton88974692013-06-21 08:58:14 -0400502}
503
Jeff Layton7012b022013-06-21 08:58:22 -0400504/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500505static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400506{
Jeff Layton7012b022013-06-21 08:58:22 -0400507 /*
508 * Avoid taking lock if already unhashed. This is safe since this check
509 * is done while holding the i_lock, and new insertions into the list
510 * also require that it be held.
511 */
512 if (hlist_unhashed(&fl->fl_link))
513 return;
514 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton139ca042013-06-21 08:58:17 -0400515 hlist_del_init(&fl->fl_link);
Jeff Layton7012b022013-06-21 08:58:22 -0400516 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton88974692013-06-21 08:58:14 -0400517}
518
Jeff Layton3999e492013-06-21 08:58:19 -0400519static unsigned long
520posix_owner_key(struct file_lock *fl)
521{
522 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
523 return fl->fl_lmops->lm_owner_key(fl);
524 return (unsigned long)fl->fl_owner;
525}
526
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500527static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400528{
Jeff Layton3999e492013-06-21 08:58:19 -0400529 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400530}
531
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500532static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400533{
Jeff Layton48f74182013-06-21 08:58:18 -0400534 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400535}
536
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537/* Remove waiter from blocker's block list.
538 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400539 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400540 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 */
Matt Mackall33443c42006-01-08 01:05:22 -0800542static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543{
Jeff Layton88974692013-06-21 08:58:14 -0400544 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 waiter->fl_next = NULL;
547}
548
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400549static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400551 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400553 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554}
555
556/* Insert waiter into blocker's block list.
557 * We use a circular list so that processes can be easily woken up in
558 * the order they blocked. The documentation doesn't require this but
559 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400560 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400561 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
Jeff Layton46dad762014-02-03 12:13:06 -0500562 * list itself is protected by the blocked_lock_lock, but by ensuring that the
Jeff Layton7b2296a2013-06-21 08:58:20 -0400563 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
Jeff Layton4e8c7652013-06-21 08:58:16 -0400564 * in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400566static void __locks_insert_block(struct file_lock *blocker,
567 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800569 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400571 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400572 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400573 locks_insert_global_blocked(waiter);
574}
575
576/* Must be called with i_lock held. */
577static void locks_insert_block(struct file_lock *blocker,
578 struct file_lock *waiter)
579{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400580 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400581 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400582 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583}
584
Jeff Layton1cb36012013-06-21 08:58:12 -0400585/*
586 * Wake up processes blocked waiting for blocker.
587 *
Jeff Layton1c8c6012013-06-21 08:58:15 -0400588 * Must be called with the inode->i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 */
590static void locks_wake_up_blocks(struct file_lock *blocker)
591{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400592 /*
593 * Avoid taking global lock if list is empty. This is safe since new
594 * blocked requests are only added to the list under the i_lock, and
595 * the i_lock is always held here. Note that removal from the fl_block
596 * list does not require the i_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400597 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400598 */
599 if (list_empty(&blocker->fl_block))
600 return;
601
Jeff Layton7b2296a2013-06-21 08:58:20 -0400602 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400604 struct file_lock *waiter;
605
606 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 struct file_lock, fl_block);
608 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400609 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
610 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 else
612 wake_up(&waiter->fl_wait);
613 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400614 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615}
616
617/* Insert file lock fl into an inode's lock list at the position indicated
618 * by pos. At the same time add the lock to the global file lock list.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400619 *
620 * Must be called with the i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 */
622static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
623{
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000624 fl->fl_nspid = get_pid(task_tgid(current));
625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 /* insert into file's list */
627 fl->fl_next = *pos;
628 *pos = fl;
Jeff Layton88974692013-06-21 08:58:14 -0400629
630 locks_insert_global_locks(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631}
632
Jeff Layton24cbe782014-02-03 12:13:06 -0500633/**
634 * locks_delete_lock - Delete a lock and then free it.
635 * @thisfl_p: pointer that points to the fl_next field of the previous
636 * inode->i_flock list entry
637 *
638 * Unlink a lock from all lists and free the namespace reference, but don't
639 * free it yet. Wake up processes that are blocked waiting for this lock and
640 * notify the FS that the lock has been cleared.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400641 *
642 * Must be called with the i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 */
Jeff Layton24cbe782014-02-03 12:13:06 -0500644static void locks_unlink_lock(struct file_lock **thisfl_p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
646 struct file_lock *fl = *thisfl_p;
647
Jeff Layton88974692013-06-21 08:58:14 -0400648 locks_delete_global_locks(fl);
649
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 *thisfl_p = fl->fl_next;
651 fl->fl_next = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000653 if (fl->fl_nspid) {
654 put_pid(fl->fl_nspid);
655 fl->fl_nspid = NULL;
656 }
657
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500659}
660
661/*
662 * Unlink a lock from all lists and free it.
663 *
664 * Must be called with i_lock held!
665 */
Jeff Laytoned9814d2014-08-11 14:20:31 -0400666static void locks_delete_lock(struct file_lock **thisfl_p,
667 struct list_head *dispose)
Jeff Layton24cbe782014-02-03 12:13:06 -0500668{
669 struct file_lock *fl = *thisfl_p;
670
671 locks_unlink_lock(thisfl_p);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400672 if (dispose)
673 list_add(&fl->fl_block, dispose);
674 else
675 locks_free_lock(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676}
677
678/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
679 * checks for shared/exclusive status of overlapping locks.
680 */
681static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
682{
683 if (sys_fl->fl_type == F_WRLCK)
684 return 1;
685 if (caller_fl->fl_type == F_WRLCK)
686 return 1;
687 return 0;
688}
689
690/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
691 * checking before calling the locks_conflict().
692 */
693static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
694{
695 /* POSIX locks owned by the same process do not conflict with
696 * each other.
697 */
698 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
699 return (0);
700
701 /* Check whether they overlap */
702 if (!locks_overlap(caller_fl, sys_fl))
703 return 0;
704
705 return (locks_conflict(caller_fl, sys_fl));
706}
707
708/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
709 * checking before calling the locks_conflict().
710 */
711static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
712{
713 /* FLOCK locks referring to the same filp do not conflict with
714 * each other.
715 */
716 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
717 return (0);
718 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
719 return 0;
720
721 return (locks_conflict(caller_fl, sys_fl));
722}
723
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400724void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500725posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726{
727 struct file_lock *cfl;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400728 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
Jeff Layton1c8c6012013-06-21 08:58:15 -0400730 spin_lock(&inode->i_lock);
Al Viro496ad9a2013-01-23 17:07:38 -0500731 for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 if (!IS_POSIX(cfl))
733 continue;
J. Bruce Fieldsb842e242007-05-10 19:02:07 -0400734 if (posix_locks_conflict(fl, cfl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735 break;
736 }
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000737 if (cfl) {
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400738 locks_copy_conflock(fl, cfl);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000739 if (cfl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -0800740 fl->fl_pid = pid_vnr(cfl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000741 } else
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400742 fl->fl_type = F_UNLCK;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400743 spin_unlock(&inode->i_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400744 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746EXPORT_SYMBOL(posix_test_lock);
747
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400748/*
749 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400751 * We attempt to detect deadlocks that are due purely to posix file
752 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400754 * We assume that a task can be waiting for at most one lock at a time.
755 * So for any acquired lock, the process holding that lock may be
756 * waiting on at most one other lock. That lock in turns may be held by
757 * someone waiting for at most one other lock. Given a requested lock
758 * caller_fl which is about to wait for a conflicting lock block_fl, we
759 * follow this chain of waiters to ensure we are not about to create a
760 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400761 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400762 * Since we do this before we ever put a process to sleep on a lock, we
763 * are ensured that there is never a cycle; that is what guarantees that
764 * the while() loop in posix_locks_deadlock() eventually completes.
