blob: 528fedfda15e6432bd69b80c7bd8faceb7d1351d [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)
Christoph Hellwig11afe9f2015-01-21 19:17:03 +0100140#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
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 Laytone084c1b2015-02-16 14:32:03 -0500684locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
Jeff Layton5263e312015-01-16 15:05:55 -0500685{
686 fl->fl_nspid = get_pid(task_tgid(current));
687 list_add_tail(&fl->fl_list, before);
688 locks_insert_global_locks(fl);
689}
690
Jeff Layton8634b512015-01-16 15:05:55 -0500691static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500692locks_unlink_lock_ctx(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693{
Jeff Layton88974692013-06-21 08:58:14 -0400694 locks_delete_global_locks(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500695 list_del_init(&fl->fl_list);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000696 if (fl->fl_nspid) {
697 put_pid(fl->fl_nspid);
698 fl->fl_nspid = NULL;
699 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500701}
702
Jeff Layton5263e312015-01-16 15:05:55 -0500703static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500704locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
Jeff Layton5263e312015-01-16 15:05:55 -0500705{
Jeff Laytone084c1b2015-02-16 14:32:03 -0500706 locks_unlink_lock_ctx(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500707 if (dispose)
708 list_add(&fl->fl_list, dispose);
709 else
710 locks_free_lock(fl);
Jeff Layton5263e312015-01-16 15:05:55 -0500711}
712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
714 * checks for shared/exclusive status of overlapping locks.
715 */
716static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
717{
718 if (sys_fl->fl_type == F_WRLCK)
719 return 1;
720 if (caller_fl->fl_type == F_WRLCK)
721 return 1;
722 return 0;
723}
724
725/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
726 * checking before calling the locks_conflict().
727 */
728static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
729{
730 /* POSIX locks owned by the same process do not conflict with
731 * each other.
732 */
733 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
734 return (0);
735
736 /* Check whether they overlap */
737 if (!locks_overlap(caller_fl, sys_fl))
738 return 0;
739
740 return (locks_conflict(caller_fl, sys_fl));
741}
742
743/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
744 * checking before calling the locks_conflict().
745 */
746static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
747{
748 /* FLOCK locks referring to the same filp do not conflict with
749 * each other.
750 */
751 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
752 return (0);
753 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
754 return 0;
755
756 return (locks_conflict(caller_fl, sys_fl));
757}
758
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400759void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500760posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761{
762 struct file_lock *cfl;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500763 struct file_lock_context *ctx;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400764 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500766 ctx = inode->i_flctx;
767 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400768 fl->fl_type = F_UNLCK;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500769 return;
770 }
771
Jeff Layton6109c852015-01-16 15:05:57 -0500772 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500773 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
774 if (posix_locks_conflict(fl, cfl)) {
775 locks_copy_conflock(fl, cfl);
776 if (cfl->fl_nspid)
777 fl->fl_pid = pid_vnr(cfl->fl_nspid);
778 goto out;
779 }
780 }
781 fl->fl_type = F_UNLCK;
782out:
Jeff Layton6109c852015-01-16 15:05:57 -0500783 spin_unlock(&ctx->flc_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400784 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786EXPORT_SYMBOL(posix_test_lock);
787
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400788/*
789 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400791 * We attempt to detect deadlocks that are due purely to posix file
792 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400794 * We assume that a task can be waiting for at most one lock at a time.
795 * So for any acquired lock, the process holding that lock may be
796 * waiting on at most one other lock. That lock in turns may be held by
797 * someone waiting for at most one other lock. Given a requested lock
798 * caller_fl which is about to wait for a conflicting lock block_fl, we
799 * follow this chain of waiters to ensure we are not about to create a
800 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400801 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400802 * Since we do this before we ever put a process to sleep on a lock, we
803 * are ensured that there is never a cycle; that is what guarantees that
804 * the while() loop in posix_locks_deadlock() eventually completes.
805 *
806 * Note: the above assumption may not be true when handling lock
807 * requests from a broken NFS client. It may also fail in the presence
808 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400809 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500810 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400811 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500812 * Because the owner is not even nominally tied to a thread of
813 * execution, the deadlock detection below can't reasonably work well. Just
814 * skip it for those.
815 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400816 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500817 * locks that just checks for the case where two tasks are attempting to
818 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400820
821#define MAX_DEADLK_ITERATIONS 10
822
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400823/* Find a lock that the owner of the given block_fl is blocking on. */
824static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
825{
826 struct file_lock *fl;
827
Jeff Layton3999e492013-06-21 08:58:19 -0400828 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400829 if (posix_same_owner(fl, block_fl))
830 return fl->fl_next;
831 }
832 return NULL;
833}
834
Jeff Layton7b2296a2013-06-21 08:58:20 -0400835/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700836static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 struct file_lock *block_fl)
838{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400839 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
Jeff Layton57b65322014-02-03 12:13:09 -0500841 /*
842 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400843 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500844 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400845 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500846 return 0;
847
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400848 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
849 if (i++ > MAX_DEADLK_ITERATIONS)
850 return 0;
851 if (posix_same_owner(caller_fl, block_fl))
852 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 }
854 return 0;
855}
856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400858 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400859 *
860 * Note that if called with an FL_EXISTS argument, the caller may determine
861 * whether or not a lock was successfully freed by testing the return
862 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800864static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800866 struct file_lock *new_fl = NULL;
Jeff Layton5263e312015-01-16 15:05:55 -0500867 struct file_lock *fl;
868 struct file_lock_context *ctx;
869 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 int error = 0;
Jeff Layton5263e312015-01-16 15:05:55 -0500871 bool found = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400872 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
Jeff Layton5263e312015-01-16 15:05:55 -0500874 ctx = locks_get_lock_context(inode);
875 if (!ctx)
876 return -ENOMEM;
877
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200878 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
879 new_fl = locks_alloc_lock();
880 if (!new_fl)
881 return -ENOMEM;
882 }
883
Jeff Layton6109c852015-01-16 15:05:57 -0500884 spin_lock(&ctx->flc_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400885 if (request->fl_flags & FL_ACCESS)
886 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400887
Jeff Layton5263e312015-01-16 15:05:55 -0500888 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 if (filp != fl->fl_file)
890 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800891 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 goto out;
Jeff Layton5263e312015-01-16 15:05:55 -0500893 found = true;
Jeff Laytone084c1b2015-02-16 14:32:03 -0500894 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 break;
896 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
Trond Myklebustf475ae92006-06-29 16:38:32 -0400898 if (request->fl_type == F_UNLCK) {
899 if ((request->fl_flags & FL_EXISTS) && !found)
900 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800901 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400902 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400904find_conflict:
Jeff Layton5263e312015-01-16 15:05:55 -0500905 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Trond Myklebust993dfa82006-03-31 02:30:55 -0800906 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 continue;
908 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700909 if (!(request->fl_flags & FL_SLEEP))
910 goto out;
911 error = FILE_LOCK_DEFERRED;
912 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 goto out;
914 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400915 if (request->fl_flags & FL_ACCESS)
916 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800917 locks_copy_lock(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -0500918 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800919 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400920 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922out:
Jeff Layton6109c852015-01-16 15:05:57 -0500923 spin_unlock(&ctx->flc_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800924 if (new_fl)
925 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400926 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 return error;
928}
929
Marc Eshel150b3932007-01-18 16:15:35 -0500930static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500932 struct file_lock *fl, *tmp;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700933 struct file_lock *new_fl = NULL;
934 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 struct file_lock *left = NULL;
936 struct file_lock *right = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500937 struct file_lock_context *ctx;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400938 int error;
939 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400940 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500942 ctx = locks_get_lock_context(inode);
943 if (!ctx)
944 return -ENOMEM;
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 /*
947 * We may need two file_lock structures for this operation,
948 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700949 *
950 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700952 if (!(request->fl_flags & FL_ACCESS) &&
953 (request->fl_type != F_UNLCK ||
954 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
955 new_fl = locks_alloc_lock();
956 new_fl2 = locks_alloc_lock();
957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
Jeff Layton6109c852015-01-16 15:05:57 -0500959 spin_lock(&ctx->flc_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400960 /*
961 * New lock request. Walk all POSIX locks and look for conflicts. If
962 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400963 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400964 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 if (request->fl_type != F_UNLCK) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500966 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 if (!IS_POSIX(fl))
968 continue;
969 if (!posix_locks_conflict(request, fl))
970 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800971 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400972 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 error = -EAGAIN;
974 if (!(request->fl_flags & FL_SLEEP))
975 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400976 /*
977 * Deadlock detection and insertion into the blocked
978 * locks list must be done while holding the same lock!
