blob: 22ac7694cc84022b060e36588030bbcd4edf21a5 [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
Jeff Layton1c8c6012013-06-21 08:58:15 -0400160/*
Jeff Layton7012b022013-06-21 08:58:22 -0400161 * The global file_lock_list is only used for displaying /proc/locks, so we
162 * keep a list on each CPU, with each list protected by its own spinlock via
163 * the file_lock_lglock. Note that alterations to the list also require that
Jeff Layton6109c852015-01-16 15:05:57 -0500164 * the relevant flc_lock is held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400165 */
Jeff Layton7012b022013-06-21 08:58:22 -0400166DEFINE_STATIC_LGLOCK(file_lock_lglock);
167static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
Jeff Layton88974692013-06-21 08:58:14 -0400168
Jeff Layton1c8c6012013-06-21 08:58:15 -0400169/*
Jeff Layton48f74182013-06-21 08:58:18 -0400170 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400171 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400172 *
173 * We hash locks by lockowner in order to optimize searching for the lock a
174 * particular lockowner is waiting on.
175 *
176 * FIXME: make this value scale via some heuristic? We generally will want more
177 * buckets when we have more lockowners holding locks, but that's a little
178 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400179 */
Jeff Layton48f74182013-06-21 08:58:18 -0400180#define BLOCKED_HASH_BITS 7
181static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400182
Jeff Layton1c8c6012013-06-21 08:58:15 -0400183/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400184 * This lock protects the blocked_hash. Generally, if you're accessing it, you
185 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400186 *
187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
188 * pointer for file_lock structures that are acting as lock requests (in
189 * contrast to those that are acting as records of acquired locks).
190 *
191 * Note that when we acquire this lock in order to change the above fields,
Jeff Layton6109c852015-01-16 15:05:57 -0500192 * we often hold the flc_lock as well. In certain cases, when reading the fields
Jeff Layton1c8c6012013-06-21 08:58:15 -0400193 * protected by this lock, we can skip acquiring it iff we already hold the
Jeff Layton6109c852015-01-16 15:05:57 -0500194 * flc_lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400195 *
196 * In particular, adding an entry to the fl_block list requires that you hold
Jeff Layton6109c852015-01-16 15:05:57 -0500197 * both the flc_lock and the blocked_lock_lock (acquired in that order).
198 * Deleting an entry from the list however only requires the file_lock_lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400199 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400200static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Jeff Layton4a075e32015-01-16 15:05:54 -0500202static struct kmem_cache *flctx_cache __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -0800203static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Jeff Layton4a075e32015-01-16 15:05:54 -0500205static struct file_lock_context *
206locks_get_lock_context(struct inode *inode)
207{
208 struct file_lock_context *new;
209
210 if (likely(inode->i_flctx))
211 goto out;
212
213 new = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
214 if (!new)
215 goto out;
216
Jeff Layton6109c852015-01-16 15:05:57 -0500217 spin_lock_init(&new->flc_lock);
Jeff Layton4a075e32015-01-16 15:05:54 -0500218 INIT_LIST_HEAD(&new->flc_flock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500219 INIT_LIST_HEAD(&new->flc_posix);
Jeff Layton8634b512015-01-16 15:05:55 -0500220 INIT_LIST_HEAD(&new->flc_lease);
Jeff Layton4a075e32015-01-16 15:05:54 -0500221
222 /*
223 * Assign the pointer if it's not already assigned. If it is, then
224 * free the context we just allocated.
225 */
226 spin_lock(&inode->i_lock);
227 if (likely(!inode->i_flctx)) {
228 inode->i_flctx = new;
229 new = NULL;
230 }
231 spin_unlock(&inode->i_lock);
232
233 if (new)
234 kmem_cache_free(flctx_cache, new);
235out:
236 return inode->i_flctx;
237}
238
239void
240locks_free_lock_context(struct file_lock_context *ctx)
241{
242 if (ctx) {
243 WARN_ON_ONCE(!list_empty(&ctx->flc_flock));
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500244 WARN_ON_ONCE(!list_empty(&ctx->flc_posix));
Jeff Layton8634b512015-01-16 15:05:55 -0500245 WARN_ON_ONCE(!list_empty(&ctx->flc_lease));
Jeff Layton4a075e32015-01-16 15:05:54 -0500246 kmem_cache_free(flctx_cache, ctx);
247 }
248}
249
Miklos Szerediee19cc42011-07-07 13:06:09 +0200250static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200251{
Jeff Layton139ca042013-06-21 08:58:17 -0400252 INIT_HLIST_NODE(&fl->fl_link);
Jeff Layton6dee60f2015-01-16 15:05:54 -0500253 INIT_LIST_HEAD(&fl->fl_list);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200254 INIT_LIST_HEAD(&fl->fl_block);
255 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200256}
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200259struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200261 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200262
263 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200264 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200265
266 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200268EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500270void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500271{
272 if (fl->fl_ops) {
273 if (fl->fl_ops->fl_release_private)
274 fl->fl_ops->fl_release_private(fl);
275 fl->fl_ops = NULL;
276 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500277
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400278 if (fl->fl_lmops) {
279 if (fl->fl_lmops->lm_put_owner)
280 fl->fl_lmops->lm_put_owner(fl);
281 fl->fl_lmops = NULL;
282 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500283}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500284EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500285
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400287void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800289 BUG_ON(waitqueue_active(&fl->fl_wait));
Jeff Layton6dee60f2015-01-16 15:05:54 -0500290 BUG_ON(!list_empty(&fl->fl_list));
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800291 BUG_ON(!list_empty(&fl->fl_block));
Jeff Layton139ca042013-06-21 08:58:17 -0400292 BUG_ON(!hlist_unhashed(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293
Trond Myklebust47831f32006-03-20 13:44:05 -0500294 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 kmem_cache_free(filelock_cache, fl);
296}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400297EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Jeff Laytoned9814d2014-08-11 14:20:31 -0400299static void
300locks_dispose_list(struct list_head *dispose)
301{
302 struct file_lock *fl;
303
304 while (!list_empty(dispose)) {
Jeff Layton6dee60f2015-01-16 15:05:54 -0500305 fl = list_first_entry(dispose, struct file_lock, fl_list);
306 list_del_init(&fl->fl_list);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400307 locks_free_lock(fl);
308 }
309}
310
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311void locks_init_lock(struct file_lock *fl)
312{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200313 memset(fl, 0, sizeof(struct file_lock));
314 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315}
316
317EXPORT_SYMBOL(locks_init_lock);
318
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319/*
320 * Initialize a new lock from an existing file_lock structure.
321 */
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400322void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323{
324 new->fl_owner = fl->fl_owner;
325 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500326 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 new->fl_flags = fl->fl_flags;
328 new->fl_type = fl->fl_type;
329 new->fl_start = fl->fl_start;
330 new->fl_end = fl->fl_end;
Kinglong Meef3282962014-08-22 10:18:43 -0400331 new->fl_lmops = fl->fl_lmops;
Trond Myklebust09969052006-03-20 13:44:38 -0500332 new->fl_ops = NULL;
Kinglong Meef3282962014-08-22 10:18:43 -0400333
334 if (fl->fl_lmops) {
335 if (fl->fl_lmops->lm_get_owner)
336 fl->fl_lmops->lm_get_owner(new, fl);
337 }
Trond Myklebust09969052006-03-20 13:44:38 -0500338}
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400339EXPORT_SYMBOL(locks_copy_conflock);
Trond Myklebust09969052006-03-20 13:44:38 -0500340
341void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
342{
Jeff Layton566709b2014-08-11 14:09:35 -0400343 /* "new" must be a freshly-initialized lock */
344 WARN_ON_ONCE(new->fl_ops);
Trond Myklebust09969052006-03-20 13:44:38 -0500345
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400346 locks_copy_conflock(new, fl);
Kinglong Meef3282962014-08-22 10:18:43 -0400347
Trond Myklebust09969052006-03-20 13:44:38 -0500348 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 new->fl_ops = fl->fl_ops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500350
Kinglong Meef3282962014-08-22 10:18:43 -0400351 if (fl->fl_ops) {
352 if (fl->fl_ops->fl_copy_lock)
353 fl->fl_ops->fl_copy_lock(new, fl);
354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355}
356
357EXPORT_SYMBOL(locks_copy_lock);
358
359static inline int flock_translate_cmd(int cmd) {
360 if (cmd & LOCK_MAND)
361 return cmd & (LOCK_MAND | LOCK_RW);
362 switch (cmd) {
363 case LOCK_SH:
364 return F_RDLCK;
365 case LOCK_EX:
366 return F_WRLCK;
367 case LOCK_UN:
368 return F_UNLCK;
369 }
370 return -EINVAL;
371}
372
373/* Fill in a file_lock structure with an appropriate FLOCK lock. */
Jeff Layton6e129d002014-09-04 10:25:06 -0400374static struct file_lock *
375flock_make_lock(struct file *filp, unsigned int cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
377 struct file_lock *fl;
378 int type = flock_translate_cmd(cmd);
Jeff Layton6e129d002014-09-04 10:25:06 -0400379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 if (type < 0)
Jeff Layton6e129d002014-09-04 10:25:06 -0400381 return ERR_PTR(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383 fl = locks_alloc_lock();
384 if (fl == NULL)
Jeff Layton6e129d002014-09-04 10:25:06 -0400385 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387 fl->fl_file = filp;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +0200388 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 fl->fl_pid = current->tgid;
390 fl->fl_flags = FL_FLOCK;
391 fl->fl_type = type;
392 fl->fl_end = OFFSET_MAX;
393
Jeff Layton6e129d002014-09-04 10:25:06 -0400394 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395}
396
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400397static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
399 switch (type) {
400 case F_RDLCK:
401 case F_WRLCK:
402 case F_UNLCK:
403 fl->fl_type = type;
404 break;
405 default:
406 return -EINVAL;
407 }
408 return 0;
409}
410
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500411static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
412 struct flock64 *l)
413{
414 switch (l->l_whence) {
415 case SEEK_SET:
416 fl->fl_start = 0;
417 break;
418 case SEEK_CUR:
419 fl->fl_start = filp->f_pos;
420 break;
421 case SEEK_END:
422 fl->fl_start = i_size_read(file_inode(filp));
423 break;
424 default:
425 return -EINVAL;
426 }
427 if (l->l_start > OFFSET_MAX - fl->fl_start)
428 return -EOVERFLOW;
429 fl->fl_start += l->l_start;
430 if (fl->fl_start < 0)
431 return -EINVAL;
432
433 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
434 POSIX-2001 defines it. */
435 if (l->l_len > 0) {
436 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
437 return -EOVERFLOW;
438 fl->fl_end = fl->fl_start + l->l_len - 1;
439
440 } else if (l->l_len < 0) {
441 if (fl->fl_start + l->l_len < 0)
442 return -EINVAL;
443 fl->fl_end = fl->fl_start - 1;
444 fl->fl_start += l->l_len;
445 } else
446 fl->fl_end = OFFSET_MAX;
447
448 fl->fl_owner = current->files;
449 fl->fl_pid = current->tgid;
450 fl->fl_file = filp;
451 fl->fl_flags = FL_POSIX;
452 fl->fl_ops = NULL;
453 fl->fl_lmops = NULL;
454
455 return assign_type(fl, l->l_type);
456}
457
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
459 * style lock.
