blob: 4667f82267476356b61fffa3834f75aa7d10bd19 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/locks.c
3 *
4 * Provide support for fcntl()'s F_GETLK, F_SETLK, and F_SETLKW calls.
5 * Doug Evans (dje@spiff.uucp), August 07, 1992
6 *
7 * Deadlock detection added.
8 * FIXME: one thing isn't handled yet:
9 * - mandatory locks (requires lots of changes elsewhere)
10 * Kelly Carmichael (kelly@[142.24.8.65]), September 17, 1994.
11 *
12 * Miscellaneous edits, and a total rewrite of posix_lock_file() code.
13 * Kai Petzke (wpp@marie.physik.tu-berlin.de), 1994
14 *
15 * Converted file_lock_table to a linked list from an array, which eliminates
16 * the limits on how many active file locks are open.
17 * Chad Page (pageone@netcom.com), November 27, 1994
18 *
19 * Removed dependency on file descriptors. dup()'ed file descriptors now
20 * get the same locks as the original file descriptors, and a close() on
21 * any file descriptor removes ALL the locks on the file for the current
22 * process. Since locks still depend on the process id, locks are inherited
23 * after an exec() but not after a fork(). This agrees with POSIX, and both
24 * BSD and SVR4 practice.
25 * Andy Walker (andy@lysaker.kvaerner.no), February 14, 1995
26 *
27 * Scrapped free list which is redundant now that we allocate locks
28 * dynamically with kmalloc()/kfree().
29 * Andy Walker (andy@lysaker.kvaerner.no), February 21, 1995
30 *
31 * Implemented two lock personalities - FL_FLOCK and FL_POSIX.
32 *
33 * FL_POSIX locks are created with calls to fcntl() and lockf() through the
34 * fcntl() system call. They have the semantics described above.
35 *
36 * FL_FLOCK locks are created with calls to flock(), through the flock()
37 * system call, which is new. Old C libraries implement flock() via fcntl()
38 * and will continue to use the old, broken implementation.
39 *
40 * FL_FLOCK locks follow the 4.4 BSD flock() semantics. They are associated
41 * with a file pointer (filp). As a result they can be shared by a parent
42 * process and its children after a fork(). They are removed when the last
43 * file descriptor referring to the file pointer is closed (unless explicitly
44 * unlocked).
45 *
46 * FL_FLOCK locks never deadlock, an existing lock is always removed before
47 * upgrading from shared to exclusive (or vice versa). When this happens
48 * any processes blocked by the current lock are woken up and allowed to
49 * run before the new lock is applied.
50 * Andy Walker (andy@lysaker.kvaerner.no), June 09, 1995
51 *
52 * Removed some race conditions in flock_lock_file(), marked other possible
53 * races. Just grep for FIXME to see them.
54 * Dmitry Gorodchanin (pgmdsg@ibi.com), February 09, 1996.
55 *
56 * Addressed Dmitry's concerns. Deadlock checking no longer recursive.
57 * Lock allocation changed to GFP_ATOMIC as we can't afford to sleep
58 * once we've checked for blocking and deadlocking.
59 * Andy Walker (andy@lysaker.kvaerner.no), April 03, 1996.
60 *
61 * Initial implementation of mandatory locks. SunOS turned out to be
62 * a rotten model, so I implemented the "obvious" semantics.
Paul Bolle395cf962011-08-15 02:02:26 +020063 * See 'Documentation/filesystems/mandatory-locking.txt' for details.
Linus Torvalds1da177e2005-04-16 15:20:36 -070064 * Andy Walker (andy@lysaker.kvaerner.no), April 06, 1996.
65 *
66 * Don't allow mandatory locks on mmap()'ed files. Added simple functions to
67 * check if a file has mandatory locks, used by mmap(), open() and creat() to
68 * see if system call should be rejected. Ref. HP-UX/SunOS/Solaris Reference
69 * Manual, Section 2.
70 * Andy Walker (andy@lysaker.kvaerner.no), April 09, 1996.
71 *
72 * Tidied up block list handling. Added '/proc/locks' interface.
73 * Andy Walker (andy@lysaker.kvaerner.no), April 24, 1996.
74 *
75 * Fixed deadlock condition for pathological code that mixes calls to
76 * flock() and fcntl().
77 * Andy Walker (andy@lysaker.kvaerner.no), April 29, 1996.
78 *
79 * Allow only one type of locking scheme (FL_POSIX or FL_FLOCK) to be in use
80 * for a given file at a time. Changed the CONFIG_LOCK_MANDATORY scheme to
81 * guarantee sensible behaviour in the case where file system modules might
82 * be compiled with different options than the kernel itself.
83 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
84 *
85 * Added a couple of missing wake_up() calls. Thanks to Thomas Meckel
86 * (Thomas.Meckel@mni.fh-giessen.de) for spotting this.
87 * Andy Walker (andy@lysaker.kvaerner.no), May 15, 1996.
88 *
89 * Changed FL_POSIX locks to use the block list in the same way as FL_FLOCK
90 * locks. Changed process synchronisation to avoid dereferencing locks that
91 * have already been freed.
92 * Andy Walker (andy@lysaker.kvaerner.no), Sep 21, 1996.
93 *
94 * Made the block list a circular list to minimise searching in the list.
95 * Andy Walker (andy@lysaker.kvaerner.no), Sep 25, 1996.
96 *
97 * Made mandatory locking a mount option. Default is not to allow mandatory
98 * locking.
99 * Andy Walker (andy@lysaker.kvaerner.no), Oct 04, 1996.
100 *
101 * Some adaptations for NFS support.
102 * Olaf Kirch (okir@monad.swb.de), Dec 1996,
103 *
104 * Fixed /proc/locks interface so that we can't overrun the buffer we are handed.
105 * Andy Walker (andy@lysaker.kvaerner.no), May 12, 1997.
106 *
107 * Use slab allocator instead of kmalloc/kfree.
108 * Use generic list implementation from <linux/list.h>.
109 * Sped up posix_locks_deadlock by only considering blocked locks.
110 * Matthew Wilcox <willy@debian.org>, March, 2000.
111 *
112 * Leases and LOCK_MAND
113 * Matthew Wilcox <willy@debian.org>, June, 2000.
114 * Stephen Rothwell <sfr@canb.auug.org.au>, June, 2000.
115 */
116
117#include <linux/capability.h>
118#include <linux/file.h>
Al Viro9f3acc32008-04-24 07:44:08 -0400119#include <linux/fdtable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#include <linux/fs.h>
121#include <linux/init.h>
122#include <linux/module.h>
123#include <linux/security.h>
124#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125#include <linux/syscalls.h>
126#include <linux/time.h>
Dipankar Sarma4fb3a532005-09-16 19:28:13 -0700127#include <linux/rcupdate.h>
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000128#include <linux/pid_namespace.h>
Jeff Layton48f74182013-06-21 08:58:18 -0400129#include <linux/hashtable.h>
Jeff Layton7012b022013-06-21 08:58:22 -0400130#include <linux/percpu.h>
131#include <linux/lglock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Jeff Layton62af4f12014-05-09 14:13:05 -0400133#define CREATE_TRACE_POINTS
134#include <trace/events/filelock.h>
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136#include <asm/uaccess.h>
137
138#define IS_POSIX(fl) (fl->fl_flags & FL_POSIX)
139#define IS_FLOCK(fl) (fl->fl_flags & FL_FLOCK)
Christoph Hellwig11afe9f2015-01-21 19:17:03 +0100140#define IS_LEASE(fl) (fl->fl_flags & (FL_LEASE|FL_DELEG|FL_LAYOUT))
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400141#define IS_OFDLCK(fl) (fl->fl_flags & FL_OFDLCK)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400143static bool lease_breaking(struct file_lock *fl)
144{
J. Bruce Fields778fc542011-07-26 18:25:49 -0400145 return fl->fl_flags & (FL_UNLOCK_PENDING | FL_DOWNGRADE_PENDING);
146}
147
148static int target_leasetype(struct file_lock *fl)
149{
150 if (fl->fl_flags & FL_UNLOCK_PENDING)
151 return F_UNLCK;
152 if (fl->fl_flags & FL_DOWNGRADE_PENDING)
153 return F_RDLCK;
154 return fl->fl_type;
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -0400155}
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157int leases_enable = 1;
158int lease_break_time = 45;
159
Jeff Layton1c8c6012013-06-21 08:58:15 -0400160/*
Jeff Layton7012b022013-06-21 08:58:22 -0400161 * The global file_lock_list is only used for displaying /proc/locks, so we
162 * keep a list on each CPU, with each list protected by its own spinlock via
163 * the file_lock_lglock. Note that alterations to the list also require that
Jeff Layton6109c852015-01-16 15:05:57 -0500164 * the relevant flc_lock is held.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400165 */
Jeff Layton7012b022013-06-21 08:58:22 -0400166DEFINE_STATIC_LGLOCK(file_lock_lglock);
167static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
Jeff Layton88974692013-06-21 08:58:14 -0400168
Jeff Layton1c8c6012013-06-21 08:58:15 -0400169/*
Jeff Layton48f74182013-06-21 08:58:18 -0400170 * The blocked_hash is used to find POSIX lock loops for deadlock detection.
Jeff Layton7b2296a2013-06-21 08:58:20 -0400171 * It is protected by blocked_lock_lock.
Jeff Layton48f74182013-06-21 08:58:18 -0400172 *
173 * We hash locks by lockowner in order to optimize searching for the lock a
174 * particular lockowner is waiting on.
175 *
176 * FIXME: make this value scale via some heuristic? We generally will want more
177 * buckets when we have more lockowners holding locks, but that's a little
178 * difficult to determine without knowing what the workload will look like.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400179 */
Jeff Layton48f74182013-06-21 08:58:18 -0400180#define BLOCKED_HASH_BITS 7
181static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
Jeff Layton88974692013-06-21 08:58:14 -0400182
Jeff Layton1c8c6012013-06-21 08:58:15 -0400183/*
Jeff Layton7b2296a2013-06-21 08:58:20 -0400184 * This lock protects the blocked_hash. Generally, if you're accessing it, you
185 * want to be holding this lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400186 *
187 * In addition, it also protects the fl->fl_block list, and the fl->fl_next
188 * pointer for file_lock structures that are acting as lock requests (in
189 * contrast to those that are acting as records of acquired locks).
190 *
191 * Note that when we acquire this lock in order to change the above fields,
Jeff Layton6109c852015-01-16 15:05:57 -0500192 * we often hold the flc_lock as well. In certain cases, when reading the fields
Jeff Layton1c8c6012013-06-21 08:58:15 -0400193 * protected by this lock, we can skip acquiring it iff we already hold the
Jeff Layton6109c852015-01-16 15:05:57 -0500194 * flc_lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400195 *
196 * In particular, adding an entry to the fl_block list requires that you hold
Jeff Layton6109c852015-01-16 15:05:57 -0500197 * both the flc_lock and the blocked_lock_lock (acquired in that order).
198 * Deleting an entry from the list however only requires the file_lock_lock.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400199 */
Jeff Layton7b2296a2013-06-21 08:58:20 -0400200static DEFINE_SPINLOCK(blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Jeff Layton4a075e32015-01-16 15:05:54 -0500202static struct kmem_cache *flctx_cache __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -0800203static struct kmem_cache *filelock_cache __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
Jeff Layton4a075e32015-01-16 15:05:54 -0500205static struct file_lock_context *
Jeff Layton5c1c6692015-04-03 09:04:03 -0400206locks_get_lock_context(struct inode *inode, int type)
Jeff Layton4a075e32015-01-16 15:05:54 -0500207{
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200208 struct file_lock_context *ctx;
Jeff Layton4a075e32015-01-16 15:05:54 -0500209
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200210 /* paired with cmpxchg() below */
211 ctx = smp_load_acquire(&inode->i_flctx);
212 if (likely(ctx) || type == F_UNLCK)
Jeff Layton4a075e32015-01-16 15:05:54 -0500213 goto out;
214
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200215 ctx = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
216 if (!ctx)
Jeff Layton4a075e32015-01-16 15:05:54 -0500217 goto out;
218
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200219 spin_lock_init(&ctx->flc_lock);
220 INIT_LIST_HEAD(&ctx->flc_flock);
221 INIT_LIST_HEAD(&ctx->flc_posix);
222 INIT_LIST_HEAD(&ctx->flc_lease);
Jeff Layton4a075e32015-01-16 15:05:54 -0500223
224 /*
225 * Assign the pointer if it's not already assigned. If it is, then
226 * free the context we just allocated.