979 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400981 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400982 if (likely(!posix_locks_deadlock(request, fl))) {
983 error = FILE_LOCK_DEFERRED;
984 __locks_insert_block(fl, request);
985 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400986 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 goto out;
988 }
989 }
990
991 /* If we're just looking for a conflict, we're done. */
992 error = 0;
993 if (request->fl_flags & FL_ACCESS)
994 goto out;
995
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500996 /* Find the first old lock with the same owner as the new lock */
997 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
998 if (posix_same_owner(request, fl))
999 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
1001
Jeff Layton1cb36012013-06-21 08:58:12 -04001002 /* Process locks with this owner. */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001003 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1004 if (!posix_same_owner(request, fl))
1005 break;
1006
1007 /* Detect adjacent or overlapping regions (if same lock type) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -07001009 /* In all comparisons of start vs end, use
1010 * "start - 1" rather than "end + 1". If end
1011 * is OFFSET_MAX, end + 1 will become negative.
1012 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (fl->fl_end < request->fl_start - 1)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001014 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 /* If the next lock in the list has entirely bigger
1016 * addresses than the new one, insert the lock here.
1017 */
Olaf Kirch449231d2005-08-25 16:25:35 -07001018 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 break;
1020
1021 /* If we come here, the new and old lock are of the
1022 * same type and adjacent or overlapping. Make one
1023 * lock yielding from the lower start address of both
1024 * locks to the higher end address.
1025 */
1026 if (fl->fl_start > request->fl_start)
1027 fl->fl_start = request->fl_start;
1028 else
1029 request->fl_start = fl->fl_start;
1030 if (fl->fl_end < request->fl_end)
1031 fl->fl_end = request->fl_end;
1032 else
1033 request->fl_end = fl->fl_end;
1034 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001035 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 continue;
1037 }
1038 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001039 added = true;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001040 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 /* Processing for different lock types is a bit
1042 * more complex.
1043 */
1044 if (fl->fl_end < request->fl_start)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001045 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 if (fl->fl_start > request->fl_end)
1047 break;
1048 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001049 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (fl->fl_start < request->fl_start)
1051 left = fl;
1052 /* If the next lock in the list has a higher end
1053 * address than the new one, insert the new one here.
1054 */
1055 if (fl->fl_end > request->fl_end) {
1056 right = fl;
1057 break;
1058 }
1059 if (fl->fl_start >= request->fl_start) {
1060 /* The new lock completely replaces an old
1061 * one (This may happen several times).
1062 */
1063 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001064 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 continue;
1066 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001067 /*
1068 * Replace the old lock with new_fl, and
1069 * remove the old one. It's safe to do the
1070 * insert here since we know that we won't be
1071 * using new_fl later, and that the lock is
1072 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001074 error = -ENOLCK;
1075 if (!new_fl)
1076 goto out;
1077 locks_copy_lock(new_fl, request);
1078 request = new_fl;
1079 new_fl = NULL;
Jeff Laytone084c1b2015-02-16 14:32:03 -05001080 locks_insert_lock_ctx(request, &fl->fl_list);
1081 locks_delete_lock_ctx(fl, &dispose);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001082 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 }
1084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001087 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001088 * The above code only modifies existing locks in case of merging or
1089 * replacing. If new lock(s) need to be inserted all modifications are
1090 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001091 */
1092 error = -ENOLCK; /* "no luck" */
1093 if (right && left == right && !new_fl2)
1094 goto out;
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 error = 0;
1097 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001098 if (request->fl_type == F_UNLCK) {
1099 if (request->fl_flags & FL_EXISTS)
1100 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001102 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001103
1104 if (!new_fl) {
1105 error = -ENOLCK;
1106 goto out;
1107 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 locks_copy_lock(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001109 locks_insert_lock_ctx(new_fl, &fl->fl_list);
Jeff Layton2e2f756f2015-02-17 17:08:23 -05001110 fl = new_fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 new_fl = NULL;
1112 }
1113 if (right) {
1114 if (left == right) {
1115 /* The new lock breaks the old one in two pieces,
1116 * so we have to use the second new lock.
1117 */
1118 left = new_fl2;
1119 new_fl2 = NULL;
1120 locks_copy_lock(left, right);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001121 locks_insert_lock_ctx(left, &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 }
1123 right->fl_start = request->fl_end + 1;
1124 locks_wake_up_blocks(right);
1125 }
1126 if (left) {
1127 left->fl_end = request->fl_start - 1;
1128 locks_wake_up_blocks(left);
1129 }
1130 out:
Jeff Layton6109c852015-01-16 15:05:57 -05001131 spin_unlock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 /*
1133 * Free any unused locks.
1134 */
1135 if (new_fl)
1136 locks_free_lock(new_fl);
1137 if (new_fl2)
1138 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001139 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 return error;
1141}
1142
1143/**
1144 * posix_lock_file - Apply a POSIX-style lock to a file
1145 * @filp: The file to apply the lock to
1146 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001147 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 *
1149 * Add a POSIX style lock to a file.
1150 * We merge adjacent & overlapping locks whenever possible.
1151 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001152 *
1153 * Note that if called with an FL_EXISTS argument, the caller may determine
1154 * whether or not a lock was successfully freed by testing the return
1155 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 */
Marc Eshel150b3932007-01-18 16:15:35 -05001157int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001158 struct file_lock *conflock)
1159{
Al Viro496ad9a2013-01-23 17:07:38 -05001160 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001161}
Marc Eshel150b3932007-01-18 16:15:35 -05001162EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163
1164/**
1165 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1166 * @filp: The file to apply the lock to
1167 * @fl: The lock to be applied
1168 *
1169 * Add a POSIX style lock to a file.
1170 * We merge adjacent & overlapping locks whenever possible.