460 */
461static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
462 struct flock *l)
463{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500464 struct flock64 ll = {
465 .l_type = l->l_type,
466 .l_whence = l->l_whence,
467 .l_start = l->l_start,
468 .l_len = l->l_len,
469 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500471 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472}
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/* default lease lock manager operations */
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400475static bool
476lease_break_callback(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
478 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400479 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480}
481
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400482static void
483lease_setup(struct file_lock *fl, void **priv)
484{
485 struct file *filp = fl->fl_file;
486 struct fasync_struct *fa = *priv;
487
488 /*
489 * fasync_insert_entry() returns the old entry if any. If there was no
490 * old entry, then it used "priv" and inserted it into the fasync list.
491 * Clear the pointer to indicate that it shouldn't be freed.
492 */
493 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
494 *priv = NULL;
495
496 __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
497}
498
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700499static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400500 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400501 .lm_change = lease_modify,
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400502 .lm_setup = lease_setup,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503};
504
505/*
506 * Initialize a lease, use the default lock manager operations
507 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400508static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400510 if (assign_type(fl, type) != 0)
511 return -EINVAL;
512
Jeff Layton7ca76312014-09-01 19:04:48 -0400513 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 fl->fl_pid = current->tgid;
515
516 fl->fl_file = filp;
517 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 fl->fl_start = 0;
519 fl->fl_end = OFFSET_MAX;
520 fl->fl_ops = NULL;
521 fl->fl_lmops = &lease_manager_ops;
522 return 0;
523}
524
525/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400526static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527{
528 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400529 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530
531 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500532 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
534 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400535 if (error) {
536 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500537 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400538 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500539 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540}
541
542/* Check if two locks overlap each other.
543 */
544static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
545{
546 return ((fl1->fl_end >= fl2->fl_start) &&
547 (fl2->fl_end >= fl1->fl_start));
548}
549
550/*
551 * Check whether two locks have the same owner.
552 */
Matt Mackall33443c42006-01-08 01:05:22 -0800553static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400555 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400557 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 return fl1->fl_owner == fl2->fl_owner;
559}
560
Jeff Layton6109c852015-01-16 15:05:57 -0500561/* Must be called with the flc_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500562static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400563{
Jeff Layton7012b022013-06-21 08:58:22 -0400564 lg_local_lock(&file_lock_lglock);
565 fl->fl_link_cpu = smp_processor_id();
566 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
567 lg_local_unlock(&file_lock_lglock);
Jeff Layton88974692013-06-21 08:58:14 -0400568}
569
Jeff Layton6109c852015-01-16 15:05:57 -0500570/* Must be called with the flc_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500571static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400572{
Jeff Layton7012b022013-06-21 08:58:22 -0400573 /*
574 * Avoid taking lock if already unhashed. This is safe since this check
Jeff Layton6109c852015-01-16 15:05:57 -0500575 * is done while holding the flc_lock, and new insertions into the list
Jeff Layton7012b022013-06-21 08:58:22 -0400576 * also require that it be held.
577 */
578 if (hlist_unhashed(&fl->fl_link))
579 return;
580 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton139ca042013-06-21 08:58:17 -0400581 hlist_del_init(&fl->fl_link);
Jeff Layton7012b022013-06-21 08:58:22 -0400582 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton88974692013-06-21 08:58:14 -0400583}
584
Jeff Layton3999e492013-06-21 08:58:19 -0400585static unsigned long
586posix_owner_key(struct file_lock *fl)
587{
588 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
589 return fl->fl_lmops->lm_owner_key(fl);
590 return (unsigned long)fl->fl_owner;
591}
592
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500593static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400594{
Jeff Layton3999e492013-06-21 08:58:19 -0400595 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400596}
597
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500598static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400599{
Jeff Layton48f74182013-06-21 08:58:18 -0400600 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400601}
602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603/* Remove waiter from blocker's block list.
604 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400605 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400606 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 */
Matt Mackall33443c42006-01-08 01:05:22 -0800608static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609{
Jeff Layton88974692013-06-21 08:58:14 -0400610 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 waiter->fl_next = NULL;
613}
614
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400615static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400617 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400619 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620}
621
622/* Insert waiter into blocker's block list.
623 * We use a circular list so that processes can be easily woken up in
624 * the order they blocked. The documentation doesn't require this but
625 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400626 *
Jeff Layton6109c852015-01-16 15:05:57 -0500627 * Must be called with both the flc_lock and blocked_lock_lock held. The
628 * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
629 * that the flc_lock is also held on insertions we can avoid taking the
630 * blocked_lock_lock in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400632static void __locks_insert_block(struct file_lock *blocker,
633 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800635 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400637 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400638 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400639 locks_insert_global_blocked(waiter);
640}
641
Jeff Layton6109c852015-01-16 15:05:57 -0500642/* Must be called with flc_lock held. */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400643static void locks_insert_block(struct file_lock *blocker,
644 struct file_lock *waiter)
645{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400646 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400647 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400648 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Jeff Layton1cb36012013-06-21 08:58:12 -0400651/*
652 * Wake up processes blocked waiting for blocker.
653 *
Jeff Layton6109c852015-01-16 15:05:57 -0500654 * Must be called with the inode->flc_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 */
656static void locks_wake_up_blocks(struct file_lock *blocker)
657{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400658 /*
659 * Avoid taking global lock if list is empty. This is safe since new
Jeff Layton6109c852015-01-16 15:05:57 -0500660 * blocked requests are only added to the list under the flc_lock, and
661 * the flc_lock is always held here. Note that removal from the fl_block
662 * list does not require the flc_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400663 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400664 */
665 if (list_empty(&blocker->fl_block))
666 return;
667
Jeff Layton7b2296a2013-06-21 08:58:20 -0400668 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400670 struct file_lock *waiter;
671
672 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 struct file_lock, fl_block);
674 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400675 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
676 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 else
678 wake_up(&waiter->fl_wait);
679 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400680 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681}
682
Jeff Layton5263e312015-01-16 15:05:55 -0500683static void
Jeff Layton9bd0f452015-01-16 15:05:57 -0500684locks_insert_lock_ctx(struct file_lock *fl, int *counter,
685 struct list_head *before)
Jeff Layton5263e312015-01-16 15:05:55 -0500686{
687 fl->fl_nspid = get_pid(task_tgid(current));
688 list_add_tail(&fl->fl_list, before);
Jeff Layton9bd0f452015-01-16 15:05:57 -0500689 ++*counter;
Jeff Layton5263e312015-01-16 15:05:55 -0500690 locks_insert_global_locks(fl);
691}
692
Jeff Layton8634b512015-01-16 15:05:55 -0500693static void
Jeff Layton9bd0f452015-01-16 15:05:57 -0500694locks_unlink_lock_ctx(struct file_lock *fl, int *counter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695{
Jeff Layton88974692013-06-21 08:58:14 -0400696 locks_delete_global_locks(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500697 list_del_init(&fl->fl_list);
Jeff Layton9bd0f452015-01-16 15:05:57 -0500698 --*counter;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000699 if (fl->fl_nspid) {
700 put_pid(fl->fl_nspid);
701 fl->fl_nspid = NULL;
702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500704}
705
Jeff Layton5263e312015-01-16 15:05:55 -0500706static void
Jeff Layton9bd0f452015-01-16 15:05:57 -0500707locks_delete_lock_ctx(struct file_lock *fl, int *counter,
708 struct list_head *dispose)
Jeff Layton5263e312015-01-16 15:05:55 -0500709{
Jeff Layton9bd0f452015-01-16 15:05:57 -0500710 locks_unlink_lock_ctx(fl, counter);
Jeff Layton8634b512015-01-16 15:05:55 -0500711 if (dispose)
712 list_add(&fl->fl_list, dispose);
713 else
714 locks_free_lock(fl);
Jeff Layton5263e312015-01-16 15:05:55 -0500715}
716
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
718 * checks for shared/exclusive status of overlapping locks.