765 *
766 * Note: the above assumption may not be true when handling lock
767 * requests from a broken NFS client. It may also fail in the presence
768 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400769 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500770 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400771 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500772 * Because the owner is not even nominally tied to a thread of
773 * execution, the deadlock detection below can't reasonably work well. Just
774 * skip it for those.
775 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400776 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500777 * locks that just checks for the case where two tasks are attempting to
778 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400780
781#define MAX_DEADLK_ITERATIONS 10
782
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400783/* Find a lock that the owner of the given block_fl is blocking on. */
784static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
785{
786 struct file_lock *fl;
787
Jeff Layton3999e492013-06-21 08:58:19 -0400788 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400789 if (posix_same_owner(fl, block_fl))
790 return fl->fl_next;
791 }
792 return NULL;
793}
794
Jeff Layton7b2296a2013-06-21 08:58:20 -0400795/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700796static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 struct file_lock *block_fl)
798{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400799 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800
Jeff Layton57b65322014-02-03 12:13:09 -0500801 /*
802 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400803 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500804 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400805 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500806 return 0;
807
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400808 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
809 if (i++ > MAX_DEADLK_ITERATIONS)
810 return 0;
811 if (posix_same_owner(caller_fl, block_fl))
812 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 }
814 return 0;
815}
816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400818 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400819 *
820 * Note that if called with an FL_EXISTS argument, the caller may determine
821 * whether or not a lock was successfully freed by testing the return
822 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800824static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800826 struct file_lock *new_fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 struct file_lock **before;
Al Viro496ad9a2013-01-23 17:07:38 -0500828 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 int error = 0;
830 int found = 0;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400831 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200833 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
834 new_fl = locks_alloc_lock();
835 if (!new_fl)
836 return -ENOMEM;
837 }
838
Jeff Layton1c8c6012013-06-21 08:58:15 -0400839 spin_lock(&inode->i_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400840 if (request->fl_flags & FL_ACCESS)
841 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400842
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 for_each_lock(inode, before) {
844 struct file_lock *fl = *before;
845 if (IS_POSIX(fl))
846 break;
847 if (IS_LEASE(fl))
848 continue;
849 if (filp != fl->fl_file)
850 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800851 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 goto out;
853 found = 1;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400854 locks_delete_lock(before, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 break;
856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Trond Myklebustf475ae92006-06-29 16:38:32 -0400858 if (request->fl_type == F_UNLCK) {
859 if ((request->fl_flags & FL_EXISTS) && !found)
860 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800861 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400862 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
864 /*
865 * If a higher-priority process was blocked on the old file lock,
866 * give it the opportunity to lock the file.
867 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200868 if (found) {
Jeff Layton1c8c6012013-06-21 08:58:15 -0400869 spin_unlock(&inode->i_lock);
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200870 cond_resched();
Jeff Layton1c8c6012013-06-21 08:58:15 -0400871 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400874find_conflict:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 for_each_lock(inode, before) {
876 struct file_lock *fl = *before;
877 if (IS_POSIX(fl))
878 break;
879 if (IS_LEASE(fl))
880 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800881 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 continue;
883 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700884 if (!(request->fl_flags & FL_SLEEP))
885 goto out;
886 error = FILE_LOCK_DEFERRED;
887 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 goto out;
889 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400890 if (request->fl_flags & FL_ACCESS)
891 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800892 locks_copy_lock(new_fl, request);
Pavel Emelyanov0e2f6db2007-09-11 15:24:01 -0700893 locks_insert_lock(before, new_fl);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800894 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400895 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897out:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400898 spin_unlock(&inode->i_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800899 if (new_fl)
900 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400901 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 return error;
903}
904
Marc Eshel150b3932007-01-18 16:15:35 -0500905static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906{
907 struct file_lock *fl;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700908 struct file_lock *new_fl = NULL;
909 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 struct file_lock *left = NULL;
911 struct file_lock *right = NULL;
912 struct file_lock **before;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400913 int error;
914 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400915 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 /*
918 * We may need two file_lock structures for this operation,
919 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700920 *
921 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700923 if (!(request->fl_flags & FL_ACCESS) &&
924 (request->fl_type != F_UNLCK ||
925 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
926 new_fl = locks_alloc_lock();
927 new_fl2 = locks_alloc_lock();
928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929
Jeff Layton1c8c6012013-06-21 08:58:15 -0400930 spin_lock(&inode->i_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400931 /*
932 * New lock request. Walk all POSIX locks and look for conflicts. If
933 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400934 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400935 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (request->fl_type != F_UNLCK) {
937 for_each_lock(inode, before) {
J. Bruce Fields526985b2006-11-14 16:54:36 -0500938 fl = *before;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 if (!IS_POSIX(fl))
940 continue;
941 if (!posix_locks_conflict(request, fl))
942 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800943 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400944 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945 error = -EAGAIN;
946 if (!(request->fl_flags & FL_SLEEP))
947 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400948 /*
949 * Deadlock detection and insertion into the blocked
950 * locks list must be done while holding the same lock!
951 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400953 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400954 if (likely(!posix_locks_deadlock(request, fl))) {
955 error = FILE_LOCK_DEFERRED;
956 __locks_insert_block(fl, request);
957 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400958 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 goto out;
960 }
961 }
962
963 /* If we're just looking for a conflict, we're done. */
964 error = 0;
965 if (request->fl_flags & FL_ACCESS)
966 goto out;
967
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 * Find the first old lock with the same owner as the new lock.
970 */
971
972 before = &inode->i_flock;
973
974 /* First skip locks owned by other processes. */
975 while ((fl = *before) && (!IS_POSIX(fl) ||
976 !posix_same_owner(request, fl))) {
977 before = &fl->fl_next;
978 }
979
Jeff Layton1cb36012013-06-21 08:58:12 -0400980 /* Process locks with this owner. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 while ((fl = *before) && posix_same_owner(request, fl)) {
982 /* Detect adjacent or overlapping regions (if same lock type)
983 */
984 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -0700985 /* In all comparisons of start vs end, use
986 * "start - 1" rather than "end + 1". If end
987 * is OFFSET_MAX, end + 1 will become negative.
988 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 if (fl->fl_end < request->fl_start - 1)
990 goto next_lock;
991 /* If the next lock in the list has entirely bigger
992 * addresses than the new one, insert the lock here.
993 */
Olaf Kirch449231d2005-08-25 16:25:35 -0700994 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 break;
996
997 /* If we come here, the new and old lock are of the
998 * same type and adjacent or overlapping. Make one
999 * lock yielding from the lower start address of both
1000 * locks to the higher end address.
1001 */
1002 if (fl->fl_start > request->fl_start)
1003 fl->fl_start = request->fl_start;
1004 else
1005 request->fl_start = fl->fl_start;
1006 if (fl->fl_end < request->fl_end)
1007 fl->fl_end = request->fl_end;
1008 else
1009 request->fl_end = fl->fl_end;
1010 if (added) {
Jeff Laytoned9814d2014-08-11 14:20:31 -04001011 locks_delete_lock(before, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 continue;
1013 }
1014 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001015 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 }
1017 else {
1018 /* Processing for different lock types is a bit
1019 * more complex.
1020 */
1021 if (fl->fl_end < request->fl_start)
1022 goto next_lock;
1023 if (fl->fl_start > request->fl_end)
1024 break;
1025 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001026 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (fl->fl_start < request->fl_start)
1028 left = fl;
1029 /* If the next lock in the list has a higher end
1030 * address than the new one, insert the new one here.
1031 */
1032 if (fl->fl_end > request->fl_end) {
1033 right = fl;
1034 break;
1035 }
1036 if (fl->fl_start >= request->fl_start) {
1037 /* The new lock completely replaces an old
1038 * one (This may happen several times).