1171 * POSIX locks are sorted by owner task, then by starting address
1172 */
1173int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1174{
1175 int error;
1176 might_sleep ();
1177 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001178 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001179 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 break;
1181 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1182 if (!error)
1183 continue;
1184
1185 locks_delete_block(fl);
1186 break;
1187 }
1188 return error;
1189}
1190EXPORT_SYMBOL(posix_lock_file_wait);
1191
1192/**
1193 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001194 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 *
1196 * Searches the inode's list of locks to find any POSIX locks which conflict.
1197 * This function is called from locks_verify_locked() only.
1198 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001199int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001201 int ret;
Jeff Laytond7a06982014-03-10 09:54:15 -04001202 struct inode *inode = file_inode(file);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001203 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 struct file_lock *fl;
1205
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001206 ctx = inode->i_flctx;
1207 if (!ctx || list_empty_careful(&ctx->flc_posix))
1208 return 0;
1209
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 /*
1211 * Search the lock list for this inode for any POSIX locks.
1212 */
Jeff Layton6109c852015-01-16 15:05:57 -05001213 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001214 ret = 0;
1215 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001216 if (fl->fl_owner != current->files &&
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001217 fl->fl_owner != file) {
1218 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 break;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001220 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 }
Jeff Layton6109c852015-01-16 15:05:57 -05001222 spin_unlock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001223 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224}
1225
1226/**
1227 * locks_mandatory_area - Check for a conflicting lock
1228 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1229 * for shared
1230 * @inode: the file to check
1231 * @filp: how the file was opened (if it was)
1232 * @offset: start of area to check
1233 * @count: length of area to check
1234 *
1235 * Searches the inode's list of locks to find any POSIX locks which conflict.
1236 * This function is called from rw_verify_area() and
1237 * locks_verify_truncate().
1238 */
1239int locks_mandatory_area(int read_write, struct inode *inode,
1240 struct file *filp, loff_t offset,
1241 size_t count)
1242{
1243 struct file_lock fl;
1244 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001245 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 fl.fl_pid = current->tgid;
1249 fl.fl_file = filp;
1250 fl.fl_flags = FL_POSIX | FL_ACCESS;
1251 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001252 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1254 fl.fl_start = offset;
1255 fl.fl_end = offset + count - 1;
1256
1257 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001258 if (filp) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001259 fl.fl_owner = filp;
Jeff Layton29723ad2014-03-10 09:54:19 -04001260 fl.fl_flags &= ~FL_SLEEP;
1261 error = __posix_lock_file(inode, &fl, NULL);
1262 if (!error)
1263 break;
1264 }
1265
1266 if (sleep)
1267 fl.fl_flags |= FL_SLEEP;
1268 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001269 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001270 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 break;
1272 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1273 if (!error) {
1274 /*
1275 * If we've been sleeping someone might have
1276 * changed the permissions behind our back.
1277 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001278 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 continue;
1280 }
1281
1282 locks_delete_block(&fl);
1283 break;
1284 }
1285
1286 return error;
1287}
1288
1289EXPORT_SYMBOL(locks_mandatory_area);
1290
J. Bruce Fields778fc542011-07-26 18:25:49 -04001291static void lease_clear_pending(struct file_lock *fl, int arg)
1292{
1293 switch (arg) {
1294 case F_UNLCK:
1295 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1296 /* fall through: */
1297 case F_RDLCK:
1298 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1299 }
1300}
1301
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302/* We already had a lease on this file; just change its type */
Jeff Layton7448cc32015-01-16 15:05:57 -05001303int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 int error = assign_type(fl, arg);
1306
1307 if (error)
1308 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001309 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001311 if (arg == F_UNLCK) {
1312 struct file *filp = fl->fl_file;
1313
1314 f_delown(filp);
1315 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001316 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1317 if (fl->fl_fasync != NULL) {
1318 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1319 fl->fl_fasync = NULL;
1320 }
Jeff Laytone084c1b2015-02-16 14:32:03 -05001321 locks_delete_lock_ctx(fl, dispose);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001322 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 return 0;
1324}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325EXPORT_SYMBOL(lease_modify);
1326
J. Bruce Fields778fc542011-07-26 18:25:49 -04001327static bool past_time(unsigned long then)
1328{
1329 if (!then)
1330 /* 0 is a special value meaning "this never expires": */
1331 return false;
1332 return time_after(jiffies, then);
1333}
1334
Jeff Laytonc45198e2014-09-01 07:12:07 -04001335static void time_out_leases(struct inode *inode, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336{
Jeff Layton8634b512015-01-16 15:05:55 -05001337 struct file_lock_context *ctx = inode->i_flctx;
1338 struct file_lock *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Jeff Layton6109c852015-01-16 15:05:57 -05001340 lockdep_assert_held(&ctx->flc_lock);
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001341
Jeff Layton8634b512015-01-16 15:05:55 -05001342 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001343 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001344 if (past_time(fl->fl_downgrade_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001345 lease_modify(fl, F_RDLCK, dispose);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001346 if (past_time(fl->fl_break_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001347 lease_modify(fl, F_UNLCK, dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
1349}
1350
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001351static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1352{
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001353 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1354 return false;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001355 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1356 return false;
1357 return locks_conflict(breaker, lease);
1358}
1359
Jeff Layton03d12dd2014-09-01 14:53:41 -04001360static bool
1361any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1362{
Jeff Layton8634b512015-01-16 15:05:55 -05001363 struct file_lock_context *ctx = inode->i_flctx;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001364 struct file_lock *fl;
1365
Jeff Layton6109c852015-01-16 15:05:57 -05001366 lockdep_assert_held(&ctx->flc_lock);
Jeff Layton03d12dd2014-09-01 14:53:41 -04001367
Jeff Layton8634b512015-01-16 15:05:55 -05001368 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Jeff Layton03d12dd2014-09-01 14:53:41 -04001369 if (leases_conflict(fl, breaker))
1370 return true;
1371 }
1372 return false;
1373}
1374
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375/**
1376 * __break_lease - revoke all outstanding leases on file
1377 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001378 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1379 * break all leases
1380 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1381 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 *
david m. richter87250dd2007-05-09 16:10:27 -04001383 * break_lease (inlined for speed) has checked there already is at least
1384 * some kind of lock (maybe a lease) on this file. Leases are broken on
1385 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 * specified %O_NONBLOCK to your open().