719 */
720static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
721{
722 if (sys_fl->fl_type == F_WRLCK)
723 return 1;
724 if (caller_fl->fl_type == F_WRLCK)
725 return 1;
726 return 0;
727}
728
729/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
730 * checking before calling the locks_conflict().
731 */
732static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
733{
734 /* POSIX locks owned by the same process do not conflict with
735 * each other.
736 */
737 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
738 return (0);
739
740 /* Check whether they overlap */
741 if (!locks_overlap(caller_fl, sys_fl))
742 return 0;
743
744 return (locks_conflict(caller_fl, sys_fl));
745}
746
747/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
748 * checking before calling the locks_conflict().
749 */
750static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
751{
752 /* FLOCK locks referring to the same filp do not conflict with
753 * each other.
754 */
755 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
756 return (0);
757 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
758 return 0;
759
760 return (locks_conflict(caller_fl, sys_fl));
761}
762
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400763void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500764posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
766 struct file_lock *cfl;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500767 struct file_lock_context *ctx;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400768 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500770 ctx = inode->i_flctx;
771 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400772 fl->fl_type = F_UNLCK;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500773 return;
774 }
775
Jeff Layton6109c852015-01-16 15:05:57 -0500776 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500777 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
778 if (posix_locks_conflict(fl, cfl)) {
779 locks_copy_conflock(fl, cfl);
780 if (cfl->fl_nspid)
781 fl->fl_pid = pid_vnr(cfl->fl_nspid);
782 goto out;
783 }
784 }
785 fl->fl_type = F_UNLCK;
786out:
Jeff Layton6109c852015-01-16 15:05:57 -0500787 spin_unlock(&ctx->flc_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400788 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790EXPORT_SYMBOL(posix_test_lock);
791
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400792/*
793 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400795 * We attempt to detect deadlocks that are due purely to posix file
796 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400798 * We assume that a task can be waiting for at most one lock at a time.
799 * So for any acquired lock, the process holding that lock may be
800 * waiting on at most one other lock. That lock in turns may be held by
801 * someone waiting for at most one other lock. Given a requested lock
802 * caller_fl which is about to wait for a conflicting lock block_fl, we
803 * follow this chain of waiters to ensure we are not about to create a
804 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400805 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400806 * Since we do this before we ever put a process to sleep on a lock, we
807 * are ensured that there is never a cycle; that is what guarantees that
808 * the while() loop in posix_locks_deadlock() eventually completes.
809 *
810 * Note: the above assumption may not be true when handling lock
811 * requests from a broken NFS client. It may also fail in the presence
812 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400813 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500814 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400815 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500816 * Because the owner is not even nominally tied to a thread of
817 * execution, the deadlock detection below can't reasonably work well. Just
818 * skip it for those.
819 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400820 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500821 * locks that just checks for the case where two tasks are attempting to
822 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400824
825#define MAX_DEADLK_ITERATIONS 10
826
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400827/* Find a lock that the owner of the given block_fl is blocking on. */
828static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
829{
830 struct file_lock *fl;
831
Jeff Layton3999e492013-06-21 08:58:19 -0400832 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400833 if (posix_same_owner(fl, block_fl))
834 return fl->fl_next;
835 }
836 return NULL;
837}
838
Jeff Layton7b2296a2013-06-21 08:58:20 -0400839/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700840static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 struct file_lock *block_fl)
842{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400843 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844
Jeff Layton57b65322014-02-03 12:13:09 -0500845 /*
846 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400847 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500848 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400849 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500850 return 0;
851
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400852 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
853 if (i++ > MAX_DEADLK_ITERATIONS)
854 return 0;
855 if (posix_same_owner(caller_fl, block_fl))
856 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857 }
858 return 0;
859}
860
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400862 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400863 *
864 * Note that if called with an FL_EXISTS argument, the caller may determine
865 * whether or not a lock was successfully freed by testing the return
866 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800868static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800870 struct file_lock *new_fl = NULL;
Jeff Layton5263e312015-01-16 15:05:55 -0500871 struct file_lock *fl;
872 struct file_lock_context *ctx;
873 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 int error = 0;
Jeff Layton5263e312015-01-16 15:05:55 -0500875 bool found = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400876 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Jeff Layton5263e312015-01-16 15:05:55 -0500878 ctx = locks_get_lock_context(inode);
879 if (!ctx)
880 return -ENOMEM;
881
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200882 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
883 new_fl = locks_alloc_lock();
884 if (!new_fl)
885 return -ENOMEM;
886 }
887
Jeff Layton6109c852015-01-16 15:05:57 -0500888 spin_lock(&ctx->flc_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400889 if (request->fl_flags & FL_ACCESS)
890 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400891
Jeff Layton5263e312015-01-16 15:05:55 -0500892 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 if (filp != fl->fl_file)
894 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800895 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 goto out;
Jeff Layton5263e312015-01-16 15:05:55 -0500897 found = true;
Jeff Layton9bd0f452015-01-16 15:05:57 -0500898 locks_delete_lock_ctx(fl, &ctx->flc_flock_cnt, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 break;
900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Trond Myklebustf475ae92006-06-29 16:38:32 -0400902 if (request->fl_type == F_UNLCK) {
903 if ((request->fl_flags & FL_EXISTS) && !found)
904 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800905 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400906 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
908 /*
909 * If a higher-priority process was blocked on the old file lock,
910 * give it the opportunity to lock the file.
911 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200912 if (found) {
Jeff Layton6109c852015-01-16 15:05:57 -0500913 spin_unlock(&ctx->flc_lock);
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200914 cond_resched();
Jeff Layton6109c852015-01-16 15:05:57 -0500915 spin_lock(&ctx->flc_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400918find_conflict:
Jeff Layton5263e312015-01-16 15:05:55 -0500919 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Trond Myklebust993dfa82006-03-31 02:30:55 -0800920 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 continue;
922 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700923 if (!(request->fl_flags & FL_SLEEP))
924 goto out;
925 error = FILE_LOCK_DEFERRED;
926 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 goto out;
928 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400929 if (request->fl_flags & FL_ACCESS)
930 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800931 locks_copy_lock(new_fl, request);
Jeff Layton9bd0f452015-01-16 15:05:57 -0500932 locks_insert_lock_ctx(new_fl, &ctx->flc_flock_cnt, &ctx->flc_flock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800933 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400934 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
936out:
Jeff Layton6109c852015-01-16 15:05:57 -0500937 spin_unlock(&ctx->flc_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800938 if (new_fl)
939 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400940 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 return error;
942}
943
Marc Eshel150b3932007-01-18 16:15:35 -0500944static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500946 struct file_lock *fl, *tmp;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700947 struct file_lock *new_fl = NULL;
948 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 struct file_lock *left = NULL;
950 struct file_lock *right = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500951 struct file_lock_context *ctx;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400952 int error;
953 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400954 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500956 ctx = locks_get_lock_context(inode);
957 if (!ctx)
958 return -ENOMEM;
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 /*
961 * We may need two file_lock structures for this operation,
962 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700963 *
964 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700966 if (!(request->fl_flags & FL_ACCESS) &&
967 (request->fl_type != F_UNLCK ||
968 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
969 new_fl = locks_alloc_lock();
970 new_fl2 = locks_alloc_lock();
971 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Jeff Layton6109c852015-01-16 15:05:57 -0500973 spin_lock(&ctx->flc_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400974 /*
975 * New lock request. Walk all POSIX locks and look for conflicts. If
976 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400977 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400978 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 if (request->fl_type != F_UNLCK) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500980 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 if (!IS_POSIX(fl))
982 continue;
983 if (!posix_locks_conflict(request, fl))
984 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800985 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400986 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 error = -EAGAIN;
988 if (!(request->fl_flags & FL_SLEEP))
989 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400990 /*
991 * Deadlock detection and insertion into the blocked
992 * locks list must be done while holding the same lock!
993 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400995 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400996 if (likely(!posix_locks_deadlock(request, fl))) {
997 error = FILE_LOCK_DEFERRED;
998 __locks_insert_block(fl, request);
999 }
Jeff Layton7b2296a2013-06-21 08:58:20 -04001000 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 goto out;
1002 }
1003 }
1004
1005 /* If we're just looking for a conflict, we're done. */
1006 error = 0;
1007 if (request->fl_flags & FL_ACCESS)
1008 goto out;
1009
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001010 /* Find the first old lock with the same owner as the new lock */
1011 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1012 if (posix_same_owner(request, fl))
1013 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 }
1015
Jeff Layton1cb36012013-06-21 08:58:12 -04001016 /* Process locks with this owner. */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001017 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1018 if (!posix_same_owner(request, fl))
1019 break;
1020
1021 /* Detect adjacent or overlapping regions (if same lock type) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -07001023 /* In all comparisons of start vs end, use
1024 * "start - 1" rather than "end + 1". If end
1025 * is OFFSET_MAX, end + 1 will become negative.
1026 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 if (fl->fl_end < request->fl_start - 1)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001028 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 /* If the next lock in the list has entirely bigger
1030 * addresses than the new one, insert the lock here.
1031 */
Olaf Kirch449231d2005-08-25 16:25:35 -07001032 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 break;
1034
1035 /* If we come here, the new and old lock are of the
1036 * same type and adjacent or overlapping. Make one
1037 * lock yielding from the lower start address of both
1038 * locks to the higher end address.
1039 */
1040 if (fl->fl_start > request->fl_start)
1041 fl->fl_start = request->fl_start;
1042 else
1043 request->fl_start = fl->fl_start;
1044 if (fl->fl_end < request->fl_end)
1045 fl->fl_end = request->fl_end;
1046 else
1047 request->fl_end = fl->fl_end;
1048 if (added) {
Jeff Layton9bd0f452015-01-16 15:05:57 -05001049 locks_delete_lock_ctx(fl, &ctx->flc_posix_cnt,
1050 &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 continue;
1052 }
1053 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001054 added = true;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001055 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 /* Processing for different lock types is a bit
1057 * more complex.