227 */
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200228 if (cmpxchg(&inode->i_flctx, NULL, ctx)) {
229 kmem_cache_free(flctx_cache, ctx);
230 ctx = smp_load_acquire(&inode->i_flctx);
231 }
Jeff Layton4a075e32015-01-16 15:05:54 -0500232out:
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200233 return ctx;
Jeff Layton4a075e32015-01-16 15:05:54 -0500234}
235
236void
237locks_free_lock_context(struct file_lock_context *ctx)
238{
239 if (ctx) {
240 WARN_ON_ONCE(!list_empty(&ctx->flc_flock));
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500241 WARN_ON_ONCE(!list_empty(&ctx->flc_posix));
Jeff Layton8634b512015-01-16 15:05:55 -0500242 WARN_ON_ONCE(!list_empty(&ctx->flc_lease));
Jeff Layton4a075e32015-01-16 15:05:54 -0500243 kmem_cache_free(flctx_cache, ctx);
244 }
245}
246
Miklos Szerediee19cc42011-07-07 13:06:09 +0200247static void locks_init_lock_heads(struct file_lock *fl)
Miklos Szeredia51cb912011-07-06 12:33:55 +0200248{
Jeff Layton139ca042013-06-21 08:58:17 -0400249 INIT_HLIST_NODE(&fl->fl_link);
Jeff Layton6dee60f2015-01-16 15:05:54 -0500250 INIT_LIST_HEAD(&fl->fl_list);
Miklos Szerediee19cc42011-07-07 13:06:09 +0200251 INIT_LIST_HEAD(&fl->fl_block);
252 init_waitqueue_head(&fl->fl_wait);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200253}
254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255/* Allocate an empty lock structure. */
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200256struct file_lock *locks_alloc_lock(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257{
Miklos Szerediee19cc42011-07-07 13:06:09 +0200258 struct file_lock *fl = kmem_cache_zalloc(filelock_cache, GFP_KERNEL);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200259
260 if (fl)
Miklos Szerediee19cc42011-07-07 13:06:09 +0200261 locks_init_lock_heads(fl);
Miklos Szeredia51cb912011-07-06 12:33:55 +0200262
263 return fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264}
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +0200265EXPORT_SYMBOL_GPL(locks_alloc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266
Felix Blyakhera9e61e22009-03-31 15:12:56 -0500267void locks_release_private(struct file_lock *fl)
Trond Myklebust47831f32006-03-20 13:44:05 -0500268{
269 if (fl->fl_ops) {
270 if (fl->fl_ops->fl_release_private)
271 fl->fl_ops->fl_release_private(fl);
272 fl->fl_ops = NULL;
273 }
Trond Myklebust47831f32006-03-20 13:44:05 -0500274
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400275 if (fl->fl_lmops) {
Jeff Laytoncae80b32015-04-03 09:04:04 -0400276 if (fl->fl_lmops->lm_put_owner) {
277 fl->fl_lmops->lm_put_owner(fl->fl_owner);
278 fl->fl_owner = NULL;
279 }
Kinglong Mee5c97d7b2014-08-22 10:18:43 -0400280 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)
Jeff Laytoncae80b32015-04-03 09:04:04 -0400335 fl->fl_lmops->lm_get_owner(fl->fl_owner);
Kinglong Meef3282962014-08-22 10:18:43 -0400336 }
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 Layton6109c852015-01-16 15:05:57 -0500560/* Must be called with the flc_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 Layton6109c852015-01-16 15:05:57 -0500569/* Must be called with the flc_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
Jeff Layton6109c852015-01-16 15:05:57 -0500574 * is done while holding the flc_lock, and new insertions into the list
Jeff Layton7012b022013-06-21 08:58:22 -0400575 * 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{
Daniel Wagner663d5af2015-04-03 09:04:03 -0400594 lockdep_assert_held(&blocked_lock_lock);
595
Jeff Layton3999e492013-06-21 08:58:19 -0400596 hash_add(blocked_hash, &waiter->fl_link, posix_owner_key(waiter));
Jeff Layton88974692013-06-21 08:58:14 -0400597}
598
Jeff Layton6ca10ed2014-02-03 12:13:07 -0500599static void locks_delete_global_blocked(struct file_lock *waiter)
Jeff Layton88974692013-06-21 08:58:14 -0400600{
Daniel Wagner663d5af2015-04-03 09:04:03 -0400601 lockdep_assert_held(&blocked_lock_lock);
602
Jeff Layton48f74182013-06-21 08:58:18 -0400603 hash_del(&waiter->fl_link);
Jeff Layton88974692013-06-21 08:58:14 -0400604}
605
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606/* Remove waiter from blocker's block list.
607 * When blocker ends up pointing to itself then the list is empty.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400608 *
Jeff Layton7b2296a2013-06-21 08:58:20 -0400609 * Must be called with blocked_lock_lock held.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 */
Matt Mackall33443c42006-01-08 01:05:22 -0800611static void __locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612{
Jeff Layton88974692013-06-21 08:58:14 -0400613 locks_delete_global_blocked(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 list_del_init(&waiter->fl_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 waiter->fl_next = NULL;
616}
617
Jeff Layton1a9e64a2013-06-21 08:58:10 -0400618static void locks_delete_block(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400620 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 __locks_delete_block(waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400622 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623}
624
625/* Insert waiter into blocker's block list.
626 * We use a circular list so that processes can be easily woken up in
627 * the order they blocked. The documentation doesn't require this but
628 * it seems like the reasonable thing to do.
Jeff Layton1c8c6012013-06-21 08:58:15 -0400629 *
Jeff Layton6109c852015-01-16 15:05:57 -0500630 * Must be called with both the flc_lock and blocked_lock_lock held. The
631 * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
632 * that the flc_lock is also held on insertions we can avoid taking the
633 * blocked_lock_lock in some cases when we see that the fl_block list is empty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400635static void __locks_insert_block(struct file_lock *blocker,
636 struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
J. Bruce Fields6dc0fe82006-03-26 01:37:24 -0800638 BUG_ON(!list_empty(&waiter->fl_block));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 waiter->fl_next = blocker;
Jeff Layton88974692013-06-21 08:58:14 -0400640 list_add_tail(&waiter->fl_block, &blocker->fl_block);
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400641 if (IS_POSIX(blocker) && !IS_OFDLCK(blocker))
Jeff Layton1c8c6012013-06-21 08:58:15 -0400642 locks_insert_global_blocked(waiter);
643}
644
Jeff Layton6109c852015-01-16 15:05:57 -0500645/* Must be called with flc_lock held. */
Jeff Layton1c8c6012013-06-21 08:58:15 -0400646static void locks_insert_block(struct file_lock *blocker,
647 struct file_lock *waiter)
648{
Jeff Layton7b2296a2013-06-21 08:58:20 -0400649 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400650 __locks_insert_block(blocker, waiter);
Jeff Layton7b2296a2013-06-21 08:58:20 -0400651 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
653
Jeff Layton1cb36012013-06-21 08:58:12 -0400654/*
655 * Wake up processes blocked waiting for blocker.
656 *
Jeff Layton6109c852015-01-16 15:05:57 -0500657 * Must be called with the inode->flc_lock held!
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 */
659static void locks_wake_up_blocks(struct file_lock *blocker)
660{
Jeff Layton4e8c7652013-06-21 08:58:16 -0400661 /*
662 * Avoid taking global lock if list is empty. This is safe since new
Jeff Layton6109c852015-01-16 15:05:57 -0500663 * blocked requests are only added to the list under the flc_lock, and
664 * the flc_lock is always held here. Note that removal from the fl_block
665 * list does not require the flc_lock, so we must recheck list_empty()
Jeff Layton7b2296a2013-06-21 08:58:20 -0400666 * after acquiring the blocked_lock_lock.
Jeff Layton4e8c7652013-06-21 08:58:16 -0400667 */
668 if (list_empty(&blocker->fl_block))
669 return;
670
Jeff Layton7b2296a2013-06-21 08:58:20 -0400671 spin_lock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 while (!list_empty(&blocker->fl_block)) {
Pavel Emelyanovf0c1cd02007-09-19 16:44:07 +0400673 struct file_lock *waiter;
674
675 waiter = list_first_entry(&blocker->fl_block,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 struct file_lock, fl_block);
677 __locks_delete_block(waiter);
J. Bruce Fields8fb47a42011-07-20 20:21:59 -0400678 if (waiter->fl_lmops && waiter->fl_lmops->lm_notify)
679 waiter->fl_lmops->lm_notify(waiter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 else
681 wake_up(&waiter->fl_wait);
682 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400683 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684}
685
Jeff Layton5263e312015-01-16 15:05:55 -0500686static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500687locks_insert_lock_ctx(struct file_lock *fl, struct list_head *before)
Jeff Layton5263e312015-01-16 15:05:55 -0500688{
689 fl->fl_nspid = get_pid(task_tgid(current));
690 list_add_tail(&fl->fl_list, before);
691 locks_insert_global_locks(fl);
692}
693
Jeff Layton8634b512015-01-16 15:05:55 -0500694static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500695locks_unlink_lock_ctx(struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Jeff Layton88974692013-06-21 08:58:14 -0400697 locks_delete_global_locks(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500698 list_del_init(&fl->fl_list);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +0000699 if (fl->fl_nspid) {
700 put_pid(fl->fl_nspid);
701 fl->fl_nspid = NULL;
702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 locks_wake_up_blocks(fl);
Jeff Layton24cbe782014-02-03 12:13:06 -0500704}
705
Jeff Layton5263e312015-01-16 15:05:55 -0500706static void
Jeff Laytone084c1b2015-02-16 14:32:03 -0500707locks_delete_lock_ctx(struct file_lock *fl, struct list_head *dispose)
Jeff Layton5263e312015-01-16 15:05:55 -0500708{
Jeff Laytone084c1b2015-02-16 14:32:03 -0500709 locks_unlink_lock_ctx(fl);
Jeff Layton8634b512015-01-16 15:05:55 -0500710 if (dispose)
711 list_add(&fl->fl_list, dispose);
712 else
713 locks_free_lock(fl);
Jeff Layton5263e312015-01-16 15:05:55 -0500714}
715
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716/* Determine if lock sys_fl blocks lock caller_fl. Common functionality
717 * checks for shared/exclusive status of overlapping locks.
718 */
719static int locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
720{
721 if (sys_fl->fl_type == F_WRLCK)
722 return 1;
723 if (caller_fl->fl_type == F_WRLCK)
724 return 1;
725 return 0;
726}
727
728/* Determine if lock sys_fl blocks lock caller_fl. POSIX specific
729 * checking before calling the locks_conflict().
730 */
731static int posix_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
732{
733 /* POSIX locks owned by the same process do not conflict with
734 * each other.
735 */
Jeff Layton9b8c8692015-04-03 09:04:02 -0400736 if (posix_same_owner(caller_fl, sys_fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 return (0);
738
739 /* Check whether they overlap */
740 if (!locks_overlap(caller_fl, sys_fl))
741 return 0;
742
743 return (locks_conflict(caller_fl, sys_fl));
744}
745
746/* Determine if lock sys_fl blocks lock caller_fl. FLOCK specific
747 * checking before calling the locks_conflict().
748 */
749static int flock_locks_conflict(struct file_lock *caller_fl, struct file_lock *sys_fl)
750{
751 /* FLOCK locks referring to the same filp do not conflict with
752 * each other.
753 */
Jeff Layton9b8c8692015-04-03 09:04:02 -0400754 if (caller_fl->fl_file == sys_fl->fl_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 return (0);
756 if ((caller_fl->fl_type & LOCK_MAND) || (sys_fl->fl_type & LOCK_MAND))
757 return 0;
758
759 return (locks_conflict(caller_fl, sys_fl));
760}
761
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400762void
Marc Eshel9d6a8c52007-02-21 00:55:18 -0500763posix_test_lock(struct file *filp, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764{
765 struct file_lock *cfl;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500766 struct file_lock_context *ctx;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400767 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
Dmitry Vyukov128a3782015-09-21 09:43:06 +0200769 ctx = smp_load_acquire(&inode->i_flctx);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500770 if (!ctx || list_empty_careful(&ctx->flc_posix)) {
J. Bruce Fields129a84d2007-05-10 18:38:43 -0400771 fl->fl_type = F_UNLCK;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500772 return;
773 }
774
Jeff Layton6109c852015-01-16 15:05:57 -0500775 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500776 list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
777 if (posix_locks_conflict(fl, cfl)) {
778 locks_copy_conflock(fl, cfl);
779 if (cfl->fl_nspid)
780 fl->fl_pid = pid_vnr(cfl->fl_nspid);
781 goto out;
782 }
783 }
784 fl->fl_type = F_UNLCK;
785out:
Jeff Layton6109c852015-01-16 15:05:57 -0500786 spin_unlock(&ctx->flc_lock);
J. Bruce Fields6d34ac12007-05-11 16:09:32 -0400787 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789EXPORT_SYMBOL(posix_test_lock);
790
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400791/*
792 * Deadlock detection:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400794 * We attempt to detect deadlocks that are due purely to posix file
795 * locks.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400797 * We assume that a task can be waiting for at most one lock at a time.
798 * So for any acquired lock, the process holding that lock may be
799 * waiting on at most one other lock. That lock in turns may be held by
800 * someone waiting for at most one other lock. Given a requested lock
801 * caller_fl which is about to wait for a conflicting lock block_fl, we
802 * follow this chain of waiters to ensure we are not about to create a
803 * cycle.
J. Bruce Fields97855b42007-10-30 11:20:02 -0400804 *
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400805 * Since we do this before we ever put a process to sleep on a lock, we
806 * are ensured that there is never a cycle; that is what guarantees that
807 * the while() loop in posix_locks_deadlock() eventually completes.
808 *
809 * Note: the above assumption may not be true when handling lock
810 * requests from a broken NFS client. It may also fail in the presence
811 * of tasks (such as posix threads) sharing the same open file table.
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400812 * To handle those cases, we just bail out after a few iterations.
Jeff Layton57b65322014-02-03 12:13:09 -0500813 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400814 * For FL_OFDLCK locks, the owner is the filp, not the files_struct.
Jeff Layton57b65322014-02-03 12:13:09 -0500815 * Because the owner is not even nominally tied to a thread of
816 * execution, the deadlock detection below can't reasonably work well. Just
817 * skip it for those.
818 *
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400819 * In principle, we could do a more limited deadlock detection on FL_OFDLCK
Jeff Layton57b65322014-02-03 12:13:09 -0500820 * locks that just checks for the case where two tasks are attempting to
821 * upgrade from read to write locks on the same inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822 */
J. Bruce Fields97855b42007-10-30 11:20:02 -0400823
824#define MAX_DEADLK_ITERATIONS 10
825
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400826/* Find a lock that the owner of the given block_fl is blocking on. */
827static struct file_lock *what_owner_is_waiting_for(struct file_lock *block_fl)
828{
829 struct file_lock *fl;
830
Jeff Layton3999e492013-06-21 08:58:19 -0400831 hash_for_each_possible(blocked_hash, fl, fl_link, posix_owner_key(block_fl)) {
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400832 if (posix_same_owner(fl, block_fl))
833 return fl->fl_next;
834 }
835 return NULL;
836}
837
Jeff Layton7b2296a2013-06-21 08:58:20 -0400838/* Must be called with the blocked_lock_lock held! */
Adrian Bunkb0904e12006-06-23 02:05:13 -0700839static int posix_locks_deadlock(struct file_lock *caller_fl,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 struct file_lock *block_fl)
841{
J. Bruce Fields97855b42007-10-30 11:20:02 -0400842 int i = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Daniel Wagner663d5af2015-04-03 09:04:03 -0400844 lockdep_assert_held(&blocked_lock_lock);
845
Jeff Layton57b65322014-02-03 12:13:09 -0500846 /*
847 * This deadlock detector can't reasonably detect deadlocks with
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400848 * FL_OFDLCK locks, since they aren't owned by a process, per-se.
