blob: d46e70567b9953eecbb2d9c41b1922a5b599578c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
Paul Bolle395cf962011-08-15 02:02:26 +020063 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -0400119#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000128#include <linux/pid_namespace.h>
Jeff Layton48f74182013-06-21 08:58:18 -0400129#include <linux/hashtable.h>
Jeff Layton7012b022013-06-21 08:58:22 -0400130#include <linux/percpu.h>
131#include <linux/lglock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Jeff Layton62af4f12014-05-09 14:13:05 -0400133#define CREATE_TRACE_POINTS
134#include <trace/events/filelock.h>
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136#include <asm/uaccess.h>
137
138#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
139#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
J. Bruce Fields617588d2011-07-01 15:18:34 -0400140#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG))
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400141#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400143static bool lease_breaking(struct file_lock *fl)
144{
J. Bruce Fields778fc542011-07-26 18:25:49 -0400145 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
146}
147
148static int target_leasetype(struct file_lock *fl)
149{
150 if (fl->fl_flags & FL_UNLOCK_PENDING)
151 return F_UNLCK;
152 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
153 return F_RDLCK;
154 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400155}
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157int leases_enable = 1;
158int lease_break_time = 45;
159
Jeff Layton1c8c6012013-06-21 08:58:15 -0400160/*
Jeff Layton7012b022013-06-21 08:58:22 -0400161 * The global file_lock_list is only used for displaying /proc/locks, so we
162 * keep a list on each CPU, with each list protected by its own spinlock via
163 * the file_lock_lglock. Note that alterations to the list also require that
164 * the relevant i_lock is held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400165 */
Jeff Layton7012b022013-06-21 08:58:22 -0400166DEFINE_STATIC_LGLOCK(file_lock_lglock);
167static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
Jeff Layton88974692013-06-21 08:58:14 -0400168
Jeff Layton1c8c6012013-06-21 08:58:15 -0400169/*
Jeff Layton48f74182013-06-21 08:58:18 -0400170 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400171 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400172 *
173 * We hash locks by lockowner in order to optimize searching for the lock a
174 * particular lockowner is waiting on.
175 *
176 * FIXME: make this value scale via some heuristic? We generally will want more
177 * buckets when we have more lockowners holding locks, but that's a little
178 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400179 */
Jeff Layton48f74182013-06-21 08:58:18 -0400180#define BLOCKED_HASH_BITS 7
181static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400182
Jeff Layton1c8c6012013-06-21 08:58:15 -0400183/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400184 * This lock protects the blocked_hash. Generally, if you're accessing it, you
185 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400186 *
187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
188 * pointer for file_lock structures that are acting as lock requests (in
189 * contrast to those that are acting as records of acquired locks).
190 *
191 * Note that when we acquire this lock in order to change the above fields,
192 * we often hold the i_lock as well. In certain cases, when reading the fields
193 * protected by this lock, we can skip acquiring it iff we already hold the
194 * i_lock.
195 *
196 * In particular, adding an entry to the fl_block list requires that you hold
197 * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
198 * an entry from the list however only requires the file_lock_lock.
199 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400200static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
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
217 INIT_LIST_HEAD(&new->flc_flock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500218 INIT_LIST_HEAD(&new->flc_posix);
Jeff Layton8634b512015-01-16 15:05:55 -0500219 INIT_LIST_HEAD(&new->flc_lease);
Jeff Layton4a075e32015-01-16 15:05:54 -0500220
221 /*
222 * Assign the pointer if it's not already assigned. If it is, then
223 * free the context we just allocated.
224 */
225 spin_lock(&inode->i_lock);
226 if (likely(!inode->i_flctx)) {
227 inode->i_flctx = new;
228 new = NULL;
229 }
230 spin_unlock(&inode->i_lock);
231
232 if (new)
233 kmem_cache_free(flctx_cache, new);
234out:
235 return inode->i_flctx;
236}
237
238void
239locks_free_lock_context(struct file_lock_context *ctx)
240{
241 if (ctx) {
242 WARN_ON_ONCE(!list_empty(&ctx->flc_flock));
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500243 WARN_ON_ONCE(!list_empty(&ctx->flc_posix));
Jeff Layton8634b512015-01-16 15:05:55 -0500244 WARN_ON_ONCE(!list_empty(&ctx->flc_lease));
Jeff Layton4a075e32015-01-16 15:05:54 -0500245 kmem_cache_free(flctx_cache, ctx);
246 }
247}
248
Miklos Szerediee19cc42011-07-07 13:06:09 +0200249static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200250{
Jeff Layton139ca042013-06-21 08:58:17 -0400251 INIT_HLIST_NODE(&fl->fl_link);
Jeff Layton6dee60f2015-01-16 15:05:54 -0500252 INIT_LIST_HEAD(&fl->fl_list);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200253 INIT_LIST_HEAD(&fl->fl_block);
254 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200255}
256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200258struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200260 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200261
262 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200263 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200264
265 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200267EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500269void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500270{
271 if (fl->fl_ops) {
272 if (fl->fl_ops->fl_release_private)
273 fl->fl_ops->fl_release_private(fl);
274 fl->fl_ops = NULL;
275 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500276
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400277 if (fl->fl_lmops) {
278 if (fl->fl_lmops->lm_put_owner)
279 fl->fl_lmops->lm_put_owner(fl);
280 fl->fl_lmops = NULL;
281 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500282}
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500283EXPORT_SYMBOL_GPL(locks_release_private);
Trond Myklebust47831f32006-03-20 13:44:05 -0500284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285/* Free a lock which is not in use. */
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400286void locks_free_lock(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287{
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800288 BUG_ON(waitqueue_active(&fl->fl_wait));
Jeff Layton6dee60f2015-01-16 15:05:54 -0500289 BUG_ON(!list_empty(&fl->fl_list));
Miklos Szeredi5ce29642006-03-31 02:30:29 -0800290 BUG_ON(!list_empty(&fl->fl_block));
Jeff Layton139ca042013-06-21 08:58:17 -0400291 BUG_ON(!hlist_unhashed(&fl->fl_link));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
Trond Myklebust47831f32006-03-20 13:44:05 -0500293 locks_release_private(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 kmem_cache_free(filelock_cache, fl);
295}
J. Bruce Fields05fa3132010-10-30 17:31:15 -0400296EXPORT_SYMBOL(locks_free_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
Jeff Laytoned9814d2014-08-11 14:20:31 -0400298static void
299locks_dispose_list(struct list_head *dispose)
300{
301 struct file_lock *fl;
302
303 while (!list_empty(dispose)) {
Jeff Layton6dee60f2015-01-16 15:05:54 -0500304 fl = list_first_entry(dispose, struct file_lock, fl_list);
305 list_del_init(&fl->fl_list);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400306 locks_free_lock(fl);
307 }
308}
309
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310void locks_init_lock(struct file_lock *fl)
311{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200312 memset(fl, 0, sizeof(struct file_lock));
313 locks_init_lock_heads(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314}
315
316EXPORT_SYMBOL(locks_init_lock);
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318/*
319 * Initialize a new lock from an existing file_lock structure.