1058 */
1059 if (fl->fl_end < request->fl_start)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001060 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 if (fl->fl_start > request->fl_end)
1062 break;
1063 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001064 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 if (fl->fl_start < request->fl_start)
1066 left = fl;
1067 /* If the next lock in the list has a higher end
1068 * address than the new one, insert the new one here.
1069 */
1070 if (fl->fl_end > request->fl_end) {
1071 right = fl;
1072 break;
1073 }
1074 if (fl->fl_start >= request->fl_start) {
1075 /* The new lock completely replaces an old
1076 * one (This may happen several times).
1077 */
1078 if (added) {
Jeff Layton9bd0f452015-01-16 15:05:57 -05001079 locks_delete_lock_ctx(fl,
1080 &ctx->flc_posix_cnt, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 continue;
1082 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001083 /*
1084 * Replace the old lock with new_fl, and
1085 * remove the old one. It's safe to do the
1086 * insert here since we know that we won't be
1087 * using new_fl later, and that the lock is
1088 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001089 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001090 error = -ENOLCK;
1091 if (!new_fl)
1092 goto out;
1093 locks_copy_lock(new_fl, request);
1094 request = new_fl;
1095 new_fl = NULL;
Jeff Layton9bd0f452015-01-16 15:05:57 -05001096 locks_insert_lock_ctx(request,
1097 &ctx->flc_posix_cnt, &fl->fl_list);
1098 locks_delete_lock_ctx(fl,
1099 &ctx->flc_posix_cnt, &dispose);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001100 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 }
1102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 }
1104
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001105 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001106 * The above code only modifies existing locks in case of merging or
1107 * replacing. If new lock(s) need to be inserted all modifications are
1108 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001109 */
1110 error = -ENOLCK; /* "no luck" */
1111 if (right && left == right && !new_fl2)
1112 goto out;
1113
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 error = 0;
1115 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001116 if (request->fl_type == F_UNLCK) {
1117 if (request->fl_flags & FL_EXISTS)
1118 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001120 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001121
1122 if (!new_fl) {
1123 error = -ENOLCK;
1124 goto out;
1125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 locks_copy_lock(new_fl, request);
Jeff Layton9bd0f452015-01-16 15:05:57 -05001127 locks_insert_lock_ctx(new_fl, &ctx->flc_posix_cnt,
1128 &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 new_fl = NULL;
1130 }
1131 if (right) {
1132 if (left == right) {
1133 /* The new lock breaks the old one in two pieces,
1134 * so we have to use the second new lock.
1135 */
1136 left = new_fl2;
1137 new_fl2 = NULL;
1138 locks_copy_lock(left, right);
Jeff Layton9bd0f452015-01-16 15:05:57 -05001139 locks_insert_lock_ctx(left, &ctx->flc_posix_cnt,
1140 &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 }
1142 right->fl_start = request->fl_end + 1;
1143 locks_wake_up_blocks(right);
1144 }
1145 if (left) {
1146 left->fl_end = request->fl_start - 1;
1147 locks_wake_up_blocks(left);
1148 }
1149 out:
Jeff Layton6109c852015-01-16 15:05:57 -05001150 spin_unlock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 /*
1152 * Free any unused locks.
1153 */
1154 if (new_fl)
1155 locks_free_lock(new_fl);
1156 if (new_fl2)
1157 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001158 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 return error;
1160}
1161
1162/**
1163 * posix_lock_file - Apply a POSIX-style lock to a file
1164 * @filp: The file to apply the lock to
1165 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001166 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 *
1168 * Add a POSIX style lock to a file.
1169 * We merge adjacent & overlapping locks whenever possible.
1170 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001171 *
1172 * Note that if called with an FL_EXISTS argument, the caller may determine
1173 * whether or not a lock was successfully freed by testing the return
1174 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 */
Marc Eshel150b3932007-01-18 16:15:35 -05001176int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001177 struct file_lock *conflock)
1178{
Al Viro496ad9a2013-01-23 17:07:38 -05001179 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001180}
Marc Eshel150b3932007-01-18 16:15:35 -05001181EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183/**
1184 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1185 * @filp: The file to apply the lock to
1186 * @fl: The lock to be applied
1187 *
1188 * Add a POSIX style lock to a file.
1189 * We merge adjacent & overlapping locks whenever possible.
1190 * POSIX locks are sorted by owner task, then by starting address
1191 */
1192int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1193{
1194 int error;
1195 might_sleep ();
1196 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001197 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001198 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 break;
1200 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1201 if (!error)
1202 continue;
1203
1204 locks_delete_block(fl);
1205 break;
1206 }
1207 return error;
1208}
1209EXPORT_SYMBOL(posix_lock_file_wait);
1210
1211/**
1212 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001213 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 *
1215 * Searches the inode's list of locks to find any POSIX locks which conflict.
1216 * This function is called from locks_verify_locked() only.
1217 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001218int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001220 int ret;
Jeff Laytond7a06982014-03-10 09:54:15 -04001221 struct inode *inode = file_inode(file);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001222 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 struct file_lock *fl;
1224
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001225 ctx = inode->i_flctx;
1226 if (!ctx || list_empty_careful(&ctx->flc_posix))
1227 return 0;
1228
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 /*
1230 * Search the lock list for this inode for any POSIX locks.
1231 */
Jeff Layton6109c852015-01-16 15:05:57 -05001232 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001233 ret = 0;
1234 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001235 if (fl->fl_owner != current->files &&
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001236 fl->fl_owner != file) {
1237 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 break;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001239 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 }
Jeff Layton6109c852015-01-16 15:05:57 -05001241 spin_unlock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001242 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243}
1244
1245/**
1246 * locks_mandatory_area - Check for a conflicting lock
1247 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1248 * for shared
1249 * @inode: the file to check
1250 * @filp: how the file was opened (if it was)
1251 * @offset: start of area to check
1252 * @count: length of area to check
1253 *
1254 * Searches the inode's list of locks to find any POSIX locks which conflict.
1255 * This function is called from rw_verify_area() and
1256 * locks_verify_truncate().
1257 */
1258int locks_mandatory_area(int read_write, struct inode *inode,
1259 struct file *filp, loff_t offset,
1260 size_t count)
1261{
1262 struct file_lock fl;
1263 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001264 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 fl.fl_pid = current->tgid;
1268 fl.fl_file = filp;
1269 fl.fl_flags = FL_POSIX | FL_ACCESS;
1270 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001271 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1273 fl.fl_start = offset;
1274 fl.fl_end = offset + count - 1;
1275
1276 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001277 if (filp) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001278 fl.fl_owner = filp;
Jeff Layton29723ad2014-03-10 09:54:19 -04001279 fl.fl_flags &= ~FL_SLEEP;
1280 error = __posix_lock_file(inode, &fl, NULL);
1281 if (!error)
1282 break;
1283 }
1284
1285 if (sleep)
1286 fl.fl_flags |= FL_SLEEP;
1287 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001288 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001289 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 break;
1291 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1292 if (!error) {
1293 /*
1294 * If we've been sleeping someone might have
1295 * changed the permissions behind our back.
1296 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001297 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 continue;
1299 }
1300
1301 locks_delete_block(&fl);
1302 break;
1303 }
1304
1305 return error;
1306}
1307
1308EXPORT_SYMBOL(locks_mandatory_area);
1309
J. Bruce Fields778fc542011-07-26 18:25:49 -04001310static void lease_clear_pending(struct file_lock *fl, int arg)
1311{
1312 switch (arg) {
1313 case F_UNLCK:
1314 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1315 /* fall through: */
1316 case F_RDLCK:
1317 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1318 }
1319}
1320
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321/* We already had a lease on this file; just change its type */
Jeff Layton7448cc32015-01-16 15:05:57 -05001322int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323{
Jeff Layton9bd0f452015-01-16 15:05:57 -05001324 struct file_lock_context *flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 int error = assign_type(fl, arg);
1326
1327 if (error)
1328 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001329 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001331 if (arg == F_UNLCK) {
1332 struct file *filp = fl->fl_file;
1333
Jeff Layton9bd0f452015-01-16 15:05:57 -05001334 flctx = file_inode(filp)->i_flctx;
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001335 f_delown(filp);
1336 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001337 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1338 if (fl->fl_fasync != NULL) {
1339 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1340 fl->fl_fasync = NULL;
1341 }
Jeff Layton9bd0f452015-01-16 15:05:57 -05001342 locks_delete_lock_ctx(fl, &flctx->flc_lease_cnt, dispose);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001343 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 return 0;
1345}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346EXPORT_SYMBOL(lease_modify);
1347
J. Bruce Fields778fc542011-07-26 18:25:49 -04001348static bool past_time(unsigned long then)
1349{
1350 if (!then)
1351 /* 0 is a special value meaning "this never expires": */
1352 return false;
1353 return time_after(jiffies, then);
1354}
1355
Jeff Laytonc45198e2014-09-01 07:12:07 -04001356static void time_out_leases(struct inode *inode, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
Jeff Layton8634b512015-01-16 15:05:55 -05001358 struct file_lock_context *ctx = inode->i_flctx;
1359 struct file_lock *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360
Jeff Layton6109c852015-01-16 15:05:57 -05001361 lockdep_assert_held(&ctx->flc_lock);
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001362
Jeff Layton8634b512015-01-16 15:05:55 -05001363 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001364 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001365 if (past_time(fl->fl_downgrade_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001366 lease_modify(fl, F_RDLCK, dispose);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001367 if (past_time(fl->fl_break_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001368 lease_modify(fl, F_UNLCK, dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 }
1370}
1371
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001372static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1373{
1374 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1375 return false;
1376 return locks_conflict(breaker, lease);
1377}
1378
Jeff Layton03d12dd2014-09-01 14:53:41 -04001379static bool
1380any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1381{
Jeff Layton8634b512015-01-16 15:05:55 -05001382 struct file_lock_context *ctx = inode->i_flctx;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001383 struct file_lock *fl;
1384
Jeff Layton6109c852015-01-16 15:05:57 -05001385 lockdep_assert_held(&ctx->flc_lock);
Jeff Layton03d12dd2014-09-01 14:53:41 -04001386
Jeff Layton8634b512015-01-16 15:05:55 -05001387 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Jeff Layton03d12dd2014-09-01 14:53:41 -04001388 if (leases_conflict(fl, breaker))
1389 return true;
1390 }
1391 return false;
1392}
1393
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394/**
1395 * __break_lease - revoke all outstanding leases on file
1396 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001397 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1398 * break all leases
1399 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1400 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 *
david m. richter87250dd2007-05-09 16:10:27 -04001402 * break_lease (inlined for speed) has checked there already is at least
1403 * some kind of lock (maybe a lease) on this file. Leases are broken on
1404 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 * specified %O_NONBLOCK to your open().