1387 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001388int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001390 int error = 0;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001391 struct file_lock *new_fl;
Jeff Layton8634b512015-01-16 15:05:55 -05001392 struct file_lock_context *ctx = inode->i_flctx;
1393 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394 unsigned long break_time;
Al Viro8737c932009-12-24 06:47:55 -05001395 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001396 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
Al Viro8737c932009-12-24 06:47:55 -05001398 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001399 if (IS_ERR(new_fl))
1400 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001401 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402
Jeff Layton8634b512015-01-16 15:05:55 -05001403 /* typically we will check that ctx is non-NULL before calling */
1404 if (!ctx) {
1405 WARN_ON_ONCE(1);
1406 return error;
1407 }
1408
Jeff Layton6109c852015-01-16 15:05:57 -05001409 spin_lock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410
Jeff Laytonc45198e2014-09-01 07:12:07 -04001411 time_out_leases(inode, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Jeff Layton03d12dd2014-09-01 14:53:41 -04001413 if (!any_leases_conflict(inode, new_fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001414 goto out;
1415
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 break_time = 0;
1417 if (lease_break_time > 0) {
1418 break_time = jiffies + lease_break_time * HZ;
1419 if (break_time == 0)
1420 break_time++; /* so that 0 means no break time */
1421 }
1422
Jeff Layton8634b512015-01-16 15:05:55 -05001423 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001424 if (!leases_conflict(fl, new_fl))
1425 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001426 if (want_write) {
1427 if (fl->fl_flags & FL_UNLOCK_PENDING)
1428 continue;
1429 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001431 } else {
Jeff Layton8634b512015-01-16 15:05:55 -05001432 if (lease_breaking(fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001433 continue;
1434 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1435 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 }
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001437 if (fl->fl_lmops->lm_break(fl))
Jeff Laytone084c1b2015-02-16 14:32:03 -05001438 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 }
1440
Jeff Layton8634b512015-01-16 15:05:55 -05001441 if (list_empty(&ctx->flc_lease))
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001442 goto out;
1443
Jeff Layton843c6b22014-09-01 14:27:43 -04001444 if (mode & O_NONBLOCK) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001445 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 error = -EWOULDBLOCK;
1447 goto out;
1448 }
1449
1450restart:
Jeff Layton8634b512015-01-16 15:05:55 -05001451 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1452 break_time = fl->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001453 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001455 if (break_time == 0)
1456 break_time++;
Jeff Layton8634b512015-01-16 15:05:55 -05001457 locks_insert_block(fl, new_fl);
Jeff Layton62af4f12014-05-09 14:13:05 -04001458 trace_break_lease_block(inode, new_fl);
Jeff Layton6109c852015-01-16 15:05:57 -05001459 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001460 locks_dispose_list(&dispose);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001461 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1462 !new_fl->fl_next, break_time);
Jeff Layton6109c852015-01-16 15:05:57 -05001463 spin_lock(&ctx->flc_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001464 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001465 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466 if (error >= 0) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001467 /*
1468 * Wait for the next conflicting lease that has not been
1469 * broken yet
1470 */
Jeff Layton03d12dd2014-09-01 14:53:41 -04001471 if (error == 0)
1472 time_out_leases(inode, &dispose);
1473 if (any_leases_conflict(inode, new_fl))
1474 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 error = 0;
1476 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477out:
Jeff Layton6109c852015-01-16 15:05:57 -05001478 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001479 locks_dispose_list(&dispose);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001480 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 return error;
1482}
1483
1484EXPORT_SYMBOL(__break_lease);
1485
1486/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001487 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 * @inode: the inode
1489 * @time: pointer to a timespec which will contain the last modified time
1490 *
1491 * This is to force NFS clients to flush their caches for files with
1492 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001493 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 */
1495void lease_get_mtime(struct inode *inode, struct timespec *time)
1496{
Jeff Laytonbfe86022014-08-22 10:18:44 -04001497 bool has_lease = false;
Jeff Layton8634b512015-01-16 15:05:55 -05001498 struct file_lock_context *ctx = inode->i_flctx;
1499 struct file_lock *fl;
Jeff Laytonbfe86022014-08-22 10:18:44 -04001500
Jeff Layton8634b512015-01-16 15:05:55 -05001501 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001502 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001503 if (!list_empty(&ctx->flc_lease)) {
1504 fl = list_first_entry(&ctx->flc_lease,
1505 struct file_lock, fl_list);
1506 if (fl->fl_type == F_WRLCK)
1507 has_lease = true;
1508 }
Jeff Layton6109c852015-01-16 15:05:57 -05001509 spin_unlock(&ctx->flc_lock);
Jeff Laytonbfe86022014-08-22 10:18:44 -04001510 }
1511
1512 if (has_lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 *time = current_fs_time(inode->i_sb);
1514 else
1515 *time = inode->i_mtime;
1516}
1517
1518EXPORT_SYMBOL(lease_get_mtime);
1519
1520/**
1521 * fcntl_getlease - Enquire what lease is currently active
1522 * @filp: the file
1523 *
1524 * The value returned by this function will be one of
1525 * (if no lease break is pending):
1526 *
1527 * %F_RDLCK to indicate a shared lease is held.
1528 *
1529 * %F_WRLCK to indicate an exclusive lease is held.
1530 *
1531 * %F_UNLCK to indicate no lease is held.
1532 *
1533 * (if a lease break is pending):
1534 *
1535 * %F_RDLCK to indicate an exclusive lease needs to be
1536 * changed to a shared lease (or removed).
1537 *
1538 * %F_UNLCK to indicate the lease needs to be removed.
1539 *
1540 * XXX: sfr & willy disagree over whether F_INPROGRESS
1541 * should be returned to userspace.
1542 */
1543int fcntl_getlease(struct file *filp)
1544{
1545 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001546 struct inode *inode = file_inode(filp);
Jeff Layton8634b512015-01-16 15:05:55 -05001547 struct file_lock_context *ctx = inode->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 int type = F_UNLCK;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001549 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550
Jeff Layton8634b512015-01-16 15:05:55 -05001551 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001552 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001553 time_out_leases(file_inode(filp), &dispose);
1554 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1555 if (fl->fl_file != filp)
1556 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001557 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 break;
1559 }
Jeff Layton6109c852015-01-16 15:05:57 -05001560 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001561 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 return type;
1564}
1565
Jeff Layton24cbe782014-02-03 12:13:06 -05001566/**
1567 * check_conflicting_open - see if the given dentry points to a file that has
1568 * an existing open that would conflict with the
1569 * desired lease.
1570 * @dentry: dentry to check
1571 * @arg: type of lease that we're trying to acquire
1572 *
1573 * Check to see if there's an existing open fd on this file that would
1574 * conflict with the lease we're trying to set.
1575 */
1576static int
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001577check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
Jeff Layton24cbe782014-02-03 12:13:06 -05001578{
1579 int ret = 0;
1580 struct inode *inode = dentry->d_inode;
1581
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001582 if (flags & FL_LAYOUT)
1583 return 0;
1584
Jeff Layton24cbe782014-02-03 12:13:06 -05001585 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1586 return -EAGAIN;
1587
1588 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1589 (atomic_read(&inode->i_count) > 1)))
1590 ret = -EAGAIN;
1591
1592 return ret;
1593}
1594
Jeff Laytone6f5c782014-08-22 10:40:25 -04001595static int
1596generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597{
Jeff Layton8634b512015-01-16 15:05:55 -05001598 struct file_lock *fl, *my_fl = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001599 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001601 struct file_lock_context *ctx;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001602 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001603 int error;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001604 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605
J. Bruce Fields096657b2010-10-30 17:31:14 -04001606 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001607 trace_generic_add_lease(inode, lease);
1608
Jeff Layton8634b512015-01-16 15:05:55 -05001609 ctx = locks_get_lock_context(inode);
1610 if (!ctx)
1611 return -ENOMEM;
1612
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001613 /*
1614 * In the delegation case we need mutual exclusion with
1615 * a number of operations that take the i_mutex. We trylock
1616 * because delegations are an optional optimization, and if
1617 * there's some chance of a conflict--we'd rather not
1618 * bother, maybe that's a sign this just isn't a good file to
1619 * hand out a delegation on.