320 */
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400321void locks_copy_conflock(struct file_lock *new, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322{
323 new->fl_owner = fl->fl_owner;
324 new->fl_pid = fl->fl_pid;
Trond Myklebust09969052006-03-20 13:44:38 -0500325 new->fl_file = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 new->fl_flags = fl->fl_flags;
327 new->fl_type = fl->fl_type;
328 new->fl_start = fl->fl_start;
329 new->fl_end = fl->fl_end;
Kinglong Meef3282962014-08-22 10:18:43 -0400330 new->fl_lmops = fl->fl_lmops;
Trond Myklebust09969052006-03-20 13:44:38 -0500331 new->fl_ops = NULL;
Kinglong Meef3282962014-08-22 10:18:43 -0400332
333 if (fl->fl_lmops) {
334 if (fl->fl_lmops->lm_get_owner)
335 fl->fl_lmops->lm_get_owner(new, fl);
336 }
Trond Myklebust09969052006-03-20 13:44:38 -0500337}
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400338EXPORT_SYMBOL(locks_copy_conflock);
Trond Myklebust09969052006-03-20 13:44:38 -0500339
340void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
341{
Jeff Layton566709b2014-08-11 14:09:35 -0400342 /* "new" must be a freshly-initialized lock */
343 WARN_ON_ONCE(new->fl_ops);
Trond Myklebust09969052006-03-20 13:44:38 -0500344
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400345 locks_copy_conflock(new, fl);
Kinglong Meef3282962014-08-22 10:18:43 -0400346
Trond Myklebust09969052006-03-20 13:44:38 -0500347 new->fl_file = fl->fl_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 new->fl_ops = fl->fl_ops;
Trond Myklebust47831f32006-03-20 13:44:05 -0500349
Kinglong Meef3282962014-08-22 10:18:43 -0400350 if (fl->fl_ops) {
351 if (fl->fl_ops->fl_copy_lock)
352 fl->fl_ops->fl_copy_lock(new, fl);
353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354}
355
356EXPORT_SYMBOL(locks_copy_lock);
357
358static inline int flock_translate_cmd(int cmd) {
359 if (cmd & LOCK_MAND)
360 return cmd & (LOCK_MAND | LOCK_RW);
361 switch (cmd) {
362 case LOCK_SH:
363 return F_RDLCK;
364 case LOCK_EX:
365 return F_WRLCK;
366 case LOCK_UN:
367 return F_UNLCK;
368 }
369 return -EINVAL;
370}
371
372/* Fill in a file_lock structure with an appropriate FLOCK lock. */
Jeff Layton6e129d002014-09-04 10:25:06 -0400373static struct file_lock *
374flock_make_lock(struct file *filp, unsigned int cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
376 struct file_lock *fl;
377 int type = flock_translate_cmd(cmd);
Jeff Layton6e129d002014-09-04 10:25:06 -0400378
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 if (type < 0)
Jeff Layton6e129d002014-09-04 10:25:06 -0400380 return ERR_PTR(type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382 fl = locks_alloc_lock();
383 if (fl == NULL)
Jeff Layton6e129d002014-09-04 10:25:06 -0400384 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386 fl->fl_file = filp;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +0200387 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 fl->fl_pid = current->tgid;
389 fl->fl_flags = FL_FLOCK;
390 fl->fl_type = type;
391 fl->fl_end = OFFSET_MAX;
392
Jeff Layton6e129d002014-09-04 10:25:06 -0400393 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394}
395
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400396static int assign_type(struct file_lock *fl, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397{
398 switch (type) {
399 case F_RDLCK:
400 case F_WRLCK:
401 case F_UNLCK:
402 fl->fl_type = type;
403 break;
404 default:
405 return -EINVAL;
406 }
407 return 0;
408}
409
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500410static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
411 struct flock64 *l)
412{
413 switch (l->l_whence) {
414 case SEEK_SET:
415 fl->fl_start = 0;
416 break;
417 case SEEK_CUR:
418 fl->fl_start = filp->f_pos;
419 break;
420 case SEEK_END:
421 fl->fl_start = i_size_read(file_inode(filp));
422 break;
423 default:
424 return -EINVAL;
425 }
426 if (l->l_start > OFFSET_MAX - fl->fl_start)
427 return -EOVERFLOW;
428 fl->fl_start += l->l_start;
429 if (fl->fl_start < 0)
430 return -EINVAL;
431
432 /* POSIX-1996 leaves the case l->l_len < 0 undefined;
433 POSIX-2001 defines it. */
434 if (l->l_len > 0) {
435 if (l->l_len - 1 > OFFSET_MAX - fl->fl_start)
436 return -EOVERFLOW;
437 fl->fl_end = fl->fl_start + l->l_len - 1;
438
439 } else if (l->l_len < 0) {
440 if (fl->fl_start + l->l_len < 0)
441 return -EINVAL;
442 fl->fl_end = fl->fl_start - 1;
443 fl->fl_start += l->l_len;
444 } else
445 fl->fl_end = OFFSET_MAX;
446
447 fl->fl_owner = current->files;
448 fl->fl_pid = current->tgid;
449 fl->fl_file = filp;
450 fl->fl_flags = FL_POSIX;
451 fl->fl_ops = NULL;
452 fl->fl_lmops = NULL;
453
454 return assign_type(fl, l->l_type);
455}
456
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457/* Verify a "struct flock" and copy it to a "struct file_lock" as a POSIX
458 * style lock.
459 */
460static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
461 struct flock *l)
462{
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500463 struct flock64 ll = {
464 .l_type = l->l_type,
465 .l_whence = l->l_whence,
466 .l_start = l->l_start,
467 .l_len = l->l_len,
468 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
J. Bruce Fieldsef12e722014-02-03 12:13:08 -0500470 return flock64_to_posix_lock(filp, fl, &ll);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
472
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473/* default lease lock manager operations */
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400474static bool
475lease_break_callback(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476{
477 kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
Jeff Layton4d01b7f2014-09-01 15:06:54 -0400478 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479}
480
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400481static void
482lease_setup(struct file_lock *fl, void **priv)
483{
484 struct file *filp = fl->fl_file;
485 struct fasync_struct *fa = *priv;
486
487 /*
488 * fasync_insert_entry() returns the old entry if any. If there was no
489 * old entry, then it used "priv" and inserted it into the fasync list.
490 * Clear the pointer to indicate that it shouldn't be freed.
491 */
492 if (!fasync_insert_entry(fa->fa_fd, filp, &fl->fl_fasync, fa))
493 *priv = NULL;
494
495 __f_setown(filp, task_pid(current), PIDTYPE_PID, 0);
496}
497
Alexey Dobriyan7b021962009-09-21 17:01:12 -0700498static const struct lock_manager_operations lease_manager_ops = {
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400499 .lm_break = lease_break_callback,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400500 .lm_change = lease_modify,
Jeff Layton1c7dd2f2014-08-22 10:55:47 -0400501 .lm_setup = lease_setup,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502};
503
504/*
505 * Initialize a lease, use the default lock manager operations
506 */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400507static int lease_init(struct file *filp, long type, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 {
Trond Myklebust75dff552006-05-07 23:02:42 -0400509 if (assign_type(fl, type) != 0)
510 return -EINVAL;
511
Jeff Layton7ca76312014-09-01 19:04:48 -0400512 fl->fl_owner = filp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 fl->fl_pid = current->tgid;
514
515 fl->fl_file = filp;
516 fl->fl_flags = FL_LEASE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 fl->fl_start = 0;
518 fl->fl_end = OFFSET_MAX;
519 fl->fl_ops = NULL;
520 fl->fl_lmops = &lease_manager_ops;
521 return 0;
522}
523
524/* Allocate a file_lock initialised to this type of lease */
J. Bruce Fields0ec4f432012-07-23 15:17:17 -0400525static struct file_lock *lease_alloc(struct file *filp, long type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526{
527 struct file_lock *fl = locks_alloc_lock();
Trond Myklebust75dff552006-05-07 23:02:42 -0400528 int error = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
530 if (fl == NULL)
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500531 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
533 error = lease_init(filp, type, fl);
Trond Myklebust75dff552006-05-07 23:02:42 -0400534 if (error) {
535 locks_free_lock(fl);
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500536 return ERR_PTR(error);
Trond Myklebust75dff552006-05-07 23:02:42 -0400537 }
J. Bruce Fieldse32b8ee2007-03-01 14:34:35 -0500538 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539}
540
541/* Check if two locks overlap each other.