1406 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001407int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001409 int error = 0;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001410 struct file_lock *new_fl;
Jeff Layton8634b512015-01-16 15:05:55 -05001411 struct file_lock_context *ctx = inode->i_flctx;
1412 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 unsigned long break_time;
Al Viro8737c932009-12-24 06:47:55 -05001414 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001415 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Al Viro8737c932009-12-24 06:47:55 -05001417 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001418 if (IS_ERR(new_fl))
1419 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001420 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Jeff Layton8634b512015-01-16 15:05:55 -05001422 /* typically we will check that ctx is non-NULL before calling */
1423 if (!ctx) {
1424 WARN_ON_ONCE(1);
1425 return error;
1426 }
1427
Jeff Layton6109c852015-01-16 15:05:57 -05001428 spin_lock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429
Jeff Laytonc45198e2014-09-01 07:12:07 -04001430 time_out_leases(inode, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431
Jeff Layton03d12dd2014-09-01 14:53:41 -04001432 if (!any_leases_conflict(inode, new_fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001433 goto out;
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 break_time = 0;
1436 if (lease_break_time > 0) {
1437 break_time = jiffies + lease_break_time * HZ;
1438 if (break_time == 0)
1439 break_time++; /* so that 0 means no break time */
1440 }
1441
Jeff Layton8634b512015-01-16 15:05:55 -05001442 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001443 if (!leases_conflict(fl, new_fl))
1444 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001445 if (want_write) {
1446 if (fl->fl_flags & FL_UNLOCK_PENDING)
1447 continue;
1448 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001450 } else {
Jeff Layton8634b512015-01-16 15:05:55 -05001451 if (lease_breaking(fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001452 continue;
1453 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1454 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 }
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001456 if (fl->fl_lmops->lm_break(fl))
Jeff Layton9bd0f452015-01-16 15:05:57 -05001457 locks_delete_lock_ctx(fl, &ctx->flc_lease_cnt,
1458 &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 }
1460
Jeff Layton8634b512015-01-16 15:05:55 -05001461 if (list_empty(&ctx->flc_lease))
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001462 goto out;
1463
Jeff Layton843c6b22014-09-01 14:27:43 -04001464 if (mode & O_NONBLOCK) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001465 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 error = -EWOULDBLOCK;
1467 goto out;
1468 }
1469
1470restart:
Jeff Layton8634b512015-01-16 15:05:55 -05001471 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1472 break_time = fl->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001473 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001475 if (break_time == 0)
1476 break_time++;
Jeff Layton8634b512015-01-16 15:05:55 -05001477 locks_insert_block(fl, new_fl);
Jeff Layton62af4f12014-05-09 14:13:05 -04001478 trace_break_lease_block(inode, new_fl);
Jeff Layton6109c852015-01-16 15:05:57 -05001479 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001480 locks_dispose_list(&dispose);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001481 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1482 !new_fl->fl_next, break_time);
Jeff Layton6109c852015-01-16 15:05:57 -05001483 spin_lock(&ctx->flc_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001484 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001485 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (error >= 0) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001487 /*
1488 * Wait for the next conflicting lease that has not been
1489 * broken yet
1490 */
Jeff Layton03d12dd2014-09-01 14:53:41 -04001491 if (error == 0)
1492 time_out_leases(inode, &dispose);
1493 if (any_leases_conflict(inode, new_fl))
1494 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 error = 0;
1496 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497out:
Jeff Layton6109c852015-01-16 15:05:57 -05001498 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001499 locks_dispose_list(&dispose);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001500 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 return error;
1502}
1503
1504EXPORT_SYMBOL(__break_lease);
1505
1506/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001507 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 * @inode: the inode
1509 * @time: pointer to a timespec which will contain the last modified time
1510 *
1511 * This is to force NFS clients to flush their caches for files with
1512 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001513 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 */
1515void lease_get_mtime(struct inode *inode, struct timespec *time)
1516{
Jeff Laytonbfe86022014-08-22 10:18:44 -04001517 bool has_lease = false;
Jeff Layton8634b512015-01-16 15:05:55 -05001518 struct file_lock_context *ctx = inode->i_flctx;
1519 struct file_lock *fl;
Jeff Laytonbfe86022014-08-22 10:18:44 -04001520
Jeff Layton8634b512015-01-16 15:05:55 -05001521 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001522 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001523 if (!list_empty(&ctx->flc_lease)) {
1524 fl = list_first_entry(&ctx->flc_lease,
1525 struct file_lock, fl_list);
1526 if (fl->fl_type == F_WRLCK)
1527 has_lease = true;
1528 }
Jeff Layton6109c852015-01-16 15:05:57 -05001529 spin_unlock(&ctx->flc_lock);
Jeff Laytonbfe86022014-08-22 10:18:44 -04001530 }
1531
1532 if (has_lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 *time = current_fs_time(inode->i_sb);
1534 else
1535 *time = inode->i_mtime;
1536}
1537
1538EXPORT_SYMBOL(lease_get_mtime);
1539
1540/**
1541 * fcntl_getlease - Enquire what lease is currently active
1542 * @filp: the file
1543 *
1544 * The value returned by this function will be one of
1545 * (if no lease break is pending):
1546 *
1547 * %F_RDLCK to indicate a shared lease is held.
1548 *
1549 * %F_WRLCK to indicate an exclusive lease is held.
1550 *
1551 * %F_UNLCK to indicate no lease is held.
1552 *
1553 * (if a lease break is pending):
1554 *
1555 * %F_RDLCK to indicate an exclusive lease needs to be
1556 * changed to a shared lease (or removed).
1557 *
1558 * %F_UNLCK to indicate the lease needs to be removed.
1559 *
1560 * XXX: sfr & willy disagree over whether F_INPROGRESS
1561 * should be returned to userspace.
1562 */
1563int fcntl_getlease(struct file *filp)
1564{
1565 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001566 struct inode *inode = file_inode(filp);
Jeff Layton8634b512015-01-16 15:05:55 -05001567 struct file_lock_context *ctx = inode->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 int type = F_UNLCK;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001569 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Jeff Layton8634b512015-01-16 15:05:55 -05001571 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001572 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001573 time_out_leases(file_inode(filp), &dispose);
1574 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1575 if (fl->fl_file != filp)
1576 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001577 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 break;
1579 }
Jeff Layton6109c852015-01-16 15:05:57 -05001580 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001581 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 return type;
1584}
1585
Jeff Layton24cbe782014-02-03 12:13:06 -05001586/**
1587 * check_conflicting_open - see if the given dentry points to a file that has
1588 * an existing open that would conflict with the
1589 * desired lease.
1590 * @dentry: dentry to check
1591 * @arg: type of lease that we're trying to acquire
1592 *
1593 * Check to see if there's an existing open fd on this file that would
1594 * conflict with the lease we're trying to set.
1595 */
1596static int
1597check_conflicting_open(const struct dentry *dentry, const long arg)
1598{
1599 int ret = 0;
1600 struct inode *inode = dentry->d_inode;
1601
1602 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1603 return -EAGAIN;
1604
1605 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1606 (atomic_read(&inode->i_count) > 1)))
1607 ret = -EAGAIN;
1608
1609 return ret;
1610}
1611
Jeff Laytone6f5c782014-08-22 10:40:25 -04001612static int
1613generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614{
Jeff Layton8634b512015-01-16 15:05:55 -05001615 struct file_lock *fl, *my_fl = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001616 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001618 struct file_lock_context *ctx;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001619 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001620 int error;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001621 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
J. Bruce Fields096657b2010-10-30 17:31:14 -04001623 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001624 trace_generic_add_lease(inode, lease);
1625
Jeff Layton8634b512015-01-16 15:05:55 -05001626 ctx = locks_get_lock_context(inode);
1627 if (!ctx)
1628 return -ENOMEM;
1629
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001630 /*
1631 * In the delegation case we need mutual exclusion with
1632 * a number of operations that take the i_mutex. We trylock
1633 * because delegations are an optional optimization, and if
1634 * there's some chance of a conflict--we'd rather not
1635 * bother, maybe that's a sign this just isn't a good file to
1636 * hand out a delegation on.