Jeff Layton57b65322014-02-03 12:13:09 -0500849 */
Jeff Laytoncff2fce2014-04-22 08:24:32 -0400850 if (IS_OFDLCK(caller_fl))
Jeff Layton57b65322014-02-03 12:13:09 -0500851 return 0;
852
J. Bruce Fieldsb5331842007-10-26 18:05:40 -0400853 while ((block_fl = what_owner_is_waiting_for(block_fl))) {
854 if (i++ > MAX_DEADLK_ITERATIONS)
855 return 0;
856 if (posix_same_owner(caller_fl, block_fl))
857 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 }
859 return 0;
860}
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/* Try to create a FLOCK lock on filp. We always insert new FLOCK locks
J. Bruce Fields02888f42007-09-12 15:45:07 -0400863 * after any leases, but before any posix locks.
Trond Myklebustf475ae92006-06-29 16:38:32 -0400864 *
865 * Note that if called with an FL_EXISTS argument, the caller may determine
866 * whether or not a lock was successfully freed by testing the return
867 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 */
Jeff Laytonbcd7f782015-07-11 06:43:02 -0400869static int flock_lock_inode(struct inode *inode, struct file_lock *request)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870{
Trond Myklebust993dfa82006-03-31 02:30:55 -0800871 struct file_lock *new_fl = NULL;
Jeff Layton5263e312015-01-16 15:05:55 -0500872 struct file_lock *fl;
873 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 int error = 0;
Jeff Layton5263e312015-01-16 15:05:55 -0500875 bool found = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400876 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877
Jeff Layton5c1c6692015-04-03 09:04:03 -0400878 ctx = locks_get_lock_context(inode, request->fl_type);
879 if (!ctx) {
880 if (request->fl_type != F_UNLCK)
881 return -ENOMEM;
882 return (request->fl_flags & FL_EXISTS) ? -ENOENT : 0;
883 }
Jeff Layton5263e312015-01-16 15:05:55 -0500884
Arnd Bergmannb89f4322010-09-18 15:09:31 +0200885 if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
886 new_fl = locks_alloc_lock();
887 if (!new_fl)
888 return -ENOMEM;
889 }
890
Jeff Layton6109c852015-01-16 15:05:57 -0500891 spin_lock(&ctx->flc_lock);
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400892 if (request->fl_flags & FL_ACCESS)
893 goto find_conflict;
Pavel Emelyanov84d535a2007-09-11 16:38:13 +0400894
Jeff Layton5263e312015-01-16 15:05:55 -0500895 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Jeff Laytonbcd7f782015-07-11 06:43:02 -0400896 if (request->fl_file != fl->fl_file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 continue;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800898 if (request->fl_type == fl->fl_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 goto out;
Jeff Layton5263e312015-01-16 15:05:55 -0500900 found = true;
Jeff Laytone084c1b2015-02-16 14:32:03 -0500901 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 break;
903 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904
Trond Myklebustf475ae92006-06-29 16:38:32 -0400905 if (request->fl_type == F_UNLCK) {
906 if ((request->fl_flags & FL_EXISTS) && !found)
907 error = -ENOENT;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800908 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -0400909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400911find_conflict:
Jeff Layton5263e312015-01-16 15:05:55 -0500912 list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
Trond Myklebust993dfa82006-03-31 02:30:55 -0800913 if (!flock_locks_conflict(request, fl))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 continue;
915 error = -EAGAIN;
Miklos Szeredibde74e42008-07-25 01:48:57 -0700916 if (!(request->fl_flags & FL_SLEEP))
917 goto out;
918 error = FILE_LOCK_DEFERRED;
919 locks_insert_block(fl, request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920 goto out;
921 }
Trond Myklebustf07f18d2006-06-29 16:38:37 -0400922 if (request->fl_flags & FL_ACCESS)
923 goto out;
Trond Myklebust993dfa82006-03-31 02:30:55 -0800924 locks_copy_lock(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -0500925 locks_insert_lock_ctx(new_fl, &ctx->flc_flock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800926 new_fl = NULL;
Kirill Korotaev9cedc192006-06-14 17:59:35 +0400927 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
929out:
Jeff Layton6109c852015-01-16 15:05:57 -0500930 spin_unlock(&ctx->flc_lock);
Trond Myklebust993dfa82006-03-31 02:30:55 -0800931 if (new_fl)
932 locks_free_lock(new_fl);
Jeff Laytoned9814d2014-08-11 14:20:31 -0400933 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 return error;
935}
936
Marc Eshel150b3932007-01-18 16:15:35 -0500937static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500939 struct file_lock *fl, *tmp;
Miklos Szeredi39005d02006-06-23 02:05:10 -0700940 struct file_lock *new_fl = NULL;
941 struct file_lock *new_fl2 = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 struct file_lock *left = NULL;
943 struct file_lock *right = NULL;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500944 struct file_lock_context *ctx;
Jeff Laytonb9746ef2013-06-21 08:58:13 -0400945 int error;
946 bool added = false;
Jeff Laytoned9814d2014-08-11 14:20:31 -0400947 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948
Jeff Layton5c1c6692015-04-03 09:04:03 -0400949 ctx = locks_get_lock_context(inode, request->fl_type);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500950 if (!ctx)
Jeff Layton5c1c6692015-04-03 09:04:03 -0400951 return (request->fl_type == F_UNLCK) ? 0 : -ENOMEM;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 /*
954 * We may need two file_lock structures for this operation,
955 * so we get them in advance to avoid races.
Miklos Szeredi39005d02006-06-23 02:05:10 -0700956 *
957 * In some cases we can be sure, that no new locks will be needed
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 */
Miklos Szeredi39005d02006-06-23 02:05:10 -0700959 if (!(request->fl_flags & FL_ACCESS) &&
960 (request->fl_type != F_UNLCK ||
961 request->fl_start != 0 || request->fl_end != OFFSET_MAX)) {
962 new_fl = locks_alloc_lock();
963 new_fl2 = locks_alloc_lock();
964 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Jeff Layton6109c852015-01-16 15:05:57 -0500966 spin_lock(&ctx->flc_lock);
Jeff Layton1cb36012013-06-21 08:58:12 -0400967 /*
968 * New lock request. Walk all POSIX locks and look for conflicts. If
969 * there are any, either return error or put the request on the
Jeff Layton48f74182013-06-21 08:58:18 -0400970 * blocker's list of waiters and the global blocked_hash.
Jeff Layton1cb36012013-06-21 08:58:12 -0400971 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 if (request->fl_type != F_UNLCK) {
Jeff Laytonbd61e0a2015-01-16 15:05:55 -0500973 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974 if (!posix_locks_conflict(request, fl))
975 continue;
Andy Adamson5842add2006-03-26 01:37:26 -0800976 if (conflock)
Kinglong Mee3fe0fff12014-08-22 10:18:42 -0400977 locks_copy_conflock(conflock, fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978 error = -EAGAIN;
979 if (!(request->fl_flags & FL_SLEEP))
980 goto out;
Jeff Layton1c8c6012013-06-21 08:58:15 -0400981 /*
982 * Deadlock detection and insertion into the blocked
983 * locks list must be done while holding the same lock!
984 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 error = -EDEADLK;
Jeff Layton7b2296a2013-06-21 08:58:20 -0400986 spin_lock(&blocked_lock_lock);
Jeff Layton1c8c6012013-06-21 08:58:15 -0400987 if (likely(!posix_locks_deadlock(request, fl))) {
988 error = FILE_LOCK_DEFERRED;
989 __locks_insert_block(fl, request);
990 }
Jeff Layton7b2296a2013-06-21 08:58:20 -0400991 spin_unlock(&blocked_lock_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 goto out;
993 }
994 }
995
996 /* If we're just looking for a conflict, we're done. */
997 error = 0;
998 if (request->fl_flags & FL_ACCESS)
999 goto out;
1000
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001001 /* Find the first old lock with the same owner as the new lock */
1002 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
1003 if (posix_same_owner(request, fl))
1004 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 }
1006
Jeff Layton1cb36012013-06-21 08:58:12 -04001007 /* Process locks with this owner. */
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001008 list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
1009 if (!posix_same_owner(request, fl))
1010 break;
1011
1012 /* Detect adjacent or overlapping regions (if same lock type) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (request->fl_type == fl->fl_type) {
Olaf Kirch449231d2005-08-25 16:25:35 -07001014 /* In all comparisons of start vs end, use
1015 * "start - 1" rather than "end + 1". If end
1016 * is OFFSET_MAX, end + 1 will become negative.
1017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 if (fl->fl_end < request->fl_start - 1)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001019 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 /* If the next lock in the list has entirely bigger
1021 * addresses than the new one, insert the lock here.
1022 */
Olaf Kirch449231d2005-08-25 16:25:35 -07001023 if (fl->fl_start - 1 > request->fl_end)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 break;
1025
1026 /* If we come here, the new and old lock are of the
1027 * same type and adjacent or overlapping. Make one
1028 * lock yielding from the lower start address of both
1029 * locks to the higher end address.
1030 */
1031 if (fl->fl_start > request->fl_start)
1032 fl->fl_start = request->fl_start;
1033 else
1034 request->fl_start = fl->fl_start;
1035 if (fl->fl_end < request->fl_end)
1036 fl->fl_end = request->fl_end;
1037 else
1038 request->fl_end = fl->fl_end;
1039 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001040 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 continue;
1042 }
1043 request = fl;
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001044 added = true;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001045 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 /* Processing for different lock types is a bit
1047 * more complex.
1048 */
1049 if (fl->fl_end < request->fl_start)
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001050 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 if (fl->fl_start > request->fl_end)
1052 break;
1053 if (request->fl_type == F_UNLCK)
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001054 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 if (fl->fl_start < request->fl_start)
1056 left = fl;
1057 /* If the next lock in the list has a higher end
1058 * address than the new one, insert the new one here.
1059 */
1060 if (fl->fl_end > request->fl_end) {
1061 right = fl;
1062 break;
1063 }
1064 if (fl->fl_start >= request->fl_start) {
1065 /* The new lock completely replaces an old
1066 * one (This may happen several times).
1067 */
1068 if (added) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001069 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 continue;
1071 }
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001072 /*
1073 * Replace the old lock with new_fl, and
1074 * remove the old one. It's safe to do the
1075 * insert here since we know that we won't be
1076 * using new_fl later, and that the lock is
1077 * just replacing an existing lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 */
Jeff Laytonb84d49f2014-08-12 08:03:49 -04001079 error = -ENOLCK;
1080 if (!new_fl)
1081 goto out;
1082 locks_copy_lock(new_fl, request);
1083 request = new_fl;
1084 new_fl = NULL;
Jeff Laytone084c1b2015-02-16 14:32:03 -05001085 locks_insert_lock_ctx(request, &fl->fl_list);
1086 locks_delete_lock_ctx(fl, &dispose);
Jeff Laytonb9746ef2013-06-21 08:58:13 -04001087 added = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 }
1089 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 }
1091
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001092 /*
Jeff Layton1cb36012013-06-21 08:58:12 -04001093 * The above code only modifies existing locks in case of merging or
1094 * replacing. If new lock(s) need to be inserted all modifications are
1095 * done below this, so it's safe yet to bail out.
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001096 */
1097 error = -ENOLCK; /* "no luck" */
1098 if (right && left == right && !new_fl2)
1099 goto out;
1100
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101 error = 0;
1102 if (!added) {
Trond Myklebustf475ae92006-06-29 16:38:32 -04001103 if (request->fl_type == F_UNLCK) {
1104 if (request->fl_flags & FL_EXISTS)
1105 error = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 goto out;
Trond Myklebustf475ae92006-06-29 16:38:32 -04001107 }
Miklos Szeredi0d9a4902006-06-23 02:05:09 -07001108
1109 if (!new_fl) {
1110 error = -ENOLCK;
1111 goto out;
1112 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 locks_copy_lock(new_fl, request);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001114 locks_insert_lock_ctx(new_fl, &fl->fl_list);
Jeff Layton2e2f756f2015-02-17 17:08:23 -05001115 fl = new_fl;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 new_fl = NULL;
1117 }
1118 if (right) {
1119 if (left == right) {
1120 /* The new lock breaks the old one in two pieces,
1121 * so we have to use the second new lock.
1122 */
1123 left = new_fl2;
1124 new_fl2 = NULL;
1125 locks_copy_lock(left, right);
Jeff Laytone084c1b2015-02-16 14:32:03 -05001126 locks_insert_lock_ctx(left, &fl->fl_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 }
1128 right->fl_start = request->fl_end + 1;
1129 locks_wake_up_blocks(right);
1130 }
1131 if (left) {
1132 left->fl_end = request->fl_start - 1;
1133 locks_wake_up_blocks(left);
1134 }
1135 out:
Jeff Layton6109c852015-01-16 15:05:57 -05001136 spin_unlock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 /*
1138 * Free any unused locks.
1139 */
1140 if (new_fl)
1141 locks_free_lock(new_fl);
1142 if (new_fl2)
1143 locks_free_lock(new_fl2);
Jeff Laytoned9814d2014-08-11 14:20:31 -04001144 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 return error;
1146}
1147
1148/**
1149 * posix_lock_file - Apply a POSIX-style lock to a file
1150 * @filp: The file to apply the lock to
1151 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05001152 * @conflock: Place to return a copy of the conflicting lock, if found.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 *
1154 * Add a POSIX style lock to a file.
1155 * We merge adjacent & overlapping locks whenever possible.