542 */
543static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
544{
545 return ((fl1->fl_end >= fl2->fl_start) &&
546 (fl2->fl_end >= fl1->fl_start));
547}
548
549/*
550 * Check whether two locks have the same owner.
551 */
Matt Mackall33443c42006-01-08 01:05:22 -0800552static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553{
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400554 if (fl1->fl_lmops && fl1->fl_lmops->lm_compare_owner)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 return fl2->fl_lmops == fl1->fl_lmops &&
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400556 fl1->fl_lmops->lm_compare_owner(fl1, fl2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 return fl1->fl_owner == fl2->fl_owner;
558}
559
Jeff Layton7012b022013-06-21 08:58:22 -0400560/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500561static void locks_insert_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400562{
Jeff Layton7012b022013-06-21 08:58:22 -0400563 lg_local_lock(&file_lock_lglock);
564 fl->fl_link_cpu = smp_processor_id();
565 hlist_add_head(&fl->fl_link, this_cpu_ptr(&file_lock_list));
566 lg_local_unlock(&file_lock_lglock);
Jeff Layton88974692013-06-21 08:58:14 -0400567}
568
Jeff Layton7012b022013-06-21 08:58:22 -0400569/* Must be called with the i_lock held! */
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500570static void locks_delete_global_locks(struct file_lock *fl)
Jeff Layton88974692013-06-21 08:58:14 -0400571{
Jeff Layton7012b022013-06-21 08:58:22 -0400572 /*
573 * Avoid taking lock if already unhashed. This is safe since this check
574 * is done while holding the i_lock, and new insertions into the list
575 * also require that it be held.
576 */
577 if (hlist_unhashed(&fl->fl_link))
578 return;
579 lg_local_lock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton139ca042013-06-21 08:58:17 -0400580 hlist_del_init(&fl->fl_link);
Jeff Layton7012b022013-06-21 08:58:22 -0400581 lg_local_unlock_cpu(&file_lock_lglock, fl->fl_link_cpu);
Jeff Layton88974692013-06-21 08:58:14 -0400582}
583
Jeff Layton3999e492013-06-21 08:58:19 -0400584static unsigned long
585posix_owner_key(struct file_lock *fl)
586{
587 if (fl->fl_lmops && fl->fl_lmops->lm_owner_key)
588 return fl->fl_lmops->lm_owner_key(fl);
589 return (unsigned long)fl->fl_owner;
590}
591
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500592static void locks_insert_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400593{
Jeff Layton3999e492013-06-21 08:58:19 -0400594 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400595}
596
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500597static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400598{
Jeff Layton48f74182013-06-21 08:58:18 -0400599 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400600}
601
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602/* Remove waiter from blocker's block list.
603 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400604 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400605 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 */
Matt Mackall33443c42006-01-08 01:05:22 -0800607static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608{
Jeff Layton88974692013-06-21 08:58:14 -0400609 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 waiter->fl_next = NULL;
612}
613
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400614static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400616 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400618 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
621/* Insert waiter into blocker's block list.
622 * We use a circular list so that processes can be easily woken up in
623 * the order they blocked. The documentation doesn't require this but
624 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400625 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400626 * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
Jeff Layton46dad762014-02-03 12:13:06 -0500627 * list itself is protected by the blocked_lock_lock, but by ensuring that the
Jeff Layton7b2296a2013-06-21 08:58:20 -0400628 * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
Jeff Layton4e8c7652013-06-21 08:58:16 -0400629 * in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400631static void __locks_insert_block(struct file_lock *blocker,
632 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800634 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400636 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400637 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400638 locks_insert_global_blocked(waiter);
639}
640
641/* Must be called with i_lock held. */
642static void locks_insert_block(struct file_lock *blocker,
643 struct file_lock *waiter)
644{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400645 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400646 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400647 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
649
Jeff Layton1cb36012013-06-21 08:58:12 -0400650/*
651 * Wake up processes blocked waiting for blocker.
652 *
Jeff Layton1c8c6012013-06-21 08:58:15 -0400653 * Must be called with the inode->i_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 */
655static void locks_wake_up_blocks(struct file_lock *blocker)
656{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400657 /*
658 * Avoid taking global lock if list is empty. This is safe since new
659 * blocked requests are only added to the list under the i_lock, and
660 * the i_lock is always held here. Note that removal from the fl_block
661 * list does not require the i_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400662 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400663 */
664 if (list_empty(&blocker->fl_block))
665 return;
666
Jeff Layton7b2296a2013-06-21 08:58:20 -0400667 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400669 struct file_lock *waiter;
670
671 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 struct file_lock, fl_block);
673 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400674 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
675 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 else
677 wake_up(&waiter->fl_wait);
678 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400679 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680}
681
Jeff Layton5263e312015-01-16 15:05:55 -0500682static void
683locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
684{
685 fl->fl_nspid = get_pid(task_tgid(current));
686 list_add_tail(&fl->fl_list, before);
687 locks_insert_global_locks(fl);
688}
689
Jeff Layton8634b512015-01-16 15:05:55 -0500690static void
691locks_unlink_lock_ctx(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Jeff Layton88974692013-06-21 08:58:14 -0400693 locks_delete_global_locks(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500694 list_del_init(&fl->fl_list);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000695 if (fl->fl_nspid) {
696 put_pid(fl->fl_nspid);
697 fl->fl_nspid = NULL;
698 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500700}
701
Jeff Layton5263e312015-01-16 15:05:55 -0500702static void
703locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
704{
Jeff Layton8634b512015-01-16 15:05:55 -0500705 locks_unlink_lock_ctx(fl);
706 if (dispose)
707 list_add(&fl->fl_list, dispose);
708 else
709 locks_free_lock(fl);
Jeff Layton5263e312015-01-16 15:05:55 -0500710}
711
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
713 * checks for shared/exclusive status of overlapping locks.
714 */
715static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
716{
717 if (sys_fl->fl_type == F_WRLCK)
718 return 1;
719 if (caller_fl->fl_type == F_WRLCK)
720 return 1;
721 return 0;
722}
723
724/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
725 * checking before calling the locks_conflict().
726 */
727static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
728{
729 /* POSIX locks owned by the same process do not conflict with
730 * each other.
731 */
732 if (!IS_POSIX(sys_fl) || posix_same_owner(caller_fl, sys_fl))
733 return (0);
734
735 /* Check whether they overlap */
736 if (!locks_overlap(caller_fl, sys_fl))
737 return 0;
738
739 return (locks_conflict(caller_fl, sys_fl));
740}
741
742/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
743 * checking before calling the locks_conflict().
744 */
745static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
746{
747 /* FLOCK locks referring to the same filp do not conflict with
748 * each other.
749 */
750 if (!IS_FLOCK(sys_fl) || (caller_fl->fl_file == sys_fl->fl_file))
751 return (0);
752 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
753 return 0;
754
755 return (locks_conflict(caller_fl, sys_fl));
756}
757
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400758void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500759posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760{
761 struct file_lock *cfl;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500762 struct file_lock_context *ctx;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400763 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500765 ctx = inode->i_flctx;
766 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400767 fl->fl_type = F_UNLCK;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500768 return;
769 }
770
771 spin_lock(&inode->i_lock);
772 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
773 if (posix_locks_conflict(fl, cfl)) {
774 locks_copy_conflock(fl, cfl);
775 if (cfl->fl_nspid)
776 fl->fl_pid = pid_vnr(cfl->fl_nspid);
777 goto out;
778 }
779 }
780 fl->fl_type = F_UNLCK;
781out:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400782 spin_unlock(&inode->i_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400783 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785EXPORT_SYMBOL(posix_test_lock);
786
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400787/*
788 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400790 * We attempt to detect deadlocks that are due purely to posix file
791 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400793 * We assume that a task can be waiting for at most one lock at a time.