1039 */
1040 if (added) {
Jeff Laytoned9814d2014-08-11 14:20:31 -04001041 locks_delete_lock(before, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 continue;
1043 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001044 /*
1045 * Replace the old lock with new_fl, and
1046 * remove the old one. It's safe to do the
1047 * insert here since we know that we won't be
1048 * using new_fl later, and that the lock is
1049 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001051 error = -ENOLCK;
1052 if (!new_fl)
1053 goto out;
1054 locks_copy_lock(new_fl, request);
1055 request = new_fl;
1056 new_fl = NULL;
Jeff Laytoned9814d2014-08-11 14:20:31 -04001057 locks_delete_lock(before, &dispose);
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001058 locks_insert_lock(before, request);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001059 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 }
1061 }
1062 /* Go on to next lock.
1063 */
1064 next_lock:
1065 before = &fl->fl_next;
1066 }
1067
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001068 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001069 * The above code only modifies existing locks in case of merging or
1070 * replacing. If new lock(s) need to be inserted all modifications are
1071 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001072 */
1073 error = -ENOLCK; /* "no luck" */
1074 if (right && left == right && !new_fl2)
1075 goto out;
1076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 error = 0;
1078 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001079 if (request->fl_type == F_UNLCK) {
1080 if (request->fl_flags & FL_EXISTS)
1081 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001083 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001084
1085 if (!new_fl) {
1086 error = -ENOLCK;
1087 goto out;
1088 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 locks_copy_lock(new_fl, request);
1090 locks_insert_lock(before, new_fl);
1091 new_fl = NULL;
1092 }
1093 if (right) {
1094 if (left == right) {
1095 /* The new lock breaks the old one in two pieces,
1096 * so we have to use the second new lock.
1097 */
1098 left = new_fl2;
1099 new_fl2 = NULL;
1100 locks_copy_lock(left, right);
1101 locks_insert_lock(before, left);
1102 }
1103 right->fl_start = request->fl_end + 1;
1104 locks_wake_up_blocks(right);
1105 }
1106 if (left) {
1107 left->fl_end = request->fl_start - 1;
1108 locks_wake_up_blocks(left);
1109 }
1110 out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001111 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 /*
1113 * Free any unused locks.
1114 */
1115 if (new_fl)
1116 locks_free_lock(new_fl);
1117 if (new_fl2)
1118 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001119 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 return error;
1121}
1122
1123/**
1124 * posix_lock_file - Apply a POSIX-style lock to a file
1125 * @filp: The file to apply the lock to
1126 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001127 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 *
1129 * Add a POSIX style lock to a file.
1130 * We merge adjacent & overlapping locks whenever possible.
1131 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001132 *
1133 * Note that if called with an FL_EXISTS argument, the caller may determine
1134 * whether or not a lock was successfully freed by testing the return
1135 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 */
Marc Eshel150b3932007-01-18 16:15:35 -05001137int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001138 struct file_lock *conflock)
1139{
Al Viro496ad9a2013-01-23 17:07:38 -05001140 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001141}
Marc Eshel150b3932007-01-18 16:15:35 -05001142EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143
1144/**
1145 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1146 * @filp: The file to apply the lock to
1147 * @fl: The lock to be applied
1148 *
1149 * Add a POSIX style lock to a file.
1150 * We merge adjacent & overlapping locks whenever possible.
1151 * POSIX locks are sorted by owner task, then by starting address
1152 */
1153int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1154{
1155 int error;
1156 might_sleep ();
1157 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001158 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001159 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 break;
1161 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1162 if (!error)
1163 continue;
1164
1165 locks_delete_block(fl);
1166 break;
1167 }
1168 return error;
1169}
1170EXPORT_SYMBOL(posix_lock_file_wait);
1171
1172/**
1173 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001174 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 *
1176 * Searches the inode's list of locks to find any POSIX locks which conflict.
1177 * This function is called from locks_verify_locked() only.
1178 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001179int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180{
Jeff Laytond7a06982014-03-10 09:54:15 -04001181 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 struct file_lock *fl;
1183
1184 /*
1185 * Search the lock list for this inode for any POSIX locks.
1186 */
Jeff Layton1c8c6012013-06-21 08:58:15 -04001187 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
1189 if (!IS_POSIX(fl))
1190 continue;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001191 if (fl->fl_owner != current->files &&
1192 fl->fl_owner != file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 break;
1194 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001195 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 return fl ? -EAGAIN : 0;
1197}
1198
1199/**
1200 * locks_mandatory_area - Check for a conflicting lock
1201 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1202 * for shared
1203 * @inode: the file to check
1204 * @filp: how the file was opened (if it was)
1205 * @offset: start of area to check
1206 * @count: length of area to check
1207 *
1208 * Searches the inode's list of locks to find any POSIX locks which conflict.
1209 * This function is called from rw_verify_area() and
1210 * locks_verify_truncate().
1211 */
1212int locks_mandatory_area(int read_write, struct inode *inode,
1213 struct file *filp, loff_t offset,
1214 size_t count)
1215{
1216 struct file_lock fl;
1217 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001218 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 fl.fl_pid = current->tgid;
1222 fl.fl_file = filp;
1223 fl.fl_flags = FL_POSIX | FL_ACCESS;
1224 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001225 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1227 fl.fl_start = offset;
1228 fl.fl_end = offset + count - 1;
1229
1230 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001231 if (filp) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001232 fl.fl_owner = filp;
Jeff Layton29723ad2014-03-10 09:54:19 -04001233 fl.fl_flags &= ~FL_SLEEP;
1234 error = __posix_lock_file(inode, &fl, NULL);
1235 if (!error)
1236 break;
1237 }
1238
1239 if (sleep)
1240 fl.fl_flags |= FL_SLEEP;
1241 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001242 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001243 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 break;
1245 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1246 if (!error) {
1247 /*
1248 * If we've been sleeping someone might have
1249 * changed the permissions behind our back.
1250 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001251 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 continue;
1253 }
1254
1255 locks_delete_block(&fl);
1256 break;
1257 }
1258
1259 return error;
1260}
1261
1262EXPORT_SYMBOL(locks_mandatory_area);
1263
J. Bruce Fields778fc542011-07-26 18:25:49 -04001264static void lease_clear_pending(struct file_lock *fl, int arg)
1265{
1266 switch (arg) {
1267 case F_UNLCK:
1268 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1269 /* fall through: */
1270 case F_RDLCK:
1271 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1272 }
1273}
1274
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275/* We already had a lease on this file; just change its type */
1276int lease_modify(struct file_lock **before, int arg)
1277{
1278 struct file_lock *fl = *before;
1279 int error = assign_type(fl, arg);
1280
1281 if (error)
1282 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001283 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001285 if (arg == F_UNLCK) {
1286 struct file *filp = fl->fl_file;
1287
1288 f_delown(filp);
1289 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001290 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1291 if (fl->fl_fasync != NULL) {
1292 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1293 fl->fl_fasync = NULL;
1294 }
Jeff Laytoned9814d2014-08-11 14:20:31 -04001295 locks_delete_lock(before, NULL);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001296 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 return 0;
1298}
1299
1300EXPORT_SYMBOL(lease_modify);
1301
J. Bruce Fields778fc542011-07-26 18:25:49 -04001302static bool past_time(unsigned long then)
1303{
1304 if (!then)
1305 /* 0 is a special value meaning "this never expires": */
1306 return false;
1307 return time_after(jiffies, then);
1308}
1309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310static void time_out_leases(struct inode *inode)
1311{
1312 struct file_lock **before;
1313 struct file_lock *fl;
1314
1315 before = &inode->i_flock;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04001316 while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001317 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001318 if (past_time(fl->fl_downgrade_time))
1319 lease_modify(before, F_RDLCK);
1320 if (past_time(fl->fl_break_time))
1321 lease_modify(before, F_UNLCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 if (fl == *before) /* lease_modify may have freed fl */
1323 before = &fl->fl_next;
1324 }
1325}
1326
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001327static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1328{
1329 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1330 return false;
1331 return locks_conflict(breaker, lease);
1332}
1333
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334/**
1335 * __break_lease - revoke all outstanding leases on file
1336 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001337 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1338 * break all leases
1339 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1340 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 *
david m. richter87250dd2007-05-09 16:10:27 -04001342 * break_lease (inlined for speed) has checked there already is at least
1343 * some kind of lock (maybe a lease) on this file. Leases are broken on
1344 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 * specified %O_NONBLOCK to your open().