1620 */
1621 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1622 return -EAGAIN;
1623
1624 if (is_deleg && arg == F_WRLCK) {
1625 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001626 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001627 WARN_ON_ONCE(1);
1628 return -EINVAL;
1629 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001630
Jeff Layton6109c852015-01-16 15:05:57 -05001631 spin_lock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001632 time_out_leases(inode, &dispose);
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001633 error = check_conflicting_open(dentry, arg, lease->fl_flags);
Jeff Layton24cbe782014-02-03 12:13:06 -05001634 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001635 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001636
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 /*
1638 * At this point, we know that if there is an exclusive
1639 * lease on this file, then we hold it on this filp
1640 * (otherwise our open of this file would have blocked).
1641 * And if we are trying to acquire an exclusive lease,
1642 * then the file is not open by anyone (including us)
1643 * except for this filp.
1644 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001645 error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001646 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001647 if (fl->fl_file == filp &&
1648 fl->fl_owner == lease->fl_owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001649 my_fl = fl;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001650 continue;
1651 }
Jeff Layton8634b512015-01-16 15:05:55 -05001652
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001653 /*
1654 * No exclusive leases if someone else has a lease on
1655 * this file:
1656 */
1657 if (arg == F_WRLCK)
1658 goto out;
1659 /*
1660 * Modifying our existing lease is OK, but no getting a
1661 * new lease if someone else is opening for write:
1662 */
1663 if (fl->fl_flags & FL_UNLOCK_PENDING)
1664 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 }
1666
Jeff Layton8634b512015-01-16 15:05:55 -05001667 if (my_fl != NULL) {
Jeff Layton0164bf02015-03-04 17:34:32 -05001668 lease = my_fl;
1669 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001670 if (error)
1671 goto out;
1672 goto out_setup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 }
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 error = -EINVAL;
1676 if (!leases_enable)
1677 goto out;
1678
Jeff Laytone084c1b2015-02-16 14:32:03 -05001679 locks_insert_lock_ctx(lease, &ctx->flc_lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001680 /*
1681 * The check in break_lease() is lockless. It's possible for another
1682 * open to race in after we did the earlier check for a conflicting
1683 * open but before the lease was inserted. Check again for a
1684 * conflicting open and cancel the lease if there is one.
1685 *
1686 * We also add a barrier here to ensure that the insertion of the lock
1687 * precedes these checks.
1688 */
1689 smp_mb();
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001690 error = check_conflicting_open(dentry, arg, lease->fl_flags);
Jeff Layton8634b512015-01-16 15:05:55 -05001691 if (error) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001692 locks_unlink_lock_ctx(lease);
Jeff Layton8634b512015-01-16 15:05:55 -05001693 goto out;
1694 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001695
1696out_setup:
1697 if (lease->fl_lmops->lm_setup)
1698 lease->fl_lmops->lm_setup(lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699out:
Jeff Layton6109c852015-01-16 15:05:57 -05001700 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001701 locks_dispose_list(&dispose);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001702 if (is_deleg)
1703 mutex_unlock(&inode->i_mutex);
Jeff Layton8634b512015-01-16 15:05:55 -05001704 if (!error && !my_fl)
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001705 *flp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 return error;
1707}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001708
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001709static int generic_delete_lease(struct file *filp, void *owner)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001710{
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001711 int error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001712 struct file_lock *fl, *victim = NULL;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001713 struct dentry *dentry = filp->f_path.dentry;
1714 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001715 struct file_lock_context *ctx = inode->i_flctx;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001716 LIST_HEAD(dispose);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001717
Jeff Layton8634b512015-01-16 15:05:55 -05001718 if (!ctx) {
1719 trace_generic_delete_lease(inode, NULL);
1720 return error;
1721 }
1722
Jeff Layton6109c852015-01-16 15:05:57 -05001723 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001724 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001725 if (fl->fl_file == filp &&
1726 fl->fl_owner == owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001727 victim = fl;
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001728 break;
Jeff Layton8634b512015-01-16 15:05:55 -05001729 }
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001730 }
Jeff Laytona9b1b452015-03-14 09:45:35 -04001731 trace_generic_delete_lease(inode, victim);
Jeff Layton8634b512015-01-16 15:05:55 -05001732 if (victim)
Jeff Layton7448cc32015-01-16 15:05:57 -05001733 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05001734 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001735 locks_dispose_list(&dispose);
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001736 return error;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001737}
1738
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739/**
1740 * generic_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001741 * @filp: file pointer
1742 * @arg: type of lease to obtain
1743 * @flp: input - file_lock to use, output - file_lock inserted
1744 * @priv: private data for lm_setup (may be NULL if lm_setup
1745 * doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 *
1747 * The (input) flp->fl_lmops->lm_break function is required
1748 * by break_lease().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001750int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1751 void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 struct dentry *dentry = filp->f_path.dentry;
1754 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001755 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001757 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001758 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001760 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761 error = security_file_lock(filp, arg);
1762 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001763 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001765 switch (arg) {
1766 case F_UNLCK:
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001767 return generic_delete_lease(filp, *priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001768 case F_RDLCK:
1769 case F_WRLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001770 if (!(*flp)->fl_lmops->lm_break) {
1771 WARN_ON_ONCE(1);
1772 return -ENOLCK;
1773 }
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001774
Jeff Laytone6f5c782014-08-22 10:40:25 -04001775 return generic_add_lease(filp, arg, flp, priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001776 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001777 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001779}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001780EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001782/**
Jeff Laytone51673a2014-08-22 18:13:28 -04001783 * vfs_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001784 * @filp: file pointer
1785 * @arg: type of lease to obtain
1786 * @lease: file_lock to use when adding a lease
1787 * @priv: private info for lm_setup when adding a lease (may be
1788 * NULL if lm_setup doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 *
Jeff Laytone51673a2014-08-22 18:13:28 -04001790 * Call this to establish a lease on the file. The "lease" argument is not
1791 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1792 * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1793 * if not, this function will return -ENOLCK (and generate a scary-looking
1794 * stack trace).
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001795 *
1796 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1797 * may be NULL if the lm_setup operation doesn't require it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001799int
1800vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001802 if (filp->f_op->setlease)
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001803 return filp->f_op->setlease(filp, arg, lease, priv);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001804 else
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001805 return generic_setlease(filp, arg, lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001807EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001809static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001811 struct file_lock *fl;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001812 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 int error;
1814
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001815 fl = lease_alloc(filp, arg);
1816 if (IS_ERR(fl))
1817 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001819 new = fasync_alloc();
1820 if (!new) {
1821 locks_free_lock(fl);
1822 return -ENOMEM;
1823 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001824 new->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001826 error = vfs_setlease(filp, arg, &fl, (void **)&new);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001827 if (fl)
1828 locks_free_lock(fl);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001829 if (new)
1830 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 return error;
1832}
1833
1834/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001835 * fcntl_setlease - sets a lease on an open file
1836 * @fd: open file descriptor
1837 * @filp: file pointer
1838 * @arg: type of lease to obtain
1839 *
1840 * Call this fcntl to establish a lease on the file.
1841 * Note that you also need to call %F_SETSIG to
1842 * receive a signal when the lease is broken.
1843 */
1844int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1845{
1846 if (arg == F_UNLCK)
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001847 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001848 return do_fcntl_add_lease(fd, filp, arg);
1849}
1850
1851/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852 * flock_lock_file_wait - Apply a FLOCK-style lock to a file
1853 * @filp: The file to apply the lock to
1854 * @fl: The lock to be applied
1855 *
1856 * Add a FLOCK style lock to a file.