1156 * POSIX locks are sorted by owner task, then by starting address
Trond Myklebustf475ae92006-06-29 16:38:32 -04001157 *
1158 * Note that if called with an FL_EXISTS argument, the caller may determine
1159 * whether or not a lock was successfully freed by testing the return
1160 * value for -ENOENT.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 */
Marc Eshel150b3932007-01-18 16:15:35 -05001162int posix_lock_file(struct file *filp, struct file_lock *fl,
Andy Adamson5842add2006-03-26 01:37:26 -08001163 struct file_lock *conflock)
1164{
Al Viro496ad9a2013-01-23 17:07:38 -05001165 return __posix_lock_file(file_inode(filp), fl, conflock);
Andy Adamson5842add2006-03-26 01:37:26 -08001166}
Marc Eshel150b3932007-01-18 16:15:35 -05001167EXPORT_SYMBOL(posix_lock_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168
1169/**
Jeff Layton29d01b22015-07-11 06:43:02 -04001170 * posix_lock_inode_wait - Apply a POSIX-style lock to a file
1171 * @inode: inode of file to which lock request should be applied
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 * @fl: The lock to be applied
1173 *
Jeff Layton29d01b22015-07-11 06:43:02 -04001174 * Variant of posix_lock_file_wait that does not take a filp, and so can be
1175 * used after the filp has already been torn down.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 */
Jeff Layton29d01b22015-07-11 06:43:02 -04001177int posix_lock_inode_wait(struct inode *inode, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178{
1179 int error;
1180 might_sleep ();
1181 for (;;) {
Jeff Layton29d01b22015-07-11 06:43:02 -04001182 error = __posix_lock_file(inode, fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001183 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 break;
1185 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1186 if (!error)
1187 continue;
1188
1189 locks_delete_block(fl);
1190 break;
1191 }
1192 return error;
1193}
Jeff Layton29d01b22015-07-11 06:43:02 -04001194EXPORT_SYMBOL(posix_lock_inode_wait);
1195
1196/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 * locks_mandatory_locked - Check for an active lock
Jeff Laytond7a06982014-03-10 09:54:15 -04001198 * @file: the file to check
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199 *
1200 * Searches the inode's list of locks to find any POSIX locks which conflict.
1201 * This function is called from locks_verify_locked() only.
1202 */
Jeff Laytond7a06982014-03-10 09:54:15 -04001203int locks_mandatory_locked(struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001205 int ret;
Jeff Laytond7a06982014-03-10 09:54:15 -04001206 struct inode *inode = file_inode(file);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001207 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 struct file_lock *fl;
1209
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001210 ctx = smp_load_acquire(&inode->i_flctx);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001211 if (!ctx || list_empty_careful(&ctx->flc_posix))
1212 return 0;
1213
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 /*
1215 * Search the lock list for this inode for any POSIX locks.
1216 */
Jeff Layton6109c852015-01-16 15:05:57 -05001217 spin_lock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001218 ret = 0;
1219 list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001220 if (fl->fl_owner != current->files &&
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001221 fl->fl_owner != file) {
1222 ret = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 break;
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001224 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 }
Jeff Layton6109c852015-01-16 15:05:57 -05001226 spin_unlock(&ctx->flc_lock);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05001227 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228}
1229
1230/**
1231 * locks_mandatory_area - Check for a conflicting lock
1232 * @read_write: %FLOCK_VERIFY_WRITE for exclusive access, %FLOCK_VERIFY_READ
1233 * for shared
1234 * @inode: the file to check
1235 * @filp: how the file was opened (if it was)
1236 * @offset: start of area to check
1237 * @count: length of area to check
1238 *
1239 * Searches the inode's list of locks to find any POSIX locks which conflict.
1240 * This function is called from rw_verify_area() and
1241 * locks_verify_truncate().
1242 */
1243int locks_mandatory_area(int read_write, struct inode *inode,
1244 struct file *filp, loff_t offset,
1245 size_t count)
1246{
1247 struct file_lock fl;
1248 int error;
Jeff Layton29723ad2014-03-10 09:54:19 -04001249 bool sleep = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250
1251 locks_init_lock(&fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 fl.fl_pid = current->tgid;
1253 fl.fl_file = filp;
1254 fl.fl_flags = FL_POSIX | FL_ACCESS;
1255 if (filp && !(filp->f_flags & O_NONBLOCK))
Jeff Layton29723ad2014-03-10 09:54:19 -04001256 sleep = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 fl.fl_type = (read_write == FLOCK_VERIFY_WRITE) ? F_WRLCK : F_RDLCK;
1258 fl.fl_start = offset;
1259 fl.fl_end = offset + count - 1;
1260
1261 for (;;) {
Jeff Layton29723ad2014-03-10 09:54:19 -04001262 if (filp) {
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02001263 fl.fl_owner = filp;
Jeff Layton29723ad2014-03-10 09:54:19 -04001264 fl.fl_flags &= ~FL_SLEEP;
1265 error = __posix_lock_file(inode, &fl, NULL);
1266 if (!error)
1267 break;
1268 }
1269
1270 if (sleep)
1271 fl.fl_flags |= FL_SLEEP;
1272 fl.fl_owner = current->files;
Marc Eshel150b3932007-01-18 16:15:35 -05001273 error = __posix_lock_file(inode, &fl, NULL);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001274 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 break;
1276 error = wait_event_interruptible(fl.fl_wait, !fl.fl_next);
1277 if (!error) {
1278 /*
1279 * If we've been sleeping someone might have
1280 * changed the permissions behind our back.
1281 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07001282 if (__mandatory_lock(inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 continue;
1284 }
1285
1286 locks_delete_block(&fl);
1287 break;
1288 }
1289
1290 return error;
1291}
1292
1293EXPORT_SYMBOL(locks_mandatory_area);
1294
J. Bruce Fields778fc542011-07-26 18:25:49 -04001295static void lease_clear_pending(struct file_lock *fl, int arg)
1296{
1297 switch (arg) {
1298 case F_UNLCK:
1299 fl->fl_flags &= ~FL_UNLOCK_PENDING;
1300 /* fall through: */
1301 case F_RDLCK:
1302 fl->fl_flags &= ~FL_DOWNGRADE_PENDING;
1303 }
1304}
1305
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306/* We already had a lease on this file; just change its type */
Jeff Layton7448cc32015-01-16 15:05:57 -05001307int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 int error = assign_type(fl, arg);
1310
1311 if (error)
1312 return error;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001313 lease_clear_pending(fl, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 locks_wake_up_blocks(fl);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001315 if (arg == F_UNLCK) {
1316 struct file *filp = fl->fl_file;
1317
1318 f_delown(filp);
1319 filp->f_owner.signum = 0;
J. Bruce Fields96d6d592012-07-27 16:18:00 -04001320 fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
1321 if (fl->fl_fasync != NULL) {
1322 printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
1323 fl->fl_fasync = NULL;
1324 }
Jeff Laytone084c1b2015-02-16 14:32:03 -05001325 locks_delete_lock_ctx(fl, dispose);
Filipe Brandenburger3b6e2722012-07-27 00:42:52 -04001326 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 return 0;
1328}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329EXPORT_SYMBOL(lease_modify);
1330
J. Bruce Fields778fc542011-07-26 18:25:49 -04001331static bool past_time(unsigned long then)
1332{
1333 if (!then)
1334 /* 0 is a special value meaning "this never expires": */
1335 return false;
1336 return time_after(jiffies, then);
1337}
1338
Jeff Laytonc45198e2014-09-01 07:12:07 -04001339static void time_out_leases(struct inode *inode, struct list_head *dispose)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340{
Jeff Layton8634b512015-01-16 15:05:55 -05001341 struct file_lock_context *ctx = inode->i_flctx;
1342 struct file_lock *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343
Jeff Layton6109c852015-01-16 15:05:57 -05001344 lockdep_assert_held(&ctx->flc_lock);
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001345
Jeff Layton8634b512015-01-16 15:05:55 -05001346 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001347 trace_time_out_leases(inode, fl);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001348 if (past_time(fl->fl_downgrade_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001349 lease_modify(fl, F_RDLCK, dispose);
J. Bruce Fields778fc542011-07-26 18:25:49 -04001350 if (past_time(fl->fl_break_time))
Jeff Layton7448cc32015-01-16 15:05:57 -05001351 lease_modify(fl, F_UNLCK, dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 }
1353}
1354
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001355static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
1356{
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001357 if ((breaker->fl_flags & FL_LAYOUT) != (lease->fl_flags & FL_LAYOUT))
1358 return false;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001359 if ((breaker->fl_flags & FL_DELEG) && (lease->fl_flags & FL_LEASE))
1360 return false;
1361 return locks_conflict(breaker, lease);
1362}
1363
Jeff Layton03d12dd2014-09-01 14:53:41 -04001364static bool
1365any_leases_conflict(struct inode *inode, struct file_lock *breaker)
1366{
Jeff Layton8634b512015-01-16 15:05:55 -05001367 struct file_lock_context *ctx = inode->i_flctx;
Jeff Layton03d12dd2014-09-01 14:53:41 -04001368 struct file_lock *fl;
1369
Jeff Layton6109c852015-01-16 15:05:57 -05001370 lockdep_assert_held(&ctx->flc_lock);
Jeff Layton03d12dd2014-09-01 14:53:41 -04001371
Jeff Layton8634b512015-01-16 15:05:55 -05001372 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Jeff Layton03d12dd2014-09-01 14:53:41 -04001373 if (leases_conflict(fl, breaker))
1374 return true;
1375 }
1376 return false;
1377}
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379/**
1380 * __break_lease - revoke all outstanding leases on file
1381 * @inode: the inode of the file to return
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001382 * @mode: O_RDONLY: break only write leases; O_WRONLY or O_RDWR:
1383 * break all leases
1384 * @type: FL_LEASE: break leases and delegations; FL_DELEG: break
1385 * only delegations
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 *
david m. richter87250dd2007-05-09 16:10:27 -04001387 * break_lease (inlined for speed) has checked there already is at least
1388 * some kind of lock (maybe a lease) on this file. Leases are broken on
1389 * a call to open() or truncate(). This function can sleep unless you
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 * specified %O_NONBLOCK to your open().
1391 */
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001392int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
J. Bruce Fields778fc542011-07-26 18:25:49 -04001394 int error = 0;
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001395 struct file_lock_context *ctx;
Yan, Zhenga9011252015-03-27 10:34:20 +08001396 struct file_lock *new_fl, *fl, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 unsigned long break_time;
Al Viro8737c932009-12-24 06:47:55 -05001398 int want_write = (mode & O_ACCMODE) != O_RDONLY;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001399 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400
Al Viro8737c932009-12-24 06:47:55 -05001401 new_fl = lease_alloc(NULL, want_write ? F_WRLCK : F_RDLCK);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001402 if (IS_ERR(new_fl))
1403 return PTR_ERR(new_fl);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001404 new_fl->fl_flags = type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405
Jeff Layton8634b512015-01-16 15:05:55 -05001406 /* typically we will check that ctx is non-NULL before calling */
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001407 ctx = smp_load_acquire(&inode->i_flctx);
Jeff Layton8634b512015-01-16 15:05:55 -05001408 if (!ctx) {
1409 WARN_ON_ONCE(1);
1410 return error;
1411 }
1412
Jeff Layton6109c852015-01-16 15:05:57 -05001413 spin_lock(&ctx->flc_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414
Jeff Laytonc45198e2014-09-01 07:12:07 -04001415 time_out_leases(inode, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Jeff Layton03d12dd2014-09-01 14:53:41 -04001417 if (!any_leases_conflict(inode, new_fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001418 goto out;
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420 break_time = 0;
1421 if (lease_break_time > 0) {
1422 break_time = jiffies + lease_break_time * HZ;
1423 if (break_time == 0)
1424 break_time++; /* so that 0 means no break time */
1425 }
1426
Yan, Zhenga9011252015-03-27 10:34:20 +08001427 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001428 if (!leases_conflict(fl, new_fl))
1429 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001430 if (want_write) {
1431 if (fl->fl_flags & FL_UNLOCK_PENDING)
1432 continue;
1433 fl->fl_flags |= FL_UNLOCK_PENDING;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 fl->fl_break_time = break_time;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001435 } else {
Jeff Layton8634b512015-01-16 15:05:55 -05001436 if (lease_breaking(fl))
J. Bruce Fields778fc542011-07-26 18:25:49 -04001437 continue;
1438 fl->fl_flags |= FL_DOWNGRADE_PENDING;
1439 fl->fl_downgrade_time = break_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 }
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001441 if (fl->fl_lmops->lm_break(fl))
Jeff Laytone084c1b2015-02-16 14:32:03 -05001442 locks_delete_lock_ctx(fl, &dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 }
1444
Jeff Layton8634b512015-01-16 15:05:55 -05001445 if (list_empty(&ctx->flc_lease))
Jeff Layton4d01b7f2014-09-01 15:06:54 -04001446 goto out;
1447
Jeff Layton843c6b22014-09-01 14:27:43 -04001448 if (mode & O_NONBLOCK) {
Jeff Layton62af4f12014-05-09 14:13:05 -04001449 trace_break_lease_noblock(inode, new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 error = -EWOULDBLOCK;
1451 goto out;
1452 }
1453
1454restart:
Jeff Layton8634b512015-01-16 15:05:55 -05001455 fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
1456 break_time = fl->fl_break_time;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001457 if (break_time != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 break_time -= jiffies;
Jeff Laytonf1c6bb22014-04-15 06:17:49 -04001459 if (break_time == 0)
1460 break_time++;
Jeff Layton8634b512015-01-16 15:05:55 -05001461 locks_insert_block(fl, new_fl);
Jeff Layton62af4f12014-05-09 14:13:05 -04001462 trace_break_lease_block(inode, new_fl);
Jeff Layton6109c852015-01-16 15:05:57 -05001463 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001464 locks_dispose_list(&dispose);
Matthew Wilcox4321e012008-01-14 21:28:30 -07001465 error = wait_event_interruptible_timeout(new_fl->fl_wait,
1466 !new_fl->fl_next, break_time);
Jeff Layton6109c852015-01-16 15:05:57 -05001467 spin_lock(&ctx->flc_lock);
Jeff Layton62af4f12014-05-09 14:13:05 -04001468 trace_break_lease_unblock(inode, new_fl);
Jeff Layton1c8c6012013-06-21 08:58:15 -04001469 locks_delete_block(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 if (error >= 0) {
J. Bruce Fields778fc542011-07-26 18:25:49 -04001471 /*
1472 * Wait for the next conflicting lease that has not been
1473 * broken yet
1474 */
Jeff Layton03d12dd2014-09-01 14:53:41 -04001475 if (error == 0)
1476 time_out_leases(inode, &dispose);
1477 if (any_leases_conflict(inode, new_fl))
1478 goto restart;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 error = 0;
1480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481out:
Jeff Layton6109c852015-01-16 15:05:57 -05001482 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001483 locks_dispose_list(&dispose);
Linus Torvalds6d4b9e32011-12-26 10:25:26 -08001484 locks_free_lock(new_fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 return error;
1486}
1487
1488EXPORT_SYMBOL(__break_lease);
1489
1490/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07001491 * lease_get_mtime - get the last modified time of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 * @inode: the inode
1493 * @time: pointer to a timespec which will contain the last modified time
1494 *
1495 * This is to force NFS clients to flush their caches for files with
1496 * exclusive leases. The justification is that if someone has an
Randy Dunlapa6b91912008-03-19 17:01:00 -07001497 * exclusive lease, then they could be modifying it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 */
1499void lease_get_mtime(struct inode *inode, struct timespec *time)
1500{
Jeff Laytonbfe86022014-08-22 10:18:44 -04001501 bool has_lease = false;
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001502 struct file_lock_context *ctx;
Jeff Layton8634b512015-01-16 15:05:55 -05001503 struct file_lock *fl;
Jeff Laytonbfe86022014-08-22 10:18:44 -04001504
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001505 ctx = smp_load_acquire(&inode->i_flctx);
Jeff Layton8634b512015-01-16 15:05:55 -05001506 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001507 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001508 if (!list_empty(&ctx->flc_lease)) {
1509 fl = list_first_entry(&ctx->flc_lease,
1510 struct file_lock, fl_list);
1511 if (fl->fl_type == F_WRLCK)
1512 has_lease = true;
1513 }
Jeff Layton6109c852015-01-16 15:05:57 -05001514 spin_unlock(&ctx->flc_lock);
Jeff Laytonbfe86022014-08-22 10:18:44 -04001515 }
1516
1517 if (has_lease)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 *time = current_fs_time(inode->i_sb);
1519 else
1520 *time = inode->i_mtime;
1521}
1522
1523EXPORT_SYMBOL(lease_get_mtime);
1524
1525/**
1526 * fcntl_getlease - Enquire what lease is currently active
1527 * @filp: the file
1528 *
1529 * The value returned by this function will be one of
1530 * (if no lease break is pending):
1531 *
1532 * %F_RDLCK to indicate a shared lease is held.