794 * So for any acquired lock, the process holding that lock may be
795 * waiting on at most one other lock. That lock in turns may be held by
796 * someone waiting for at most one other lock. Given a requested lock
797 * caller_fl which is about to wait for a conflicting lock block_fl, we
798 * follow this chain of waiters to ensure we are not about to create a
799 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400800 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400801 * Since we do this before we ever put a process to sleep on a lock, we
802 * are ensured that there is never a cycle; that is what guarantees that
803 * the while() loop in posix_locks_deadlock() eventually completes.
804 *
805 * Note: the above assumption may not be true when handling lock
806 * requests from a broken NFS client. It may also fail in the presence
807 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400808 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500809 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400810 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500811 * Because the owner is not even nominally tied to a thread of
812 * execution, the deadlock detection below can't reasonably work well. Just
813 * skip it for those.
814 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400815 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500816 * locks that just checks for the case where two tasks are attempting to
817 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400819
820#define MAX_DEADLK_ITERATIONS 10
821
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400822/* Find a lock that the owner of the given block_fl is blocking on. */
823static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
824{
825 struct file_lock *fl;
826
Jeff Layton3999e492013-06-21 08:58:19 -0400827 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400828 if (posix_same_owner(fl, block_fl))
829 return fl->fl_next;
830 }
831 return NULL;
832}
833
Jeff Layton7b2296a2013-06-21 08:58:20 -0400834/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700835static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 struct file_lock *block_fl)
837{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400838 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839
Jeff Layton57b65322014-02-03 12:13:09 -0500840 /*
841 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400842 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500843 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400844 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500845 return 0;
846
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400847 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
848 if (i++ > MAX_DEADLK_ITERATIONS)
849 return 0;
850 if (posix_same_owner(caller_fl, block_fl))
851 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 }
853 return 0;
854}
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400857 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400858 *
859 * Note that if called with an FL_EXISTS argument, the caller may determine
860 * whether or not a lock was successfully freed by testing the return
861 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 */
Trond Myklebust993dfa82006-03-31 02:30:55 -0800863static int flock_lock_file(struct file *filp, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800865 struct file_lock *new_fl = NULL;
Jeff Layton5263e312015-01-16 15:05:55 -0500866 struct file_lock *fl;
867 struct file_lock_context *ctx;
868 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 int error = 0;
Jeff Layton5263e312015-01-16 15:05:55 -0500870 bool found = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400871 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Jeff Layton5263e312015-01-16 15:05:55 -0500873 ctx = locks_get_lock_context(inode);
874 if (!ctx)
875 return -ENOMEM;
876
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200877 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
878 new_fl = locks_alloc_lock();
879 if (!new_fl)
880 return -ENOMEM;
881 }
882
Jeff Layton1c8c6012013-06-21 08:58:15 -0400883 spin_lock(&inode->i_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400884 if (request->fl_flags & FL_ACCESS)
885 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400886
Jeff Layton5263e312015-01-16 15:05:55 -0500887 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 if (filp != fl->fl_file)
889 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800890 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 goto out;
Jeff Layton5263e312015-01-16 15:05:55 -0500892 found = true;
893 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 break;
895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
Trond Myklebustf475ae92006-06-29 16:38:32 -0400897 if (request->fl_type == F_UNLCK) {
898 if ((request->fl_flags & FL_EXISTS) && !found)
899 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800900 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400901 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
903 /*
904 * If a higher-priority process was blocked on the old file lock,
905 * give it the opportunity to lock the file.
906 */
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200907 if (found) {
Jeff Layton1c8c6012013-06-21 08:58:15 -0400908 spin_unlock(&inode->i_lock);
Frederic Weisbeckerdef01bc2009-07-16 15:44:29 +0200909 cond_resched();
Jeff Layton1c8c6012013-06-21 08:58:15 -0400910 spin_lock(&inode->i_lock);
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700912
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400913find_conflict:
Jeff Layton5263e312015-01-16 15:05:55 -0500914 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Trond Myklebust993dfa82006-03-31 02:30:55 -0800915 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916 continue;
917 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700918 if (!(request->fl_flags & FL_SLEEP))
919 goto out;
920 error = FILE_LOCK_DEFERRED;
921 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 goto out;
923 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400924 if (request->fl_flags & FL_ACCESS)
925 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800926 locks_copy_lock(new_fl, request);
Jeff Layton5263e312015-01-16 15:05:55 -0500927 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800928 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400929 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
931out:
Jeff Layton1c8c6012013-06-21 08:58:15 -0400932 spin_unlock(&inode->i_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800933 if (new_fl)
934 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400935 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 return error;
937}
938
Marc Eshel150b3932007-01-18 16:15:35 -0500939static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500941 struct file_lock *fl, *tmp;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700942 struct file_lock *new_fl = NULL;
943 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 struct file_lock *left = NULL;
945 struct file_lock *right = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500946 struct file_lock_context *ctx;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400947 int error;
948 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400949 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500951 ctx = locks_get_lock_context(inode);
952 if (!ctx)
953 return -ENOMEM;
954
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 /*
956 * We may need two file_lock structures for this operation,
957 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700958 *
959 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700961 if (!(request->fl_flags & FL_ACCESS) &&
962 (request->fl_type != F_UNLCK ||
963 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
964 new_fl = locks_alloc_lock();
965 new_fl2 = locks_alloc_lock();
966 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967
Jeff Layton1c8c6012013-06-21 08:58:15 -0400968 spin_lock(&inode->i_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400969 /*
970 * New lock request. Walk all POSIX locks and look for conflicts. If
971 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400972 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400973 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (request->fl_type != F_UNLCK) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500975 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 if (!IS_POSIX(fl))
977 continue;
978 if (!posix_locks_conflict(request, fl))
979 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800980 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400981 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 error = -EAGAIN;
983 if (!(request->fl_flags & FL_SLEEP))
984 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400985 /*
986 * Deadlock detection and insertion into the blocked
987 * locks list must be done while holding the same lock!
988 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400990 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400991 if (likely(!posix_locks_deadlock(request, fl))) {
992 error = FILE_LOCK_DEFERRED;
993 __locks_insert_block(fl, request);
994 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400995 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 goto out;
997 }
998 }
999
1000 /* If we're just looking for a conflict, we're done. */
1001 error = 0;
1002 if (request->fl_flags & FL_ACCESS)
1003 goto out;
1004
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001005 /* Find the first old lock with the same owner as the new lock */
1006 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1007 if (posix_same_owner(request, fl))
1008 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
1010
Jeff Layton1cb36012013-06-21 08:58:12 -04001011 /* Process locks with this owner. */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001012 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1013 if (!posix_same_owner(request, fl))
1014 break;
1015
1016 /* Detect adjacent or overlapping regions (if same lock type) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -07001018 /* In all comparisons of start vs end, use
1019 * "start - 1" rather than "end + 1". If end
1020 * is OFFSET_MAX, end + 1 will become negative.
1021 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 if (fl->fl_end < request->fl_start - 1)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001023 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 /* If the next lock in the list has entirely bigger
1025 * addresses than the new one, insert the lock here.
1026 */
Olaf Kirch449231d2005-08-25 16:25:35 -07001027 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 break;
1029
1030 /* If we come here, the new and old lock are of the
1031 * same type and adjacent or overlapping. Make one
1032 * lock yielding from the lower start address of both
1033 * locks to the higher end address.