1857 */
1858int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
1859{
1860 int error;
1861 might_sleep();
1862 for (;;) {
1863 error = flock_lock_file(filp, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001864 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 break;
1866 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1867 if (!error)
1868 continue;
1869
1870 locks_delete_block(fl);
1871 break;
1872 }
1873 return error;
1874}
1875
1876EXPORT_SYMBOL(flock_lock_file_wait);
1877
1878/**
1879 * sys_flock: - flock() system call.
1880 * @fd: the file descriptor to lock.
1881 * @cmd: the type of lock to apply.
1882 *
1883 * Apply a %FL_FLOCK style lock to an open file descriptor.
1884 * The @cmd can be one of
1885 *
1886 * %LOCK_SH -- a shared lock.
1887 *
1888 * %LOCK_EX -- an exclusive lock.
1889 *
1890 * %LOCK_UN -- remove an existing lock.
1891 *
1892 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1893 *
1894 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1895 * processes read and write access respectively.
1896 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001897SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898{
Al Viro2903ff02012-08-28 12:52:22 -04001899 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 struct file_lock *lock;
1901 int can_sleep, unlock;
1902 int error;
1903
1904 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001905 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 goto out;
1907
1908 can_sleep = !(cmd & LOCK_NB);
1909 cmd &= ~LOCK_NB;
1910 unlock = (cmd == LOCK_UN);
1911
Al Viroaeb5d722008-09-02 15:28:45 -04001912 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001913 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 goto out_putf;
1915
Jeff Layton6e129d002014-09-04 10:25:06 -04001916 lock = flock_make_lock(f.file, cmd);
1917 if (IS_ERR(lock)) {
1918 error = PTR_ERR(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 goto out_putf;
Jeff Layton6e129d002014-09-04 10:25:06 -04001920 }
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (can_sleep)
1923 lock->fl_flags |= FL_SLEEP;
1924
Al Viro2903ff02012-08-28 12:52:22 -04001925 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 if (error)
1927 goto out_free;
1928
Al Viro72c2d532013-09-22 16:27:52 -04001929 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001930 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 (can_sleep) ? F_SETLKW : F_SETLK,
1932 lock);
1933 else
Al Viro2903ff02012-08-28 12:52:22 -04001934 error = flock_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935
1936 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001937 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938
1939 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001940 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 out:
1942 return error;
1943}
1944
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001945/**
1946 * vfs_test_lock - test file byte range lock
1947 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001948 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001949 *
1950 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1951 * setting conf->fl_type to something other than F_UNLCK.
1952 */
1953int vfs_test_lock(struct file *filp, struct file_lock *fl)
1954{
Al Viro72c2d532013-09-22 16:27:52 -04001955 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001956 return filp->f_op->lock(filp, F_GETLK, fl);
1957 posix_test_lock(filp, fl);
1958 return 0;
1959}
1960EXPORT_SYMBOL_GPL(vfs_test_lock);
1961
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001962static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1963{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001964 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001965#if BITS_PER_LONG == 32
1966 /*
1967 * Make sure we can represent the posix lock via
1968 * legacy 32bit flock.
1969 */
1970 if (fl->fl_start > OFFT_OFFSET_MAX)
1971 return -EOVERFLOW;
1972 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
1973 return -EOVERFLOW;
1974#endif
1975 flock->l_start = fl->fl_start;
1976 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1977 fl->fl_end - fl->fl_start + 1;
1978 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04001979 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001980 return 0;
1981}
1982
1983#if BITS_PER_LONG == 32
1984static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
1985{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001986 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001987 flock->l_start = fl->fl_start;
1988 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
1989 fl->fl_end - fl->fl_start + 1;
1990 flock->l_whence = 0;
1991 flock->l_type = fl->fl_type;
1992}
1993#endif
1994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995/* Report the first existing lock that would conflict with l.
1996 * This implements the F_GETLK command of fcntl().
1997 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05001998int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002000 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 struct flock flock;
2002 int error;
2003
2004 error = -EFAULT;
2005 if (copy_from_user(&flock, l, sizeof(flock)))
2006 goto out;
2007 error = -EINVAL;
2008 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2009 goto out;
2010
2011 error = flock_to_posix_lock(filp, &file_lock, &flock);
2012 if (error)
2013 goto out;
2014
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002015 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002016 error = -EINVAL;
2017 if (flock.l_pid != 0)
2018 goto out;
2019
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002020 cmd = F_GETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002021 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002022 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002023 }
2024
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002025 error = vfs_test_lock(filp, &file_lock);
2026 if (error)
2027 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002029 flock.l_type = file_lock.fl_type;
2030 if (file_lock.fl_type != F_UNLCK) {
2031 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002032 if (error)
Kinglong Meef3282962014-08-22 10:18:43 -04002033 goto rel_priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 }
2035 error = -EFAULT;
2036 if (!copy_to_user(l, &flock, sizeof(flock)))
2037 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002038rel_priv:
2039 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040out:
2041 return error;
2042}
2043
Marc Eshel7723ec92007-01-18 15:08:55 -05002044/**
2045 * vfs_lock_file - file byte range lock
2046 * @filp: The file to apply the lock to
2047 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2048 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05002049 * @conf: Place to return a copy of the conflicting lock, if found.
2050 *
2051 * A caller that doesn't care about the conflicting lock may pass NULL
2052 * as the final argument.
2053 *
2054 * If the filesystem defines a private ->lock() method, then @conf will
2055 * be left unchanged; so a caller that cares should initialize it to
2056 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05002057 *
2058 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2059 * locks, the ->lock() interface may return asynchronously, before the lock has
2060 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002061 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05002062 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2063 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002064 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05002065 * request completes.
2066 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07002067 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2068 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05002069 * nonzero return code and the file system should release the lock. The file
2070 * system is also responsible to keep a corresponding posix lock when it
2071 * grants a lock so the VFS can find out which locks are locally held and do
2072 * the correct lock cleanup when required.
2073 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002074 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002075 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002076 */
Marc Eshel150b3932007-01-18 16:15:35 -05002077int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002078{
Al Viro72c2d532013-09-22 16:27:52 -04002079 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002080 return filp->f_op->lock(filp, cmd, fl);
2081 else
Marc Eshel150b3932007-01-18 16:15:35 -05002082 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002083}
2084EXPORT_SYMBOL_GPL(vfs_lock_file);
2085
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002086static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2087 struct file_lock *fl)
2088{
2089 int error;
2090
2091 error = security_file_lock(filp, fl->fl_type);
2092 if (error)
2093 return error;
2094
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002095 for (;;) {
2096 error = vfs_lock_file(filp, cmd, fl, NULL);
2097 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002098 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002099 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2100 if (!error)
2101 continue;
2102
2103 locks_delete_block(fl);
2104 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002105 }
2106
2107 return error;
2108}
2109
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002110/* Ensure that fl->fl_filp has compatible f_mode for F_SETLK calls */
2111static int
2112check_fmode_for_setlk(struct file_lock *fl)
2113{
2114 switch (fl->fl_type) {
2115 case F_RDLCK:
2116 if (!(fl->fl_file->f_mode & FMODE_READ))
2117 return -EBADF;
2118 break;
2119 case F_WRLCK:
2120 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2121 return -EBADF;
2122 }
2123 return 0;
2124}
2125
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126/* Apply the lock described by l to an open file descriptor.