1637 */
1638 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1639 return -EAGAIN;
1640
1641 if (is_deleg && arg == F_WRLCK) {
1642 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001643 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001644 WARN_ON_ONCE(1);
1645 return -EINVAL;
1646 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001647
Jeff Layton6109c852015-01-16 15:05:57 -05001648 spin_lock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001649 time_out_leases(inode, &dispose);
Jeff Layton24cbe782014-02-03 12:13:06 -05001650 error = check_conflicting_open(dentry, arg);
1651 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001652 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001653
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 /*
1655 * At this point, we know that if there is an exclusive
1656 * lease on this file, then we hold it on this filp
1657 * (otherwise our open of this file would have blocked).
1658 * And if we are trying to acquire an exclusive lease,
1659 * then the file is not open by anyone (including us)
1660 * except for this filp.
1661 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001662 error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001663 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001664 if (fl->fl_file == filp &&
1665 fl->fl_owner == lease->fl_owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001666 my_fl = fl;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001667 continue;
1668 }
Jeff Layton8634b512015-01-16 15:05:55 -05001669
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001670 /*
1671 * No exclusive leases if someone else has a lease on
1672 * this file:
1673 */
1674 if (arg == F_WRLCK)
1675 goto out;
1676 /*
1677 * Modifying our existing lease is OK, but no getting a
1678 * new lease if someone else is opening for write:
1679 */
1680 if (fl->fl_flags & FL_UNLOCK_PENDING)
1681 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 }
1683
Jeff Layton8634b512015-01-16 15:05:55 -05001684 if (my_fl != NULL) {
Jeff Layton7448cc32015-01-16 15:05:57 -05001685 error = lease->fl_lmops->lm_change(my_fl, arg, &dispose);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001686 if (error)
1687 goto out;
1688 goto out_setup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 }
1690
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691 error = -EINVAL;
1692 if (!leases_enable)
1693 goto out;
1694
Jeff Layton9bd0f452015-01-16 15:05:57 -05001695 locks_insert_lock_ctx(lease, &ctx->flc_lease_cnt, &ctx->flc_lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001696 /*
1697 * The check in break_lease() is lockless. It's possible for another
1698 * open to race in after we did the earlier check for a conflicting
1699 * open but before the lease was inserted. Check again for a
1700 * conflicting open and cancel the lease if there is one.
1701 *
1702 * We also add a barrier here to ensure that the insertion of the lock
1703 * precedes these checks.
1704 */
1705 smp_mb();
1706 error = check_conflicting_open(dentry, arg);
Jeff Layton8634b512015-01-16 15:05:55 -05001707 if (error) {
Jeff Layton9bd0f452015-01-16 15:05:57 -05001708 locks_unlink_lock_ctx(lease, &ctx->flc_lease_cnt);
Jeff Layton8634b512015-01-16 15:05:55 -05001709 goto out;
1710 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001711
1712out_setup:
1713 if (lease->fl_lmops->lm_setup)
1714 lease->fl_lmops->lm_setup(lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715out:
Jeff Layton6109c852015-01-16 15:05:57 -05001716 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001717 locks_dispose_list(&dispose);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001718 if (is_deleg)
1719 mutex_unlock(&inode->i_mutex);
Jeff Layton8634b512015-01-16 15:05:55 -05001720 if (!error && !my_fl)
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001721 *flp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 return error;
1723}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001724
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001725static int generic_delete_lease(struct file *filp, void *owner)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001726{
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001727 int error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001728 struct file_lock *fl, *victim = NULL;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001729 struct dentry *dentry = filp->f_path.dentry;
1730 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001731 struct file_lock_context *ctx = inode->i_flctx;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001732 LIST_HEAD(dispose);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001733
Jeff Layton8634b512015-01-16 15:05:55 -05001734 if (!ctx) {
1735 trace_generic_delete_lease(inode, NULL);
1736 return error;
1737 }
1738
Jeff Layton6109c852015-01-16 15:05:57 -05001739 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001740 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001741 if (fl->fl_file == filp &&
1742 fl->fl_owner == owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001743 victim = fl;
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001744 break;
Jeff Layton8634b512015-01-16 15:05:55 -05001745 }
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001746 }
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001747 trace_generic_delete_lease(inode, fl);
Jeff Layton8634b512015-01-16 15:05:55 -05001748 if (victim)
Jeff Layton7448cc32015-01-16 15:05:57 -05001749 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05001750 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001751 locks_dispose_list(&dispose);
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001752 return error;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001753}
1754
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755/**
1756 * generic_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001757 * @filp: file pointer
1758 * @arg: type of lease to obtain
1759 * @flp: input - file_lock to use, output - file_lock inserted
1760 * @priv: private data for lm_setup (may be NULL if lm_setup
1761 * doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 *
1763 * The (input) flp->fl_lmops->lm_break function is required
1764 * by break_lease().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001766int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1767 void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 struct dentry *dentry = filp->f_path.dentry;
1770 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001771 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001773 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001774 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001776 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 error = security_file_lock(filp, arg);
1778 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001779 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001781 switch (arg) {
1782 case F_UNLCK:
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001783 return generic_delete_lease(filp, *priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001784 case F_RDLCK:
1785 case F_WRLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001786 if (!(*flp)->fl_lmops->lm_break) {
1787 WARN_ON_ONCE(1);
1788 return -ENOLCK;
1789 }
Jeff Laytone6f5c782014-08-22 10:40:25 -04001790 return generic_add_lease(filp, arg, flp, priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001791 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001792 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001795EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001797/**
Jeff Laytone51673a2014-08-22 18:13:28 -04001798 * vfs_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001799 * @filp: file pointer
1800 * @arg: type of lease to obtain
1801 * @lease: file_lock to use when adding a lease
1802 * @priv: private info for lm_setup when adding a lease (may be
1803 * NULL if lm_setup doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 *
Jeff Laytone51673a2014-08-22 18:13:28 -04001805 * Call this to establish a lease on the file. The "lease" argument is not
1806 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1807 * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1808 * if not, this function will return -ENOLCK (and generate a scary-looking
1809 * stack trace).
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001810 *
1811 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1812 * may be NULL if the lm_setup operation doesn't require it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001814int
1815vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001817 if (filp->f_op->setlease)
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001818 return filp->f_op->setlease(filp, arg, lease, priv);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001819 else
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001820 return generic_setlease(filp, arg, lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001822EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001824static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001826 struct file_lock *fl;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001827 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 int error;
1829
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001830 fl = lease_alloc(filp, arg);
1831 if (IS_ERR(fl))
1832 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001834 new = fasync_alloc();
1835 if (!new) {
1836 locks_free_lock(fl);
1837 return -ENOMEM;
1838 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001839 new->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001841 error = vfs_setlease(filp, arg, &fl, (void **)&new);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001842 if (fl)
1843 locks_free_lock(fl);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001844 if (new)
1845 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 return error;
1847}
1848
1849/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001850 * fcntl_setlease - sets a lease on an open file
1851 * @fd: open file descriptor
1852 * @filp: file pointer
1853 * @arg: type of lease to obtain
1854 *
1855 * Call this fcntl to establish a lease on the file.
1856 * Note that you also need to call %F_SETSIG to
1857 * receive a signal when the lease is broken.
1858 */
1859int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1860{
1861 if (arg == F_UNLCK)
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001862 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001863 return do_fcntl_add_lease(fd, filp, arg);
1864}
1865
1866/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1868 * @filp: The file to apply the lock to
1869 * @fl: The lock to be applied
1870 *
1871 * Add a FLOCK style lock to a file.
1872 */
1873int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1874{
1875 int error;
1876 might_sleep();
1877 for (;;) {
1878 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001879 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 break;
1881 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1882 if (!error)
1883 continue;
1884
1885 locks_delete_block(fl);
1886 break;
1887 }
1888 return error;
1889}
1890
1891EXPORT_SYMBOL(flock_lock_file_wait);
1892
1893/**
1894 * sys_flock: - flock() system call.
1895 * @fd: the file descriptor to lock.
1896 * @cmd: the type of lock to apply.
1897 *
1898 * Apply a %FL_FLOCK style lock to an open file descriptor.
1899 * The @cmd can be one of
1900 *
1901 * %LOCK_SH -- a shared lock.
1902 *
1903 * %LOCK_EX -- an exclusive lock.
1904 *
1905 * %LOCK_UN -- remove an existing lock.
1906 *
1907 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1908 *
1909 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1910 * processes read and write access respectively.
1911 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001912SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913{
Al Viro2903ff02012-08-28 12:52:22 -04001914 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 struct file_lock *lock;
1916 int can_sleep, unlock;
1917 int error;
1918
1919 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001920 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921 goto out;
1922
1923 can_sleep = !(cmd & LOCK_NB);
1924 cmd &= ~LOCK_NB;
1925 unlock = (cmd == LOCK_UN);
1926
Al Viroaeb5d722008-09-02 15:28:45 -04001927 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001928 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 goto out_putf;
1930
Jeff Layton6e129d002014-09-04 10:25:06 -04001931 lock = flock_make_lock(f.file, cmd);
1932 if (IS_ERR(lock)) {
1933 error = PTR_ERR(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 goto out_putf;
Jeff Layton6e129d002014-09-04 10:25:06 -04001935 }
1936
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 if (can_sleep)
1938 lock->fl_flags |= FL_SLEEP;
1939
Al Viro2903ff02012-08-28 12:52:22 -04001940 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 if (error)
1942 goto out_free;
1943
Al Viro72c2d532013-09-22 16:27:52 -04001944 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001945 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 (can_sleep) ? F_SETLKW : F_SETLK,
1947 lock);
1948 else
Al Viro2903ff02012-08-28 12:52:22 -04001949 error = flock_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950
1951 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001952 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001955 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 out:
1957 return error;
1958}
1959
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001960/**
1961 * vfs_test_lock - test file byte range lock
1962 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001963 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001964 *
1965 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1966 * setting conf->fl_type to something other than F_UNLCK.