1034 */
1035 if (fl->fl_start > request->fl_start)
1036 fl->fl_start = request->fl_start;
1037 else
1038 request->fl_start = fl->fl_start;
1039 if (fl->fl_end < request->fl_end)
1040 fl->fl_end = request->fl_end;
1041 else
1042 request->fl_end = fl->fl_end;
1043 if (added) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001044 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 continue;
1046 }
1047 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001048 added = true;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001049 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /* Processing for different lock types is a bit
1051 * more complex.
1052 */
1053 if (fl->fl_end < request->fl_start)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001054 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (fl->fl_start > request->fl_end)
1056 break;
1057 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001058 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 if (fl->fl_start < request->fl_start)
1060 left = fl;
1061 /* If the next lock in the list has a higher end
1062 * address than the new one, insert the new one here.
1063 */
1064 if (fl->fl_end > request->fl_end) {
1065 right = fl;
1066 break;
1067 }
1068 if (fl->fl_start >= request->fl_start) {
1069 /* The new lock completely replaces an old
1070 * one (This may happen several times).
1071 */
1072 if (added) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001073 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 continue;
1075 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001076 /*
1077 * Replace the old lock with new_fl, and
1078 * remove the old one. It's safe to do the
1079 * insert here since we know that we won't be
1080 * using new_fl later, and that the lock is
1081 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001083 error = -ENOLCK;
1084 if (!new_fl)
1085 goto out;
1086 locks_copy_lock(new_fl, request);
1087 request = new_fl;
1088 new_fl = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001089 locks_insert_lock_ctx(request, &fl->fl_list);
1090 locks_delete_lock_ctx(fl, &dispose);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001091 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092 }
1093 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 }
1095
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001096 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001097 * The above code only modifies existing locks in case of merging or
1098 * replacing. If new lock(s) need to be inserted all modifications are
1099 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001100 */
1101 error = -ENOLCK; /* "no luck" */
1102 if (right && left == right && !new_fl2)
1103 goto out;
1104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 error = 0;
1106 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001107 if (request->fl_type == F_UNLCK) {
1108 if (request->fl_flags & FL_EXISTS)
1109 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001111 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001112
1113 if (!new_fl) {
1114 error = -ENOLCK;
1115 goto out;
1116 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 locks_copy_lock(new_fl, request);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001118 locks_insert_lock_ctx(new_fl, &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 new_fl = NULL;
1120 }
1121 if (right) {
1122 if (left == right) {
1123 /* The new lock breaks the old one in two pieces,
1124 * so we have to use the second new lock.
1125 */
1126 left = new_fl2;
1127 new_fl2 = NULL;
1128 locks_copy_lock(left, right);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001129 locks_insert_lock_ctx(left, &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
1131 right->fl_start = request->fl_end + 1;
1132 locks_wake_up_blocks(right);
1133 }
1134 if (left) {
1135 left->fl_end = request->fl_start - 1;
1136 locks_wake_up_blocks(left);
1137 }
1138 out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001139 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 /*
1141 * Free any unused locks.
1142 */
1143 if (new_fl)
1144 locks_free_lock(new_fl);
1145 if (new_fl2)
1146 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001147 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 return error;
1149}
1150
1151/**
1152 * posix_lock_file - Apply a POSIX-style lock to a file
1153 * @filp: The file to apply the lock to
1154 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001155 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 *
1157 * Add a POSIX style lock to a file.
1158 * We merge adjacent & overlapping locks whenever possible.
1159 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001160 *
1161 * Note that if called with an FL_EXISTS argument, the caller may determine
1162 * whether or not a lock was successfully freed by testing the return
1163 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 */
Marc Eshel150b3932007-01-18 16:15:35 -05001165int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001166 struct file_lock *conflock)
1167{
Al Viro496ad9a2013-01-23 17:07:38 -05001168 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001169}
Marc Eshel150b3932007-01-18 16:15:35 -05001170EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171
1172/**
1173 * posix_lock_file_wait - Apply a POSIX-style lock to a file
1174 * @filp: The file to apply the lock to
1175 * @fl: The lock to be applied
1176 *
1177 * Add a POSIX style lock to a file.
1178 * We merge adjacent & overlapping locks whenever possible.
1179 * POSIX locks are sorted by owner task, then by starting address
1180 */
1181int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
1182{
1183 int error;
1184 might_sleep ();
1185 for (;;) {
Marc Eshel150b3932007-01-18 16:15:35 -05001186 error = posix_lock_file(filp, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001187 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 break;
1189 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1190 if (!error)
1191 continue;
1192
1193 locks_delete_block(fl);
1194 break;
1195 }
1196 return error;
1197}
1198EXPORT_SYMBOL(posix_lock_file_wait);
1199
1200/**
1201 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001202 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 *
1204 * Searches the inode's list of locks to find any POSIX locks which conflict.
1205 * This function is called from locks_verify_locked() only.
1206 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001207int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001209 int ret;
Jeff Laytond7a06982014-03-10 09:54:15 -04001210 struct inode *inode = file_inode(file);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001211 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 struct file_lock *fl;
1213
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001214 ctx = inode->i_flctx;
1215 if (!ctx || list_empty_careful(&ctx->flc_posix))
1216 return 0;
1217
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 /*
1219 * Search the lock list for this inode for any POSIX locks.
1220 */
Jeff Layton1c8c6012013-06-21 08:58:15 -04001221 spin_lock(&inode->i_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001222 ret = 0;
1223 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001224 if (fl->fl_owner != current->files &&
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001225 fl->fl_owner != file) {
1226 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 break;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 }
Jeff Layton1c8c6012013-06-21 08:58:15 -04001230 spin_unlock(&inode->i_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001231 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232}
1233
1234/**
1235 * locks_mandatory_area - Check for a conflicting lock
1236 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1237 * for shared
1238 * @inode: the file to check
1239 * @filp: how the file was opened (if it was)
1240 * @offset: start of area to check
1241 * @count: length of area to check
1242 *
1243 * Searches the inode's list of locks to find any POSIX locks which conflict.
1244 * This function is called from rw_verify_area() and
1245 * locks_verify_truncate().
1246 */
1247int locks_mandatory_area(int read_write, struct inode *inode,
1248 struct file *filp, loff_t offset,
1249 size_t count)
1250{
1251 struct file_lock fl;
1252 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001253 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254
1255 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 fl.fl_pid = current->tgid;
1257 fl.fl_file = filp;
1258 fl.fl_flags = FL_POSIX | FL_ACCESS;
1259 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001260 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1262 fl.fl_start = offset;
1263 fl.fl_end = offset + count - 1;
1264
1265 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001266 if (filp) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001267 fl.fl_owner = filp;
Jeff Layton29723ad2014-03-10 09:54:19 -04001268 fl.fl_flags &= ~FL_SLEEP;
1269 error = __posix_lock_file(inode, &fl, NULL);
1270 if (!error)
1271 break;
1272 }
1273
1274 if (sleep)
1275 fl.fl_flags |= FL_SLEEP;
1276 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001277 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001278 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 break;
1280 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1281 if (!error) {
1282 /*
1283 * If we've been sleeping someone might have
1284 * changed the permissions behind our back.