1533 *
1534 * %F_WRLCK to indicate an exclusive lease is held.
1535 *
1536 * %F_UNLCK to indicate no lease is held.
1537 *
1538 * (if a lease break is pending):
1539 *
1540 * %F_RDLCK to indicate an exclusive lease needs to be
1541 * changed to a shared lease (or removed).
1542 *
1543 * %F_UNLCK to indicate the lease needs to be removed.
1544 *
1545 * XXX: sfr & willy disagree over whether F_INPROGRESS
1546 * should be returned to userspace.
1547 */
1548int fcntl_getlease(struct file *filp)
1549{
1550 struct file_lock *fl;
Jeff Layton1c8c6012013-06-21 08:58:15 -04001551 struct inode *inode = file_inode(filp);
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001552 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 int type = F_UNLCK;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001554 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001556 ctx = smp_load_acquire(&inode->i_flctx);
Jeff Layton8634b512015-01-16 15:05:55 -05001557 if (ctx && !list_empty_careful(&ctx->flc_lease)) {
Jeff Layton6109c852015-01-16 15:05:57 -05001558 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001559 time_out_leases(file_inode(filp), &dispose);
1560 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
1561 if (fl->fl_file != filp)
1562 continue;
J. Bruce Fields778fc542011-07-26 18:25:49 -04001563 type = target_leasetype(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 break;
1565 }
Jeff Layton6109c852015-01-16 15:05:57 -05001566 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001567 locks_dispose_list(&dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569 return type;
1570}
1571
Jeff Layton24cbe782014-02-03 12:13:06 -05001572/**
1573 * check_conflicting_open - see if the given dentry points to a file that has
1574 * an existing open that would conflict with the
1575 * desired lease.
1576 * @dentry: dentry to check
1577 * @arg: type of lease that we're trying to acquire
Randy Dunlap7fadc592015-08-09 18:43:17 -07001578 * @flags: current lock flags
Jeff Layton24cbe782014-02-03 12:13:06 -05001579 *
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
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001584check_conflicting_open(const struct dentry *dentry, const long arg, int flags)
Jeff Layton24cbe782014-02-03 12:13:06 -05001585{
1586 int ret = 0;
1587 struct inode *inode = dentry->d_inode;
1588
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001589 if (flags & FL_LAYOUT)
1590 return 0;
1591
Jeff Layton24cbe782014-02-03 12:13:06 -05001592 if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
1593 return -EAGAIN;
1594
1595 if ((arg == F_WRLCK) && ((d_count(dentry) > 1) ||
1596 (atomic_read(&inode->i_count) > 1)))
1597 ret = -EAGAIN;
1598
1599 return ret;
1600}
1601
Jeff Laytone6f5c782014-08-22 10:40:25 -04001602static int
1603generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604{
Jeff Layton8634b512015-01-16 15:05:55 -05001605 struct file_lock *fl, *my_fl = NULL, *lease;
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001606 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607 struct inode *inode = dentry->d_inode;
Jeff Layton8634b512015-01-16 15:05:55 -05001608 struct file_lock_context *ctx;
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001609 bool is_deleg = (*flp)->fl_flags & FL_DELEG;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001610 int error;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001611 LIST_HEAD(dispose);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612
J. Bruce Fields096657b2010-10-30 17:31:14 -04001613 lease = *flp;
Jeff Layton62af4f12014-05-09 14:13:05 -04001614 trace_generic_add_lease(inode, lease);
1615
Jeff Layton5c1c6692015-04-03 09:04:03 -04001616 /* Note that arg is never F_UNLCK here */
1617 ctx = locks_get_lock_context(inode, arg);
Jeff Layton8634b512015-01-16 15:05:55 -05001618 if (!ctx)
1619 return -ENOMEM;
1620
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001621 /*
1622 * In the delegation case we need mutual exclusion with
1623 * a number of operations that take the i_mutex. We trylock
1624 * because delegations are an optional optimization, and if
1625 * there's some chance of a conflict--we'd rather not
1626 * bother, maybe that's a sign this just isn't a good file to
1627 * hand out a delegation on.
1628 */
1629 if (is_deleg && !mutex_trylock(&inode->i_mutex))
1630 return -EAGAIN;
1631
1632 if (is_deleg && arg == F_WRLCK) {
1633 /* Write delegations are not currently supported: */
Dan Carpenter4fdb7932013-11-13 10:56:27 +03001634 mutex_unlock(&inode->i_mutex);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001635 WARN_ON_ONCE(1);
1636 return -EINVAL;
1637 }
J. Bruce Fields096657b2010-10-30 17:31:14 -04001638
Jeff Layton6109c852015-01-16 15:05:57 -05001639 spin_lock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001640 time_out_leases(inode, &dispose);
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001641 error = check_conflicting_open(dentry, arg, lease->fl_flags);
Jeff Layton24cbe782014-02-03 12:13:06 -05001642 if (error)
J. Bruce Fields096657b2010-10-30 17:31:14 -04001643 goto out;
Pavel Emelyanov85c59582007-09-20 12:45:02 +04001644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 /*
1646 * At this point, we know that if there is an exclusive
1647 * lease on this file, then we hold it on this filp
1648 * (otherwise our open of this file would have blocked).
1649 * And if we are trying to acquire an exclusive lease,
1650 * then the file is not open by anyone (including us)
1651 * except for this filp.
1652 */
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001653 error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001654 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001655 if (fl->fl_file == filp &&
1656 fl->fl_owner == lease->fl_owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001657 my_fl = fl;
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001658 continue;
1659 }
Jeff Layton8634b512015-01-16 15:05:55 -05001660
J. Bruce Fieldsc1f24ef2011-08-19 10:59:49 -04001661 /*
1662 * No exclusive leases if someone else has a lease on
1663 * this file:
1664 */
1665 if (arg == F_WRLCK)
1666 goto out;
1667 /*
1668 * Modifying our existing lease is OK, but no getting a
1669 * new lease if someone else is opening for write:
1670 */
1671 if (fl->fl_flags & FL_UNLOCK_PENDING)
1672 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 }
1674
Jeff Layton8634b512015-01-16 15:05:55 -05001675 if (my_fl != NULL) {
Jeff Layton0164bf02015-03-04 17:34:32 -05001676 lease = my_fl;
1677 error = lease->fl_lmops->lm_change(lease, arg, &dispose);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001678 if (error)
1679 goto out;
1680 goto out_setup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 }
1682
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 error = -EINVAL;
1684 if (!leases_enable)
1685 goto out;
1686
Jeff Laytone084c1b2015-02-16 14:32:03 -05001687 locks_insert_lock_ctx(lease, &ctx->flc_lease);
Jeff Layton24cbe782014-02-03 12:13:06 -05001688 /*
1689 * The check in break_lease() is lockless. It's possible for another
1690 * open to race in after we did the earlier check for a conflicting
1691 * open but before the lease was inserted. Check again for a
1692 * conflicting open and cancel the lease if there is one.
1693 *
1694 * We also add a barrier here to ensure that the insertion of the lock
1695 * precedes these checks.
1696 */
1697 smp_mb();
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001698 error = check_conflicting_open(dentry, arg, lease->fl_flags);
Jeff Layton8634b512015-01-16 15:05:55 -05001699 if (error) {
Jeff Laytone084c1b2015-02-16 14:32:03 -05001700 locks_unlink_lock_ctx(lease);
Jeff Layton8634b512015-01-16 15:05:55 -05001701 goto out;
1702 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001703
1704out_setup:
1705 if (lease->fl_lmops->lm_setup)
1706 lease->fl_lmops->lm_setup(lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707out:
Jeff Layton6109c852015-01-16 15:05:57 -05001708 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001709 locks_dispose_list(&dispose);
J. Bruce Fieldsdf4e8d22012-03-05 13:18:59 -05001710 if (is_deleg)
1711 mutex_unlock(&inode->i_mutex);
Jeff Layton8634b512015-01-16 15:05:55 -05001712 if (!error && !my_fl)
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001713 *flp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 return error;
1715}
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001716
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001717static int generic_delete_lease(struct file *filp, void *owner)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001718{
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001719 int error = -EAGAIN;
Jeff Layton8634b512015-01-16 15:05:55 -05001720 struct file_lock *fl, *victim = NULL;
Benjamin Coddington6ca7d912015-10-15 09:07:07 -04001721 struct inode *inode = file_inode(filp);
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001722 struct file_lock_context *ctx;
Jeff Laytonc45198e2014-09-01 07:12:07 -04001723 LIST_HEAD(dispose);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001724
Dmitry Vyukov128a3782015-09-21 09:43:06 +02001725 ctx = smp_load_acquire(&inode->i_flctx);
Jeff Layton8634b512015-01-16 15:05:55 -05001726 if (!ctx) {
1727 trace_generic_delete_lease(inode, NULL);
1728 return error;
1729 }
1730
Jeff Layton6109c852015-01-16 15:05:57 -05001731 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05001732 list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001733 if (fl->fl_file == filp &&
1734 fl->fl_owner == owner) {
Jeff Layton8634b512015-01-16 15:05:55 -05001735 victim = fl;
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001736 break;
Jeff Layton8634b512015-01-16 15:05:55 -05001737 }
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001738 }
Jeff Laytona9b1b452015-03-14 09:45:35 -04001739 trace_generic_delete_lease(inode, victim);
Jeff Layton8634b512015-01-16 15:05:55 -05001740 if (victim)
Jeff Layton7448cc32015-01-16 15:05:57 -05001741 error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05001742 spin_unlock(&ctx->flc_lock);
Jeff Laytonc45198e2014-09-01 07:12:07 -04001743 locks_dispose_list(&dispose);
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001744 return error;
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001745}
1746
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747/**
1748 * generic_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001749 * @filp: file pointer
1750 * @arg: type of lease to obtain
1751 * @flp: input - file_lock to use, output - file_lock inserted
1752 * @priv: private data for lm_setup (may be NULL if lm_setup
1753 * doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754 *
1755 * The (input) flp->fl_lmops->lm_break function is required
1756 * by break_lease().