2127 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2128 */
Peter Staubachc2936212005-07-27 11:45:09 -07002129int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2130 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131{
2132 struct file_lock *file_lock = locks_alloc_lock();
2133 struct flock flock;
2134 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002135 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 int error;
2137
2138 if (file_lock == NULL)
2139 return -ENOLCK;
2140
2141 /*
2142 * This might block, so we do it before checking the inode.
2143 */
2144 error = -EFAULT;
2145 if (copy_from_user(&flock, l, sizeof(flock)))
2146 goto out;
2147
Al Viro496ad9a2013-01-23 17:07:38 -05002148 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
2150 /* Don't allow mandatory locks on files that may be memory mapped
2151 * and shared.
2152 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002153 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154 error = -EAGAIN;
2155 goto out;
2156 }
2157
Peter Staubachc2936212005-07-27 11:45:09 -07002158again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 error = flock_to_posix_lock(filp, file_lock, &flock);
2160 if (error)
2161 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002162
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002163 error = check_fmode_for_setlk(file_lock);
2164 if (error)
2165 goto out;
2166
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002167 /*
2168 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002169 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002170 */
2171 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002172 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002173 error = -EINVAL;
2174 if (flock.l_pid != 0)
2175 goto out;
2176
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002177 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002178 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002179 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002180 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002181 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002182 error = -EINVAL;
2183 if (flock.l_pid != 0)
2184 goto out;
2185
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002186 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002187 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002188 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002189 /* Fallthrough */
2190 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 file_lock->fl_flags |= FL_SLEEP;
2192 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002193
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002194 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
Peter Staubachc2936212005-07-27 11:45:09 -07002196 /*
2197 * Attempt to detect a close/fcntl race and recover by
2198 * releasing the lock that was just acquired.
2199 */
Al Viro0b2bac22008-05-06 13:58:34 -04002200 /*
2201 * we need that spin_lock here - it prevents reordering between
Jeff Layton8116bf42015-01-21 20:44:01 -05002202 * update of i_flctx->flc_posix and check for it done in close().
Al Viro0b2bac22008-05-06 13:58:34 -04002203 * rcu_read_lock() wouldn't do.
2204 */
2205 spin_lock(&current->files->file_lock);
2206 f = fcheck(fd);
2207 spin_unlock(&current->files->file_lock);
2208 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002209 flock.l_type = F_UNLCK;
2210 goto again;
2211 }
2212
2213out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002214 locks_free_lock(file_lock);
2215 return error;
2216}
2217
2218#if BITS_PER_LONG == 32
2219/* Report the first existing lock that would conflict with l.
2220 * This implements the F_GETLK command of fcntl().
2221 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002222int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002224 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225 struct flock64 flock;
2226 int error;
2227
2228 error = -EFAULT;
2229 if (copy_from_user(&flock, l, sizeof(flock)))
2230 goto out;
2231 error = -EINVAL;
2232 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2233 goto out;
2234
2235 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2236 if (error)
2237 goto out;
2238
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002239 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002240 error = -EINVAL;
2241 if (flock.l_pid != 0)
2242 goto out;
2243
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002244 cmd = F_GETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002245 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002246 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002247 }
2248
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002249 error = vfs_test_lock(filp, &file_lock);
2250 if (error)
2251 goto out;
2252
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002253 flock.l_type = file_lock.fl_type;
2254 if (file_lock.fl_type != F_UNLCK)
2255 posix_lock_to_flock64(&flock, &file_lock);
2256
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 error = -EFAULT;
2258 if (!copy_to_user(l, &flock, sizeof(flock)))
2259 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002260
2261 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262out:
2263 return error;
2264}
2265
2266/* Apply the lock described by l to an open file descriptor.
2267 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2268 */
Peter Staubachc2936212005-07-27 11:45:09 -07002269int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2270 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271{
2272 struct file_lock *file_lock = locks_alloc_lock();
2273 struct flock64 flock;
2274 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002275 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 int error;
2277
2278 if (file_lock == NULL)
2279 return -ENOLCK;
2280
2281 /*
2282 * This might block, so we do it before checking the inode.
2283 */
2284 error = -EFAULT;
2285 if (copy_from_user(&flock, l, sizeof(flock)))
2286 goto out;
2287
Al Viro496ad9a2013-01-23 17:07:38 -05002288 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289
2290 /* Don't allow mandatory locks on files that may be memory mapped
2291 * and shared.
2292 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002293 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 error = -EAGAIN;
2295 goto out;
2296 }
2297
Peter Staubachc2936212005-07-27 11:45:09 -07002298again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 error = flock64_to_posix_lock(filp, file_lock, &flock);
2300 if (error)
2301 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002302
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002303 error = check_fmode_for_setlk(file_lock);
2304 if (error)
2305 goto out;
2306
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002307 /*
2308 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002309 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002310 */
2311 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002312 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002313 error = -EINVAL;
2314 if (flock.l_pid != 0)
2315 goto out;
2316
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002317 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002318 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002319 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002320 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002321 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002322 error = -EINVAL;
2323 if (flock.l_pid != 0)
2324 goto out;
2325
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002326 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002327 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002328 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002329 /* Fallthrough */
2330 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 file_lock->fl_flags |= FL_SLEEP;
2332 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002333
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002334 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335
Peter Staubachc2936212005-07-27 11:45:09 -07002336 /*
2337 * Attempt to detect a close/fcntl race and recover by
2338 * releasing the lock that was just acquired.
2339 */
Al Viro0b2bac22008-05-06 13:58:34 -04002340 spin_lock(&current->files->file_lock);
2341 f = fcheck(fd);
2342 spin_unlock(&current->files->file_lock);
2343 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002344 flock.l_type = F_UNLCK;
2345 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346 }
2347
2348out:
2349 locks_free_lock(file_lock);
2350 return error;
2351}
2352#endif /* BITS_PER_LONG == 32 */
2353
2354/*
2355 * This function is called when the file is being removed
2356 * from the task's fd array. POSIX locks belonging to this task
2357 * are deleted at this time.
2358 */
2359void locks_remove_posix(struct file *filp, fl_owner_t owner)
2360{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002361 struct file_lock lock;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002362 struct file_lock_context *ctx = file_inode(filp)->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
2364 /*
2365 * If there are no locks held on this file, we don't need to call
2366 * posix_lock_file(). Another process could be setting a lock on this
2367 * file at the same time, but we wouldn't remove that lock anyway.