1346 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001347int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001349 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 struct file_lock *new_fl, *flock;
1351 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 unsigned long break_time;
1353 int i_have_this_lease = 0;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001354 bool lease_conflict = false;
Al Viro8737c932009-12-24 06:47:55 -05001355 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Al Viro8737c932009-12-24 06:47:55 -05001357 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001358 if (IS_ERR(new_fl))
1359 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001360 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361
Jeff Layton1c8c6012013-06-21 08:58:15 -04001362 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363
1364 time_out_leases(inode);
1365
1366 flock = inode->i_flock;
1367 if ((flock == NULL) || !IS_LEASE(flock))
1368 goto out;
1369
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001370 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
1371 if (leases_conflict(fl, new_fl)) {
1372 lease_conflict = true;
1373 if (fl->fl_owner == current->files)
1374 i_have_this_lease = 1;
1375 }
1376 }
1377 if (!lease_conflict)
J. Bruce Fields778fc542011-07-26 18:25:49 -04001378 goto out;
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 break_time = 0;
1381 if (lease_break_time > 0) {
1382 break_time = jiffies + lease_break_time * HZ;
1383 if (break_time == 0)
1384 break_time++; /* so that 0 means no break time */
1385 }
1386
1387 for (fl = flock; fl && IS_LEASE(fl); fl = fl->fl_next) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001388 if (!leases_conflict(fl, new_fl))
1389 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001390 if (want_write) {
1391 if (fl->fl_flags & FL_UNLOCK_PENDING)
1392 continue;
1393 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001395 } else {
1396 if (lease_breaking(flock))
1397 continue;
1398 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1399 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 }
J. Bruce Fields778fc542011-07-26 18:25:49 -04001401 fl->fl_lmops->lm_break(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 }
1403
1404 if (i_have_this_lease || (mode & O_NONBLOCK)) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001405 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 error = -EWOULDBLOCK;
1407 goto out;
1408 }
1409
1410restart:
1411 break_time = flock->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001412 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001414 if (break_time == 0)
1415 break_time++;
Matthew Wilcox4321e012008-01-14 21:28:30 -07001416 locks_insert_block(flock, new_fl);
Jeff Layton62af4f12014-05-09 14:13:05 -04001417 trace_break_lease_block(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001418 spin_unlock(&inode->i_lock);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001419 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1420 !new_fl->fl_next, break_time);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001421 spin_lock(&inode->i_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001422 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001423 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (error >= 0) {
1425 if (error == 0)
1426 time_out_leases(inode);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001427 /*
1428 * Wait for the next conflicting lease that has not been
1429 * broken yet
1430 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 for (flock = inode->i_flock; flock && IS_LEASE(flock);
1432 flock = flock->fl_next) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001433 if (leases_conflict(new_fl, flock))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 goto restart;
1435 }
1436 error = 0;
1437 }
1438
1439out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001440 spin_unlock(&inode->i_lock);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001441 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 return error;
1443}
1444
1445EXPORT_SYMBOL(__break_lease);
1446
1447/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001448 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 * @inode: the inode
1450 * @time: pointer to a timespec which will contain the last modified time
1451 *
1452 * This is to force NFS clients to flush their caches for files with
1453 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001454 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 */
1456void lease_get_mtime(struct inode *inode, struct timespec *time)
1457{
1458 struct file_lock *flock = inode->i_flock;
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04001459 if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 *time = current_fs_time(inode->i_sb);
1461 else
1462 *time = inode->i_mtime;
1463}
1464
1465EXPORT_SYMBOL(lease_get_mtime);
1466
1467/**
1468 * fcntl_getlease - Enquire what lease is currently active
1469 * @filp: the file
1470 *
1471 * The value returned by this function will be one of
1472 * (if no lease break is pending):
1473 *
1474 * %F_RDLCK to indicate a shared lease is held.
1475 *
1476 * %F_WRLCK to indicate an exclusive lease is held.
1477 *
1478 * %F_UNLCK to indicate no lease is held.
1479 *
1480 * (if a lease break is pending):
1481 *
1482 * %F_RDLCK to indicate an exclusive lease needs to be
1483 * changed to a shared lease (or removed).
1484 *
1485 * %F_UNLCK to indicate the lease needs to be removed.
1486 *
1487 * XXX: sfr & willy disagree over whether F_INPROGRESS
1488 * should be returned to userspace.
1489 */
1490int fcntl_getlease(struct file *filp)
1491{
1492 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001493 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 int type = F_UNLCK;
1495
Jeff Layton1c8c6012013-06-21 08:58:15 -04001496 spin_lock(&inode->i_lock);
Al Viro496ad9a2013-01-23 17:07:38 -05001497 time_out_leases(file_inode(filp));
1498 for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 fl = fl->fl_next) {
1500 if (fl->fl_file == filp) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001501 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 break;
1503 }
1504 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001505 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506 return type;
1507}
1508
Jeff Layton24cbe782014-02-03 12:13:06 -05001509/**
1510 * check_conflicting_open - see if the given dentry points to a file that has
1511 * an existing open that would conflict with the
1512 * desired lease.
1513 * @dentry: dentry to check
1514 * @arg: type of lease that we're trying to acquire
1515 *
1516 * Check to see if there's an existing open fd on this file that would
1517 * conflict with the lease we're trying to set.
1518 */
1519static int
1520check_conflicting_open(const struct dentry *dentry, const long arg)
1521{
1522 int ret = 0;
1523 struct inode *inode = dentry->d_inode;
1524
1525 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1526 return -EAGAIN;
1527
1528 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1529 (atomic_read(&inode->i_count) > 1)))
1530 ret = -EAGAIN;
1531
1532 return ret;
1533}
1534
Jeff Laytond4f22d12013-06-21 08:58:11 -04001535static int generic_add_lease(struct file *filp, long arg, struct file_lock **flp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536{
KAMBAROV, ZAUR7eaae282005-07-07 17:57:06 -07001537 struct file_lock *fl, **before, **my_before = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001538 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 struct inode *inode = dentry->d_inode;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001540 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001541 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
J. Bruce Fields096657b2010-10-30 17:31:14 -04001543 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001544 trace_generic_add_lease(inode, lease);
1545
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001546 /*
1547 * In the delegation case we need mutual exclusion with
1548 * a number of operations that take the i_mutex. We trylock
1549 * because delegations are an optional optimization, and if
1550 * there's some chance of a conflict--we'd rather not
1551 * bother, maybe that's a sign this just isn't a good file to
1552 * hand out a delegation on.
1553 */
1554 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1555 return -EAGAIN;
1556
1557 if (is_deleg && arg == F_WRLCK) {
1558 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001559 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001560 WARN_ON_ONCE(1);
1561 return -EINVAL;
1562 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001563
Jeff Layton24cbe782014-02-03 12:13:06 -05001564 error = check_conflicting_open(dentry, arg);
1565 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001566 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001567
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 /*
1569 * At this point, we know that if there is an exclusive
1570 * lease on this file, then we hold it on this filp
1571 * (otherwise our open of this file would have blocked).
1572 * And if we are trying to acquire an exclusive lease,
1573 * then the file is not open by anyone (including us)
1574 * except for this filp.