1285 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001286 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 continue;
1288 }
1289
1290 locks_delete_block(&fl);
1291 break;
1292 }
1293
1294 return error;
1295}
1296
1297EXPORT_SYMBOL(locks_mandatory_area);
1298
J. Bruce Fields778fc542011-07-26 18:25:49 -04001299static void lease_clear_pending(struct file_lock *fl, int arg)
1300{
1301 switch (arg) {
1302 case F_UNLCK:
1303 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1304 /* fall through: */
1305 case F_RDLCK:
1306 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1307 }
1308}
1309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310/* We already had a lease on this file; just change its type */
Jeff Laytonc45198e2014-09-01 07:12:07 -04001311int lease_modify(struct file_lock **before, int arg, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312{
1313 struct file_lock *fl = *before;
1314 int error = assign_type(fl, arg);
1315
1316 if (error)
1317 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001318 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001320 if (arg == F_UNLCK) {
1321 struct file *filp = fl->fl_file;
1322
1323 f_delown(filp);
1324 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001325 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1326 if (fl->fl_fasync != NULL) {
1327 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1328 fl->fl_fasync = NULL;
1329 }
Jeff Layton8634b512015-01-16 15:05:55 -05001330 locks_delete_lock_ctx(fl, dispose);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001331 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 return 0;
1333}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334EXPORT_SYMBOL(lease_modify);
1335
J. Bruce Fields778fc542011-07-26 18:25:49 -04001336static bool past_time(unsigned long then)
1337{
1338 if (!then)
1339 /* 0 is a special value meaning "this never expires": */
1340 return false;
1341 return time_after(jiffies, then);
1342}
1343
Jeff Laytonc45198e2014-09-01 07:12:07 -04001344static void time_out_leases(struct inode *inode, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
Jeff Layton8634b512015-01-16 15:05:55 -05001346 struct file_lock_context *ctx = inode->i_flctx;
1347 struct file_lock *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001349 lockdep_assert_held(&inode->i_lock);
1350
Jeff Layton8634b512015-01-16 15:05:55 -05001351 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001352 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001353 if (past_time(fl->fl_downgrade_time))
Jeff Layton8634b512015-01-16 15:05:55 -05001354 lease_modify(&fl, F_RDLCK, dispose);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001355 if (past_time(fl->fl_break_time))
Jeff Layton8634b512015-01-16 15:05:55 -05001356 lease_modify(&fl, F_UNLCK, dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 }
1358}
1359
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001360static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1361{
1362 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1363 return false;
1364 return locks_conflict(breaker, lease);
1365}
1366
Jeff Layton03d12dd2014-09-01 14:53:41 -04001367static bool
1368any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1369{
Jeff Layton8634b512015-01-16 15:05:55 -05001370 struct file_lock_context *ctx = inode->i_flctx;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001371 struct file_lock *fl;
1372
1373 lockdep_assert_held(&inode->i_lock);
1374
Jeff Layton8634b512015-01-16 15:05:55 -05001375 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Jeff Layton03d12dd2014-09-01 14:53:41 -04001376 if (leases_conflict(fl, breaker))
1377 return true;
1378 }
1379 return false;
1380}
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382/**
1383 * __break_lease - revoke all outstanding leases on file
1384 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001385 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1386 * break all leases
1387 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1388 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 *
david m. richter87250dd2007-05-09 16:10:27 -04001390 * break_lease (inlined for speed) has checked there already is at least
1391 * some kind of lock (maybe a lease) on this file. Leases are broken on
1392 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 * specified %O_NONBLOCK to your open().
1394 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001395int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001397 int error = 0;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001398 struct file_lock *new_fl;
Jeff Layton8634b512015-01-16 15:05:55 -05001399 struct file_lock_context *ctx = inode->i_flctx;
1400 struct file_lock *fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 unsigned long break_time;
Al Viro8737c932009-12-24 06:47:55 -05001402 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001403 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404
Al Viro8737c932009-12-24 06:47:55 -05001405 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001406 if (IS_ERR(new_fl))
1407 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001408 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409
Jeff Layton8634b512015-01-16 15:05:55 -05001410 /* typically we will check that ctx is non-NULL before calling */
1411 if (!ctx) {
1412 WARN_ON_ONCE(1);
1413 return error;
1414 }
1415
Jeff Layton1c8c6012013-06-21 08:58:15 -04001416 spin_lock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
Jeff Laytonc45198e2014-09-01 07:12:07 -04001418 time_out_leases(inode, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419
Jeff Layton03d12dd2014-09-01 14:53:41 -04001420 if (!any_leases_conflict(inode, new_fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001421 goto out;
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 break_time = 0;
1424 if (lease_break_time > 0) {
1425 break_time = jiffies + lease_break_time * HZ;
1426 if (break_time == 0)
1427 break_time++; /* so that 0 means no break time */
1428 }
1429
Jeff Layton8634b512015-01-16 15:05:55 -05001430 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001431 if (!leases_conflict(fl, new_fl))
1432 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001433 if (want_write) {
1434 if (fl->fl_flags & FL_UNLOCK_PENDING)
1435 continue;
1436 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001438 } else {
Jeff Layton8634b512015-01-16 15:05:55 -05001439 if (lease_breaking(fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001440 continue;
1441 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1442 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001444 if (fl->fl_lmops->lm_break(fl))
Jeff Layton8634b512015-01-16 15:05:55 -05001445 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 }
1447
Jeff Layton8634b512015-01-16 15:05:55 -05001448 if (list_empty(&ctx->flc_lease))
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001449 goto out;
1450
Jeff Layton843c6b22014-09-01 14:27:43 -04001451 if (mode & O_NONBLOCK) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001452 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 error = -EWOULDBLOCK;
1454 goto out;
1455 }
1456
1457restart:
Jeff Layton8634b512015-01-16 15:05:55 -05001458 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1459 break_time = fl->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001460 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001462 if (break_time == 0)
1463 break_time++;
Jeff Layton8634b512015-01-16 15:05:55 -05001464 locks_insert_block(fl, new_fl);
Jeff Layton62af4f12014-05-09 14:13:05 -04001465 trace_break_lease_block(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001466 spin_unlock(&inode->i_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001467 locks_dispose_list(&dispose);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001468 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1469 !new_fl->fl_next, break_time);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001470 spin_lock(&inode->i_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001471 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001472 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (error >= 0) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001474 /*
1475 * Wait for the next conflicting lease that has not been
1476 * broken yet
1477 */
Jeff Layton03d12dd2014-09-01 14:53:41 -04001478 if (error == 0)
1479 time_out_leases(inode, &dispose);
1480 if (any_leases_conflict(inode, new_fl))
1481 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 error = 0;
1483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484out:
Jeff Layton1c8c6012013-06-21 08:58:15 -04001485 spin_unlock(&inode->i_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001486 locks_dispose_list(&dispose);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001487 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 return error;
1489}
1490
1491EXPORT_SYMBOL(__break_lease);
1492
1493/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001494 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 * @inode: the inode
1496 * @time: pointer to a timespec which will contain the last modified time
1497 *
1498 * This is to force NFS clients to flush their caches for files with
1499 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001500 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 */
1502void lease_get_mtime(struct inode *inode, struct timespec *time)
1503{
Jeff Laytonbfe86022014-08-22 10:18:44 -04001504 bool has_lease = false;
Jeff Layton8634b512015-01-16 15:05:55 -05001505 struct file_lock_context *ctx = inode->i_flctx;
1506 struct file_lock *fl;
Jeff Laytonbfe86022014-08-22 10:18:44 -04001507
Jeff Layton8634b512015-01-16 15:05:55 -05001508 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Laytonbfe86022014-08-22 10:18:44 -04001509 spin_lock(&inode->i_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001510 if (!list_empty(&ctx->flc_lease)) {
1511 fl = list_first_entry(&ctx->flc_lease,
1512 struct file_lock, fl_list);
1513 if (fl->fl_type == F_WRLCK)
1514 has_lease = true;
1515 }
Jeff Laytonbfe86022014-08-22 10:18:44 -04001516 spin_unlock(&inode->i_lock);
1517 }
1518
1519 if (has_lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 *time = current_fs_time(inode->i_sb);
1521 else
1522 *time = inode->i_mtime;
1523}
1524
1525EXPORT_SYMBOL(lease_get_mtime);
1526
1527/**
1528 * fcntl_getlease - Enquire what lease is currently active
1529 * @filp: the file
1530 *
1531 * The value returned by this function will be one of
1532 * (if no lease break is pending):
1533 *
1534 * %F_RDLCK to indicate a shared lease is held.