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001758int generic_setlease(struct file *filp, long arg, struct file_lock **flp,
1759 void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
Benjamin Coddington6ca7d912015-10-15 09:07:07 -04001761 struct inode *inode = file_inode(filp);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001762 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763
Eric W. Biederman8e96e3b2012-03-03 21:17:15 -08001764 if ((!uid_eq(current_fsuid(), inode->i_uid)) && !capable(CAP_LEASE))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001765 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (!S_ISREG(inode->i_mode))
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001767 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 error = security_file_lock(filp, arg);
1769 if (error)
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001770 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001772 switch (arg) {
1773 case F_UNLCK:
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001774 return generic_delete_lease(filp, *priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001775 case F_RDLCK:
1776 case F_WRLCK:
Jeff Layton0efaa7e2014-08-22 10:18:45 -04001777 if (!(*flp)->fl_lmops->lm_break) {
1778 WARN_ON_ONCE(1);
1779 return -ENOLCK;
1780 }
Christoph Hellwig11afe9f2015-01-21 19:17:03 +01001781
Jeff Laytone6f5c782014-08-22 10:40:25 -04001782 return generic_add_lease(filp, arg, flp, priv);
J. Bruce Fields8335ebd2011-09-21 08:34:32 -04001783 default:
Dave Jones8d657eb2012-07-13 13:35:36 -04001784 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786}
Christoph Hellwig0af1a452007-07-31 00:39:22 -07001787EXPORT_SYMBOL(generic_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788
Arnd Bergmannb89f4322010-09-18 15:09:31 +02001789/**
Jeff Laytone51673a2014-08-22 18:13:28 -04001790 * vfs_setlease - sets a lease on an open file
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001791 * @filp: file pointer
1792 * @arg: type of lease to obtain
1793 * @lease: file_lock to use when adding a lease
1794 * @priv: private info for lm_setup when adding a lease (may be
1795 * NULL if lm_setup doesn't require it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796 *
Jeff Laytone51673a2014-08-22 18:13:28 -04001797 * Call this to establish a lease on the file. The "lease" argument is not
1798 * used for F_UNLCK requests and may be NULL. For commands that set or alter
1799 * an existing lease, the (*lease)->fl_lmops->lm_break operation must be set;
1800 * if not, this function will return -ENOLCK (and generate a scary-looking
1801 * stack trace).
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001802 *
1803 * The "priv" pointer is passed directly to the lm_setup function as-is. It
1804 * may be NULL if the lm_setup operation doesn't require it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 */
Jeff Laytone6f5c782014-08-22 10:40:25 -04001806int
1807vfs_setlease(struct file *filp, long arg, struct file_lock **lease, void **priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001809 if (filp->f_op->setlease)
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001810 return filp->f_op->setlease(filp, arg, lease, priv);
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001811 else
Jeff Laytonf82b4b62014-08-22 18:50:48 -04001812 return generic_setlease(filp, arg, lease, priv);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813}
J. Bruce Fieldsa9933ce2007-06-07 17:09:49 -04001814EXPORT_SYMBOL_GPL(vfs_setlease);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001816static int do_fcntl_add_lease(unsigned int fd, struct file *filp, long arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817{
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001818 struct file_lock *fl;
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001819 struct fasync_struct *new;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 int error;
1821
Arnd Bergmannc5b1f0d2010-10-27 15:46:08 +02001822 fl = lease_alloc(filp, arg);
1823 if (IS_ERR(fl))
1824 return PTR_ERR(fl);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001826 new = fasync_alloc();
1827 if (!new) {
1828 locks_free_lock(fl);
1829 return -ENOMEM;
1830 }
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001831 new->fa_fd = fd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
Jeff Layton1c7dd2f2014-08-22 10:55:47 -04001833 error = vfs_setlease(filp, arg, &fl, (void **)&new);
Jeff Layton2dfb9282014-08-11 18:14:12 -04001834 if (fl)
1835 locks_free_lock(fl);
Linus Torvaldsf7347ce2010-10-27 12:38:12 -04001836 if (new)
1837 fasync_free(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 return error;
1839}
1840
1841/**
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001842 * fcntl_setlease - sets a lease on an open file
1843 * @fd: open file descriptor
1844 * @filp: file pointer
1845 * @arg: type of lease to obtain
1846 *
1847 * Call this fcntl to establish a lease on the file.
1848 * Note that you also need to call %F_SETSIG to
1849 * receive a signal when the lease is broken.
1850 */
1851int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
1852{
1853 if (arg == F_UNLCK)
Christoph Hellwig2ab99ee2015-01-21 19:14:02 +01001854 return vfs_setlease(filp, F_UNLCK, NULL, (void **)&filp);
J. Bruce Fields0ceaf6c2010-10-30 17:31:13 -04001855 return do_fcntl_add_lease(fd, filp, arg);
1856}
1857
1858/**
Jeff Layton29d01b22015-07-11 06:43:02 -04001859 * flock_lock_inode_wait - Apply a FLOCK-style lock to a file
1860 * @inode: inode of the file to apply to
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 * @fl: The lock to be applied
1862 *
Jeff Layton29d01b22015-07-11 06:43:02 -04001863 * Apply a FLOCK style lock request to an inode.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 */
Jeff Layton29d01b22015-07-11 06:43:02 -04001865int flock_lock_inode_wait(struct inode *inode, struct file_lock *fl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866{
1867 int error;
1868 might_sleep();
1869 for (;;) {
Jeff Layton29d01b22015-07-11 06:43:02 -04001870 error = flock_lock_inode(inode, fl);
Miklos Szeredibde74e42008-07-25 01:48:57 -07001871 if (error != FILE_LOCK_DEFERRED)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 break;
1873 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
1874 if (!error)
1875 continue;
1876
1877 locks_delete_block(fl);
1878 break;
1879 }
1880 return error;
1881}
Jeff Layton29d01b22015-07-11 06:43:02 -04001882EXPORT_SYMBOL(flock_lock_inode_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Jeff Layton29d01b22015-07-11 06:43:02 -04001884/**
Benjamin Coddingtone55c34a2015-10-22 13:38:13 -04001885 * locks_lock_inode_wait - Apply a lock to an inode
1886 * @inode: inode of the file to apply to
1887 * @fl: The lock to be applied
1888 *
1889 * Apply a POSIX or FLOCK style lock request to an inode.
1890 */
1891int locks_lock_inode_wait(struct inode *inode, struct file_lock *fl)
1892{
1893 int res = 0;
1894 switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) {
1895 case FL_POSIX:
1896 res = posix_lock_inode_wait(inode, fl);
1897 break;
1898 case FL_FLOCK:
1899 res = flock_lock_inode_wait(inode, fl);
1900 break;
1901 default:
1902 BUG();
1903 }
1904 return res;
1905}
1906EXPORT_SYMBOL(locks_lock_inode_wait);
1907
1908/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 * sys_flock: - flock() system call.
1910 * @fd: the file descriptor to lock.
1911 * @cmd: the type of lock to apply.
1912 *
1913 * Apply a %FL_FLOCK style lock to an open file descriptor.
1914 * The @cmd can be one of
1915 *
1916 * %LOCK_SH -- a shared lock.
1917 *
1918 * %LOCK_EX -- an exclusive lock.
1919 *
1920 * %LOCK_UN -- remove an existing lock.
1921 *
1922 * %LOCK_MAND -- a `mandatory' flock. This exists to emulate Windows Share Modes.
1923 *
1924 * %LOCK_MAND can be combined with %LOCK_READ or %LOCK_WRITE to allow other
1925 * processes read and write access respectively.
1926 */
Heiko Carstens002c8972009-01-14 14:14:18 +01001927SYSCALL_DEFINE2(flock, unsigned int, fd, unsigned int, cmd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928{
Al Viro2903ff02012-08-28 12:52:22 -04001929 struct fd f = fdget(fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930 struct file_lock *lock;
1931 int can_sleep, unlock;
1932 int error;
1933
1934 error = -EBADF;
Al Viro2903ff02012-08-28 12:52:22 -04001935 if (!f.file)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 goto out;
1937
1938 can_sleep = !(cmd & LOCK_NB);
1939 cmd &= ~LOCK_NB;
1940 unlock = (cmd == LOCK_UN);
1941
Al Viroaeb5d722008-09-02 15:28:45 -04001942 if (!unlock && !(cmd & LOCK_MAND) &&
Al Viro2903ff02012-08-28 12:52:22 -04001943 !(f.file->f_mode & (FMODE_READ|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 goto out_putf;
1945
Jeff Layton6e129d002014-09-04 10:25:06 -04001946 lock = flock_make_lock(f.file, cmd);
1947 if (IS_ERR(lock)) {
1948 error = PTR_ERR(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 goto out_putf;
Jeff Layton6e129d002014-09-04 10:25:06 -04001950 }
1951
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 if (can_sleep)
1953 lock->fl_flags |= FL_SLEEP;
1954
Al Viro2903ff02012-08-28 12:52:22 -04001955 error = security_file_lock(f.file, lock->fl_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 if (error)
1957 goto out_free;
1958
Al Viro72c2d532013-09-22 16:27:52 -04001959 if (f.file->f_op->flock)
Al Viro2903ff02012-08-28 12:52:22 -04001960 error = f.file->f_op->flock(f.file,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961 (can_sleep) ? F_SETLKW : F_SETLK,
1962 lock);
1963 else
Benjamin Coddington4f656362015-10-22 13:38:14 -04001964 error = locks_lock_file_wait(f.file, lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965
1966 out_free:
Trond Myklebust993dfa82006-03-31 02:30:55 -08001967 locks_free_lock(lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968
1969 out_putf:
Al Viro2903ff02012-08-28 12:52:22 -04001970 fdput(f);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 out:
1972 return error;
1973}
1974
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001975/**
1976 * vfs_test_lock - test file byte range lock
1977 * @filp: The file to test lock for
J. Bruce Fields6924c552007-05-11 16:22:50 -04001978 * @fl: The lock to test; also used to hold result
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001979 *
1980 * Returns -ERRNO on failure. Indicates presence of conflicting lock by
1981 * setting conf->fl_type to something other than F_UNLCK.
1982 */
1983int vfs_test_lock(struct file *filp, struct file_lock *fl)
1984{
Al Viro72c2d532013-09-22 16:27:52 -04001985 if (filp->f_op->lock)
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05001986 return filp->f_op->lock(filp, F_GETLK, fl);
1987 posix_test_lock(filp, fl);
1988 return 0;
1989}
1990EXPORT_SYMBOL_GPL(vfs_test_lock);
1991
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001992static int posix_lock_to_flock(struct flock *flock, struct file_lock *fl)
1993{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04001994 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05001995#if BITS_PER_LONG == 32
1996 /*
1997 * Make sure we can represent the posix lock via
1998 * legacy 32bit flock.
1999 */
2000 if (fl->fl_start > OFFT_OFFSET_MAX)
2001 return -EOVERFLOW;
2002 if (fl->fl_end != OFFSET_MAX && fl->fl_end > OFFT_OFFSET_MAX)
2003 return -EOVERFLOW;
2004#endif
2005 flock->l_start = fl->fl_start;
2006 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2007 fl->fl_end - fl->fl_start + 1;
2008 flock->l_whence = 0;
J. Bruce Fields129a84d2007-05-10 18:38:43 -04002009 flock->l_type = fl->fl_type;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002010 return 0;
2011}
2012
2013#if BITS_PER_LONG == 32
2014static void posix_lock_to_flock64(struct flock64 *flock, struct file_lock *fl)
2015{
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002016 flock->l_pid = IS_OFDLCK(fl) ? -1 : fl->fl_pid;
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002017 flock->l_start = fl->fl_start;
2018 flock->l_len = fl->fl_end == OFFSET_MAX ? 0 :
2019 fl->fl_end - fl->fl_start + 1;
2020 flock->l_whence = 0;
2021 flock->l_type = fl->fl_type;
2022}
2023#endif
2024
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025/* Report the first existing lock that would conflict with l.
2026 * This implements the F_GETLK command of fcntl().
2027 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002028int fcntl_getlk(struct file *filp, unsigned int cmd, struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002030 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 struct flock flock;
2032 int error;
2033
2034 error = -EFAULT;
2035 if (copy_from_user(&flock, l, sizeof(flock)))
2036 goto out;
2037 error = -EINVAL;
2038 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2039 goto out;
2040
2041 error = flock_to_posix_lock(filp, &file_lock, &flock);
2042 if (error)
2043 goto out;
2044
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002045 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002046 error = -EINVAL;
2047 if (flock.l_pid != 0)
2048 goto out;
2049
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002050 cmd = F_GETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002051 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002052 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002053 }
2054
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002055 error = vfs_test_lock(filp, &file_lock);
2056 if (error)
2057 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002059 flock.l_type = file_lock.fl_type;
2060 if (file_lock.fl_type != F_UNLCK) {
2061 error = posix_lock_to_flock(&flock, &file_lock);
J. Bruce Fieldsc2fa1b82007-02-20 16:10:11 -05002062 if (error)
Kinglong Meef3282962014-08-22 10:18:43 -04002063 goto rel_priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 }
2065 error = -EFAULT;
2066 if (!copy_to_user(l, &flock, sizeof(flock)))
2067 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002068rel_priv:
2069 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070out:
2071 return error;
2072}
2073
Marc Eshel7723ec92007-01-18 15:08:55 -05002074/**
2075 * vfs_lock_file - file byte range lock
2076 * @filp: The file to apply the lock to
2077 * @cmd: type of locking operation (F_SETLK, F_GETLK, etc.)
2078 * @fl: The lock to be applied
Marc Eshel150b3932007-01-18 16:15:35 -05002079 * @conf: Place to return a copy of the conflicting lock, if found.
2080 *
2081 * A caller that doesn't care about the conflicting lock may pass NULL
2082 * as the final argument.
2083 *
2084 * If the filesystem defines a private ->lock() method, then @conf will
2085 * be left unchanged; so a caller that cares should initialize it to
2086 * some acceptable default.
Marc Eshel2beb6612006-12-05 23:31:28 -05002087 *
2088 * To avoid blocking kernel daemons, such as lockd, that need to acquire POSIX
2089 * locks, the ->lock() interface may return asynchronously, before the lock has
2090 * been granted or denied by the underlying filesystem, if (and only if)
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002091 * lm_grant is set. Callers expecting ->lock() to return asynchronously
Marc Eshel2beb6612006-12-05 23:31:28 -05002092 * will only use F_SETLK, not F_SETLKW; they will set FL_SLEEP if (and only if)
2093 * the request is for a blocking lock. When ->lock() does return asynchronously,
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002094 * it must return FILE_LOCK_DEFERRED, and call ->lm_grant() when the lock
Marc Eshel2beb6612006-12-05 23:31:28 -05002095 * request completes.
2096 * If the request is for non-blocking lock the file system should return
Miklos Szeredibde74e42008-07-25 01:48:57 -07002097 * FILE_LOCK_DEFERRED then try to get the lock and call the callback routine
2098 * with the result. If the request timed out the callback routine will return a
Marc Eshel2beb6612006-12-05 23:31:28 -05002099 * nonzero return code and the file system should release the lock. The file
2100 * system is also responsible to keep a corresponding posix lock when it
2101 * grants a lock so the VFS can find out which locks are locally held and do
2102 * the correct lock cleanup when required.