1575 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001576 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577 for (before = &inode->i_flock;
1578 ((fl = *before) != NULL) && IS_LEASE(fl);
1579 before = &fl->fl_next) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001580 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581 my_before = before;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001582 continue;
1583 }
1584 /*
1585 * No exclusive leases if someone else has a lease on
1586 * this file:
1587 */
1588 if (arg == F_WRLCK)
1589 goto out;
1590 /*
1591 * Modifying our existing lease is OK, but no getting a
1592 * new lease if someone else is opening for write:
1593 */
1594 if (fl->fl_flags & FL_UNLOCK_PENDING)
1595 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 }
1597
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (my_before != NULL) {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001599 error = lease->fl_lmops->lm_change(my_before, arg);
Christoph Hellwig51ee4b82010-10-31 08:35:10 -04001600 if (!error)
1601 *flp = *my_before;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 goto out;
1603 }
1604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 error = -EINVAL;
1606 if (!leases_enable)
1607 goto out;
1608
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001609 locks_insert_lock(before, lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001610 /*
1611 * The check in break_lease() is lockless. It's possible for another
1612 * open to race in after we did the earlier check for a conflicting
1613 * open but before the lease was inserted. Check again for a
1614 * conflicting open and cancel the lease if there is one.
1615 *
1616 * We also add a barrier here to ensure that the insertion of the lock
1617 * precedes these checks.
1618 */
1619 smp_mb();
1620 error = check_conflicting_open(dentry, arg);
1621 if (error)
Jeff Laytonf39b9132014-09-09 16:00:51 -04001622 locks_unlink_lock(before);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623out:
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001624 if (is_deleg)
1625 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 return error;
1627}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001628
Jeff Laytond4f22d12013-06-21 08:58:11 -04001629static int generic_delete_lease(struct file *filp, struct file_lock **flp)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001630{
1631 struct file_lock *fl, **before;
1632 struct dentry *dentry = filp->f_path.dentry;
1633 struct inode *inode = dentry->d_inode;
1634
Jeff Layton62af4f12014-05-09 14:13:05 -04001635 trace_generic_delete_lease(inode, *flp);
1636
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001637 for (before = &inode->i_flock;
1638 ((fl = *before) != NULL) && IS_LEASE(fl);
1639 before = &fl->fl_next) {
1640 if (fl->fl_file != filp)
1641 continue;
1642 return (*flp)->fl_lmops->lm_change(before, F_UNLCK);
1643 }
1644 return -EAGAIN;
1645}
1646
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647/**
1648 * generic_setlease - sets a lease on an open file
1649 * @filp: file pointer
1650 * @arg: type of lease to obtain
1651 * @flp: input - file_lock to use, output - file_lock inserted
1652 *
1653 * The (input) flp->fl_lmops->lm_break function is required
1654 * by break_lease().
1655 *
Jeff Layton1c8c6012013-06-21 08:58:15 -04001656 * Called with inode->i_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 */
1658int generic_setlease(struct file *filp, long arg, struct file_lock **flp)
1659{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 struct dentry *dentry = filp->f_path.dentry;
1661 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001662 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001664 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001665 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001667 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 error = security_file_lock(filp, arg);
1669 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001670 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
1672 time_out_leases(inode);
1673
1674 BUG_ON(!(*flp)->fl_lmops->lm_break);
1675
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001676 switch (arg) {
1677 case F_UNLCK:
1678 return generic_delete_lease(filp, flp);
1679 case F_RDLCK:
1680 case F_WRLCK:
1681 return generic_add_lease(filp, arg, flp);
1682 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001683 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001686EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001688static int __vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
1689{
Al Viro72c2d532013-09-22 16:27:52 -04001690 if (filp->f_op->setlease)
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001691 return filp->f_op->setlease(filp, arg, lease);
1692 else
1693 return generic_setlease(filp, arg, lease);
1694}
1695
1696/**
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001697 * vfs_setlease - sets a lease on an open file
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 * @filp: file pointer
1699 * @arg: type of lease to obtain
1700 * @lease: file_lock to use
1701 *
1702 * Call this to establish a lease on the file.
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04001703 * The (*lease)->fl_lmops->lm_break operation must be set; if not,
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001704 * break_lease will oops!
1705 *
1706 * This will call the filesystem's setlease file method, if
1707 * defined. Note that there is no getlease method; instead, the
1708 * filesystem setlease method should call back to setlease() to
1709 * add a lease to the inode's lease list, where fcntl_getlease() can
1710 * find it. Since fcntl_getlease() only reports whether the current
1711 * task holds a lease, a cluster filesystem need only do this for
1712 * leases held by processes on this node.
1713 *
1714 * There is also no break_lease method; filesystems that
Adam Buchbinderc9404c92009-12-18 15:40:42 -05001715 * handle their own leases should break leases themselves from the
J. Bruce Fieldsf9ffed22006-11-14 15:51:40 -05001716 * filesystem's open, create, and (on truncate) setattr methods.
1717 *
1718 * Warning: the only current setlease methods exist only to disable
1719 * leases in certain cases. More vfs changes may be required to
1720 * allow a full filesystem lease implementation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 */
1722
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001723int vfs_setlease(struct file *filp, long arg, struct file_lock **lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724{
Jeff Layton1c8c6012013-06-21 08:58:15 -04001725 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 int error;
1727
Jeff Layton1c8c6012013-06-21 08:58:15 -04001728 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001729 error = __vfs_setlease(filp, arg, lease);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001730 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731
1732 return error;
1733}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001734EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001736static int do_fcntl_delete_lease(struct file *filp)
1737{
1738 struct file_lock fl, *flp = &fl;
1739
1740 lease_init(filp, F_UNLCK, flp);
1741
1742 return vfs_setlease(filp, F_UNLCK, &flp);
1743}
1744
1745static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746{
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001747 struct file_lock *fl, *ret;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001748 struct inode *inode = file_inode(filp);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001749 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 int error;
1751
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001752 fl = lease_alloc(filp, arg);
1753 if (IS_ERR(fl))
1754 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001756 new = fasync_alloc();
1757 if (!new) {
1758 locks_free_lock(fl);
1759 return -ENOMEM;
1760 }
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001761 ret = fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001762 spin_lock(&inode->i_lock);
J. Bruce Fields8896b932010-11-03 18:09:18 -04001763 error = __vfs_setlease(filp, arg, &ret);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001764 if (error)
1765 goto out_unlock;
1766 if (ret == fl)
1767 fl = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001769 /*
1770 * fasync_insert_entry() returns the old entry if any.
1771 * If there was no old entry, then it used 'new' and
1772 * inserted it into the fasync list. Clear new so that
1773 * we don't release it here.
1774 */
J. Bruce Fields3df057a2010-11-03 16:49:44 -04001775 if (!fasync_insert_entry(fd, filp, &ret->fl_fasync, new))
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001776 new = NULL;
1777
J. Bruce Fields8896b932010-11-03 18:09:18 -04001778 error = __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001779out_unlock:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001780 spin_unlock(&inode->i_lock);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001781 if (fl)
1782 locks_free_lock(fl);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001783 if (new)
1784 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return error;
1786}
1787
1788/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001789 * fcntl_setlease - sets a lease on an open file
1790 * @fd: open file descriptor
1791 * @filp: file pointer
1792 * @arg: type of lease to obtain
1793 *
1794 * Call this fcntl to establish a lease on the file.
1795 * Note that you also need to call %F_SETSIG to
1796 * receive a signal when the lease is broken.
1797 */
1798int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1799{
1800 if (arg == F_UNLCK)
1801 return do_fcntl_delete_lease(filp);
1802 return do_fcntl_add_lease(fd, filp, arg);
1803}
1804
1805/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1807 * @filp: The file to apply the lock to
1808 * @fl: The lock to be applied
1809 *
1810 * Add a FLOCK style lock to a file.