2368 */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002369 if (!ctx || list_empty(&ctx->flc_posix))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 return;
2371
2372 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002373 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 lock.fl_start = 0;
2375 lock.fl_end = OFFSET_MAX;
2376 lock.fl_owner = owner;
2377 lock.fl_pid = current->tgid;
2378 lock.fl_file = filp;
2379 lock.fl_ops = NULL;
2380 lock.fl_lmops = NULL;
2381
Marc Eshel150b3932007-01-18 16:15:35 -05002382 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2385 lock.fl_ops->fl_release_private(&lock);
2386}
2387
2388EXPORT_SYMBOL(locks_remove_posix);
2389
Jeff Layton3d8e5602015-01-16 15:05:58 -05002390/* The i_flctx must be valid when calling into here */
Jeff Laytondd459bb2015-01-16 15:05:54 -05002391static void
2392locks_remove_flock(struct file *filp)
2393{
2394 struct file_lock fl = {
2395 .fl_owner = filp,
2396 .fl_pid = current->tgid,
2397 .fl_file = filp,
2398 .fl_flags = FL_FLOCK,
2399 .fl_type = F_UNLCK,
2400 .fl_end = OFFSET_MAX,
2401 };
Jeff Layton5263e312015-01-16 15:05:55 -05002402 struct file_lock_context *flctx = file_inode(filp)->i_flctx;
Jeff Laytondd459bb2015-01-16 15:05:54 -05002403
Jeff Layton3d8e5602015-01-16 15:05:58 -05002404 if (list_empty(&flctx->flc_flock))
Jeff Laytondd459bb2015-01-16 15:05:54 -05002405 return;
2406
2407 if (filp->f_op->flock)
2408 filp->f_op->flock(filp, F_SETLKW, &fl);
2409 else
2410 flock_lock_file(filp, &fl);
2411
2412 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2413 fl.fl_ops->fl_release_private(&fl);
2414}
2415
Jeff Layton3d8e5602015-01-16 15:05:58 -05002416/* The i_flctx must be valid when calling into here */
Jeff Layton8634b512015-01-16 15:05:55 -05002417static void
2418locks_remove_lease(struct file *filp)
2419{
2420 struct inode *inode = file_inode(filp);
2421 struct file_lock_context *ctx = inode->i_flctx;
2422 struct file_lock *fl, *tmp;
2423 LIST_HEAD(dispose);
2424
Jeff Layton3d8e5602015-01-16 15:05:58 -05002425 if (list_empty(&ctx->flc_lease))
Jeff Layton8634b512015-01-16 15:05:55 -05002426 return;
2427
Jeff Layton6109c852015-01-16 15:05:57 -05002428 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002429 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
Jeff Laytonc4e136c2015-02-16 19:37:42 -05002430 if (filp == fl->fl_file)
2431 lease_modify(fl, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05002432 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002433 locks_dispose_list(&dispose);
2434}
2435
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436/*
2437 * This function is called on the last close of an open file.
2438 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002439void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440{
Jeff Layton3d8e5602015-01-16 15:05:58 -05002441 if (!file_inode(filp)->i_flctx)
2442 return;
2443
Jeff Laytondd459bb2015-01-16 15:05:54 -05002444 /* remove any OFD locks */
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002445 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002446
Jeff Laytondd459bb2015-01-16 15:05:54 -05002447 /* remove flock locks */
2448 locks_remove_flock(filp);
2449
Jeff Layton8634b512015-01-16 15:05:55 -05002450 /* remove any leases */
2451 locks_remove_lease(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452}
2453
2454/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456 * @waiter: the lock which was waiting
2457 *
2458 * lockd needs to block waiting for locks.
2459 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002460int
Jeff Laytonf891a292013-06-21 08:58:09 -04002461posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002463 int status = 0;
2464
Jeff Layton7b2296a2013-06-21 08:58:20 -04002465 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002466 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002468 else
2469 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002470 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002471 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473EXPORT_SYMBOL(posix_unblock_lock);
2474
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002475/**
2476 * vfs_cancel_lock - file byte range unblock lock
2477 * @filp: The file to apply the unblock to
2478 * @fl: The lock to be unblocked
2479 *
2480 * Used by lock managers to cancel blocked requests
2481 */
2482int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2483{
Al Viro72c2d532013-09-22 16:27:52 -04002484 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002485 return filp->f_op->lock(filp, F_CANCELLK, fl);
2486 return 0;
2487}
2488
2489EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2490
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002491#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002492#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002493#include <linux/seq_file.h>
2494
Jeff Layton7012b022013-06-21 08:58:22 -04002495struct locks_iterator {
2496 int li_cpu;
2497 loff_t li_pos;
2498};
2499
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002500static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002501 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502{
2503 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002504 unsigned int fl_pid;
2505
2506 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002507 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002508 else
2509 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
2511 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002512 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513
Jerome Marchand99dc8292010-10-26 14:22:33 -07002514 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002516 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002517 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002518 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002519 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002520 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002521 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002522
2523 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002525 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 } else if (IS_FLOCK(fl)) {
2527 if (fl->fl_type & LOCK_MAND) {
Fabian Frederick5315c262014-05-09 14:13:05 -04002528 seq_puts(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002530 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531 }
2532 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002533 if (fl->fl_flags & FL_DELEG)
2534 seq_puts(f, "DELEG ");
2535 else
2536 seq_puts(f, "LEASE ");
2537
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002538 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002539 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002541 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002543 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002545 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546 }
2547 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002548 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 (fl->fl_type & LOCK_READ)
2550 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2551 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2552 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002553 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002554 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002555 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2556 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 }
2558 if (inode) {
2559#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002560 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 inode->i_sb->s_id, inode->i_ino);
2562#else
2563 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002564 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 MAJOR(inode->i_sb->s_dev),
2566 MINOR(inode->i_sb->s_dev), inode->i_ino);
2567#endif
2568 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002569 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 }
2571 if (IS_POSIX(fl)) {
2572 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002573 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002575 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002577 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 }
2579}
2580
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002581static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582{
Jeff Layton7012b022013-06-21 08:58:22 -04002583 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002584 struct file_lock *fl, *bfl;
2585
Jeff Layton139ca042013-06-21 08:58:17 -04002586 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002587
Jeff Layton7012b022013-06-21 08:58:22 -04002588 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002589
2590 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002591 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002592
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002593 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594}
2595
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002596static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002597 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598{
Jeff Layton7012b022013-06-21 08:58:22 -04002599 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002600
Jeff Layton7012b022013-06-21 08:58:22 -04002601 iter->li_pos = *pos + 1;
2602 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002603 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002604 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605}
2606
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002607static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2608{
Jeff Layton7012b022013-06-21 08:58:22 -04002609 struct locks_iterator *iter = f->private;
2610
2611 ++iter->li_pos;
2612 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002613}
2614
2615static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002616 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002617{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002618 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002619 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002620}
2621
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002622static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002623 .start = locks_start,
2624 .next = locks_next,
2625 .stop = locks_stop,
2626 .show = locks_show,
2627};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002628
2629static int locks_open(struct inode *inode, struct file *filp)
2630{
Jeff Layton7012b022013-06-21 08:58:22 -04002631 return seq_open_private(filp, &locks_seq_operations,
2632 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002633}
2634
2635static const struct file_operations proc_locks_operations = {
2636 .open = locks_open,
2637 .read = seq_read,
2638 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002639 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002640};
2641
2642static int __init proc_locks_init(void)
2643{
2644 proc_create("locks", 0, NULL, &proc_locks_operations);
2645 return 0;
2646}
2647module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002648#endif
2649
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650static int __init filelock_init(void)
2651{
Jeff Layton7012b022013-06-21 08:58:22 -04002652 int i;
2653
Jeff Layton4a075e32015-01-16 15:05:54 -05002654 flctx_cache = kmem_cache_create("file_lock_ctx",
2655 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2656
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002658 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2659
Jeff Layton7012b022013-06-21 08:58:22 -04002660 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2661
2662 for_each_possible_cpu(i)
2663 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2664
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 return 0;
2666}
2667
2668core_initcall(filelock_init);