1535 *
1536 * %F_WRLCK to indicate an exclusive lease is held.
1537 *
1538 * %F_UNLCK to indicate no lease is held.
1539 *
1540 * (if a lease break is pending):
1541 *
1542 * %F_RDLCK to indicate an exclusive lease needs to be
1543 * changed to a shared lease (or removed).
1544 *
1545 * %F_UNLCK to indicate the lease needs to be removed.
1546 *
1547 * XXX: sfr & willy disagree over whether F_INPROGRESS
1548 * should be returned to userspace.
1549 */
1550int fcntl_getlease(struct file *filp)
1551{
1552 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001553 struct inode *inode = file_inode(filp);
Jeff Layton8634b512015-01-16 15:05:55 -05001554 struct file_lock_context *ctx = inode->i_flctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 int type = F_UNLCK;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001556 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
Jeff Layton8634b512015-01-16 15:05:55 -05001558 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
1559 spin_lock(&inode->i_lock);
1560 time_out_leases(file_inode(filp), &dispose);
1561 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1562 if (fl->fl_file != filp)
1563 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001564 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565 break;
1566 }
Jeff Layton8634b512015-01-16 15:05:55 -05001567 spin_unlock(&inode->i_lock);
1568 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return type;
1571}
1572
Jeff Layton24cbe782014-02-03 12:13:06 -05001573/**
1574 * check_conflicting_open - see if the given dentry points to a file that has
1575 * an existing open that would conflict with the
1576 * desired lease.
1577 * @dentry: dentry to check
1578 * @arg: type of lease that we're trying to acquire
1579 *
1580 * Check to see if there's an existing open fd on this file that would
1581 * conflict with the lease we're trying to set.
1582 */
1583static int
1584check_conflicting_open(const struct dentry *dentry, const long arg)
1585{
1586 int ret = 0;
1587 struct inode *inode = dentry->d_inode;
1588
1589 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1590 return -EAGAIN;
1591
1592 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1593 (atomic_read(&inode->i_count) > 1)))
1594 ret = -EAGAIN;
1595
1596 return ret;
1597}
1598
Jeff Laytone6f5c782014-08-22 10:40:25 -04001599static int
1600generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601{
Jeff Layton8634b512015-01-16 15:05:55 -05001602 struct file_lock *fl, *my_fl = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001603 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001605 struct file_lock_context *ctx;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001606 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001607 int error;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001608 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609
J. Bruce Fields096657b2010-10-30 17:31:14 -04001610 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001611 trace_generic_add_lease(inode, lease);
1612
Jeff Layton8634b512015-01-16 15:05:55 -05001613 ctx = locks_get_lock_context(inode);
1614 if (!ctx)
1615 return -ENOMEM;
1616
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001617 /*
1618 * In the delegation case we need mutual exclusion with
1619 * a number of operations that take the i_mutex. We trylock
1620 * because delegations are an optional optimization, and if
1621 * there's some chance of a conflict--we'd rather not
1622 * bother, maybe that's a sign this just isn't a good file to
1623 * hand out a delegation on.
1624 */
1625 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1626 return -EAGAIN;
1627
1628 if (is_deleg && arg == F_WRLCK) {
1629 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001630 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001631 WARN_ON_ONCE(1);
1632 return -EINVAL;
1633 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001634
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001635 spin_lock(&inode->i_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001636 time_out_leases(inode, &dispose);
Jeff Layton24cbe782014-02-03 12:13:06 -05001637 error = check_conflicting_open(dentry, arg);
1638 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001639 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 /*
1642 * At this point, we know that if there is an exclusive
1643 * lease on this file, then we hold it on this filp
1644 * (otherwise our open of this file would have blocked).
1645 * And if we are trying to acquire an exclusive lease,
1646 * then the file is not open by anyone (including us)
1647 * except for this filp.
1648 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001649 error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001650 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001651 if (fl->fl_file == filp) {
Jeff Layton8634b512015-01-16 15:05:55 -05001652 my_fl = fl;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001653 continue;
1654 }
Jeff Layton8634b512015-01-16 15:05:55 -05001655
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001656 /*
1657 * No exclusive leases if someone else has a lease on
1658 * this file:
1659 */
1660 if (arg == F_WRLCK)
1661 goto out;
1662 /*
1663 * Modifying our existing lease is OK, but no getting a
1664 * new lease if someone else is opening for write:
1665 */
1666 if (fl->fl_flags & FL_UNLOCK_PENDING)
1667 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 }
1669
Jeff Layton8634b512015-01-16 15:05:55 -05001670 if (my_fl != NULL) {
1671 error = lease->fl_lmops->lm_change(&my_fl, arg, &dispose);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001672 if (error)
1673 goto out;
1674 goto out_setup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 }
1676
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 error = -EINVAL;
1678 if (!leases_enable)
1679 goto out;
1680
Jeff Layton8634b512015-01-16 15:05:55 -05001681 locks_insert_lock_ctx(lease, &ctx->flc_lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001682 /*
1683 * The check in break_lease() is lockless. It's possible for another
1684 * open to race in after we did the earlier check for a conflicting
1685 * open but before the lease was inserted. Check again for a
1686 * conflicting open and cancel the lease if there is one.
1687 *
1688 * We also add a barrier here to ensure that the insertion of the lock
1689 * precedes these checks.
1690 */
1691 smp_mb();
1692 error = check_conflicting_open(dentry, arg);
Jeff Layton8634b512015-01-16 15:05:55 -05001693 if (error) {
1694 locks_unlink_lock_ctx(lease);
1695 goto out;
1696 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001697
1698out_setup:
1699 if (lease->fl_lmops->lm_setup)
1700 lease->fl_lmops->lm_setup(lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701out:
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001702 spin_unlock(&inode->i_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001703 locks_dispose_list(&dispose);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001704 if (is_deleg)
1705 mutex_unlock(&inode->i_mutex);
Jeff Layton8634b512015-01-16 15:05:55 -05001706 if (!error && !my_fl)
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001707 *flp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 return error;
1709}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001710
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001711static int generic_delete_lease(struct file *filp)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001712{
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001713 int error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001714 struct file_lock *fl, *victim = NULL;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001715 struct dentry *dentry = filp->f_path.dentry;
1716 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001717 struct file_lock_context *ctx = inode->i_flctx;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001718 LIST_HEAD(dispose);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001719
Jeff Layton8634b512015-01-16 15:05:55 -05001720 if (!ctx) {
1721 trace_generic_delete_lease(inode, NULL);
1722 return error;
1723 }
1724
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001725 spin_lock(&inode->i_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001726 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1727 if (fl->fl_file == filp) {
1728 victim = fl;
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001729 break;
Jeff Layton8634b512015-01-16 15:05:55 -05001730 }
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001731 }
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001732 trace_generic_delete_lease(inode, fl);
Jeff Layton8634b512015-01-16 15:05:55 -05001733 if (victim)
1734 error = fl->fl_lmops->lm_change(&victim, F_UNLCK, &dispose);
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001735 spin_unlock(&inode->i_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001736 locks_dispose_list(&dispose);
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001737 return error;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001738}
1739
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740/**
1741 * generic_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001742 * @filp: file pointer
1743 * @arg: type of lease to obtain
1744 * @flp: input - file_lock to use, output - file_lock inserted
1745 * @priv: private data for lm_setup (may be NULL if lm_setup
1746 * doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 *
1748 * The (input) flp->fl_lmops->lm_break function is required
1749 * by break_lease().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001751int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1752 void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 struct dentry *dentry = filp->f_path.dentry;
1755 struct inode *inode = dentry->d_inode;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001756 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001758 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001759 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001761 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 error = security_file_lock(filp, arg);
1763 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001764 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001766 switch (arg) {
1767 case F_UNLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001768 return generic_delete_lease(filp);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001769 case F_RDLCK:
1770 case F_WRLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001771 if (!(*flp)->fl_lmops->lm_break) {
1772 WARN_ON_ONCE(1);
1773 return -ENOLCK;
1774 }
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)
Jeff Laytone6f5c782014-08-22 10:40:25 -04001847 return vfs_setlease(filp, F_UNLCK, NULL, NULL);
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
2202 * update of inode->i_flock and check for it done in close().