1811 */
1812int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1813{
1814 int error;
1815 might_sleep();
1816 for (;;) {
1817 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001818 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 break;
1820 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1821 if (!error)
1822 continue;
1823
1824 locks_delete_block(fl);
1825 break;
1826 }
1827 return error;
1828}
1829
1830EXPORT_SYMBOL(flock_lock_file_wait);
1831
1832/**
1833 * sys_flock: - flock() system call.
1834 * @fd: the file descriptor to lock.
1835 * @cmd: the type of lock to apply.
1836 *
1837 * Apply a %FL_FLOCK style lock to an open file descriptor.
1838 * The @cmd can be one of
1839 *
1840 * %LOCK_SH -- a shared lock.
1841 *
1842 * %LOCK_EX -- an exclusive lock.
1843 *
1844 * %LOCK_UN -- remove an existing lock.
1845 *
1846 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1847 *
1848 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1849 * processes read and write access respectively.
1850 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001851SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852{
Al Viro2903ff02012-08-28 12:52:22 -04001853 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 struct file_lock *lock;
1855 int can_sleep, unlock;
1856 int error;
1857
1858 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001859 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 goto out;
1861
1862 can_sleep = !(cmd & LOCK_NB);
1863 cmd &= ~LOCK_NB;
1864 unlock = (cmd == LOCK_UN);
1865
Al Viroaeb5d722008-09-02 15:28:45 -04001866 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001867 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 goto out_putf;
1869
Al Viro2903ff02012-08-28 12:52:22 -04001870 error = flock_make_lock(f.file, &lock, cmd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871 if (error)
1872 goto out_putf;
1873 if (can_sleep)
1874 lock->fl_flags |= FL_SLEEP;
1875
Al Viro2903ff02012-08-28 12:52:22 -04001876 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 if (error)
1878 goto out_free;
1879
Al Viro72c2d532013-09-22 16:27:52 -04001880 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001881 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 (can_sleep) ? F_SETLKW : F_SETLK,
1883 lock);
1884 else
Al Viro2903ff02012-08-28 12:52:22 -04001885 error = flock_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886
1887 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001888 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001891 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 out:
1893 return error;
1894}
1895
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001896/**
1897 * vfs_test_lock - test file byte range lock
1898 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001899 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001900 *
1901 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1902 * setting conf->fl_type to something other than F_UNLCK.
1903 */
1904int vfs_test_lock(struct file *filp, struct file_lock *fl)
1905{
Al Viro72c2d532013-09-22 16:27:52 -04001906 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001907 return filp->f_op->lock(filp, F_GETLK, fl);
1908 posix_test_lock(filp, fl);
1909 return 0;
1910}
1911EXPORT_SYMBOL_GPL(vfs_test_lock);
1912
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001913static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1914{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001915 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001916#if BITS_PER_LONG == 32
1917 /*
1918 * Make sure we can represent the posix lock via
1919 * legacy 32bit flock.
1920 */
1921 if (fl->fl_start > OFFT_OFFSET_MAX)
1922 return -EOVERFLOW;
1923 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1924 return -EOVERFLOW;
1925#endif
1926 flock->l_start = fl->fl_start;
1927 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1928 fl->fl_end - fl->fl_start + 1;
1929 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001930 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001931 return 0;
1932}
1933
1934#if BITS_PER_LONG == 32
1935static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1936{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001937 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001938 flock->l_start = fl->fl_start;
1939 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1940 fl->fl_end - fl->fl_start + 1;
1941 flock->l_whence = 0;
1942 flock->l_type = fl->fl_type;
1943}
1944#endif
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946/* Report the first existing lock that would conflict with l.
1947 * This implements the F_GETLK command of fcntl().
1948 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05001949int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001951 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 struct flock flock;
1953 int error;
1954
1955 error = -EFAULT;
1956 if (copy_from_user(&flock, l, sizeof(flock)))
1957 goto out;
1958 error = -EINVAL;
1959 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
1960 goto out;
1961
1962 error = flock_to_posix_lock(filp, &file_lock, &flock);
1963 if (error)
1964 goto out;
1965
Jeff Layton0d3f7a22014-04-22 08:23:58 -04001966 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05001967 error = -EINVAL;
1968 if (flock.l_pid != 0)
1969 goto out;
1970
Jeff Layton5d50ffd2014-02-03 12:13:10 -05001971 cmd = F_GETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001972 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001973 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05001974 }
1975
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001976 error = vfs_test_lock(filp, &file_lock);
1977 if (error)
1978 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979
Marc Eshel9d6a8c52007-02-21 00:55:18 -05001980 flock.l_type = file_lock.fl_type;
1981 if (file_lock.fl_type != F_UNLCK) {
1982 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001983 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001985 }
1986 error = -EFAULT;
1987 if (!copy_to_user(l, &flock, sizeof(flock)))
1988 error = 0;
1989out:
1990 return error;
1991}
1992
Marc Eshel7723ec92007-01-18 15:08:55 -05001993/**
1994 * vfs_lock_file - file byte range lock
1995 * @filp: The file to apply the lock to
1996 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
1997 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001998 * @conf: Place to return a copy of the conflicting lock, if found.
1999 *
2000 * A caller that doesn't care about the conflicting lock may pass NULL
2001 * as the final argument.
2002 *
2003 * If the filesystem defines a private ->lock() method, then @conf will
2004 * be left unchanged; so a caller that cares should initialize it to
2005 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05002006 *
2007 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2008 * locks, the ->lock() interface may return asynchronously, before the lock has
2009 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002010 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05002011 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2012 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002013 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05002014 * request completes.
2015 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07002016 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2017 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05002018 * nonzero return code and the file system should release the lock. The file
2019 * system is also responsible to keep a corresponding posix lock when it
2020 * grants a lock so the VFS can find out which locks are locally held and do
2021 * the correct lock cleanup when required.
2022 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002023 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002024 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002025 */
Marc Eshel150b3932007-01-18 16:15:35 -05002026int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002027{
Al Viro72c2d532013-09-22 16:27:52 -04002028 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002029 return filp->f_op->lock(filp, cmd, fl);
2030 else
Marc Eshel150b3932007-01-18 16:15:35 -05002031 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002032}
2033EXPORT_SYMBOL_GPL(vfs_lock_file);
2034
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002035static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2036 struct file_lock *fl)
2037{
2038 int error;
2039
2040 error = security_file_lock(filp, fl->fl_type);
2041 if (error)
2042 return error;
2043
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002044 for (;;) {
2045 error = vfs_lock_file(filp, cmd, fl, NULL);
2046 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002047 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002048 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2049 if (!error)
2050 continue;
2051
2052 locks_delete_block(fl);
2053 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002054 }
2055
2056 return error;
2057}
2058
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002059/* Ensure that fl->fl_filp has compatible f_mode for F_SETLK calls */
2060static int
2061check_fmode_for_setlk(struct file_lock *fl)
2062{
2063 switch (fl->fl_type) {
2064 case F_RDLCK:
2065 if (!(fl->fl_file->f_mode & FMODE_READ))
2066 return -EBADF;
2067 break;
2068 case F_WRLCK:
2069 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2070 return -EBADF;
2071 }
2072 return 0;
2073}
2074
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075/* Apply the lock described by l to an open file descriptor.
2076 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2077 */
Peter Staubachc2936212005-07-27 11:45:09 -07002078int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2079 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002080{
2081 struct file_lock *file_lock = locks_alloc_lock();
2082 struct flock flock;
2083 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002084 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 int error;
2086
2087 if (file_lock == NULL)
2088 return -ENOLCK;
2089
2090 /*
2091 * This might block, so we do it before checking the inode.
2092 */
2093 error = -EFAULT;
2094 if (copy_from_user(&flock, l, sizeof(flock)))
2095 goto out;
2096
Al Viro496ad9a2013-01-23 17:07:38 -05002097 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098
2099 /* Don't allow mandatory locks on files that may be memory mapped
2100 * and shared.