1967 */
1968int vfs_test_lock(struct file *filp, struct file_lock *fl)
1969{
Al Viro72c2d532013-09-22 16:27:52 -04001970 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001971 return filp->f_op->lock(filp, F_GETLK, fl);
1972 posix_test_lock(filp, fl);
1973 return 0;
1974}
1975EXPORT_SYMBOL_GPL(vfs_test_lock);
1976
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001977static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1978{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001979 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001980#if BITS_PER_LONG == 32
1981 /*
1982 * Make sure we can represent the posix lock via
1983 * legacy 32bit flock.
1984 */
1985 if (fl->fl_start > OFFT_OFFSET_MAX)
1986 return -EOVERFLOW;
1987 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1988 return -EOVERFLOW;
1989#endif
1990 flock->l_start = fl->fl_start;
1991 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1992 fl->fl_end - fl->fl_start + 1;
1993 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001994 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001995 return 0;
1996}
1997
1998#if BITS_PER_LONG == 32
1999static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2000{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002001 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002002 flock->l_start = fl->fl_start;
2003 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2004 fl->fl_end - fl->fl_start + 1;
2005 flock->l_whence = 0;
2006 flock->l_type = fl->fl_type;
2007}
2008#endif
2009
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010/* Report the first existing lock that would conflict with l.
2011 * This implements the F_GETLK command of fcntl().
2012 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002013int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002015 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 struct flock flock;
2017 int error;
2018
2019 error = -EFAULT;
2020 if (copy_from_user(&flock, l, sizeof(flock)))
2021 goto out;
2022 error = -EINVAL;
2023 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2024 goto out;
2025
2026 error = flock_to_posix_lock(filp, &file_lock, &flock);
2027 if (error)
2028 goto out;
2029
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002030 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002031 error = -EINVAL;
2032 if (flock.l_pid != 0)
2033 goto out;
2034
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002035 cmd = F_GETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002036 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002037 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002038 }
2039
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002040 error = vfs_test_lock(filp, &file_lock);
2041 if (error)
2042 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002044 flock.l_type = file_lock.fl_type;
2045 if (file_lock.fl_type != F_UNLCK) {
2046 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002047 if (error)
Kinglong Meef3282962014-08-22 10:18:43 -04002048 goto rel_priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050 error = -EFAULT;
2051 if (!copy_to_user(l, &flock, sizeof(flock)))
2052 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002053rel_priv:
2054 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002055out:
2056 return error;
2057}
2058
Marc Eshel7723ec92007-01-18 15:08:55 -05002059/**
2060 * vfs_lock_file - file byte range lock
2061 * @filp: The file to apply the lock to
2062 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2063 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05002064 * @conf: Place to return a copy of the conflicting lock, if found.
2065 *
2066 * A caller that doesn't care about the conflicting lock may pass NULL
2067 * as the final argument.
2068 *
2069 * If the filesystem defines a private ->lock() method, then @conf will
2070 * be left unchanged; so a caller that cares should initialize it to
2071 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05002072 *
2073 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2074 * locks, the ->lock() interface may return asynchronously, before the lock has
2075 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002076 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05002077 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2078 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002079 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05002080 * request completes.
2081 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07002082 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2083 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05002084 * nonzero return code and the file system should release the lock. The file
2085 * system is also responsible to keep a corresponding posix lock when it
2086 * grants a lock so the VFS can find out which locks are locally held and do
2087 * the correct lock cleanup when required.
2088 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002089 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002090 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002091 */
Marc Eshel150b3932007-01-18 16:15:35 -05002092int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002093{
Al Viro72c2d532013-09-22 16:27:52 -04002094 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002095 return filp->f_op->lock(filp, cmd, fl);
2096 else
Marc Eshel150b3932007-01-18 16:15:35 -05002097 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002098}
2099EXPORT_SYMBOL_GPL(vfs_lock_file);
2100
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002101static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2102 struct file_lock *fl)
2103{
2104 int error;
2105
2106 error = security_file_lock(filp, fl->fl_type);
2107 if (error)
2108 return error;
2109
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002110 for (;;) {
2111 error = vfs_lock_file(filp, cmd, fl, NULL);
2112 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002113 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002114 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2115 if (!error)
2116 continue;
2117
2118 locks_delete_block(fl);
2119 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002120 }
2121
2122 return error;
2123}
2124
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002125/* Ensure that fl->fl_filp has compatible f_mode for F_SETLK calls */
2126static int
2127check_fmode_for_setlk(struct file_lock *fl)
2128{
2129 switch (fl->fl_type) {
2130 case F_RDLCK:
2131 if (!(fl->fl_file->f_mode & FMODE_READ))
2132 return -EBADF;
2133 break;
2134 case F_WRLCK:
2135 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2136 return -EBADF;
2137 }
2138 return 0;
2139}
2140
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141/* Apply the lock described by l to an open file descriptor.
2142 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2143 */
Peter Staubachc2936212005-07-27 11:45:09 -07002144int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2145 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
2147 struct file_lock *file_lock = locks_alloc_lock();
2148 struct flock flock;
2149 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002150 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002151 int error;
2152
2153 if (file_lock == NULL)
2154 return -ENOLCK;
2155
2156 /*
2157 * This might block, so we do it before checking the inode.
2158 */
2159 error = -EFAULT;
2160 if (copy_from_user(&flock, l, sizeof(flock)))
2161 goto out;
2162
Al Viro496ad9a2013-01-23 17:07:38 -05002163 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164
2165 /* Don't allow mandatory locks on files that may be memory mapped
2166 * and shared.
2167 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002168 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 error = -EAGAIN;
2170 goto out;
2171 }
2172
Peter Staubachc2936212005-07-27 11:45:09 -07002173again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 error = flock_to_posix_lock(filp, file_lock, &flock);
2175 if (error)
2176 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002177
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002178 error = check_fmode_for_setlk(file_lock);
2179 if (error)
2180 goto out;
2181
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002182 /*
2183 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002184 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002185 */
2186 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002187 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002188 error = -EINVAL;
2189 if (flock.l_pid != 0)
2190 goto out;
2191
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002192 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002193 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002194 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002195 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002196 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002197 error = -EINVAL;
2198 if (flock.l_pid != 0)
2199 goto out;
2200
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002201 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002202 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002203 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002204 /* Fallthrough */
2205 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 file_lock->fl_flags |= FL_SLEEP;
2207 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002208
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002209 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210
Peter Staubachc2936212005-07-27 11:45:09 -07002211 /*
2212 * Attempt to detect a close/fcntl race and recover by
2213 * releasing the lock that was just acquired.
2214 */
Al Viro0b2bac22008-05-06 13:58:34 -04002215 /*
2216 * we need that spin_lock here - it prevents reordering between
Jeff Layton8116bf42015-01-21 20:44:01 -05002217 * update of i_flctx->flc_posix and check for it done in close().
Al Viro0b2bac22008-05-06 13:58:34 -04002218 * rcu_read_lock() wouldn't do.
2219 */
2220 spin_lock(&current->files->file_lock);
2221 f = fcheck(fd);
2222 spin_unlock(&current->files->file_lock);
2223 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002224 flock.l_type = F_UNLCK;
2225 goto again;
2226 }
2227
2228out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 locks_free_lock(file_lock);
2230 return error;
2231}
2232
2233#if BITS_PER_LONG == 32
2234/* Report the first existing lock that would conflict with l.
2235 * This implements the F_GETLK command of fcntl().
2236 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002237int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002239 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002240 struct flock64 flock;
2241 int error;
2242
2243 error = -EFAULT;
2244 if (copy_from_user(&flock, l, sizeof(flock)))
2245 goto out;
2246 error = -EINVAL;
2247 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2248 goto out;
2249
2250 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2251 if (error)
2252 goto out;
2253
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002254 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002255 error = -EINVAL;
2256 if (flock.l_pid != 0)
2257 goto out;
2258
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002259 cmd = F_GETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002260 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002261 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002262 }
2263
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002264 error = vfs_test_lock(filp, &file_lock);
2265 if (error)
2266 goto out;
2267
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002268 flock.l_type = file_lock.fl_type;
2269 if (file_lock.fl_type != F_UNLCK)
2270 posix_lock_to_flock64(&flock, &file_lock);
2271
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 error = -EFAULT;
2273 if (!copy_to_user(l, &flock, sizeof(flock)))
2274 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002275
2276 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277out:
2278 return error;
2279}
2280
2281/* Apply the lock described by l to an open file descriptor.
2282 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2283 */
Peter Staubachc2936212005-07-27 11:45:09 -07002284int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2285 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286{
2287 struct file_lock *file_lock = locks_alloc_lock();
2288 struct flock64 flock;
2289 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002290 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 int error;
2292
2293 if (file_lock == NULL)
2294 return -ENOLCK;
2295
2296 /*
2297 * This might block, so we do it before checking the inode.
2298 */
2299 error = -EFAULT;
2300 if (copy_from_user(&flock, l, sizeof(flock)))
2301 goto out;
2302
Al Viro496ad9a2013-01-23 17:07:38 -05002303 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
2305 /* Don't allow mandatory locks on files that may be memory mapped
2306 * and shared.