2203 * 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 Laytondd459bb2015-01-16 15:05:54 -05002390static void
2391locks_remove_flock(struct file *filp)
2392{
2393 struct file_lock fl = {
2394 .fl_owner = filp,
2395 .fl_pid = current->tgid,
2396 .fl_file = filp,
2397 .fl_flags = FL_FLOCK,
2398 .fl_type = F_UNLCK,
2399 .fl_end = OFFSET_MAX,
2400 };
Jeff Layton5263e312015-01-16 15:05:55 -05002401 struct file_lock_context *flctx = file_inode(filp)->i_flctx;
Jeff Laytondd459bb2015-01-16 15:05:54 -05002402
Jeff Layton5263e312015-01-16 15:05:55 -05002403 if (!flctx || list_empty(&flctx->flc_flock))
Jeff Laytondd459bb2015-01-16 15:05:54 -05002404 return;
2405
2406 if (filp->f_op->flock)
2407 filp->f_op->flock(filp, F_SETLKW, &fl);
2408 else
2409 flock_lock_file(filp, &fl);
2410
2411 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2412 fl.fl_ops->fl_release_private(&fl);
2413}
2414
Jeff Layton8634b512015-01-16 15:05:55 -05002415static void
2416locks_remove_lease(struct file *filp)
2417{
2418 struct inode *inode = file_inode(filp);
2419 struct file_lock_context *ctx = inode->i_flctx;
2420 struct file_lock *fl, *tmp;
2421 LIST_HEAD(dispose);
2422
2423 if (!ctx || list_empty(&ctx->flc_lease))
2424 return;
2425
2426 spin_lock(&inode->i_lock);
2427 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
2428 lease_modify(&fl, F_UNLCK, &dispose);
2429 spin_unlock(&inode->i_lock);
2430 locks_dispose_list(&dispose);
2431}
2432
Linus Torvalds1da177e2005-04-16 15:20:36 -07002433/*
2434 * This function is called on the last close of an open file.
2435 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002436void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437{
Jeff Laytondd459bb2015-01-16 15:05:54 -05002438 /* remove any OFD locks */
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002439 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002440
Jeff Laytondd459bb2015-01-16 15:05:54 -05002441 /* remove flock locks */
2442 locks_remove_flock(filp);
2443
Jeff Layton8634b512015-01-16 15:05:55 -05002444 /* remove any leases */
2445 locks_remove_lease(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446}
2447
2448/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450 * @waiter: the lock which was waiting
2451 *
2452 * lockd needs to block waiting for locks.
2453 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002454int
Jeff Laytonf891a292013-06-21 08:58:09 -04002455posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002457 int status = 0;
2458
Jeff Layton7b2296a2013-06-21 08:58:20 -04002459 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002460 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002462 else
2463 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002464 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002465 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467EXPORT_SYMBOL(posix_unblock_lock);
2468
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002469/**
2470 * vfs_cancel_lock - file byte range unblock lock
2471 * @filp: The file to apply the unblock to
2472 * @fl: The lock to be unblocked
2473 *
2474 * Used by lock managers to cancel blocked requests
2475 */
2476int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2477{
Al Viro72c2d532013-09-22 16:27:52 -04002478 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002479 return filp->f_op->lock(filp, F_CANCELLK, fl);
2480 return 0;
2481}
2482
2483EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2484
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002485#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002486#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002487#include <linux/seq_file.h>
2488
Jeff Layton7012b022013-06-21 08:58:22 -04002489struct locks_iterator {
2490 int li_cpu;
2491 loff_t li_pos;
2492};
2493
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002494static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002495 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496{
2497 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002498 unsigned int fl_pid;
2499
2500 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002501 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002502 else
2503 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504
2505 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002506 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507
Jerome Marchand99dc8292010-10-26 14:22:33 -07002508 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002509 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002510 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002511 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002512 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002513 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002514 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002515 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002516
2517 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002519 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 } else if (IS_FLOCK(fl)) {
2521 if (fl->fl_type & LOCK_MAND) {
Fabian Frederick5315c262014-05-09 14:13:05 -04002522 seq_puts(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002524 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525 }
2526 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002527 if (fl->fl_flags & FL_DELEG)
2528 seq_puts(f, "DELEG ");
2529 else
2530 seq_puts(f, "LEASE ");
2531
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002532 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002533 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002535 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002537 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002539 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540 }
2541 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002542 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002543 (fl->fl_type & LOCK_READ)
2544 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2545 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2546 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002547 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002548 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002549 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2550 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 }
2552 if (inode) {
2553#ifdef WE_CAN_BREAK_LSLK_NOW
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002554 seq_printf(f, "%d %s:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 inode->i_sb->s_id, inode->i_ino);
2556#else
2557 /* userspace relies on this representation of dev_t ;-( */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002558 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 MAJOR(inode->i_sb->s_dev),
2560 MINOR(inode->i_sb->s_dev), inode->i_ino);
2561#endif
2562 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002563 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 }
2565 if (IS_POSIX(fl)) {
2566 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002567 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002569 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002571 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 }
2573}
2574
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002575static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576{
Jeff Layton7012b022013-06-21 08:58:22 -04002577 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002578 struct file_lock *fl, *bfl;
2579
Jeff Layton139ca042013-06-21 08:58:17 -04002580 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002581
Jeff Layton7012b022013-06-21 08:58:22 -04002582 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002583
2584 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002585 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002586
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002587 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002588}
2589
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002590static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002591 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592{
Jeff Layton7012b022013-06-21 08:58:22 -04002593 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002594
Jeff Layton7012b022013-06-21 08:58:22 -04002595 iter->li_pos = *pos + 1;
2596 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002597 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002598 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599}
2600
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002601static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2602{
Jeff Layton7012b022013-06-21 08:58:22 -04002603 struct locks_iterator *iter = f->private;
2604
2605 ++iter->li_pos;
2606 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002607}
2608
2609static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002610 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002611{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002612 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002613 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002614}
2615
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002616static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002617 .start = locks_start,
2618 .next = locks_next,
2619 .stop = locks_stop,
2620 .show = locks_show,
2621};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002622
2623static int locks_open(struct inode *inode, struct file *filp)
2624{
Jeff Layton7012b022013-06-21 08:58:22 -04002625 return seq_open_private(filp, &locks_seq_operations,
2626 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002627}
2628
2629static const struct file_operations proc_locks_operations = {
2630 .open = locks_open,
2631 .read = seq_read,
2632 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002633 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002634};
2635
2636static int __init proc_locks_init(void)
2637{
2638 proc_create("locks", 0, NULL, &proc_locks_operations);
2639 return 0;
2640}
2641module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002642#endif
2643
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644static int __init filelock_init(void)
2645{
Jeff Layton7012b022013-06-21 08:58:22 -04002646 int i;
2647
Jeff Layton4a075e32015-01-16 15:05:54 -05002648 flctx_cache = kmem_cache_create("file_lock_ctx",
2649 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2650
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002652 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2653
Jeff Layton7012b022013-06-21 08:58:22 -04002654 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2655
2656 for_each_possible_cpu(i)
2657 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2658
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659 return 0;
2660}
2661
2662core_initcall(filelock_init);