2101 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002102 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103 error = -EAGAIN;
2104 goto out;
2105 }
2106
Peter Staubachc2936212005-07-27 11:45:09 -07002107again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 error = flock_to_posix_lock(filp, file_lock, &flock);
2109 if (error)
2110 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002111
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002112 error = check_fmode_for_setlk(file_lock);
2113 if (error)
2114 goto out;
2115
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002116 /*
2117 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002118 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002119 */
2120 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002121 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002122 error = -EINVAL;
2123 if (flock.l_pid != 0)
2124 goto out;
2125
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002126 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002127 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002128 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002129 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002130 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002131 error = -EINVAL;
2132 if (flock.l_pid != 0)
2133 goto out;
2134
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002135 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002136 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002137 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002138 /* Fallthrough */
2139 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 file_lock->fl_flags |= FL_SLEEP;
2141 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002142
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002143 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
Peter Staubachc2936212005-07-27 11:45:09 -07002145 /*
2146 * Attempt to detect a close/fcntl race and recover by
2147 * releasing the lock that was just acquired.
2148 */
Al Viro0b2bac22008-05-06 13:58:34 -04002149 /*
2150 * we need that spin_lock here - it prevents reordering between
2151 * update of inode->i_flock and check for it done in close().
2152 * rcu_read_lock() wouldn't do.
2153 */
2154 spin_lock(&current->files->file_lock);
2155 f = fcheck(fd);
2156 spin_unlock(&current->files->file_lock);
2157 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002158 flock.l_type = F_UNLCK;
2159 goto again;
2160 }
2161
2162out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002163 locks_free_lock(file_lock);
2164 return error;
2165}
2166
2167#if BITS_PER_LONG == 32
2168/* Report the first existing lock that would conflict with l.
2169 * This implements the F_GETLK command of fcntl().
2170 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002171int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002173 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 struct flock64 flock;
2175 int error;
2176
2177 error = -EFAULT;
2178 if (copy_from_user(&flock, l, sizeof(flock)))
2179 goto out;
2180 error = -EINVAL;
2181 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2182 goto out;
2183
2184 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2185 if (error)
2186 goto out;
2187
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002188 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002189 error = -EINVAL;
2190 if (flock.l_pid != 0)
2191 goto out;
2192
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002193 cmd = F_GETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002194 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002195 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002196 }
2197
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002198 error = vfs_test_lock(filp, &file_lock);
2199 if (error)
2200 goto out;
2201
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002202 flock.l_type = file_lock.fl_type;
2203 if (file_lock.fl_type != F_UNLCK)
2204 posix_lock_to_flock64(&flock, &file_lock);
2205
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 error = -EFAULT;
2207 if (!copy_to_user(l, &flock, sizeof(flock)))
2208 error = 0;
2209
2210out:
2211 return error;
2212}
2213
2214/* Apply the lock described by l to an open file descriptor.
2215 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2216 */
Peter Staubachc2936212005-07-27 11:45:09 -07002217int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2218 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219{
2220 struct file_lock *file_lock = locks_alloc_lock();
2221 struct flock64 flock;
2222 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002223 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 int error;
2225
2226 if (file_lock == NULL)
2227 return -ENOLCK;
2228
2229 /*
2230 * This might block, so we do it before checking the inode.
2231 */
2232 error = -EFAULT;
2233 if (copy_from_user(&flock, l, sizeof(flock)))
2234 goto out;
2235
Al Viro496ad9a2013-01-23 17:07:38 -05002236 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237
2238 /* Don't allow mandatory locks on files that may be memory mapped
2239 * and shared.
2240 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002241 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 error = -EAGAIN;
2243 goto out;
2244 }
2245
Peter Staubachc2936212005-07-27 11:45:09 -07002246again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 error = flock64_to_posix_lock(filp, file_lock, &flock);
2248 if (error)
2249 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002250
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002251 error = check_fmode_for_setlk(file_lock);
2252 if (error)
2253 goto out;
2254
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002255 /*
2256 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002257 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002258 */
2259 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002260 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002261 error = -EINVAL;
2262 if (flock.l_pid != 0)
2263 goto out;
2264
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002265 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002266 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002267 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002268 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002269 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002270 error = -EINVAL;
2271 if (flock.l_pid != 0)
2272 goto out;
2273
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002274 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002275 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002276 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002277 /* Fallthrough */
2278 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279 file_lock->fl_flags |= FL_SLEEP;
2280 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002281
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002282 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
Peter Staubachc2936212005-07-27 11:45:09 -07002284 /*
2285 * Attempt to detect a close/fcntl race and recover by
2286 * releasing the lock that was just acquired.
2287 */
Al Viro0b2bac22008-05-06 13:58:34 -04002288 spin_lock(&current->files->file_lock);
2289 f = fcheck(fd);
2290 spin_unlock(&current->files->file_lock);
2291 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002292 flock.l_type = F_UNLCK;
2293 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 }
2295
2296out:
2297 locks_free_lock(file_lock);
2298 return error;
2299}
2300#endif /* BITS_PER_LONG == 32 */
2301
2302/*
2303 * This function is called when the file is being removed
2304 * from the task's fd array. POSIX locks belonging to this task
2305 * are deleted at this time.
2306 */
2307void locks_remove_posix(struct file *filp, fl_owner_t owner)
2308{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002309 struct file_lock lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310
2311 /*
2312 * If there are no locks held on this file, we don't need to call
2313 * posix_lock_file(). Another process could be setting a lock on this
2314 * file at the same time, but we wouldn't remove that lock anyway.
2315 */
Al Viro496ad9a2013-01-23 17:07:38 -05002316 if (!file_inode(filp)->i_flock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 return;
2318
2319 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002320 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 lock.fl_start = 0;
2322 lock.fl_end = OFFSET_MAX;
2323 lock.fl_owner = owner;
2324 lock.fl_pid = current->tgid;
2325 lock.fl_file = filp;
2326 lock.fl_ops = NULL;
2327 lock.fl_lmops = NULL;
2328
Marc Eshel150b3932007-01-18 16:15:35 -05002329 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2332 lock.fl_ops->fl_release_private(&lock);
2333}
2334
2335EXPORT_SYMBOL(locks_remove_posix);
2336
2337/*
2338 * This function is called on the last close of an open file.
2339 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002340void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341{
Al Viro496ad9a2013-01-23 17:07:38 -05002342 struct inode * inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 struct file_lock *fl;
2344 struct file_lock **before;
Jeff Laytoned9814d2014-08-11 14:20:31 -04002345 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346
2347 if (!inode->i_flock)
2348 return;
2349
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002350 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002351
Al Viro72c2d532013-09-22 16:27:52 -04002352 if (filp->f_op->flock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 struct file_lock fl = {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002354 .fl_owner = filp,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 .fl_pid = current->tgid,
2356 .fl_file = filp,
2357 .fl_flags = FL_FLOCK,
2358 .fl_type = F_UNLCK,
2359 .fl_end = OFFSET_MAX,
2360 };
2361 filp->f_op->flock(filp, F_SETLKW, &fl);
Trond Myklebust80fec4c2005-06-22 17:16:31 +00002362 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2363 fl.fl_ops->fl_release_private(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364 }
2365
Jeff Layton1c8c6012013-06-21 08:58:15 -04002366 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 before = &inode->i_flock;
2368
2369 while ((fl = *before) != NULL) {
2370 if (fl->fl_file == filp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371 if (IS_LEASE(fl)) {
2372 lease_modify(before, F_UNLCK);
2373 continue;
2374 }
Jeff Layton8c3cac52014-02-03 12:13:07 -05002375
2376 /*
2377 * There's a leftover lock on the list of a type that
2378 * we didn't expect to see. Most likely a classic
2379 * POSIX lock that ended up not getting released
2380 * properly, or that raced onto the list somehow. Log
2381 * some info about it and then just remove it from
2382 * the list.