2103 * The underlying filesystem must not drop the kernel lock or call
J. Bruce Fields8fb47a42011-07-20 20:21:59 -04002104 * ->lm_grant() before returning to the caller with a FILE_LOCK_DEFERRED
Marc Eshel2beb6612006-12-05 23:31:28 -05002105 * return code.
Marc Eshel7723ec92007-01-18 15:08:55 -05002106 */
Marc Eshel150b3932007-01-18 16:15:35 -05002107int vfs_lock_file(struct file *filp, unsigned int cmd, struct file_lock *fl, struct file_lock *conf)
Marc Eshel7723ec92007-01-18 15:08:55 -05002108{
Al Viro72c2d532013-09-22 16:27:52 -04002109 if (filp->f_op->lock)
Marc Eshel7723ec92007-01-18 15:08:55 -05002110 return filp->f_op->lock(filp, cmd, fl);
2111 else
Marc Eshel150b3932007-01-18 16:15:35 -05002112 return posix_lock_file(filp, fl, conf);
Marc Eshel7723ec92007-01-18 15:08:55 -05002113}
2114EXPORT_SYMBOL_GPL(vfs_lock_file);
2115
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002116static int do_lock_file_wait(struct file *filp, unsigned int cmd,
2117 struct file_lock *fl)
2118{
2119 int error;
2120
2121 error = security_file_lock(filp, fl->fl_type);
2122 if (error)
2123 return error;
2124
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002125 for (;;) {
2126 error = vfs_lock_file(filp, cmd, fl, NULL);
2127 if (error != FILE_LOCK_DEFERRED)
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002128 break;
Miklos Szeredi764c76b2008-07-25 01:48:58 -07002129 error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
2130 if (!error)
2131 continue;
2132
2133 locks_delete_block(fl);
2134 break;
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002135 }
2136
2137 return error;
2138}
2139
Benjamin Coddington6ca7d912015-10-15 09:07:07 -04002140/* Ensure that fl->fl_file has compatible f_mode for F_SETLK calls */
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002141static int
2142check_fmode_for_setlk(struct file_lock *fl)
2143{
2144 switch (fl->fl_type) {
2145 case F_RDLCK:
2146 if (!(fl->fl_file->f_mode & FMODE_READ))
2147 return -EBADF;
2148 break;
2149 case F_WRLCK:
2150 if (!(fl->fl_file->f_mode & FMODE_WRITE))
2151 return -EBADF;
2152 }
2153 return 0;
2154}
2155
Linus Torvalds1da177e2005-04-16 15:20:36 -07002156/* Apply the lock described by l to an open file descriptor.
2157 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2158 */
Peter Staubachc2936212005-07-27 11:45:09 -07002159int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
2160 struct flock __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161{
2162 struct file_lock *file_lock = locks_alloc_lock();
2163 struct flock flock;
2164 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002165 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 int error;
2167
2168 if (file_lock == NULL)
2169 return -ENOLCK;
2170
2171 /*
2172 * This might block, so we do it before checking the inode.
2173 */
2174 error = -EFAULT;
2175 if (copy_from_user(&flock, l, sizeof(flock)))
2176 goto out;
2177
Al Viro496ad9a2013-01-23 17:07:38 -05002178 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179
2180 /* Don't allow mandatory locks on files that may be memory mapped
2181 * and shared.
2182 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002183 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 error = -EAGAIN;
2185 goto out;
2186 }
2187
Peter Staubachc2936212005-07-27 11:45:09 -07002188again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 error = flock_to_posix_lock(filp, file_lock, &flock);
2190 if (error)
2191 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002192
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002193 error = check_fmode_for_setlk(file_lock);
2194 if (error)
2195 goto out;
2196
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002197 /*
2198 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002199 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002200 */
2201 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002202 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002203 error = -EINVAL;
2204 if (flock.l_pid != 0)
2205 goto out;
2206
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002207 cmd = F_SETLK;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002208 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002209 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002210 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002211 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002212 error = -EINVAL;
2213 if (flock.l_pid != 0)
2214 goto out;
2215
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002216 cmd = F_SETLKW;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002217 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002218 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002219 /* Fallthrough */
2220 case F_SETLKW:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221 file_lock->fl_flags |= FL_SLEEP;
2222 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002223
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002224 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225
Peter Staubachc2936212005-07-27 11:45:09 -07002226 /*
2227 * Attempt to detect a close/fcntl race and recover by
2228 * releasing the lock that was just acquired.
2229 */
Al Viro0b2bac22008-05-06 13:58:34 -04002230 /*
2231 * we need that spin_lock here - it prevents reordering between
Jeff Layton8116bf42015-01-21 20:44:01 -05002232 * update of i_flctx->flc_posix and check for it done in close().
Al Viro0b2bac22008-05-06 13:58:34 -04002233 * rcu_read_lock() wouldn't do.
2234 */
2235 spin_lock(&current->files->file_lock);
2236 f = fcheck(fd);
2237 spin_unlock(&current->files->file_lock);
2238 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002239 flock.l_type = F_UNLCK;
2240 goto again;
2241 }
2242
2243out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 locks_free_lock(file_lock);
2245 return error;
2246}
2247
2248#if BITS_PER_LONG == 32
2249/* Report the first existing lock that would conflict with l.
2250 * This implements the F_GETLK command of fcntl().
2251 */
Jeff Laytonc1e62b82014-02-03 12:13:09 -05002252int fcntl_getlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253{
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002254 struct file_lock file_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 struct flock64 flock;
2256 int error;
2257
2258 error = -EFAULT;
2259 if (copy_from_user(&flock, l, sizeof(flock)))
2260 goto out;
2261 error = -EINVAL;
2262 if ((flock.l_type != F_RDLCK) && (flock.l_type != F_WRLCK))
2263 goto out;
2264
2265 error = flock64_to_posix_lock(filp, &file_lock, &flock);
2266 if (error)
2267 goto out;
2268
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002269 if (cmd == F_OFD_GETLK) {
Jeff Layton90478932014-03-04 10:30:23 -05002270 error = -EINVAL;
2271 if (flock.l_pid != 0)
2272 goto out;
2273
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002274 cmd = F_GETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002275 file_lock.fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002276 file_lock.fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002277 }
2278
J. Bruce Fields3ee17ab2007-02-21 00:58:50 -05002279 error = vfs_test_lock(filp, &file_lock);
2280 if (error)
2281 goto out;
2282
Marc Eshel9d6a8c52007-02-21 00:55:18 -05002283 flock.l_type = file_lock.fl_type;
2284 if (file_lock.fl_type != F_UNLCK)
2285 posix_lock_to_flock64(&flock, &file_lock);
2286
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 error = -EFAULT;
2288 if (!copy_to_user(l, &flock, sizeof(flock)))
2289 error = 0;
Kinglong Meef3282962014-08-22 10:18:43 -04002290
2291 locks_release_private(&file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292out:
2293 return error;
2294}
2295
2296/* Apply the lock described by l to an open file descriptor.
2297 * This implements both the F_SETLK and F_SETLKW commands of fcntl().
2298 */
Peter Staubachc2936212005-07-27 11:45:09 -07002299int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
2300 struct flock64 __user *l)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301{
2302 struct file_lock *file_lock = locks_alloc_lock();
2303 struct flock64 flock;
2304 struct inode *inode;
Al Viro0b2bac22008-05-06 13:58:34 -04002305 struct file *f;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306 int error;
2307
2308 if (file_lock == NULL)
2309 return -ENOLCK;
2310
2311 /*
2312 * This might block, so we do it before checking the inode.
2313 */
2314 error = -EFAULT;
2315 if (copy_from_user(&flock, l, sizeof(flock)))
2316 goto out;
2317
Al Viro496ad9a2013-01-23 17:07:38 -05002318 inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
2320 /* Don't allow mandatory locks on files that may be memory mapped
2321 * and shared.
2322 */
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002323 if (mandatory_lock(inode) && mapping_writably_mapped(filp->f_mapping)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 error = -EAGAIN;
2325 goto out;
2326 }
2327
Peter Staubachc2936212005-07-27 11:45:09 -07002328again:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329 error = flock64_to_posix_lock(filp, file_lock, &flock);
2330 if (error)
2331 goto out;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002332
Jeff Laytoncf01f4e2014-05-09 11:41:54 -04002333 error = check_fmode_for_setlk(file_lock);
2334 if (error)
2335 goto out;
2336
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002337 /*
2338 * If the cmd is requesting file-private locks, then set the
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002339 * FL_OFDLCK flag and override the owner.
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002340 */
2341 switch (cmd) {
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002342 case F_OFD_SETLK:
Jeff Layton90478932014-03-04 10:30:23 -05002343 error = -EINVAL;
2344 if (flock.l_pid != 0)
2345 goto out;
2346
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002347 cmd = F_SETLK64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002348 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002349 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002350 break;
Jeff Layton0d3f7a22014-04-22 08:23:58 -04002351 case F_OFD_SETLKW:
Jeff Layton90478932014-03-04 10:30:23 -05002352 error = -EINVAL;
2353 if (flock.l_pid != 0)
2354 goto out;
2355
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002356 cmd = F_SETLKW64;
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002357 file_lock->fl_flags |= FL_OFDLCK;
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002358 file_lock->fl_owner = filp;
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002359 /* Fallthrough */
2360 case F_SETLKW64:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 file_lock->fl_flags |= FL_SLEEP;
2362 }
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002363
Miklos Szeredib648a6d2008-07-25 01:48:57 -07002364 error = do_lock_file_wait(filp, cmd, file_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002365
Peter Staubachc2936212005-07-27 11:45:09 -07002366 /*
2367 * Attempt to detect a close/fcntl race and recover by
2368 * releasing the lock that was just acquired.
2369 */
Al Viro0b2bac22008-05-06 13:58:34 -04002370 spin_lock(&current->files->file_lock);
2371 f = fcheck(fd);
2372 spin_unlock(&current->files->file_lock);
2373 if (!error && f != filp && flock.l_type != F_UNLCK) {
Peter Staubachc2936212005-07-27 11:45:09 -07002374 flock.l_type = F_UNLCK;
2375 goto again;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376 }
2377
2378out:
2379 locks_free_lock(file_lock);
2380 return error;
2381}
2382#endif /* BITS_PER_LONG == 32 */
2383
2384/*
2385 * This function is called when the file is being removed
2386 * from the task's fd array. POSIX locks belonging to this task
2387 * are deleted at this time.
2388 */
2389void locks_remove_posix(struct file *filp, fl_owner_t owner)
2390{
Miklos Szerediff7b86b2006-06-23 02:05:11 -07002391 struct file_lock lock;
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002392 struct file_lock_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
2394 /*
2395 * If there are no locks held on this file, we don't need to call
2396 * posix_lock_file(). Another process could be setting a lock on this
2397 * file at the same time, but we wouldn't remove that lock anyway.
2398 */
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002399 ctx = smp_load_acquire(&file_inode(filp)->i_flctx);
Jeff Laytonbd61e0a2015-01-16 15:05:55 -05002400 if (!ctx || list_empty(&ctx->flc_posix))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 return;
2402
2403 lock.fl_type = F_UNLCK;
Miklos Szeredi75e1fcc2006-06-23 02:05:12 -07002404 lock.fl_flags = FL_POSIX | FL_CLOSE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 lock.fl_start = 0;
2406 lock.fl_end = OFFSET_MAX;
2407 lock.fl_owner = owner;
2408 lock.fl_pid = current->tgid;
2409 lock.fl_file = filp;
2410 lock.fl_ops = NULL;
2411 lock.fl_lmops = NULL;
2412
Marc Eshel150b3932007-01-18 16:15:35 -05002413 vfs_lock_file(filp, F_SETLK, &lock, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415 if (lock.fl_ops && lock.fl_ops->fl_release_private)
2416 lock.fl_ops->fl_release_private(&lock);
2417}
2418
2419EXPORT_SYMBOL(locks_remove_posix);
2420
Jeff Layton3d8e5602015-01-16 15:05:58 -05002421/* The i_flctx must be valid when calling into here */
Jeff Laytondd459bb2015-01-16 15:05:54 -05002422static void
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002423locks_remove_flock(struct file *filp, struct file_lock_context *flctx)
Jeff Laytondd459bb2015-01-16 15:05:54 -05002424{
2425 struct file_lock fl = {
2426 .fl_owner = filp,
2427 .fl_pid = current->tgid,
2428 .fl_file = filp,
2429 .fl_flags = FL_FLOCK,
2430 .fl_type = F_UNLCK,
2431 .fl_end = OFFSET_MAX,
2432 };
Jeff Laytonbcd7f782015-07-11 06:43:02 -04002433 struct inode *inode = file_inode(filp);
Jeff Laytondd459bb2015-01-16 15:05:54 -05002434
Jeff Layton3d8e5602015-01-16 15:05:58 -05002435 if (list_empty(&flctx->flc_flock))
Jeff Laytondd459bb2015-01-16 15:05:54 -05002436 return;
2437
2438 if (filp->f_op->flock)
2439 filp->f_op->flock(filp, F_SETLKW, &fl);
2440 else
Jeff Laytonbcd7f782015-07-11 06:43:02 -04002441 flock_lock_inode(inode, &fl);
Jeff Laytondd459bb2015-01-16 15:05:54 -05002442
2443 if (fl.fl_ops && fl.fl_ops->fl_release_private)
2444 fl.fl_ops->fl_release_private(&fl);
2445}
2446
Jeff Layton3d8e5602015-01-16 15:05:58 -05002447/* The i_flctx must be valid when calling into here */
Jeff Layton8634b512015-01-16 15:05:55 -05002448static void
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002449locks_remove_lease(struct file *filp, struct file_lock_context *ctx)
Jeff Layton8634b512015-01-16 15:05:55 -05002450{
Jeff Layton8634b512015-01-16 15:05:55 -05002451 struct file_lock *fl, *tmp;
2452 LIST_HEAD(dispose);
2453
Jeff Layton3d8e5602015-01-16 15:05:58 -05002454 if (list_empty(&ctx->flc_lease))
Jeff Layton8634b512015-01-16 15:05:55 -05002455 return;
2456
Jeff Layton6109c852015-01-16 15:05:57 -05002457 spin_lock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002458 list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
Jeff Laytonc4e136c2015-02-16 19:37:42 -05002459 if (filp == fl->fl_file)
2460 lease_modify(fl, F_UNLCK, &dispose);
Jeff Layton6109c852015-01-16 15:05:57 -05002461 spin_unlock(&ctx->flc_lock);
Jeff Layton8634b512015-01-16 15:05:55 -05002462 locks_dispose_list(&dispose);
2463}
2464
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465/*
2466 * This function is called on the last close of an open file.