2307 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002308 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 error = -EAGAIN;
2310 goto out;
2311 }
2312
Peter Staubachc2936212005-07-27 11:45:09 -07002313again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 error = flock64_to_posix_lock(filp, file_lock, &flock);
2315 if (error)
2316 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002317
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002318 error = check_fmode_for_setlk(file_lock);
2319 if (error)
2320 goto out;
2321
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002322 /*
2323 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002324 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002325 */
2326 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002327 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002328 error = -EINVAL;
2329 if (flock.l_pid != 0)
2330 goto out;
2331
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002332 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002333 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002334 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002335 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002336 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002337 error = -EINVAL;
2338 if (flock.l_pid != 0)
2339 goto out;
2340
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002341 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002342 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002343 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002344 /* Fallthrough */
2345 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 file_lock->fl_flags |= FL_SLEEP;
2347 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002348
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002349 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350
Peter Staubachc2936212005-07-27 11:45:09 -07002351 /*
2352 * Attempt to detect a close/fcntl race and recover by
2353 * releasing the lock that was just acquired.
2354 */
Al Viro0b2bac22008-05-06 13:58:34 -04002355 spin_lock(&current->files->file_lock);
2356 f = fcheck(fd);
2357 spin_unlock(&current->files->file_lock);
2358 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002359 flock.l_type = F_UNLCK;
2360 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 }
2362
2363out:
2364 locks_free_lock(file_lock);
2365 return error;
2366}
2367#endif /* BITS_PER_LONG == 32 */
2368
2369/*
2370 * This function is called when the file is being removed
2371 * from the task's fd array. POSIX locks belonging to this task
2372 * are deleted at this time.
2373 */
2374void locks_remove_posix(struct file *filp, fl_owner_t owner)
2375{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002376 struct file_lock lock;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002377 struct file_lock_context *ctx = file_inode(filp)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378
2379 /*
2380 * If there are no locks held on this file, we don't need to call
2381 * posix_lock_file(). Another process could be setting a lock on this
2382 * file at the same time, but we wouldn't remove that lock anyway.
2383 */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002384 if (!ctx || list_empty(&ctx->flc_posix))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 return;
2386
2387 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002388 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389 lock.fl_start = 0;
2390 lock.fl_end = OFFSET_MAX;
2391 lock.fl_owner = owner;
2392 lock.fl_pid = current->tgid;
2393 lock.fl_file = filp;
2394 lock.fl_ops = NULL;
2395 lock.fl_lmops = NULL;
2396
Marc Eshel150b3932007-01-18 16:15:35 -05002397 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2400 lock.fl_ops->fl_release_private(&lock);
2401}
2402
2403EXPORT_SYMBOL(locks_remove_posix);
2404
Jeff Layton3d8e5602015-01-16 15:05:58 -05002405/* The i_flctx must be valid when calling into here */
Jeff Laytondd459bb2015-01-16 15:05:54 -05002406static void
2407locks_remove_flock(struct file *filp)
2408{
2409 struct file_lock fl = {
2410 .fl_owner = filp,
2411 .fl_pid = current->tgid,
2412 .fl_file = filp,
2413 .fl_flags = FL_FLOCK,
2414 .fl_type = F_UNLCK,
2415 .fl_end = OFFSET_MAX,
2416 };
Jeff Layton5263e312015-01-16 15:05:55 -05002417 struct file_lock_context *flctx = file_inode(filp)->i_flctx;
Jeff Laytondd459bb2015-01-16 15:05:54 -05002418
Jeff Layton3d8e5602015-01-16 15:05:58 -05002419 if (list_empty(&flctx->flc_flock))
Jeff Laytondd459bb2015-01-16 15:05:54 -05002420 return;
2421
2422 if (filp->f_op->flock)
2423 filp->f_op->flock(filp, F_SETLKW, &fl);
2424 else
2425 flock_lock_file(filp, &fl);
2426
2427 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2428 fl.fl_ops->fl_release_private(&fl);
2429}
2430
Jeff Layton3d8e5602015-01-16 15:05:58 -05002431/* The i_flctx must be valid when calling into here */
Jeff Layton8634b512015-01-16 15:05:55 -05002432static void
2433locks_remove_lease(struct file *filp)
2434{
2435 struct inode *inode = file_inode(filp);
2436 struct file_lock_context *ctx = inode->i_flctx;
2437 struct file_lock *fl, *tmp;
2438 LIST_HEAD(dispose);
2439
Jeff Layton3d8e5602015-01-16 15:05:58 -05002440 if (list_empty(&ctx->flc_lease))
Jeff Layton8634b512015-01-16 15:05:55 -05002441 return;
2442
Jeff Layton6109c852015-01-16 15:05:57 -05002443 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002444 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
Jeff Layton7448cc32015-01-16 15:05:57 -05002445 lease_modify(fl, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05002446 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002447 locks_dispose_list(&dispose);
2448}
2449
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450/*
2451 * This function is called on the last close of an open file.
2452 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002453void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454{
Jeff Layton3d8e5602015-01-16 15:05:58 -05002455 if (!file_inode(filp)->i_flctx)
2456 return;
2457
Jeff Laytondd459bb2015-01-16 15:05:54 -05002458 /* remove any OFD locks */
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002459 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002460
Jeff Laytondd459bb2015-01-16 15:05:54 -05002461 /* remove flock locks */
2462 locks_remove_flock(filp);
2463
Jeff Layton8634b512015-01-16 15:05:55 -05002464 /* remove any leases */
2465 locks_remove_lease(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466}
2467
2468/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 * @waiter: the lock which was waiting
2471 *
2472 * lockd needs to block waiting for locks.
2473 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002474int
Jeff Laytonf891a292013-06-21 08:58:09 -04002475posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002477 int status = 0;
2478
Jeff Layton7b2296a2013-06-21 08:58:20 -04002479 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002480 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002482 else
2483 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002484 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002485 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487EXPORT_SYMBOL(posix_unblock_lock);
2488
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002489/**
2490 * vfs_cancel_lock - file byte range unblock lock
2491 * @filp: The file to apply the unblock to
2492 * @fl: The lock to be unblocked
2493 *
2494 * Used by lock managers to cancel blocked requests
2495 */
2496int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2497{
Al Viro72c2d532013-09-22 16:27:52 -04002498 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002499 return filp->f_op->lock(filp, F_CANCELLK, fl);
2500 return 0;
2501}
2502
2503EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2504
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002505#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002506#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002507#include <linux/seq_file.h>
2508
Jeff Layton7012b022013-06-21 08:58:22 -04002509struct locks_iterator {
2510 int li_cpu;
2511 loff_t li_pos;
2512};
2513
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002514static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002515 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516{
2517 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002518 unsigned int fl_pid;
2519
2520 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002521 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002522 else
2523 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524
2525 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002526 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
Jerome Marchand99dc8292010-10-26 14:22:33 -07002528 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002530 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002531 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002532 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002533 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002534 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002535 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002536
2537 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002539 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 } else if (IS_FLOCK(fl)) {
2541 if (fl->fl_type & LOCK_MAND) {
Fabian Frederick5315c262014-05-09 14:13:05 -04002542 seq_puts(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002544 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 }
2546 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002547 if (fl->fl_flags & FL_DELEG)
2548 seq_puts(f, "DELEG ");
2549 else
2550 seq_puts(f, "LEASE ");
2551
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002552 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002553 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002555 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002557 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002559 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 }
2561 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002562 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 (fl->fl_type & LOCK_READ)
2564 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2565 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2566 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002567 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002568 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002569 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2570 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 }
2572 if (inode) {
2573#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002574 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 inode->i_sb->s_id, inode->i_ino);
2576#else
2577 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002578 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002579 MAJOR(inode->i_sb->s_dev),
2580 MINOR(inode->i_sb->s_dev), inode->i_ino);
2581#endif
2582 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002583 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584 }
2585 if (IS_POSIX(fl)) {
2586 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002587 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002589 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002590 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002591 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 }
2593}
2594
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002595static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
Jeff Layton7012b022013-06-21 08:58:22 -04002597 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002598 struct file_lock *fl, *bfl;
2599
Jeff Layton139ca042013-06-21 08:58:17 -04002600 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002601
Jeff Layton7012b022013-06-21 08:58:22 -04002602 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002603
2604 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002605 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002606
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002607 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002608}
2609
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002610static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002611 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612{
Jeff Layton7012b022013-06-21 08:58:22 -04002613 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002614
Jeff Layton7012b022013-06-21 08:58:22 -04002615 iter->li_pos = *pos + 1;
2616 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002617 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002618 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619}
2620
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002621static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2622{
Jeff Layton7012b022013-06-21 08:58:22 -04002623 struct locks_iterator *iter = f->private;
2624
2625 ++iter->li_pos;
2626 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002627}
2628
2629static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002630 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002631{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002632 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002633 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002634}
2635
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002636static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002637 .start = locks_start,
2638 .next = locks_next,
2639 .stop = locks_stop,
2640 .show = locks_show,
2641};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002642
2643static int locks_open(struct inode *inode, struct file *filp)
2644{
Jeff Layton7012b022013-06-21 08:58:22 -04002645 return seq_open_private(filp, &locks_seq_operations,
2646 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002647}
2648
2649static const struct file_operations proc_locks_operations = {
2650 .open = locks_open,
2651 .read = seq_read,
2652 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002653 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002654};
2655
2656static int __init proc_locks_init(void)
2657{
2658 proc_create("locks", 0, NULL, &proc_locks_operations);
2659 return 0;
2660}
2661module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002662#endif
2663
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664static int __init filelock_init(void)
2665{
Jeff Layton7012b022013-06-21 08:58:22 -04002666 int i;
2667
Jeff Layton4a075e32015-01-16 15:05:54 -05002668 flctx_cache = kmem_cache_create("file_lock_ctx",
2669 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2670
Linus Torvalds1da177e2005-04-16 15:20:36 -07002671 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002672 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2673
Jeff Layton7012b022013-06-21 08:58:22 -04002674 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2675
2676 for_each_possible_cpu(i)
2677 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2678
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 return 0;
2680}
2681
2682core_initcall(filelock_init);