2383 */
2384 WARN(!IS_FLOCK(fl),
2385 "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
2386 MAJOR(inode->i_sb->s_dev),
2387 MINOR(inode->i_sb->s_dev), inode->i_ino,
2388 fl->fl_type, fl->fl_flags,
2389 fl->fl_start, fl->fl_end);
2390
Jeff Laytoned9814d2014-08-11 14:20:31 -04002391 locks_delete_lock(before, &dispose);
Jeff Layton8c3cac52014-02-03 12:13:07 -05002392 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 }
2394 before = &fl->fl_next;
2395 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002396 spin_unlock(&inode->i_lock);
Jeff Laytoned9814d2014-08-11 14:20:31 -04002397 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398}
2399
2400/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002402 * @waiter: the lock which was waiting
2403 *
2404 * lockd needs to block waiting for locks.
2405 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002406int
Jeff Laytonf891a292013-06-21 08:58:09 -04002407posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002408{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002409 int status = 0;
2410
Jeff Layton7b2296a2013-06-21 08:58:20 -04002411 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002412 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002413 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002414 else
2415 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002416 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002417 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419EXPORT_SYMBOL(posix_unblock_lock);
2420
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002421/**
2422 * vfs_cancel_lock - file byte range unblock lock
2423 * @filp: The file to apply the unblock to
2424 * @fl: The lock to be unblocked
2425 *
2426 * Used by lock managers to cancel blocked requests
2427 */
2428int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2429{
Al Viro72c2d532013-09-22 16:27:52 -04002430 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002431 return filp->f_op->lock(filp, F_CANCELLK, fl);
2432 return 0;
2433}
2434
2435EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2436
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002437#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002438#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002439#include <linux/seq_file.h>
2440
Jeff Layton7012b022013-06-21 08:58:22 -04002441struct locks_iterator {
2442 int li_cpu;
2443 loff_t li_pos;
2444};
2445
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002446static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002447 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
2449 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002450 unsigned int fl_pid;
2451
2452 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002453 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002454 else
2455 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456
2457 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002458 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459
Jerome Marchand99dc8292010-10-26 14:22:33 -07002460 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002462 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002463 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002464 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002465 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002466 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002467 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002468
2469 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002471 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 } else if (IS_FLOCK(fl)) {
2473 if (fl->fl_type & LOCK_MAND) {
Fabian Frederick5315c262014-05-09 14:13:05 -04002474 seq_puts(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002476 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 }
2478 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002479 if (fl->fl_flags & FL_DELEG)
2480 seq_puts(f, "DELEG ");
2481 else
2482 seq_puts(f, "LEASE ");
2483
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002484 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002485 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002487 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002489 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002491 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 }
2493 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002494 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 (fl->fl_type & LOCK_READ)
2496 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2497 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2498 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002499 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002500 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002501 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2502 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503 }
2504 if (inode) {
2505#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002506 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 inode->i_sb->s_id, inode->i_ino);
2508#else
2509 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002510 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002511 MAJOR(inode->i_sb->s_dev),
2512 MINOR(inode->i_sb->s_dev), inode->i_ino);
2513#endif
2514 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002515 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 }
2517 if (IS_POSIX(fl)) {
2518 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002519 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002521 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002523 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 }
2525}
2526
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002527static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528{
Jeff Layton7012b022013-06-21 08:58:22 -04002529 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002530 struct file_lock *fl, *bfl;
2531
Jeff Layton139ca042013-06-21 08:58:17 -04002532 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002533
Jeff Layton7012b022013-06-21 08:58:22 -04002534 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002535
2536 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002537 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002538
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002539 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540}
2541
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002542static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002543 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544{
Jeff Layton7012b022013-06-21 08:58:22 -04002545 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002546
Jeff Layton7012b022013-06-21 08:58:22 -04002547 iter->li_pos = *pos + 1;
2548 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002549 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002550 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551}
2552
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002553static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2554{
Jeff Layton7012b022013-06-21 08:58:22 -04002555 struct locks_iterator *iter = f->private;
2556
2557 ++iter->li_pos;
2558 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002559}
2560
2561static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002562 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002563{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002564 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002565 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002566}
2567
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002568static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002569 .start = locks_start,
2570 .next = locks_next,
2571 .stop = locks_stop,
2572 .show = locks_show,
2573};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002574
2575static int locks_open(struct inode *inode, struct file *filp)
2576{
Jeff Layton7012b022013-06-21 08:58:22 -04002577 return seq_open_private(filp, &locks_seq_operations,
2578 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002579}
2580
2581static const struct file_operations proc_locks_operations = {
2582 .open = locks_open,
2583 .read = seq_read,
2584 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002585 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002586};
2587
2588static int __init proc_locks_init(void)
2589{
2590 proc_create("locks", 0, NULL, &proc_locks_operations);
2591 return 0;
2592}
2593module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002594#endif
2595
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596/**
2597 * lock_may_read - checks that the region is free of locks
2598 * @inode: the inode that is being read
2599 * @start: the first byte to read
2600 * @len: the number of bytes to read
2601 *
2602 * Emulates Windows locking requirements. Whole-file
2603 * mandatory locks (share modes) can prohibit a read and
2604 * byte-range POSIX locks can prohibit a read if they overlap.
2605 *
2606 * N.B. this function is only ever called
2607 * from knfsd and ownership of locks is never checked.
2608 */
2609int lock_may_read(struct inode *inode, loff_t start, unsigned long len)
2610{
2611 struct file_lock *fl;
2612 int result = 1;
Jeff Layton1c8c6012013-06-21 08:58:15 -04002613
2614 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2616 if (IS_POSIX(fl)) {
2617 if (fl->fl_type == F_RDLCK)
2618 continue;
2619 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2620 continue;
2621 } else if (IS_FLOCK(fl)) {
2622 if (!(fl->fl_type & LOCK_MAND))
2623 continue;
2624 if (fl->fl_type & LOCK_READ)
2625 continue;
2626 } else
2627 continue;
2628 result = 0;
2629 break;
2630 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002631 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 return result;
2633}
2634
2635EXPORT_SYMBOL(lock_may_read);
2636
2637/**
2638 * lock_may_write - checks that the region is free of locks
2639 * @inode: the inode that is being written
2640 * @start: the first byte to write
2641 * @len: the number of bytes to write
2642 *
2643 * Emulates Windows locking requirements. Whole-file
2644 * mandatory locks (share modes) can prohibit a write and
2645 * byte-range POSIX locks can prohibit a write if they overlap.
2646 *
2647 * N.B. this function is only ever called
2648 * from knfsd and ownership of locks is never checked.
2649 */
2650int lock_may_write(struct inode *inode, loff_t start, unsigned long len)
2651{
2652 struct file_lock *fl;
2653 int result = 1;
Jeff Layton1c8c6012013-06-21 08:58:15 -04002654
2655 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
2657 if (IS_POSIX(fl)) {
2658 if ((fl->fl_end < start) || (fl->fl_start > (start + len)))
2659 continue;
2660 } else if (IS_FLOCK(fl)) {
2661 if (!(fl->fl_type & LOCK_MAND))
2662 continue;
2663 if (fl->fl_type & LOCK_WRITE)
2664 continue;
2665 } else
2666 continue;
2667 result = 0;
2668 break;
2669 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04002670 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 return result;
2672}
2673
2674EXPORT_SYMBOL(lock_may_write);
2675
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676static int __init filelock_init(void)
2677{
Jeff Layton7012b022013-06-21 08:58:22 -04002678 int i;
2679
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002681 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2682
Jeff Layton7012b022013-06-21 08:58:22 -04002683 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2684
2685 for_each_possible_cpu(i)
2686 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2687
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 return 0;
2689}
2690
2691core_initcall(filelock_init);