2467 */
Jeff Layton78ed8a12014-02-03 12:13:08 -05002468void locks_remove_file(struct file *filp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469{
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002470 struct file_lock_context *ctx;
2471
2472 ctx = smp_load_acquire(&file_inode(filp)->i_flctx);
2473 if (!ctx)
Jeff Layton3d8e5602015-01-16 15:05:58 -05002474 return;
2475
Jeff Laytondd459bb2015-01-16 15:05:54 -05002476 /* remove any OFD locks */
Christoph Hellwig73a8f5f2014-07-13 17:00:38 +02002477 locks_remove_posix(filp, filp);
Jeff Layton5d50ffd2014-02-03 12:13:10 -05002478
Jeff Laytondd459bb2015-01-16 15:05:54 -05002479 /* remove flock locks */
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002480 locks_remove_flock(filp, ctx);
Jeff Laytondd459bb2015-01-16 15:05:54 -05002481
Jeff Layton8634b512015-01-16 15:05:55 -05002482 /* remove any leases */
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002483 locks_remove_lease(filp, ctx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484}
2485
2486/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 * posix_unblock_lock - stop waiting for a file lock
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 * @waiter: the lock which was waiting
2489 *
2490 * lockd needs to block waiting for locks.
2491 */
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002492int
Jeff Laytonf891a292013-06-21 08:58:09 -04002493posix_unblock_lock(struct file_lock *waiter)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494{
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002495 int status = 0;
2496
Jeff Layton7b2296a2013-06-21 08:58:20 -04002497 spin_lock(&blocked_lock_lock);
J. Bruce Fields5996a292006-01-03 09:55:44 +01002498 if (waiter->fl_next)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002499 __locks_delete_block(waiter);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002500 else
2501 status = -ENOENT;
Jeff Layton7b2296a2013-06-21 08:58:20 -04002502 spin_unlock(&blocked_lock_lock);
J. Bruce Fields64a318e2006-01-03 09:55:46 +01002503 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505EXPORT_SYMBOL(posix_unblock_lock);
2506
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002507/**
2508 * vfs_cancel_lock - file byte range unblock lock
2509 * @filp: The file to apply the unblock to
2510 * @fl: The lock to be unblocked
2511 *
2512 * Used by lock managers to cancel blocked requests
2513 */
2514int vfs_cancel_lock(struct file *filp, struct file_lock *fl)
2515{
Al Viro72c2d532013-09-22 16:27:52 -04002516 if (filp->f_op->lock)
Marc Eshel9b9d2ab2007-01-18 17:52:58 -05002517 return filp->f_op->lock(filp, F_CANCELLK, fl);
2518 return 0;
2519}
2520
2521EXPORT_SYMBOL_GPL(vfs_cancel_lock);
2522
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002523#ifdef CONFIG_PROC_FS
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002524#include <linux/proc_fs.h>
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002525#include <linux/seq_file.h>
2526
Jeff Layton7012b022013-06-21 08:58:22 -04002527struct locks_iterator {
2528 int li_cpu;
2529 loff_t li_pos;
2530};
2531
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002532static void lock_get_status(struct seq_file *f, struct file_lock *fl,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002533 loff_t id, char *pfx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534{
2535 struct inode *inode = NULL;
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002536 unsigned int fl_pid;
2537
2538 if (fl->fl_nspid)
Pavel Emelyanov6c5f3e72008-02-08 04:19:20 -08002539 fl_pid = pid_vnr(fl->fl_nspid);
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002540 else
2541 fl_pid = fl->fl_pid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542
2543 if (fl->fl_file != NULL)
Al Viro496ad9a2013-01-23 17:07:38 -05002544 inode = file_inode(fl->fl_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545
Jerome Marchand99dc8292010-10-26 14:22:33 -07002546 seq_printf(f, "%lld:%s ", id, pfx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 if (IS_POSIX(fl)) {
Jeff Laytonc918d422014-02-03 12:13:09 -05002548 if (fl->fl_flags & FL_ACCESS)
Fabian Frederick5315c262014-05-09 14:13:05 -04002549 seq_puts(f, "ACCESS");
Jeff Laytoncff2fce2014-04-22 08:24:32 -04002550 else if (IS_OFDLCK(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002551 seq_puts(f, "OFDLCK");
Jeff Laytonc918d422014-02-03 12:13:09 -05002552 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002553 seq_puts(f, "POSIX ");
Jeff Laytonc918d422014-02-03 12:13:09 -05002554
2555 seq_printf(f, " %s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 (inode == NULL) ? "*NOINODE*" :
Pavel Emelyanova16877c2007-10-01 14:41:11 -07002557 mandatory_lock(inode) ? "MANDATORY" : "ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 } else if (IS_FLOCK(fl)) {
2559 if (fl->fl_type & LOCK_MAND) {
Fabian Frederick5315c262014-05-09 14:13:05 -04002560 seq_puts(f, "FLOCK MSNFS ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002562 seq_puts(f, "FLOCK ADVISORY ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 }
2564 } else if (IS_LEASE(fl)) {
Jeff Layton8144f1f2014-08-11 13:36:54 -04002565 if (fl->fl_flags & FL_DELEG)
2566 seq_puts(f, "DELEG ");
2567 else
2568 seq_puts(f, "LEASE ");
2569
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002570 if (lease_breaking(fl))
Fabian Frederick5315c262014-05-09 14:13:05 -04002571 seq_puts(f, "BREAKING ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 else if (fl->fl_file)
Fabian Frederick5315c262014-05-09 14:13:05 -04002573 seq_puts(f, "ACTIVE ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 else
Fabian Frederick5315c262014-05-09 14:13:05 -04002575 seq_puts(f, "BREAKER ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002577 seq_puts(f, "UNKNOWN UNKNOWN ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 }
2579 if (fl->fl_type & LOCK_MAND) {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002580 seq_printf(f, "%s ",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 (fl->fl_type & LOCK_READ)
2582 ? (fl->fl_type & LOCK_WRITE) ? "RW " : "READ "
2583 : (fl->fl_type & LOCK_WRITE) ? "WRITE" : "NONE ");
2584 } else {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002585 seq_printf(f, "%s ",
J. Bruce Fieldsab83fa42011-07-26 20:10:51 -04002586 (lease_breaking(fl))
Jeff Layton0ee5c6d2012-08-02 15:46:30 -04002587 ? (fl->fl_type == F_UNLCK) ? "UNLCK" : "READ "
2588 : (fl->fl_type == F_WRLCK) ? "WRITE" : "READ ");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 }
2590 if (inode) {
Jeff Layton36488882015-04-03 09:04:04 -04002591 /* userspace relies on this representation of dev_t */
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002592 seq_printf(f, "%d %02x:%02x:%ld ", fl_pid,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593 MAJOR(inode->i_sb->s_dev),
2594 MINOR(inode->i_sb->s_dev), inode->i_ino);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 } else {
Vitaliy Gusevab1f1612008-01-17 00:07:08 +00002596 seq_printf(f, "%d <none>:0 ", fl_pid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 }
2598 if (IS_POSIX(fl)) {
2599 if (fl->fl_end == OFFSET_MAX)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002600 seq_printf(f, "%Ld EOF\n", fl->fl_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601 else
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002602 seq_printf(f, "%Ld %Ld\n", fl->fl_start, fl->fl_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 } else {
Fabian Frederick5315c262014-05-09 14:13:05 -04002604 seq_puts(f, "0 EOF\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 }
2606}
2607
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002608static int locks_show(struct seq_file *f, void *v)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609{
Jeff Layton7012b022013-06-21 08:58:22 -04002610 struct locks_iterator *iter = f->private;
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002611 struct file_lock *fl, *bfl;
2612
Jeff Layton139ca042013-06-21 08:58:17 -04002613 fl = hlist_entry(v, struct file_lock, fl_link);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002614
Jeff Layton7012b022013-06-21 08:58:22 -04002615 lock_get_status(f, fl, iter->li_pos, "");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002616
2617 list_for_each_entry(bfl, &fl->fl_block, fl_block)
Jeff Layton7012b022013-06-21 08:58:22 -04002618 lock_get_status(f, bfl, iter->li_pos, " ->");
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002619
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002620 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621}
2622
Andrey Vagin6c8c9032015-04-16 12:49:38 -07002623static void __show_fd_locks(struct seq_file *f,
2624 struct list_head *head, int *id,
2625 struct file *filp, struct files_struct *files)
2626{
2627 struct file_lock *fl;
2628
2629 list_for_each_entry(fl, head, fl_list) {
2630
2631 if (filp != fl->fl_file)
2632 continue;
2633 if (fl->fl_owner != files &&
2634 fl->fl_owner != filp)
2635 continue;
2636
2637 (*id)++;
2638 seq_puts(f, "lock:\t");
2639 lock_get_status(f, fl, *id, "");
2640 }
2641}
2642
2643void show_fd_locks(struct seq_file *f,
2644 struct file *filp, struct files_struct *files)
2645{
2646 struct inode *inode = file_inode(filp);
2647 struct file_lock_context *ctx;
2648 int id = 0;
2649
Dmitry Vyukov128a3782015-09-21 09:43:06 +02002650 ctx = smp_load_acquire(&inode->i_flctx);
Andrey Vagin6c8c9032015-04-16 12:49:38 -07002651 if (!ctx)
2652 return;
2653
2654 spin_lock(&ctx->flc_lock);
2655 __show_fd_locks(f, &ctx->flc_flock, &id, filp, files);
2656 __show_fd_locks(f, &ctx->flc_posix, &id, filp, files);
2657 __show_fd_locks(f, &ctx->flc_lease, &id, filp, files);
2658 spin_unlock(&ctx->flc_lock);
2659}
2660
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002661static void *locks_start(struct seq_file *f, loff_t *pos)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002662 __acquires(&blocked_lock_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663{
Jeff Layton7012b022013-06-21 08:58:22 -04002664 struct locks_iterator *iter = f->private;
Jerome Marchand99dc8292010-10-26 14:22:33 -07002665
Jeff Layton7012b022013-06-21 08:58:22 -04002666 iter->li_pos = *pos + 1;
2667 lg_global_lock(&file_lock_lglock);
Jeff Layton7b2296a2013-06-21 08:58:20 -04002668 spin_lock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002669 return seq_hlist_start_percpu(&file_lock_list, &iter->li_cpu, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670}
2671
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002672static void *locks_next(struct seq_file *f, void *v, loff_t *pos)
2673{
Jeff Layton7012b022013-06-21 08:58:22 -04002674 struct locks_iterator *iter = f->private;
2675
2676 ++iter->li_pos;
2677 return seq_hlist_next_percpu(v, &file_lock_list, &iter->li_cpu, pos);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002678}
2679
2680static void locks_stop(struct seq_file *f, void *v)
Jeff Laytonb03dfdec2014-02-03 12:13:07 -05002681 __releases(&blocked_lock_lock)
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002682{
Jeff Layton7b2296a2013-06-21 08:58:20 -04002683 spin_unlock(&blocked_lock_lock);
Jeff Layton7012b022013-06-21 08:58:22 -04002684 lg_global_unlock(&file_lock_lglock);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002685}
2686
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002687static const struct seq_operations locks_seq_operations = {
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002688 .start = locks_start,
2689 .next = locks_next,
2690 .stop = locks_stop,
2691 .show = locks_show,
2692};
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002693
2694static int locks_open(struct inode *inode, struct file *filp)
2695{
Jeff Layton7012b022013-06-21 08:58:22 -04002696 return seq_open_private(filp, &locks_seq_operations,
2697 sizeof(struct locks_iterator));
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002698}
2699
2700static const struct file_operations proc_locks_operations = {
2701 .open = locks_open,
2702 .read = seq_read,
2703 .llseek = seq_lseek,
Jerome Marchand99dc8292010-10-26 14:22:33 -07002704 .release = seq_release_private,
Alexey Dobriyand8ba7a32008-10-04 22:34:18 +04002705};
2706
2707static int __init proc_locks_init(void)
2708{
2709 proc_create("locks", 0, NULL, &proc_locks_operations);
2710 return 0;
2711}
2712module_init(proc_locks_init);
Pavel Emelyanov7f8ada92007-10-01 14:41:15 -07002713#endif
2714
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715static int __init filelock_init(void)
2716{
Jeff Layton7012b022013-06-21 08:58:22 -04002717 int i;
2718
Jeff Layton4a075e32015-01-16 15:05:54 -05002719 flctx_cache = kmem_cache_create("file_lock_ctx",
2720 sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
2721
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 filelock_cache = kmem_cache_create("file_lock_cache",
Miklos Szerediee19cc42011-07-07 13:06:09 +02002723 sizeof(struct file_lock), 0, SLAB_PANIC, NULL);
2724
Jeff Layton7012b022013-06-21 08:58:22 -04002725 lg_lock_init(&file_lock_lglock, "file_lock_lglock");
2726
2727 for_each_possible_cpu(i)
2728 INIT_HLIST_HEAD(per_cpu_ptr(&file_lock_list, i));
2729
Linus Torvalds1da177e2005-04-16 15:20:36 -07002730 return 0;
2731}
2732
2733core_initcall(filelock_init);